Module:Citation/CS1: Difference between revisions

From Zoophilia Wiki
Jump to navigationJump to search
meta>Trappist the monk
Undid revision 913826030 by Ivanvector (talk) broke the cs1|2 templates;
m 1 revision imported
 
(11 intermediate revisions by 8 users not shown)
Line 1: Line 1:
require ('strict');


--[[--------------------------< F O R W A R D  D E C L A R A T I O N S >--------------------------------------
--[[--------------------------< F O R W A R D  D E C L A R A T I O N S >--------------------------------------
]]


local dates, year_date_check, reformat_dates, date_hyphen_to_dash, -- functions in Module:Citation/CS1/Date_validation
each of these counts against the Lua upvalue limit
date_name_xlate


local is_set, in_array, substitute, error_comment, set_error, select_one, -- functions in Module:Citation/CS1/Utilities
]]
add_maint_cat, wrap_style, safe_for_italics, is_wikilink, make_wikilink,
strip_apostrophe_markup;


local z ={}; -- tables in Module:Citation/CS1/Utilities
local validation; -- functions in Module:Citation/CS1/Date_validation


local extract_ids, extract_id_access_levels, build_id_list, is_embargoed; -- functions in Module:Citation/CS1/Identifiers
local utilities; -- functions in Module:Citation/CS1/Utilities
 
local z = {}; -- table of tables in Module:Citation/CS1/Utilities
local make_coins_title, get_coins_pages, COinS; -- functions in Module:Citation/CS1/COinS


local identifiers; -- functions and tables in Module:Citation/CS1/Identifiers
local metadata; -- functions in Module:Citation/CS1/COinS
local cfg = {}; -- table of configuration tables that are defined in Module:Citation/CS1/Configuration
local cfg = {}; -- table of configuration tables that are defined in Module:Citation/CS1/Configuration
local whitelist = {}; -- table of tables listing valid template parameter names; defined in Module:Citation/CS1/Whitelist
local whitelist = {}; -- table of tables listing valid template parameter names; defined in Module:Citation/CS1/Whitelist




--[[--------------------------< P A G E  S C O P E  V A R I A B L E S >--------------------------------------
--[[------------------< P A G E  S C O P E  V A R I A B L E S >---------------


delare variables here that have page-wide scope that are not brought in from other modules; thatare created here
declare variables here that have page-wide scope that are not brought in from
and used here
other modules; that are created here and used here


]]
]]


local added_deprecated_cat; -- boolean flag so that the category is added only once
local added_deprecated_cat; -- Boolean flag so that the category is added only once
local added_prop_cats = {}; -- list of property categories that have been added to z.properties_cats
local added_vanc_errs; -- Boolean flag so we only emit one Vancouver error / category
local added_vanc_errs; -- boolean flag so we only emit one Vancouver error / category
local added_generic_name_errs; -- Boolean flag so we only emit one generic name error / category and stop testing names once an error is encountered
 
local added_numeric_name_errs; -- Boolean flag so we only emit one numeric name error / category and stop testing names once an error is encountered
local added_numeric_name_maint; -- Boolean flag so we only emit one numeric name maint category and stop testing names once a category has been emitted
local Frame; -- holds the module's frame table
local Frame; -- holds the module's frame table
local is_preview_mode; -- true when article is in preview mode; false when using 'Preview page with this template' (previewing the module)
local is_sandbox; -- true when using sandbox modules to render citation




Line 48: Line 49:
local i = 1;
local i = 1;
while i <= count do -- loop through all items in list
while i <= count do -- loop through all items in list
if is_set( list[i] ) then
if utilities.is_set( list[i] ) then
return list[i]; -- return the first set list member
return list[i]; -- return the first set list member
end
end
i = i + 1; -- point to next
i = i + 1; -- point to next
end
end
--[[--------------------------< A D D _ P R O P _ C A T >--------------------------------------------------------
Adds a category to z.properties_cats using names from the configuration file with additional text if any.
foreign_lang_source and foreign_lang_source_2 keys have a language code appended to them so that multiple languages
may be categorized but multiples of the same language are not categorized.
added_prop_cats is a table declared in page scope variables above
]]
local function add_prop_cat (key, arguments)
if not added_prop_cats [key] then
added_prop_cats [key] = true; -- note that we've added this category
key = key:gsub ('(foreign_lang_source_?2?)%a%a%a?', '%1'); -- strip lang code from keyname
table.insert( z.properties_cats, substitute (cfg.prop_cats [key], arguments)); -- make name then add to table
end
end
end
end
Line 81: Line 62:
To prevent duplication, added_vanc_errs is nil until an error message is emitted.
To prevent duplication, added_vanc_errs is nil until an error message is emitted.


added_vanc_errs is a boolean declared in page scope variables above
added_vanc_errs is a Boolean declared in page scope variables above


]]
]]


local function add_vanc_error (source)
local function add_vanc_error (source, position)
if not added_vanc_errs then
if added_vanc_errs then return end
added_vanc_errs = true; -- note that we've added this category
table.insert( z.message_tail, { set_error( 'vancouver', {source}, true ) } );
added_vanc_errs = true; -- note that we've added this category
end
utilities.set_message ('err_vancouver', {source, position});
end
end


Line 95: Line 76:
--[[--------------------------< I S _ S C H E M E >------------------------------------------------------------
--[[--------------------------< I S _ S C H E M E >------------------------------------------------------------


does this thing that purports to be a uri scheme seem to be a valid scheme?  The scheme is checked to see if it
does this thing that purports to be a URI scheme seem to be a valid scheme?  The scheme is checked to see if it
is in agreement with http://tools.ietf.org/html/std66#section-3.1 which says:
is in agreement with http://tools.ietf.org/html/std66#section-3.1 which says:
Scheme names consist of a sequence of characters beginning with a
Scheme names consist of a sequence of characters beginning with a
Line 118: Line 99:
Single character names are generally reserved; see https://tools.ietf.org/html/draft-ietf-dnsind-iana-dns-01#page-15;
Single character names are generally reserved; see https://tools.ietf.org/html/draft-ietf-dnsind-iana-dns-01#page-15;
see also [[Single-letter second-level domain]]
see also [[Single-letter second-level domain]]
list of tlds: https://www.iana.org/domains/root/db
list of TLDs: https://www.iana.org/domains/root/db


rfc952 (modified by rfc 1123) requires the first and last character of a hostname to be a letter or a digit.  Between
RFC 952 (modified by RFC 1123) requires the first and last character of a hostname to be a letter or a digit.  Between
the first and last characters the name may use letters, digits, and the hyphen.
the first and last characters the name may use letters, digits, and the hyphen.


Also allowed are IPv4 addresses. IPv6 not supported
Also allowed are IPv4 addresses. IPv6 not supported


domain is expected to be stripped of any path so that the last character in the last character of the tld.  tld
domain is expected to be stripped of any path so that the last character in the last character of the TLD.  tld
is two or more alpha characters.  Any preceding '//' (from splitting a url with a scheme) will be stripped
is two or more alpha characters.  Any preceding '//' (from splitting a URL with a scheme) will be stripped
here.  Perhaps not necessary but retained incase it is necessary for IPv4 dot decimal.
here.  Perhaps not necessary but retained in case it is necessary for IPv4 dot decimal.


There are several tests:
There are several tests:
the first character of the whole domain name including subdomains must be a letter or a digit
the first character of the whole domain name including subdomains must be a letter or a digit
internationalized domain name (ascii characters with .xn-- ASCII Compatible Encoding (ACE) prefix xn-- in the tld) see https://tools.ietf.org/html/rfc3490
internationalized domain name (ASCII characters with .xn-- ASCII Compatible Encoding (ACE) prefix xn-- in the TLD) see https://tools.ietf.org/html/rfc3490
single-letter/digit second-level domains in the .org, .cash, and .today TLDs
single-letter/digit second-level domains in the .org, .cash, and .today TLDs
q, x, and z SL domains in the .com TLD
q, x, and z SL domains in the .com TLD
Line 140: Line 121:
IPv4 dot-decimal address format; TLD not allowed
IPv4 dot-decimal address format; TLD not allowed


returns true if domain appears to be a proper name and tld or IPv4 address, else false
returns true if domain appears to be a proper name and TLD or IPv4 address, else false


]=]
]=]
Line 155: Line 136:
end
end


if domain:match ('^%a+:') then -- hack to detect things that look like s:Page:Title where Page: is namespace at wikisource
if domain:match ('^%a+:') then -- hack to detect things that look like s:Page:Title where Page: is namespace at Wikisource
return false;
return false;
end
end


local patterns = { -- patterns that look like urls
local patterns = { -- patterns that look like URLs
'%f[%w][%w][%w%-]+[%w]%.%a%a+$', -- three or more character hostname.hostname or hostname.tld
'%f[%w][%w][%w%-]+[%w]%.%a%a+$', -- three or more character hostname.hostname or hostname.tld
'%f[%w][%w][%w%-]+[%w]%.xn%-%-[%w]+$', -- internationalized domain name with ACE prefix
'%f[%w][%w][%w%-]+[%w]%.xn%-%-[%w]+$', -- internationalized domain name with ACE prefix
'%f[%a][qxz]%.com$', -- assigned one character .com hostname (x.com times out 2015-12-10)
'%f[%a][qxz]%.com$', -- assigned one character .com hostname (x.com times out 2015-12-10)
'%f[%a][iq]%.net$', -- assigned one character .net hostname (q.net registered but not active 2015-12-10)
'%f[%a][iq]%.net$', -- assigned one character .net hostname (q.net registered but not active 2015-12-10)
'%f[%w][%w]%.%a%a$', -- one character hostname and cctld (2 chars)
'%f[%w][%w]%.%a%a$', -- one character hostname and ccTLD (2 chars)
'%f[%w][%w][%w]%.%a%a+$', -- two character hostname and tld
'%f[%w][%w][%w]%.%a%a+$', -- two character hostname and TLD
'^%d%d?%d?%.%d%d?%d?%.%d%d?%d?%.%d%d?%d?', -- IPv4 address
'^%d%d?%d?%.%d%d?%d?%.%d%d?%d?%.%d%d?%d?', -- IPv4 address
}
}
Line 171: Line 152:
for _, pattern in ipairs (patterns) do -- loop through the patterns list
for _, pattern in ipairs (patterns) do -- loop through the patterns list
if domain:match (pattern) then
if domain:match (pattern) then
return true; -- if a match then we think that this thing that purports to be a url is a url
return true; -- if a match then we think that this thing that purports to be a URL is a URL
end
end
end
end


for _, d in ipairs ({'cash', 'company', 'today', 'org'}) do -- look for single letter second level domain names for these top level domains
for _, d in ipairs (cfg.single_letter_2nd_lvl_domains_t) do -- look for single letter second level domain names for these top level domains
if domain:match ('%f[%w][%w]%.' .. d) then
if domain:match ('%f[%w][%w]%.' .. d) then
return true
return true
Line 186: Line 167:
--[[--------------------------< I S _ U R L >------------------------------------------------------------------
--[[--------------------------< I S _ U R L >------------------------------------------------------------------


returns true if the scheme and domain parts of a url appear to be a valid url; else false.
returns true if the scheme and domain parts of a URL appear to be a valid URL; else false.


This function is the last step in the validation process.  This function is separate because there are cases that
This function is the last step in the validation process.  This function is separate because there are cases that
Line 195: Line 176:


local function is_url (scheme, domain)
local function is_url (scheme, domain)
if is_set (scheme) then -- if scheme is set check it and domain
if utilities.is_set (scheme) then -- if scheme is set check it and domain
return is_scheme (scheme) and is_domain_name (domain);
return is_scheme (scheme) and is_domain_name (domain);
else
else
return is_domain_name (domain); -- scheme not set when url is protocol relative
return is_domain_name (domain); -- scheme not set when URL is protocol-relative
end
end
end
end
Line 205: Line 186:
--[[--------------------------< S P L I T _ U R L >------------------------------------------------------------
--[[--------------------------< S P L I T _ U R L >------------------------------------------------------------


Split a url into a scheme, authority indicator, and domain.
Split a URL into a scheme, authority indicator, and domain.


First remove Fully Qualified Domain Name terminator (a dot following tld) (if any) and any path(/), query(?) or fragment(#).
First remove Fully Qualified Domain Name terminator (a dot following TLD) (if any) and any path(/), query(?) or fragment(#).


If protocol relative url, return nil scheme and domain else return nil for both scheme and domain.
If protocol-relative URL, return nil scheme and domain else return nil for both scheme and domain.


When not protocol relative, get scheme, authority indicator, and domain.  If there is an authority indicator (one
When not protocol-relative, get scheme, authority indicator, and domain.  If there is an authority indicator (one
or more '/' characters immediately following the scheme's colon), make sure that there are only 2.
or more '/' characters immediately following the scheme's colon), make sure that there are only 2.
Any URL that does not have news: scheme must have authority indicator (//).  TODO: are there other common schemes
like news: that don't use authority indicator?


Strip off any port and path;
Strip off any port and path;
Line 223: Line 207:
url_str = url_str:gsub ('([%a%d])%.?[/%?#].*$', '%1'); -- strip FQDN terminator and path(/), query(?), fragment (#) (the capture prevents false replacement of '//')
url_str = url_str:gsub ('([%a%d])%.?[/%?#].*$', '%1'); -- strip FQDN terminator and path(/), query(?), fragment (#) (the capture prevents false replacement of '//')


if url_str:match ('^//%S*') then -- if there is what appears to be a protocol relative url
if url_str:match ('^//%S*') then -- if there is what appears to be a protocol-relative URL
domain = url_str:match ('^//(%S*)')
domain = url_str:match ('^//(%S*)')
elseif url_str:match ('%S-:/*%S+') then -- if there is what appears to be a scheme, optional authority indicator, and domain name
elseif url_str:match ('%S-:/*%S+') then -- if there is what appears to be a scheme, optional authority indicator, and domain name
scheme, authority, domain = url_str:match ('(%S-:)(/*)(%S+)'); -- extract the scheme, authority indicator, and domain portions
scheme, authority, domain = url_str:match ('(%S-:)(/*)(%S+)'); -- extract the scheme, authority indicator, and domain portions
authority = authority:gsub ('//', '', 1); -- replace place 1 pair of '/' with nothing;
if utilities.is_set (authority) then
if is_set(authority) then -- if anything left (1 or 3+ '/' where authority should be) then
authority = authority:gsub ('//', '', 1); -- replace place 1 pair of '/' with nothing;
return scheme; -- return scheme only making domain nil which will cause an error message
if utilities.is_set(authority) then -- if anything left (1 or 3+ '/' where authority should be) then
return scheme; -- return scheme only making domain nil which will cause an error message
end
else
if not scheme:match ('^news:') then -- except for news:..., MediaWiki won't link URLs that do not have authority indicator; TODO: a better way to do this test?
return scheme; -- return scheme only making domain nil which will cause an error message
end
end
end
domain = domain:gsub ('(%a):%d+', '%1'); -- strip port number if present
domain = domain:gsub ('(%a):%d+', '%1'); -- strip port number if present
Line 240: Line 230:
--[[--------------------------< L I N K _ P A R A M _ O K >---------------------------------------------------
--[[--------------------------< L I N K _ P A R A M _ O K >---------------------------------------------------


checks the content of |title-link=, |series-link=, |author-link= etc for properly formatted content: no wikilinks, no urls
checks the content of |title-link=, |series-link=, |author-link=, etc. for properly formatted content: no wikilinks, no URLs


Link parameters are to hold the title of a wikipedia article so none of the WP:TITLESPECIALCHARACTERS are allowed:
Link parameters are to hold the title of a Wikipedia article, so none of the WP:TITLESPECIALCHARACTERS are allowed:
# < > [ ] | { } _
# < > [ ] | { } _
except the underscore which is used as a space in wiki urls and # which is used for section links
except the underscore which is used as a space in wiki URLs and # which is used for section links


returns false when the value contains any of these characters.
returns false when the value contains any of these characters.


When there are no illegal characters, this function returns TRUE if value DOES NOT appear to be a valid url (the
When there are no illegal characters, this function returns TRUE if value DOES NOT appear to be a valid URL (the
|<param>-link= parameter is ok); else false when value appears to be a valid url (the |<param>-link= parameter is NOT ok).
|<param>-link= parameter is ok); else false when value appears to be a valid URL (the |<param>-link= parameter is NOT ok).


]]
]]
Line 259: Line 249:
end
end


scheme, domain = split_url (value); -- get scheme or nil and domain or nil from url;  
scheme, domain = split_url (value); -- get scheme or nil and domain or nil from URL;  
return not is_url (scheme, domain); -- return true if value DOES NOT appear to be a valid url
return not is_url (scheme, domain); -- return true if value DOES NOT appear to be a valid URL
end
end


Line 268: Line 258:
Use link_param_ok() to validate |<param>-link= value and its matching |<title>= value.
Use link_param_ok() to validate |<param>-link= value and its matching |<title>= value.


|<title>= may be wikilinked but not when |<param>-link= has a value.  This function emits an error message when
|<title>= may be wiki-linked but not when |<param>-link= has a value.  This function emits an error message when
that condition exists
that condition exists
check <link> for inter-language interwiki-link prefix.  prefix must be a MediaWiki-recognized language
code and must begin with a colon.


]]
]]


local function link_title_ok (link, lorig, title, torig)
local function link_title_ok (link, lorig, title, torig)
local orig;
local orig;
 
if utilities.is_set (link) then -- don't bother if <param>-link doesn't have a value
if is_set (link) then -- don't bother if <param>-link doesn't have a value
if not link_param_ok (link) then -- check |<param>-link= markup
if not link_param_ok (link) then -- check |<param>-link= markup
orig = lorig; -- identify the failing link parameter
orig = lorig; -- identify the failing link parameter
elseif title:find ('%[%[') then -- check |title= for wikilink markup
elseif title:find ('%[%[') then -- check |title= for wikilink markup
orig = torig; -- identify the failing |title= parameter
orig = torig; -- identify the failing |title= parameter
elseif link:match ('^%a+:') then -- if the link is what looks like an interwiki
local prefix = link:match ('^(%a+):'):lower(); -- get the interwiki prefix
if cfg.inter_wiki_map[prefix] then -- if prefix is in the map, must have preceding colon
orig = lorig; -- flag as error
end
end
end
end
end


if is_set (orig) then
if utilities.is_set (orig) then
table.insert( z.message_tail, { set_error( 'bad_paramlink', orig)}); -- url or wikilink in |title= with |title-link=;
link = ''; -- unset
utilities.set_message ('err_bad_paramlink', orig); -- URL or wikilink in |title= with |title-link=;
end
end
return link; -- link if ok, empty string else
end
end


Line 294: Line 295:
Determines whether a URL string appears to be valid.
Determines whether a URL string appears to be valid.


First we test for space characters.  If any are found, return false.  Then split the url into scheme and domain
First we test for space characters.  If any are found, return false.  Then split the URL into scheme and domain
portions, or for protocol relative (//example.com) urls, just the domain.  Use is_url() to validate the two
portions, or for protocol-relative (//example.com) URLs, just the domain.  Use is_url() to validate the two
portions of the url.  If both are valid, or for protocol relative if domain is valid, return true, else false.
portions of the URL.  If both are valid, or for protocol-relative if domain is valid, return true, else false.


Because it is different from a standard url, and because this module used external_link() to make external links
Because it is different from a standard URL, and because this module used external_link() to make external links
that work for standard and news: links, we validate newsgroup names here.  The specification for a newsgroup name
that work for standard and news: links, we validate newsgroup names here.  The specification for a newsgroup name
is at https://tools.ietf.org/html/rfc5536#section-3.1.4
is at https://tools.ietf.org/html/rfc5536#section-3.1.4
Line 305: Line 306:


local function check_url( url_str )
local function check_url( url_str )
if nil == url_str:match ("^%S+$") then -- if there are any spaces in |url=value it can't be a proper url
if nil == url_str:match ("^%S+$") then -- if there are any spaces in |url=value it can't be a proper URL
return false;
return false;
end
end
local scheme, domain;
local scheme, domain;


scheme, domain = split_url (url_str); -- get scheme or nil and domain or nil from url;
scheme, domain = split_url (url_str); -- get scheme or nil and domain or nil from URL;
if 'news:' == scheme then -- special case for newsgroups
if 'news:' == scheme then -- special case for newsgroups
Line 316: Line 317:
end
end
return is_url (scheme, domain); -- return true if value appears to be a valid url
return is_url (scheme, domain); -- return true if value appears to be a valid URL
end
end


Line 323: Line 324:


Return true if a parameter value has a string that begins and ends with square brackets [ and ] and the first
Return true if a parameter value has a string that begins and ends with square brackets [ and ] and the first
non-space characters following the opening bracket appear to be a url.  The test will also find external wikilinks
non-space characters following the opening bracket appear to be a URL.  The test will also find external wikilinks
that use protocol relative urls. Also finds bare urls.
that use protocol-relative URLs. Also finds bare URLs.


The frontier pattern prevents a match on interwiki links which are similar to scheme:path urls.  The tests that
The frontier pattern prevents a match on interwiki-links which are similar to scheme:path URLs.  The tests that
find bracketed urls are required because the parameters that call this test (currently |title=, |chapter=, |work=,
find bracketed URLs are required because the parameters that call this test (currently |title=, |chapter=, |work=,
and |publisher=) may have wikilinks and there are articles or redirects like '//Hus' so, while uncommon, |title=[[//Hus]]
and |publisher=) may have wikilinks and there are articles or redirects like '//Hus' so, while uncommon, |title=[[//Hus]]
is possible as might be [[en://Hus]].
is possible as might be [[en://Hus]].
Line 336: Line 337:
local scheme, domain;
local scheme, domain;


if value:match ('%f[%[]%[%a%S*:%S+.*%]') then -- if ext wikilink with scheme and domain: [xxxx://yyyyy.zzz]
if value:match ('%f[%[]%[%a%S*:%S+.*%]') then -- if ext. wikilink with scheme and domain: [xxxx://yyyyy.zzz]
scheme, domain = split_url (value:match ('%f[%[]%[(%a%S*:%S+).*%]'));
scheme, domain = split_url (value:match ('%f[%[]%[(%a%S*:%S+).*%]'));
elseif value:match ('%f[%[]%[//%S+.*%]') then -- if protocol relative ext wikilink: [//yyyyy.zzz]
elseif value:match ('%f[%[]%[//%S+.*%]') then -- if protocol-relative ext. wikilink: [//yyyyy.zzz]
scheme, domain = split_url (value:match ('%f[%[]%[(//%S+).*%]'));
scheme, domain = split_url (value:match ('%f[%[]%[(//%S+).*%]'));
elseif value:match ('%a%S*:%S+') then -- if bare url with scheme; may have leading or trailing plain text
elseif value:match ('%a%S*:%S+') then -- if bare URL with scheme; may have leading or trailing plain text
scheme, domain = split_url (value:match ('(%a%S*:%S+)'));
scheme, domain = split_url (value:match ('(%a%S*:%S+)'));
elseif value:match ('//%S+') then -- if protocol relative bare url: //yyyyy.zzz; may have leading or trailing plain text
elseif value:match ('//%S+') then -- if protocol-relative bare URL: //yyyyy.zzz; may have leading or trailing plain text
scheme, domain = split_url (value:match ('(//%S+)')); -- what is left should be the domain
scheme, domain = split_url (value:match ('(//%S+)')); -- what is left should be the domain
else
else
return false; -- didn't find anything that is obviously a url
return false; -- didn't find anything that is obviously a URL
end
end


return is_url (scheme, domain); -- return true if value appears to be a valid url
return is_url (scheme, domain); -- return true if value appears to be a valid URL
end
end


Line 358: Line 359:
]]
]]


local function check_for_url (parameter_list)
local function check_for_url (parameter_list, error_list)
local error_message = '';
for k, v in pairs (parameter_list) do -- for each parameter in the list
for k, v in pairs (parameter_list) do -- for each parameter in the list
if is_parameter_ext_wikilink (v) then -- look at the value; if there is a url add an error message
if is_parameter_ext_wikilink (v) then -- look at the value; if there is a URL add an error message
if is_set(error_message) then -- once we've added the first portion of the error message ...
table.insert (error_list, utilities.wrap_style ('parameter', k));
error_message=error_message .. ", "; -- ... add a comma space separator
end
error_message=error_message .. "&#124;" .. k .. "="; -- add the failed parameter
end
end
end
if is_set (error_message) then -- done looping, if there is an error message, display it
table.insert( z.message_tail, { set_error( 'param_has_ext_link', {error_message}, true ) } );
end
end
end
end
Line 382: Line 376:
local function safe_for_url( str )
local function safe_for_url( str )
if str:match( "%[%[.-%]%]" ) ~= nil then  
if str:match( "%[%[.-%]%]" ) ~= nil then  
table.insert( z.message_tail, { set_error( 'wikilink_in_url', {}, true ) } );
utilities.set_message ('err_wikilink_in_url', {});
end
end
Line 398: Line 392:
]]
]]


local function external_link( URL, label, source, access)
local function external_link (URL, label, source, access)
local error_str = "";
local err_msg = '';
local domain;
local domain;
local path;
local path;
local base_url;
local base_url;


if not is_set( label ) then
if not utilities.is_set (label) then
label = URL;
label = URL;
if is_set( source ) then
if utilities.is_set (source) then
error_str = set_error( 'bare_url_missing_title', { wrap_style ('parameter', source) }, false, " " );
utilities.set_message ('err_bare_url_missing_title', {utilities.wrap_style ('parameter', source)});
else
else
error( cfg.messages["bare_url_no_origin"] );
error (cfg.messages["bare_url_no_origin"]); -- programmer error; valid parameter name does not have matching meta-parameter
end
end
end
end
if not check_url( URL ) then
if not check_url (URL) then
error_str = set_error( 'bad_url', {wrap_style ('parameter', source)}, false, " " ) .. error_str;
utilities.set_message ('err_bad_url', {utilities.wrap_style ('parameter', source)});
end
end
domain, path = URL:match ('^([/%.%-%+:%a%d]+)([/%?#].*)$'); -- split the url into scheme plus domain and path
domain, path = URL:match ('^([/%.%-%+:%a%d]+)([/%?#].*)$'); -- split the URL into scheme plus domain and path
if path then -- if there is a path portion
if path then -- if there is a path portion
path = path:gsub ('[%[%]]', {['[']='%5b',[']']='%5d'}); -- replace '[' and ']' with their percent encoded values
path = path:gsub ('[%[%]]', {['['] = '%5b', [']'] = '%5d'}); -- replace '[' and ']' with their percent-encoded values
URL = table.concat ({domain, path}); -- and reassemble
URL = table.concat ({domain, path}); -- and reassemble
end
end


base_url = table.concat({ "[", URL, " ", safe_for_url (label), "]" }); -- assemble a wikimarkup url
base_url = table.concat ({ "[", URL, " ", safe_for_url (label), "]" }); -- assemble a wiki-markup URL


if is_set (access) then -- access level (subscription, registration, limited)
if utilities.is_set (access) then -- access level (subscription, registration, limited)
base_url = substitute (cfg.presentation['ext-link-access-signal'], {cfg.presentation[access].class, cfg.presentation[access].title, base_url}); -- add the appropriate icon
base_url = utilities.substitute (cfg.presentation['ext-link-access-signal'], {cfg.presentation[access].class, cfg.presentation[access].title, base_url}); -- add the appropriate icon
end
end
 
return table.concat ({base_url, error_str});
return base_url;
end
end


Line 438: Line 432:
parameters in the citation.
parameters in the citation.


added_deprecated_cat is a boolean declared in page scope variables above
added_deprecated_cat is a Boolean declared in page scope variables above


]]
]]
Line 445: Line 439:
if not added_deprecated_cat then
if not added_deprecated_cat then
added_deprecated_cat = true; -- note that we've added this category
added_deprecated_cat = true; -- note that we've added this category
table.insert( z.message_tail, { set_error( 'deprecated_params', {name}, true ) } ); -- add error message
utilities.set_message ('err_deprecated_params', {name}); -- add error message
end
end
end
end
Line 452: Line 446:
--[=[-------------------------< K E R N _ Q U O T E S >--------------------------------------------------------
--[=[-------------------------< K E R N _ Q U O T E S >--------------------------------------------------------


Apply kerning to open the space between the quote mark provided by the Module and a leading or trailing quote
Apply kerning to open the space between the quote mark provided by the module and a leading or trailing quote
mark contained in a |title= or |chapter= parameter's value.
mark contained in a |title= or |chapter= parameter's value.


Line 458: Line 452:
"'Unkerned title with leading and trailing single quote marks'"
"'Unkerned title with leading and trailing single quote marks'"
" 'Kerned title with leading and trailing single quote marks' " (in real life the kerning isn't as wide as this example)
" 'Kerned title with leading and trailing single quote marks' " (in real life the kerning isn't as wide as this example)
Double single quotes (italic or bold wikimarkup) are not kerned.
Double single quotes (italic or bold wiki-markup) are not kerned.


Replaces unicode quotemarks in plain text or in the label portion of a [[L|D]] style wikilink with typewriter
Replaces Unicode quote marks in plain text or in the label portion of a [[L|D]] style wikilink with typewriter
quote marks regardless of the need for kerning.  Unicode quote marks are not replaced in simple [[D]] wikilinks.
quote marks regardless of the need for kerning.  Unicode quote marks are not replaced in simple [[D]] wikilinks.


Call this function for chapter titles, for website titles, etc; not for book titles.
Call this function for chapter titles, for website titles, etc.; not for book titles.


]=]
]=]


local function kern_quotes (str)
local function kern_quotes (str)
local cap='';
local cap = '';
local cap2='';
local wl_type, label, link;
local wl_type, label, link;


wl_type, label, link = is_wikilink (str); -- wl_type is: 0, no wl (text in label variable); 1, [[D]]; 2, [[L|D]]
wl_type, label, link = utilities.is_wikilink (str); -- wl_type is: 0, no wl (text in label variable); 1, [[D]]; 2, [[L|D]]
if 1 == wl_type then -- [[D]] simple wikilink with or without quote marks
if 1 == wl_type then -- [[D]] simple wikilink with or without quote marks
if mw.ustring.match (str, '%[%[[\"“”\'‘’].+[\"“”\'‘’]%]%]') then -- leading and trailing quote marks
if mw.ustring.match (str, '%[%[[\"“”\'‘’].+[\"“”\'‘’]%]%]') then -- leading and trailing quote marks
str = substitute (cfg.presentation['kern-wl-both'], str);
str = utilities.substitute (cfg.presentation['kern-left'], str);
str = utilities.substitute (cfg.presentation['kern-right'], str);
elseif mw.ustring.match (str, '%[%[[\"“”\'‘’].+%]%]') then -- leading quote marks
elseif mw.ustring.match (str, '%[%[[\"“”\'‘’].+%]%]') then -- leading quote marks
str = substitute (cfg.presentation['kern-wl-left'], str);
str = utilities.substitute (cfg.presentation['kern-left'], str);
elseif mw.ustring.match (str, '%[%[.+[\"“”\'‘’]%]%]') then -- trailing quote marks
elseif mw.ustring.match (str, '%[%[.+[\"“”\'‘’]%]%]') then -- trailing quote marks
str = substitute (cfg.presentation['kern-wl-right'], str);
str = utilities.substitute (cfg.presentation['kern-right'], str);
end
end


else -- plain text or [[L|D]]; text in label variable
else -- plain text or [[L|D]]; text in label variable
label= mw.ustring.gsub (label, '[“”]', '\"'); -- replace “” (U+201C & U+201D) with " (typewriter double quote mark)
label = mw.ustring.gsub (label, '[“”]', '\"'); -- replace “” (U+201C & U+201D) with " (typewriter double quote mark)
label= mw.ustring.gsub (label, '[‘’]', '\''); -- replace ‘’ (U+2018 & U+2019) with ' (typewriter single quote mark)
label = mw.ustring.gsub (label, '[‘’]', '\''); -- replace ‘’ (U+2018 & U+2019) with ' (typewriter single quote mark)


cap, cap2 = mw.ustring.match (label, "^([\"\'])([^\'].+)"); -- match leading double or single quote but not doubled single quotes (italic markup)
cap = mw.ustring.match (label, "^([\"\'][^\'].+)"); -- match leading double or single quote but not doubled single quotes (italic markup)
if is_set (cap) then
if utilities.is_set (cap) then
label = substitute (cfg.presentation['kern-left'], {cap, cap2});
label = utilities.substitute (cfg.presentation['kern-left'], cap);
end
end
cap, cap2 = mw.ustring.match (label, "^(.+[^\'])([\"\'])$") -- match trailing double or single quote but not doubled single quotes (italic markup)
cap = mw.ustring.match (label, "^(.+[^\'][\"\'])$") -- match trailing double or single quote but not doubled single quotes (italic markup)
if is_set (cap) then
if utilities.is_set (cap) then
label = substitute (cfg.presentation['kern-right'], {cap, cap2});
label = utilities.substitute (cfg.presentation['kern-right'], cap);
end
end
if 2 == wl_type then
if 2 == wl_type then
str = make_wikilink (link, label); -- reassemble the wikilink
str = utilities.make_wikilink (link, label); -- reassemble the wikilink
else
else
str = label;
str = label;
Line 509: Line 503:
--[[--------------------------< F O R M A T _ S C R I P T _ V A L U E >----------------------------------------
--[[--------------------------< F O R M A T _ S C R I P T _ V A L U E >----------------------------------------


|script-title= holds title parameters that are not written in Latin based scripts: Chinese, Japanese, Arabic, Hebrew, etc. These scripts should
|script-title= holds title parameters that are not written in Latin-based scripts: Chinese, Japanese, Arabic, Hebrew, etc. These scripts should
not be italicized and may be written right-to-left.  The value supplied by |script-title= is concatenated onto Title after Title has been wrapped
not be italicized and may be written right-to-left.  The value supplied by |script-title= is concatenated onto Title after Title has been wrapped
in italic markup.
in italic markup.


Regardless of language, all values provided by |script-title= are wrapped in <bdi>...</bdi> tags to isolate rtl languages from the English left to right.
Regardless of language, all values provided by |script-title= are wrapped in <bdi>...</bdi> tags to isolate RTL languages from the English left to right.


|script-title= provides a unique feature.  The value in |script-title= may be prefixed with a two-character ISO639-1 language code and a colon:
|script-title= provides a unique feature.  The value in |script-title= may be prefixed with a two-character ISO 639-1 language code and a colon:
|script-title=ja:*** *** (where * represents a Japanese character)
|script-title=ja:*** *** (where * represents a Japanese character)
Spaces between the two-character code and the colon and the colon and the first script character are allowed:
Spaces between the two-character code and the colon and the colon and the first script character are allowed:
Line 523: Line 517:
Spaces preceding the prefix are allowed: |script-title = ja:*** ***
Spaces preceding the prefix are allowed: |script-title = ja:*** ***


The prefix is checked for validity.  If it is a valid ISO639-1 language code, the lang attribute (lang="ja") is added to the <bdi> tag so that browsers can
The prefix is checked for validity.  If it is a valid ISO 639-1 language code, the lang attribute (lang="ja") is added to the <bdi> tag so that browsers can
know the language the tag contains.  This may help the browser render the script more correctly.  If the prefix is invalid, the lang attribute
know the language the tag contains.  This may help the browser render the script more correctly.  If the prefix is invalid, the lang attribute
is not added.  At this time there is no error message for this condition.
is not added.  At this time there is no error message for this condition.
Line 534: Line 528:
local lang=''; -- initialize to empty string
local lang=''; -- initialize to empty string
local name;
local name;
if script_value:match('^%l%l%s*:') then -- if first 3 non-space characters are script language prefix
if script_value:match('^%l%l%l?%s*:') then -- if first 3 or 4 non-space characters are script language prefix
lang = script_value:match('^(%l%l)%s*:%s*%S.*'); -- get the language prefix or nil if there is no script
lang = script_value:match('^(%l%l%l?)%s*:%s*%S.*'); -- get the language prefix or nil if there is no script
if not is_set (lang) then
if not utilities.is_set (lang) then
table.insert( z.message_tail, { set_error( 'script_parameter', {script_param, 'missing title part'}, true ) } ); -- prefix without 'title'; add error message
utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['missing title part']}); -- prefix without 'title'; add error message
return ''; -- script_value was just the prefix so return empty string
return ''; -- script_value was just the prefix so return empty string
end
end
-- if we get this far we have prefix and script
-- if we get this far we have prefix and script
name = cfg.lang_code_remap[lang] or mw.language.fetchLanguageName( lang, "en" ); -- get language name so that we can use it to categorize
name = cfg.lang_tag_remap[lang] or mw.language.fetchLanguageName( lang, cfg.this_wiki_code ); -- get language name so that we can use it to categorize
if is_set (name) then -- is prefix a proper ISO 639-1 language code?
if utilities.is_set (name) then -- is prefix a proper ISO 639-1 language code?
script_value = script_value:gsub ('^%l%l%s*:%s*', ''); -- strip prefix from script
script_value = script_value:gsub ('^%l+%s*:%s*', ''); -- strip prefix from script
-- is prefix one of these language codes?
-- is prefix one of these language codes?
if in_array (lang, cfg.script_lang_codes) then
if utilities.in_array (lang, cfg.script_lang_codes) then
add_prop_cat ('script_with_name', {name, lang})
utilities.add_prop_cat ('script', {name, lang})
else
else
table.insert( z.message_tail, { set_error( 'script_parameter', {script_param, 'unknown language code'}, true ) } ); -- unknown script-language; add error message
utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['unknown language code']}); -- unknown script-language; add error message
end
end
lang = ' lang="' .. lang .. '" '; -- convert prefix into a lang attribute
lang = ' lang="' .. lang .. '" '; -- convert prefix into a lang attribute
else
else
table.insert( z.message_tail, { set_error( 'script_parameter', {script_param, 'invalid language code'}, true ) } ); -- invalid language code; add error message
utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['invalid language code']}); -- invalid language code; add error message
lang = ''; -- invalid so set lang to empty string
lang = ''; -- invalid so set lang to empty string
end
end
else
else
table.insert( z.message_tail, { set_error( 'script_parameter', {script_param, 'missing prefix'}, true ) } ); -- no language code prefix; add error message
utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['missing prefix']}); -- no language code prefix; add error message
end
end
script_value = substitute (cfg.presentation['bdi'], {lang, script_value}); -- isolate in case script is rtl
script_value = utilities.substitute (cfg.presentation['bdi'], {lang, script_value}); -- isolate in case script is RTL


return script_value;
return script_value;
Line 566: Line 560:
--[[--------------------------< S C R I P T _ C O N C A T E N A T E >------------------------------------------
--[[--------------------------< S C R I P T _ C O N C A T E N A T E >------------------------------------------


Initially for |title= and |script-title=, this function concatenates those two parameter values after the script value has been  
Initially for |title= and |script-title=, this function concatenates those two parameter values after the script
wrapped in <bdi> tags.
value has been wrapped in <bdi> tags.
 
]]
]]


local function script_concatenate (title, script, script_param)
local function script_concatenate (title, script, script_param)
if is_set (script) then
if utilities.is_set (script) then
script = format_script_value (script, script_param); -- <bdi> tags, lang atribute, categorization, etc; returns empty string on error
script = format_script_value (script, script_param); -- <bdi> tags, lang attribute, categorization, etc.; returns empty string on error
if is_set (script) then
if utilities.is_set (script) then
title = title .. ' ' .. script; -- concatenate title and script title
title = title .. ' ' .. script; -- concatenate title and script title
end
end
Line 590: Line 585:


local function wrap_msg (key, str, lower)
local function wrap_msg (key, str, lower)
if not is_set( str ) then
if not utilities.is_set ( str ) then
return "";
return "";
end
end
Line 596: Line 591:
local msg;
local msg;
msg = cfg.messages[key]:lower(); -- set the message to lower case before  
msg = cfg.messages[key]:lower(); -- set the message to lower case before  
return substitute( msg, str ); -- including template text
return utilities.substitute ( msg, str ); -- including template text
else
else
return substitute( cfg.messages[key], str );
return utilities.substitute ( cfg.messages[key], str );
end
end
end
end




--[[--------------------------< W I K I S O U R C E _ U R L _ M A K E >----------------------------------------
--[[----------------< W I K I S O U R C E _ U R L _ M A K E >-------------------


makes a wikisource url from wikisource interwiki link.  returns the url and appropriate label; nil else.
Makes a Wikisource URL from Wikisource interwiki-link.  Returns the URL and appropriate
label; nil else.


str is the value assigned to |chapter= (or aliases) or |title= or |title-link=
str is the value assigned to |chapter= (or aliases) or |title= or |title-link=
Line 614: Line 610:
local wl_type, D, L;
local wl_type, D, L;
local ws_url, ws_label;
local ws_url, ws_label;
local wikisource_prefix = table.concat ({'https://', cfg.this_wiki_code, '.wikisource.org/wiki/'});


wl_type, D, L = is_wikilink (str); -- wl_type is 0 (not a wikilink), 1 (simple wikilink), 2 (complex wikilink)
wl_type, D, L = utilities.is_wikilink (str); -- wl_type is 0 (not a wikilink), 1 (simple wikilink), 2 (complex wikilink)


if 0 == wl_type then -- not a wikilink; might be from |title-link=
if 0 == wl_type then -- not a wikilink; might be from |title-link=
str = D:match ('^[Ww]ikisource:(.+)') or D:match ('^[Ss]:(.+)'); -- article title from interwiki link with long-form or short-form namespace
str = D:match ('^[Ww]ikisource:(.+)') or D:match ('^[Ss]:(.+)'); -- article title from interwiki link with long-form or short-form namespace
if is_set (str) then
if utilities.is_set (str) then
ws_url = table.concat ({ -- build a wikisource url
ws_url = table.concat ({ -- build a Wikisource URL
'https://en.wikisource.org/wiki/', -- prefix
wikisource_prefix, -- prefix
str, -- article title
str, -- article title
});
});
ws_label = str; -- label for the url
ws_label = str; -- label for the URL
end
end
elseif 1 == wl_type then -- simple wikilink: [[Wikisource:ws article]]
elseif 1 == wl_type then -- simple wikilink: [[Wikisource:ws article]]
str = D:match ('^[Ww]ikisource:(.+)') or D:match ('^[Ss]:(.+)'); -- article title from interwiki link with long-form or short-form namespace
str = D:match ('^[Ww]ikisource:(.+)') or D:match ('^[Ss]:(.+)'); -- article title from interwiki link with long-form or short-form namespace
if is_set (str) then
if utilities.is_set (str) then
ws_url = table.concat ({ -- build a wikisource url
ws_url = table.concat ({ -- build a Wikisource URL
'https://en.wikisource.org/wiki/', -- prefix
wikisource_prefix, -- prefix
str, -- article title
str, -- article title
});
});
ws_label = str; -- label for the url
ws_label = str; -- label for the URL
end
end
elseif 2 == wl_type then -- non-so-simple wikilink: [[Wikisource:ws article|displayed text]] ([[L|D]])
elseif 2 == wl_type then -- non-so-simple wikilink: [[Wikisource:ws article|displayed text]] ([[L|D]])
str = L:match ('^[Ww]ikisource:(.+)') or L:match ('^[Ss]:(.+)'); -- article title from interwiki link with long-form or short-form namespace
str = L:match ('^[Ww]ikisource:(.+)') or L:match ('^[Ss]:(.+)'); -- article title from interwiki link with long-form or short-form namespace
if is_set (str) then
if utilities.is_set (str) then
ws_label = D; -- get ws article name from display portion of interwiki link
ws_label = D; -- get ws article name from display portion of interwiki link
ws_url = table.concat ({ -- build a wikisource url
ws_url = table.concat ({ -- build a Wikisource URL
'https://en.wikisource.org/wiki/', -- prefix
wikisource_prefix, -- prefix
str, -- article title without namespace from link portion of wikilink
str, -- article title without namespace from link portion of wikilink
});
});
end
end
end
end
 
if ws_url then
if ws_url then
ws_url = mw.uri.encode (ws_url, 'WIKI'); -- make a usable url
ws_url = mw.uri.encode (ws_url, 'WIKI'); -- make a usable URL
ws_url = ws_url:gsub ('%%23', '#'); -- undo percent encoding of anchor
ws_url = ws_url:gsub ('%%23', '#'); -- undo percent-encoding of fragment marker
end
end


return ws_url, ws_label, L or D; -- return proper url or nil and a label or nil
return ws_url, ws_label, L or D; -- return proper URL or nil and a label or nil
end
end




--[[--------------------------< F O R M A T _ P E R I O D I C A L >--------------------------------------------
--[[----------------< F O R M A T _ P E R I O D I C A L >-----------------------


Format the four periodical parameters: |script-<periodical>=, |<periodical>=, and |trans-<periodical>= into a single Periodical meta-
Format the three periodical parameters: |script-<periodical>=, |<periodical>=,
parameter.
and |trans-<periodical>= into a single Periodical meta-parameter.


]]
]]


local function format_periodical (script_periodical, script_periodical_source, periodical, trans_periodical)
local function format_periodical (script_periodical, script_periodical_source, periodical, trans_periodical)
local periodical_error = '';


if not is_set (periodical) then
if not utilities.is_set (periodical) then
periodical = ''; -- to be safe for concatenation
periodical = ''; -- to be safe for concatenation
else
else
periodical = wrap_style ('italic-title', periodical); -- style  
periodical = utilities.wrap_style ('italic-title', periodical); -- style  
end
end


periodical = script_concatenate (periodical, script_periodical, script_periodical_source); -- <bdi> tags, lang atribute, categorization, etc; must be done after title is wrapped
periodical = script_concatenate (periodical, script_periodical, script_periodical_source); -- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped


if is_set (trans_periodical) then
if utilities.is_set (trans_periodical) then
trans_periodical = wrap_style ('trans-italic-title', trans_periodical);
trans_periodical = utilities.wrap_style ('trans-italic-title', trans_periodical);
if is_set (periodical) then
if utilities.is_set (periodical) then
periodical = periodical .. ' ' .. trans_periodical;
periodical = periodical .. ' ' .. trans_periodical;
else -- here when transchapter without chapter or script-chapter
else -- here when trans-periodical without periodical or script-periodical
periodical = trans_periodical;
periodical = trans_periodical;
periodical_error = ' ' .. set_error ('trans_missing_title', {'periodical'});
utilities.set_message ('err_trans_missing_title', {'periodical'});
end
end
end
end


return periodical .. periodical_error;
return periodical;
end
end




--[[--------------------------< F O R M A T _ C H A P T E R _ T I T L E >--------------------------------------
--[[------------------< F O R M A T _ C H A P T E R _ T I T L E >---------------


Format the four chapter parameters: |script-chapter=, |chapter=, |trans-chapter=, and |chapter-url= into a single Chapter meta-
Format the four chapter parameters: |script-chapter=, |chapter=, |trans-chapter=,
parameter (chapter_url_source used for error messages).
and |chapter-url= into a single chapter meta- parameter (chapter_url_source used
for error messages).


]]
]]


local function format_chapter_title (scriptchapter, script_chapter_source, chapter, transchapter, chapterurl, chapter_url_source, no_quotes, access)
local function format_chapter_title (script_chapter, script_chapter_source, chapter, chapter_source, trans_chapter, trans_chapter_source, chapter_url, chapter_url_source, no_quotes, access)
local chapter_error = '';
local ws_url, ws_label, L = wikisource_url_make (chapter); -- make a wikisource URL and label from a wikisource interwiki link
 
local ws_url, ws_label, L = wikisource_url_make (chapter); -- make a wikisource url and label from a wikisource interwiki link
if ws_url then
if ws_url then
ws_label = ws_label:gsub ('_', ''); -- replace underscore separaters with space characters
ws_label = ws_label:gsub ('_', ' '); -- replace underscore separators with space characters
chapter = ws_label;
chapter = ws_label;
end
end


if not is_set (chapter) then
if not utilities.is_set (chapter) then
chapter = ''; -- to be safe for concatenation
chapter = ''; -- to be safe for concatenation
else
else
if false == no_quotes then
if false == no_quotes then
chapter = kern_quotes (chapter); -- if necessary, separate chapter title's leading and trailing quote marks from Module provided quote marks
chapter = kern_quotes (chapter); -- if necessary, separate chapter title's leading and trailing quote marks from module provided quote marks
chapter = wrap_style ('quoted-title', chapter);
chapter = utilities.wrap_style ('quoted-title', chapter);
end
end
end
end


chapter = script_concatenate (chapter, scriptchapter, script_chapter_source); -- <bdi> tags, lang atribute, categorization, etc; must be done after title is wrapped
chapter = script_concatenate (chapter, script_chapter, script_chapter_source); -- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped


if is_set (chapterurl) then
if utilities.is_set (chapter_url) then
chapter = external_link (chapterurl, chapter, chapter_url_source, access); -- adds bare_url_missing_title error if appropriate
chapter = external_link (chapter_url, chapter, chapter_url_source, access); -- adds bare_url_missing_title error if appropriate
elseif ws_url then
elseif ws_url then
chapter = external_link (ws_url, chapter .. '&nbsp;', 'ws link in chapter'); -- adds bare_url_missing_title error if appropriate; space char to move icon away from chap text; TODO: better way to do this?
chapter = external_link (ws_url, chapter .. '&nbsp;', 'ws link in chapter'); -- adds bare_url_missing_title error if appropriate; space char to move icon away from chap text; TODO: better way to do this?
chapter = substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, chapter});
chapter = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, chapter});
end
end


if is_set (transchapter) then
if utilities.is_set (trans_chapter) then
transchapter = wrap_style ('trans-quoted-title', transchapter);
trans_chapter = utilities.wrap_style ('trans-quoted-title', trans_chapter);
if is_set (chapter) then
if utilities.is_set (chapter) then
chapter = chapter .. ' ' .. transchapter;
chapter = chapter .. ' ' .. trans_chapter;
else -- here when transchapter without chapter or script-chapter
else -- here when trans_chapter without chapter or script-chapter
chapter = transchapter;
chapter = trans_chapter;
chapter_error = ' ' .. set_error ('trans_missing_title', {'chapter'});
chapter_source = trans_chapter_source:match ('trans%-?(.+)'); -- when no chapter, get matching name from trans-<param>
utilities.set_message ('err_trans_missing_title', {chapter_source});
end
end
end
end


-- if is_set (chapterurl) then
return chapter;
-- chapter = external_link (chapterurl, chapter, chapter_url_source, access); -- adds bare_url_missing_title error if appropriate
-- end
 
return chapter .. chapter_error;
end
end




--[[--------------------------< H A S _ I N V I S I B L E _ C H A R S >----------------------------------------
--[[----------------< H A S _ I N V I S I B L E _ C H A R S >-------------------


This function searches a parameter's value for nonprintable or invisible characters. The search stops at the
This function searches a parameter's value for non-printable or invisible characters.
first match.
The search stops at the first match.


This function will detect the visible replacement character when it is part of the wikisource.
This function will detect the visible replacement character when it is part of the Wikisource.


Detects but ignores nowiki and math stripmarkers.  Also detects other named stripmarkers (gallery, math, pre, ref)
Detects but ignores nowiki and math stripmarkers.  Also detects other named stripmarkers
and identifies them with a slightly different error message. See also coins_cleanup().
(gallery, math, pre, ref) and identifies them with a slightly different error message.
See also coins_cleanup().


Output of this function is an error message that identifies the character or the Unicode group, or the stripmarker
Output of this function is an error message that identifies the character or the
that was detected along with its position (or, for multi-byte characters, the position of its first byte) in the
Unicode group, or the stripmarker that was detected along with its position (or,
parameter value.
for multi-byte characters, the position of its first byte) in the parameter value.


]]
]]
Line 757: Line 750:
local function has_invisible_chars (param, v)
local function has_invisible_chars (param, v)
local position = ''; -- position of invisible char or starting position of stripmarker
local position = ''; -- position of invisible char or starting position of stripmarker
local dummy; -- end of matching string; not used but required to hold end position when a capture is returned
local capture; -- used by stripmarker detection to hold name of the stripmarker
local capture; -- used by stripmarker detection to hold name of the stripmarker
local i=1;
local stripmarker; -- boolean set true when a stripmarker is found
local stripmarker, apostrophe;
 
capture = string.match (v, '[%w%p ]*'); -- test for values that are simple ASCII text and bypass other tests if true
capture = string.match (v, '[%w%p ]*'); -- Test for values that are simple ASCII text and bypass other tests if true
if capture == v then -- if same there are no Unicode characters
if capture == v then -- if same there are no unicode characters
return;
return;
end
end


while cfg.invisible_chars[i] do
for _, invisible_char in ipairs (cfg.invisible_chars) do
local char=cfg.invisible_chars[i][1] -- the character or group name
local char_name = invisible_char[1]; -- the character or group name
local pattern=cfg.invisible_chars[i][2] -- the pattern used to find it
local pattern = invisible_char[2]; -- the pattern used to find it
position, dummy, capture = mw.ustring.find (v, pattern) -- see if the parameter value contains characters that match the pattern
position, _, capture = mw.ustring.find (v, pattern); -- see if the parameter value contains characters that match the pattern
if position and (char == 'zero width joiner') then -- if we found a zero width joiner character
if position and (cfg.invisible_defs.zwj == capture) then -- if we found a zero-width joiner character
if mw.ustring.find (v, cfg.indic_script) then -- its ok if one of the indic scripts
if mw.ustring.find (v, cfg.indic_script) then -- it's ok if one of the Indic scripts
position = nil; -- unset position
elseif cfg.emoji_t[mw.ustring.codepoint (v, position+1)] then -- is zwj followed by a character listed in emoji{}?
position = nil; -- unset position
position = nil; -- unset position
end
end
Line 780: Line 773:
if position then
if position then
if 'nowiki' == capture or 'math' == capture or -- nowiki and math stripmarkers (not an error condition)
if 'nowiki' == capture or 'math' == capture or -- nowiki and math stripmarkers (not an error condition)
('templatestyles' == capture and in_array (param, {'id', 'quote'})) then -- templatestyles stripmarker allowed in these parameters
('templatestyles' == capture and utilities.in_array (param, {'id', 'quote'})) then -- templatestyles stripmarker allowed in these parameters
stripmarker = true; -- set a flag
stripmarker = true; -- set a flag
elseif true == stripmarker and 'delete' == char then -- because stripmakers begin and end with the delete char, assume that we've found one end of a stripmarker
elseif true == stripmarker and cfg.invisible_defs.del == capture then -- because stripmakers begin and end with the delete char, assume that we've found one end of a stripmarker
position = nil; -- unset
position = nil; -- unset
else
else
local err_msg;
local err_msg;
if capture then
if capture and not (cfg.invisible_defs.del == capture or cfg.invisible_defs.zwj == capture) then
err_msg = capture .. ' ' .. char;
err_msg = capture .. ' ' .. char_name;
else
else
err_msg = char .. ' ' .. 'character';
err_msg = char_name .. ' ' .. 'character';
end
end


table.insert( z.message_tail, { set_error( 'invisible_char', {err_msg, wrap_style ('parameter', param), position}, true ) } ); -- add error message
utilities.set_message ('err_invisible_char', {err_msg, utilities.wrap_style ('parameter', param), position}); -- add error message
return; -- and done with this parameter
return; -- and done with this parameter
end
end
end
end
i=i+1; -- bump our index
end
end
end
end




--[[--------------------------< A R G U M E N T _ W R A P P E R >----------------------------------------------
--[[-------------------< A R G U M E N T _ W R A P P E R >----------------------


Argument wrapper.  This function provides support for argument mapping defined in the configuration file so that
Argument wrapper.  This function provides support for argument mapping defined
multiple names can be transparently aliased to single internal variable.
in the configuration file so that multiple names can be transparently aliased to
single internal variable.


]]
]]


local function argument_wrapper( args )
local function argument_wrapper ( args )
local origin = {};
local origin = {};
return setmetatable({
return setmetatable({
ORIGIN = function( self, k )
ORIGIN = function ( self, k )
local dummy = self[k]; --force the variable to be loaded.
local dummy = self[k]; -- force the variable to be loaded.
return origin[k];
return origin[k];
end
end
Line 826: Line 819:
if type( list ) == 'table' then
if type( list ) == 'table' then
v, origin[k] = select_one( args, list, 'redundant_parameters' );
v, origin[k] = utilities.select_one ( args, list, 'err_redundant_parameters' );
if origin[k] == nil then
if origin[k] == nil then
origin[k] = ''; -- Empty string, not nil
origin[k] = ''; -- Empty string, not nil
end
end
elseif list ~= nil then
elseif list ~= nil then
Line 840: Line 833:
-- Empty strings, not nil;
-- Empty strings, not nil;
if v == nil then
if v == nil then
v = cfg.defaults[k] or '';
v = '';
origin[k] = '';
origin[k] = '';
end
end
Line 851: Line 844:




--[[--------------------------< N O W R A P _ D A T E >--------------------------------------------------------
--[[--------------------------< N O W R A P _ D A T E >-------------------------


When date is YYYY-MM-DD format wrap in nowrap span: <span ...>YYYY-MM-DD</span>. When date is DD MMMM YYYY or is
When date is YYYY-MM-DD format wrap in nowrap span: <span ...>YYYY-MM-DD</span>.
MMMM DD, YYYY then wrap in nowrap span: <span ...>DD MMMM</span> YYYY or <span ...>MMMM DD,</span> YYYY
When date is DD MMMM YYYY or is MMMM DD, YYYY then wrap in nowrap span:
<span ...>DD MMMM</span> YYYY or <span ...>MMMM DD,</span> YYYY


DOES NOT yet support MMMM YYYY or any of the date ranges.
DOES NOT yet support MMMM YYYY or any of the date ranges.
Line 861: Line 855:


local function nowrap_date (date)
local function nowrap_date (date)
local cap='';
local cap = '';
local cap2='';
local cap2 = '';


if date:match("^%d%d%d%d%-%d%d%-%d%d$") then
if date:match("^%d%d%d%d%-%d%d%-%d%d$") then
date = substitute (cfg.presentation['nowrap1'], date);
date = utilities.substitute (cfg.presentation['nowrap1'], date);
elseif date:match("^%a+%s*%d%d?,%s+%d%d%d%d$") or date:match ("^%d%d?%s*%a+%s+%d%d%d%d$") then
elseif date:match("^%a+%s*%d%d?,%s+%d%d%d%d$") or date:match ("^%d%d?%s*%a+%s+%d%d%d%d$") then
cap, cap2 = string.match (date, "^(.*)%s+(%d%d%d%d)$");
cap, cap2 = string.match (date, "^(.*)%s+(%d%d%d%d)$");
date = substitute (cfg.presentation['nowrap2'], {cap, cap2});
date = utilities.substitute (cfg.presentation['nowrap2'], {cap, cap2});
end
end
Line 876: Line 870:




--[[--------------------------< S E T _ T I T L E T Y P E >----------------------------------------------------
--[[--------------------------< S E T _ T I T L E T Y P E >---------------------


This function sets default title types (equivalent to the citation including |type=<default value>) for those templates that have defaults.
This function sets default title types (equivalent to the citation including
Also handles the special case where it is desirable to omit the title type from the rendered citation (|type=none).
|type=<default value>) for those templates that have defaults. Also handles the
special case where it is desirable to omit the title type from the rendered citation
(|type=none).


]]
]]


local function set_titletype (cite_class, title_type)
local function set_titletype (cite_class, title_type)
if is_set(title_type) then
if utilities.is_set (title_type) then
if "none" == title_type then
if 'none' == cfg.keywords_xlate[title_type] then
title_type = ""; -- if |type=none then type parameter not displayed
title_type = ''; -- if |type=none then type parameter not displayed
end
end
return title_type; -- if |type= has been set to any other value use that value
return title_type; -- if |type= has been set to any other value use that value
Line 895: Line 891:




--[[--------------------------< H Y P H E N _ T O _ D A S H >--------------------------------------------------
--[[--------------------------< S A F E _ J O I N >-----------------------------
 
Converts a hyphen to a dash under certain conditions.  The hyphen must separate like items; unlike items are
returned unmodified.  These forms are modified:
letter - letter (A - B)
digit - digit (4-5)
digit separator digit - digit separator digit (4.1-4.5 or 4-1-4-5)
letterdigit - letterdigit (A1-A5) (an optional separator between letter and digit is supported – a.1-a.5 or a-1-a-5)
digitletter - digitletter (5a - 5d) (an optional separator between letter and digit is supported – 5.a-5.d or 5-a-5-d)
 
any other forms are returned unmodified.
 
str may be a comma- or semicolon-separated list
 
]]
 
local function hyphen_to_dash( str )
if not is_set (str) then
return str;
end
str, count = str:gsub ('^%(%((.+)%)%)$', '%1'); -- remove accept-this-as-written markup when it wraps all of str
if 0 ~= count then -- non-zero when markup removed; zero else
return str; -- nothing to do, we're done
end
str = str:gsub ('&[nm]dash;', {['&ndash;'] = '–', ['&mdash;'] = '—'}); -- replace &mdash; and &ndash; entities  with their characters; semicolon mucks up the text.split
local out = {};
local list = mw.text.split (str, '%s*[,;]%s*'); -- split str at comma or semicolon separators if there are any
 
for _, item in ipairs (list) do -- for each item in the list
if mw.ustring.match (item, '^%w*[%.%-]?%w+%s*[%-–—]%s*%w*[%.%-]?%w+$') then -- if a hyphenated range or has endash or emdash separators
if item:match ('%a+[%.%-]?%d+%s*%-%s*%a+[%.%-]?%d+') or -- letterdigit hyphen letterdigit (optional separator between letter and digit)
item:match ('%d+[%.%-]?%a+%s*%-%s*%d+[%.%-]?%a+') or -- digitletter hyphen digitletter (optional separator between digit and letter)
item:match ('%d+[%.%-]%d+%s*%-%s*%d+[%.%-]%d+') or -- digit separator digit hyphen digit separator digit
item:match ('%d+%s*%-%s*%d+') or -- digit hyphen digit
item:match ('%a+%s*%-%s*%a+') then -- letter hyphen letter
item = item:gsub ('(%w*[%.%-]?%w+)%s*%-%s*(%w*[%.%-]?%w+)', '%1–%2'); -- replace hyphen, remove extraneous space characters
else
item = mw.ustring.gsub (item, '%s*[–—]%s*', '–'); -- for endash or emdash separated ranges, replace em with en, remove extraneous white space
end
end
item = item:gsub ('^%(%((.+)%)%)$', '%1'); -- remove the accept-this-as-written markup
table.insert (out, item); -- add the (possibly modified) item to the output table
end
 
return table.concat (out, ', '); -- concatenate the output table into a comma separated string
end
 
 
--[[--------------------------< S A F E _ J O I N >------------------------------------------------------------


Joins a sequence of strings together while checking for duplicate separation characters.
Joins a sequence of strings together while checking for duplicate separation characters.
Line 953: Line 898:


local function safe_join( tbl, duplicate_char )
local function safe_join( tbl, duplicate_char )
local f = {}; -- create a function table appropriate to type of 'dupicate character'
local f = {}; -- create a function table appropriate to type of 'duplicate character'
if 1 == #duplicate_char then -- for single byte ascii characters use the string library functions
if 1 == #duplicate_char then -- for single byte ASCII characters use the string library functions
f.gsub=string.gsub
f.gsub = string.gsub
f.match=string.match
f.match = string.match
f.sub=string.sub
f.sub = string.sub
else -- for multi-byte characters use the ustring library functions
else -- for multi-byte characters use the ustring library functions
f.gsub=mw.ustring.gsub
f.gsub = mw.ustring.gsub
f.match=mw.ustring.match
f.match = mw.ustring.match
f.sub=mw.ustring.sub
f.sub = mw.ustring.sub
end
end


Line 974: Line 919:
str = value; -- assign value to it (first time through the loop)
str = value; -- assign value to it (first time through the loop)
elseif value ~= '' then
elseif value ~= '' then
if value:sub(1,1) == '<' then -- Special case of values enclosed in spans and other markup.
if value:sub(1, 1) == '<' then -- special case of values enclosed in spans and other markup.
comp = value:gsub( "%b<>", "" ); -- remove html markup (<span>string</span> -> string)
comp = value:gsub( "%b<>", "" ); -- remove HTML markup (<span>string</span> -> string)
else
else
comp = value;
comp = value;
end
end
-- typically duplicate_char is sepc
-- typically duplicate_char is sepc
if f.sub(comp, 1,1) == duplicate_char then -- is first character same as duplicate_char? why test first character?
if f.sub(comp, 1, 1) == duplicate_char then -- is first character same as duplicate_char? why test first character?
--  Because individual string segments often (always?) begin with terminal punct for the
--  Because individual string segments often (always?) begin with terminal punct for the
--  preceding segment: 'First element' .. 'sepc next element' .. etc?
--  preceding segment: 'First element' .. 'sepc next element' .. etc.?
trim = false;
trim = false;
end_chr = f.sub(str, -1,-1); -- get the last character of the output string
end_chr = f.sub(str, -1, -1); -- get the last character of the output string
-- str = str .. "<HERE(enchr=" .. end_chr.. ")" -- debug stuff?
-- str = str .. "<HERE(enchr=" .. end_chr .. ")" -- debug stuff?
if end_chr == duplicate_char then -- if same as separator
if end_chr == duplicate_char then -- if same as separator
str = f.sub(str, 1,-2); -- remove it
str = f.sub(str, 1, -2); -- remove it
elseif end_chr == "'" then -- if it might be wikimarkup
elseif end_chr == "'" then -- if it might be wiki-markup
if f.sub(str, -3,-1) == duplicate_char .. "''" then -- if last three chars of str are sepc''  
if f.sub(str, -3, -1) == duplicate_char .. "''" then -- if last three chars of str are sepc''  
str = f.sub(str, 1, -4) .. "''"; -- remove them and add back ''
str = f.sub(str, 1, -4) .. "''"; -- remove them and add back ''
elseif  f.sub(str, -5,-1) == duplicate_char .. "]]''" then -- if last five chars of str are sepc]]''  
elseif  f.sub(str, -5, -1) == duplicate_char .. "]]''" then -- if last five chars of str are sepc]]''  
trim = true; -- why? why do this and next differently from previous?
trim = true; -- why? why do this and next differently from previous?
elseif f.sub(str, -4,-1) == duplicate_char .. "]''" then -- if last four chars of str are sepc]''  
elseif f.sub(str, -4, -1) == duplicate_char .. "]''" then -- if last four chars of str are sepc]''  
trim = true; -- same question
trim = true; -- same question
end
end
elseif end_chr == "]" then -- if it might be wikimarkup
elseif end_chr == "]" then -- if it might be wiki-markup
if f.sub(str, -3,-1) == duplicate_char .. "]]" then -- if last three chars of str are sepc]] wikilink  
if f.sub(str, -3, -1) == duplicate_char .. "]]" then -- if last three chars of str are sepc]] wikilink  
trim = true;
trim = true;
elseif f.sub(str, -3,-1) == duplicate_char .. '"]' then -- if last three chars of str are sepc"] quoted external link  
elseif f.sub(str, -3, -1) == duplicate_char .. '"]' then -- if last three chars of str are sepc"] quoted external link  
trim = true;
trim = true;
elseif  f.sub(str, -2,-1) == duplicate_char .. "]" then -- if last two chars of str are sepc] external link
elseif  f.sub(str, -2, -1) == duplicate_char .. "]" then -- if last two chars of str are sepc] external link
trim = true;
trim = true;
elseif f.sub(str, -4,-1) == duplicate_char .. "'']" then -- normal case when |url=something & |title=Title.
elseif f.sub(str, -4, -1) == duplicate_char .. "'']" then -- normal case when |url=something & |title=Title.
trim = true;
trim = true;
end
end
elseif end_chr == " " then -- if last char of output string is a space
elseif end_chr == " " then -- if last char of output string is a space
if f.sub(str, -2,-1) == duplicate_char .. " " then -- if last two chars of str are <sepc><space>
if f.sub(str, -2, -1) == duplicate_char .. " " then -- if last two chars of str are <sepc><space>
str = f.sub(str, 1,-3); -- remove them both
str = f.sub(str, 1, -3); -- remove them both
end
end
end
end


if trim then
if trim then
if value ~= comp then -- value does not equal comp when value contains html markup
if value ~= comp then -- value does not equal comp when value contains HTML markup
local dup2 = duplicate_char;
local dup2 = duplicate_char;
if f.match(dup2, "%A" ) then dup2 = "%" .. dup2; end -- if duplicate_char not a letter then escape it
if f.match(dup2, "%A" ) then dup2 = "%" .. dup2; end -- if duplicate_char not a letter then escape it
value = f.gsub(value, "(%b<>)" .. dup2, "%1", 1 ) -- remove duplicate_char if it follows html markup
value = f.gsub(value, "(%b<>)" .. dup2, "%1", 1 ) -- remove duplicate_char if it follows HTML markup
else
else
value = f.sub(value, 2, -1 ); -- remove duplicate_char when it is first character
value = f.sub(value, 2, -1 ); -- remove duplicate_char when it is first character
Line 1,023: Line 968:
end
end
end
end
str = str .. value; --add it to the output string
str = str .. value; -- add it to the output string
end
end
end
end
Line 1,030: Line 975:




--[[--------------------------< I S _ S U F F I X >------------------------------------------------------------
--[[--------------------------< I S _ S U F F I X >-----------------------------


returns true is suffix is properly formed Jr, Sr, or ordinal in the range 2–9. Puncutation not allowed.
returns true if suffix is properly formed Jr, Sr, or ordinal in the range 1–9.
Puncutation not allowed.


]]
]]


local function is_suffix (suffix)
local function is_suffix (suffix)
if in_array (suffix, {'Jr', 'Sr', '2nd', '3rd'}) or suffix:match ('^%dth$') then
if utilities.in_array (suffix, {'Jr', 'Sr', 'Jnr', 'Snr', '1st', '2nd', '3rd'}) or suffix:match ('^%dth$') then
return true;
return true;
end
end
Line 1,044: Line 990:




--[[--------------------------< I S _ G O O D _ V A N C _ N A M E >--------------------------------------------
--[[--------------------< I S _ G O O D _ V A N C _ N A M E >-------------------


For Vancouver Style, author/editor names are supposed to be rendered in Latin (read ASCII) characters.  When a name
For Vancouver style, author/editor names are supposed to be rendered in Latin
uses characters that contain diacritical marks, those characters are to converted to the corresponding Latin character.
(read ASCII) characters.  When a name uses characters that contain diacritical
When a name is written using a non-Latin alphabet or logogram, that name is to be transliterated into Latin characters.
marks, those characters are to be converted to the corresponding Latin
These things are not currently possible in this module so are left to the editor to do.
character. When a name is written using a non-Latin alphabet or logogram, that
name is to be transliterated into Latin characters. The module doesn't do this
so editors may/must.


This test allows |first= and |last= names to contain any of the letters defined in the four Unicode Latin character sets
This test allows |first= and |last= names to contain any of the letters defined
in the four Unicode Latin character sets
[http://www.unicode.org/charts/PDF/U0000.pdf C0 Controls and Basic Latin] 0041–005A, 0061–007A
[http://www.unicode.org/charts/PDF/U0000.pdf C0 Controls and Basic Latin] 0041–005A, 0061–007A
[http://www.unicode.org/charts/PDF/U0080.pdf C1 Controls and Latin-1 Supplement] 00C0–00D6, 00D8–00F6, 00F8–00FF
[http://www.unicode.org/charts/PDF/U0080.pdf C1 Controls and Latin-1 Supplement] 00C0–00D6, 00D8–00F6, 00F8–00FF
Line 1,057: Line 1,006:
[http://www.unicode.org/charts/PDF/U0180.pdf Latin Extended-B] 0180–01BF, 01C4–024F
[http://www.unicode.org/charts/PDF/U0180.pdf Latin Extended-B] 0180–01BF, 01C4–024F


|lastn= also allowed to contain hyphens, spaces, and apostrophes. (http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35029/)
|lastn= also allowed to contain hyphens, spaces, and apostrophes.
(http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35029/)
|firstn= also allowed to contain hyphens, spaces, apostrophes, and periods
|firstn= also allowed to contain hyphens, spaces, apostrophes, and periods


This original test:
This original test:
if nil == mw.ustring.find (last, "^[A-Za-zÀ-ÖØ-öø-ƿDŽ-ɏ%-%s%']*$") or nil == mw.ustring.find (first, "^[A-Za-zÀ-ÖØ-öø-ƿDŽ-ɏ%-%s%'%.]+[2-6%a]*$") then
if nil == mw.ustring.find (last, "^[A-Za-zÀ-ÖØ-öø-ƿDŽ-ɏ%-%s%']*$")
was written ouside of the code editor and pasted here because the code editor gets confused between character insertion point and cursor position.
or nil == mw.ustring.find (first, "^[A-Za-zÀ-ÖØ-öø-ƿDŽ-ɏ%-%s%'%.]+[2-6%a]*$") then
The test has been rewritten to use decimal character escape sequence for the individual bytes of the unicode characters so that it is not necessary
was written outside of the code editor and pasted here because the code editor
to use an external editor to maintain this code.
gets confused between character insertion point and cursor position. The test has
been rewritten to use decimal character escape sequence for the individual bytes
of the Unicode characters so that it is not necessary to use an external editor
to maintain this code.


\195\128-\195\150 – À-Ö (U+00C0–U+00D6 – C0 controls)
\195\128-\195\150 – À-Ö (U+00C0–U+00D6 – C0 controls)
Line 1,073: Line 1,026:
]]
]]


local function is_good_vanc_name (last, first)
local function is_good_vanc_name (last, first, suffix, position)
local first, suffix = first:match ('(.-),?%s*([%dJS][%drndth]+)%.?$') or first; -- if first has something that looks like a generational suffix, get it
if not suffix then
 
if first:find ('[,%s]') then -- when there is a space or comma, might be first name/initials + generational suffix
if is_set (suffix) then
first = first:match ('(.-)[,%s]+'); -- get name/initials
suffix = first:match ('[,%s]+(.+)$'); -- get generational suffix
end
end
if utilities.is_set (suffix) then
if not is_suffix (suffix) then
if not is_suffix (suffix) then
add_vanc_error ('suffix');
add_vanc_error (cfg.err_msg_supl.suffix, position);
return false; -- not a name with an appropriate suffix
return false; -- not a name with an appropriate suffix
end
end
Line 1,084: Line 1,041:
if nil == mw.ustring.find (last, "^[A-Za-z\195\128-\195\150\195\152-\195\182\195\184-\198\191\199\132-\201\143%-%s%']*$") or
if nil == mw.ustring.find (last, "^[A-Za-z\195\128-\195\150\195\152-\195\182\195\184-\198\191\199\132-\201\143%-%s%']*$") or
nil == mw.ustring.find (first, "^[A-Za-z\195\128-\195\150\195\152-\195\182\195\184-\198\191\199\132-\201\143%-%s%'%.]*$") then
nil == mw.ustring.find (first, "^[A-Za-z\195\128-\195\150\195\152-\195\182\195\184-\198\191\199\132-\201\143%-%s%'%.]*$") then
add_vanc_error ('non-Latin character');
add_vanc_error (cfg.err_msg_supl['non-Latin char'], position);
return false; -- not a string of latin characters; Vancouver requires Romanization
return false; -- not a string of Latin characters; Vancouver requires Romanization
end;
end;
return true;
return true;
Line 1,093: Line 1,050:
--[[--------------------------< R E D U C E _ T O _ I N I T I A L S >------------------------------------------
--[[--------------------------< R E D U C E _ T O _ I N I T I A L S >------------------------------------------


Attempts to convert names to initials in support of |name-list-format=vanc.   
Attempts to convert names to initials in support of |name-list-style=vanc.   


Names in |firstn= may be separated by spaces or hyphens, or for initials, a period. See http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35062/.
Names in |firstn= may be separated by spaces or hyphens, or for initials, a period.
See http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35062/.


Vancouver style requires family rank designations (Jr, II, III, etc) to be rendered as Jr, 2nd, 3rd, etc.  See http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35085/.
Vancouver style requires family rank designations (Jr, II, III, etc.) to be rendered
This code only accepts and understands generational suffix in the Vancouver format because Roman numerals look like, and can be mistaken for, initials.
as Jr, 2nd, 3rd, etc.  See http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35085/.
This code only accepts and understands generational suffix in the Vancouver format
because Roman numerals look like, and can be mistaken for, initials.


This function uses ustring functions because firstname initials may be any of the unicode Latin characters accepted by is_good_vanc_name ().
This function uses ustring functions because firstname initials may be any of the
Unicode Latin characters accepted by is_good_vanc_name ().


]]
]]


local function reduce_to_initials(first)
local function reduce_to_initials(first, position)
local name, suffix = mw.ustring.match(first, "^(%u+) ([%dJS][%drndth]+)$");
local name, suffix = mw.ustring.match(first, "^(%u+) ([%dJS][%drndth]+)$");


if not name then -- if not initials and a suffix
if not name then -- if not initials and a suffix
name = mw.ustring.match(first, "^(%u+)$"); -- is it just intials?
name = mw.ustring.match(first, "^(%u+)$"); -- is it just initials?
end
end


Line 1,117: Line 1,078:
return first; -- one or two initials and a valid suffix so nothing to do
return first; -- one or two initials and a valid suffix so nothing to do
else
else
add_vanc_error ('suffix'); -- one or two initials with invalid suffix so error message
add_vanc_error (cfg.err_msg_supl.suffix, position); -- one or two initials with invalid suffix so error message
return first; -- and return first unmolested
return first; -- and return first unmolested
end
end
Line 1,125: Line 1,086:
end
end
end -- if here then name has 3 or more uppercase letters so treat them as a word
end -- if here then name has 3 or more uppercase letters so treat them as a word


local initials, names = {}, {}; -- tables to hold name parts and initials
local initials, names = {}, {}; -- tables to hold name parts and initials
Line 1,141: Line 1,101:
end
end
if 3 > i then
if 3 > i then
table.insert (initials, mw.ustring.sub(names[i],1,1)); -- insert the intial at end of initials table
table.insert (initials, mw.ustring.sub(names[i], 1, 1)); -- insert the initial at end of initials table
end
end
i = i+1; -- bump the counter
i = i + 1; -- bump the counter
end
end
Line 1,150: Line 1,110:




--[[--------------------------< L I S T _ P E O P L E >-------------------------------------------------------
--[[--------------------------< I N T E R W I K I _ P R E F I X E N _ G E T >----------------------------------
 
extract interwiki prefixen from <value>.  Returns two one or two values:
false – no prefixen
nil – prefix exists but not recognized
project prefix, language prefix – when value has either of:
:<project>:<language>:<article>
:<language>:<project>:<article>
project prefix, nil – when <value> has only a known single-letter prefix
nil, language prefix – when <value> has only a known language prefix


Formats a list of people (e.g. authors / editors)
accepts single-letter project prefixen: 'd' (wikidata), 's' (wikisource), and 'w' (wikipedia) prefixes; at this
writing, the other single-letter prefixen (b (wikibook), c (commons), m (meta), n (wikinews), q (wikiquote), and
v (wikiversity)) are not supported.


]]
]]


local function list_people(control, people, etal)
local function interwiki_prefixen_get (value, is_link)
local sep;
if not value:find (':%l+:') then -- if no prefix
local namesep;
return false; -- abandon; boolean here to distinguish from nil fail returns later
local format = control.format
end
local maximum = control.maximum
local lastauthoramp = control.lastauthoramp;
local text = {}


if 'vanc' == format then -- Vancouver-like author/editor name styling?
local prefix_patterns_linked_t = { -- sequence of valid interwiki and inter project prefixen
sep = cfg.presentation['sep_nl_vanc']; -- name-list separator between authors is a comma
'^%[%[:([dsw]):(%l%l+):', -- wikilinked; project and language prefixes
'^%[%[:(%l%l+):([dsw]):', -- wikilinked; language and project prefixes
'^%[%[:([dsw]):', -- wikilinked; project prefix
'^%[%[:(%l%l+):', -- wikilinked; language prefix
}
local prefix_patterns_unlinked_t = { -- sequence of valid interwiki and inter project prefixen
'^:([dsw]):(%l%l+):', -- project and language prefixes
'^:(%l%l+):([dsw]):', -- language and project prefixes
'^:([dsw]):', -- project prefix
'^:(%l%l+):', -- language prefix
}
local cap1, cap2;
for _, pattern in ipairs ((is_link and prefix_patterns_linked_t) or prefix_patterns_unlinked_t) do
cap1, cap2 = value:match (pattern);
if cap1 then
break; -- found a match so stop looking
end
end
if cap1 and cap2 then -- when both then :project:language: or :language:project: (both forms allowed)
if 1 == #cap1 then -- length == 1 then :project:language:
if cfg.inter_wiki_map[cap2] then -- is language prefix in the interwiki map?
return cap1, cap2; -- return interwiki project and interwiki language
end
else -- here when :language:project:
if cfg.inter_wiki_map[cap1] then -- is language prefix in the interwiki map?
return cap2, cap1; -- return interwiki project and interwiki language
end
end
return nil; -- unknown interwiki language
elseif not (cap1 or cap2) then -- both are nil?
return nil; -- we got something that looks like a project prefix but isn't; return fail
elseif 1 == #cap1 then -- here when one capture
return cap1, nil; -- length is 1 so return project, nil language
else -- here when one capture and its length it more than 1
if cfg.inter_wiki_map[cap1] then -- is language prefix in the interwiki map?
return nil, cap1; -- return nil project, language
end
end
end
 
 
--[[--------------------------< L I S T _ P E O P L E >--------------------------
 
Formats a list of people (authors, contributors, editors, interviewers, translators)
 
names in the list will be linked when
|<name>-link= has a value
|<name>-mask- does NOT have a value; masked names are presumed to have been
rendered previously so should have been linked there
 
when |<name>-mask=0, the associated name is not rendered
 
]]
 
local function list_people (control, people, etal)
local sep;
local namesep;
local format = control.format;
local maximum = control.maximum;
local name_list = {};
 
if 'vanc' == format then -- Vancouver-like name styling?
sep = cfg.presentation['sep_nl_vanc']; -- name-list separator between names is a comma
namesep = cfg.presentation['sep_name_vanc']; -- last/first separator is a space
namesep = cfg.presentation['sep_name_vanc']; -- last/first separator is a space
else
else
sep = cfg.presentation['sep_nl']; -- name-list separator between authors is a semicolon
sep = cfg.presentation['sep_nl']; -- name-list separator between names is a semicolon
namesep = cfg.presentation['sep_name']; -- last/first separator is <comma><space>
namesep = cfg.presentation['sep_name']; -- last/first separator is <comma><space>
end
end
if sep:sub(-1,-1) ~= " " then sep = sep .. " " end
if sep:sub (-1, -1) ~= " " then sep = sep .. " " end
if is_set (maximum) and maximum < 1 then return "", 0; end -- returned 0 is for EditorCount; not used for authors
if utilities.is_set (maximum) and maximum < 1 then return "", 0; end -- returned 0 is for EditorCount; not used for other names
for i,person in ipairs(people) do
for i, person in ipairs (people) do
if is_set(person.last) then
if utilities.is_set (person.last) then
local mask = person.mask
local mask = person.mask;
local one
local one;
local sep_one = sep;
local sep_one = sep;
if is_set (maximum) and i > maximum then
 
if utilities.is_set (maximum) and i > maximum then
etal = true;
etal = true;
break;
break;
elseif (mask ~= nil) then
end
local n = tonumber(mask)
if (n ~= nil) then
if mask then
one = string.rep("&mdash;",n)
local n = tonumber (mask); -- convert to a number if it can be converted; nil else
if n then
one = 0 ~= n and string.rep("&mdash;", n) or nil; -- make a string of (n > 0) mdashes, nil else, to replace name
person.link = nil; -- don't create link to name if name is replaces with mdash string or has been set nil
else
else
one = mask;
one = mask; -- replace name with mask text (must include name-list separator)
sep_one = " ";
sep_one = " "; -- modify name-list separator
end
end
else
else
one = person.last
one = person.last; -- get surname
local first = person.first
local first = person.first -- get given name
if is_set(first) then
if utilities.is_set (first) then
if ( "vanc" == format ) then -- if vancouver format
if ("vanc" == format) then -- if Vancouver format
one = one:gsub ('%.', ''); -- remove periods from surnames (http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35029/)
one = one:gsub ('%.', ''); -- remove periods from surnames (http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35029/)
if not person.corporate and is_good_vanc_name (one, first) then -- and name is all Latin characters; corporate authors not tested
if not person.corporate and is_good_vanc_name (one, first, nil, i) then -- and name is all Latin characters; corporate authors not tested
first = reduce_to_initials(first) -- attempt to convert first name(s) to initials
first = reduce_to_initials (first, i); -- attempt to convert first name(s) to initials
end
end
end
end
one = one .. namesep .. first;
one = one .. namesep .. first;
end
end
if is_set(person.link) and person.link ~= control.page_name then
one = make_wikilink (person.link, one); -- link author/editor if this page is not the author's/editor's page
end
end
end
table.insert( text, one )
if utilities.is_set (person.link) then
table.insert( text, sep_one )
one = utilities.make_wikilink (person.link, one); -- link author/editor
end
end
end
 
if one then -- if <one> has a value (name, mdash replacement, or mask text replacement)
local proj, tag = interwiki_prefixen_get (one, true); -- get the interwiki prefixen if present


local count = #text / 2; -- (number of names + number of separators) divided by 2
if 'w' == proj and ('Wikipedia' == mw.site.namespaces.Project['name']) then
if count > 0 then  
proj = nil; -- for stuff like :w:de:<article>, :w is unnecessary TODO: maint cat?
if count > 1 and is_set(lastauthoramp) and not etal then
end
text[#text-2] = " & "; -- replace last separator with ampersand text
if proj then
end
proj = ({['d'] = 'Wikidata', ['s'] = 'Wikisource', ['w'] = 'Wikipedia'})[proj]; -- :w (wikipedia) for linking from a non-wikipedia project
text[#text] = nil; -- erase the last separator
if proj then  
one = one .. utilities.wrap_style ('interproj', proj); -- add resized leading space, brackets, static text, language name
tag = nil; -- unset; don't do both project and language
end
end
if tag == cfg.this_wiki_code then
tag = nil; -- stuff like :en:<article> at en.wiki is pointless TODO: maint cat?
end
if tag then
local lang = cfg.lang_tag_remap[tag] or cfg.mw_languages_by_tag_t[tag];
if lang then -- error messaging done in extract_names() where we know parameter names
one = one .. utilities.wrap_style ('interwiki', lang); -- add resized leading space, brackets, static text, language name
end
end
 
table.insert (name_list, one); -- add it to the list of names
table.insert (name_list, sep_one); -- add the proper name-list separator
end
end
end
end
 
local result = table.concat(text) -- construct list
local count = #name_list / 2; -- (number of names + number of separators) divided by 2
if etal and is_set (result) then -- etal may be set by |display-authors=etal but we might not have a last-first list
if 0 < count then
result = result .. sep .. ' ' .. cfg.messages['et al']; -- we've go a last-first list and etal so add et al.
if 1 < count and not etal then
if 'amp' == format then
name_list[#name_list-2] = " & "; -- replace last separator with ampersand text
elseif 'and' == format then
if 2 == count then
name_list[#name_list-2] = cfg.presentation.sep_nl_and; -- replace last separator with 'and' text
else
name_list[#name_list-2] = cfg.presentation.sep_nl_end; -- replace last separator with '(sep) and' text
end
end
end
name_list[#name_list] = nil; -- erase the last separator
end
 
local result = table.concat (name_list); -- construct list
if etal and utilities.is_set (result) then -- etal may be set by |display-authors=etal but we might not have a last-first list
result = result .. sep .. cfg.messages['et al']; -- we've got a last-first list and etal so add et al.
end
end
return result, count
return result, count; -- return name-list string and count of number of names (count used for editor names only)
end
end




--[[--------------------------< A N C H O R _ I D >------------------------------------------------------------
--[[--------------------< M A K E _ C I T E R E F _ I D >-----------------------


Generates a CITEREF anchor ID if we have at least one name or a date.  Otherwise returns an empty string.
Generates a CITEREF anchor ID if we have at least one name or a date.  Otherwise
returns an empty string.


namelist is one of the contributor-, author-, or editor-name lists chosen in that order.  year is Year or anchor_year.
namelist is one of the contributor-, author-, or editor-name lists chosen in that
order.  year is Year or anchor_year.


]]
]]
local function anchor_id (namelist, year)
 
local names={}; -- a table for the one to four names and year
local function make_citeref_id (namelist, year)
for i,v in ipairs (namelist) do -- loop through the list and take up to the first four last names
local names={}; -- a table for the one to four names and year
names[i] = v.last  
for i,v in ipairs (namelist) do -- loop through the list and take up to the first four last names
if i == 4 then break end -- if four then done
names[i] = v.last
if i == 4 then break end -- if four then done
end
end
table.insert (names, year); -- add the year at the end
table.insert (names, year); -- add the year at the end
local id = table.concat(names); -- concatenate names and year for CITEREF id
local id = table.concat(names); -- concatenate names and year for CITEREF id
if is_set (id) then -- if concatenation is not an empty string
if utilities.is_set (id) then -- if concatenation is not an empty string
return "CITEREF" .. id; -- add the CITEREF portion
return "CITEREF" .. id; -- add the CITEREF portion
else
else
return ''; -- return an empty string; no reason to include CITEREF id in this citation
return ''; -- return an empty string; no reason to include CITEREF id in this citation
end
end
end
end




--[[--------------------------< N A M E _ H A S _ E T A L >----------------------------------------------------
--[[--------------------------< C I T E _ C L A S S _A T T R I B U T E _M A K E >------------------------------


Evaluates the content of name parameters (author, editor, etc) for variations on the theme of et al.  If found,
construct <cite> tag class attribute for this citation.
the et al. is removed, a flag is set to true and the function returns the modified name and the flag.


This function never sets the flag to false but returns it's previous state because it may have been set by
<cite_class> – config.CitationClass from calling template
previous passes through this function or by the associated |display-<names>=etal parameter
<mode> – value from |mode= parameter
 
]]
 
local function cite_class_attribute_make (cite_class, mode)
local class_t = {};
table.insert (class_t, 'citation'); -- required for blue highlight
if 'citation' ~= cite_class then
table.insert (class_t, cite_class); -- identify this template for user css
table.insert (class_t, utilities.is_set (mode) and mode or 'cs1'); -- identify the citation style for user css or javascript
else
table.insert (class_t, utilities.is_set (mode) and mode or 'cs2'); -- identify the citation style for user css or javascript
end
for _, prop_key in ipairs (z.prop_keys_t) do
table.insert (class_t, prop_key); -- identify various properties for user css or javascript
end
 
return table.concat (class_t, ' '); -- make a big string and done
end
 
 
--[[---------------------< N A M E _ H A S _ E T A L >--------------------------
 
Evaluates the content of name parameters (author, editor, etc.) for variations on
the theme of et al.  If found, the et al. is removed, a flag is set to true and
the function returns the modified name and the flag.
 
This function never sets the flag to false but returns its previous state because
it may have been set by previous passes through this function or by the associated
|display-<names>=etal parameter


]]
]]
Line 1,264: Line 1,365:
local function name_has_etal (name, etal, nocat, param)
local function name_has_etal (name, etal, nocat, param)


if is_set (name) then -- name can be nil in which case just return
if utilities.is_set (name) then -- name can be nil in which case just return
local patterns = cfg.et_al_patterns; --get patterns from configuration
local patterns = cfg.et_al_patterns; -- get patterns from configuration
for _, pattern in ipairs (patterns) do -- loop through all of the patterns
for _, pattern in ipairs (patterns) do -- loop through all of the patterns
Line 1,272: Line 1,373:
etal = true; -- set flag (may have been set previously here or by |display-<names>=etal)
etal = true; -- set flag (may have been set previously here or by |display-<names>=etal)
if not nocat then -- no categorization for |vauthors=
if not nocat then -- no categorization for |vauthors=
table.insert( z.message_tail, {set_error ('etal', {param})}); -- and set an error if not added
utilities.set_message ('err_etal', {param}); -- and set an error if not added
end
end
end
end
Line 1,278: Line 1,379:
end
end


return name, etal; --
return name, etal;
end
end




--[[--------------------------< N A M E _ H A S _ E D _ M A R K U P >------------------------------------------
--[[---------------------< N A M E _ I S _ N U M E R I C >----------------------
 
Add an error message and category when <name> parameter value does not contain letters. 
 
Add a maintenance category when <name> parameter value has numeric characters mixed with characters that are
not numeric characters; could be letters and/or punctuation characters.
 
This function will only emit one error and one maint message for the current template.  Does not emit both error
and maint messages/categories for the same parameter value.


Evaluates the content of author and editor parameters for extranious editor annotations: ed, ed., eds, (Ed.), etc.
returns nothing
These annotation do not belong in author parameters and are redundant in editor parameters.  If found, the function
adds the editor markup maintenance category.


]]
]]


local function name_has_ed_markup (name, list_name)
local function name_is_numeric (name, name_alias, list_name)
local _, pattern;
local patterns = {
local patterns = cfg.editor_markup_patterns; -- get patterns from configuration
'^%D+%d', -- <name> must have digits preceded by other characters
'^%D*%d+%D+', -- <name> must have digits followed by other characters
}


if is_set (name) then
if not added_numeric_name_errs and mw.ustring.match (name, '^[%A]+$') then -- if we have not already set an error message and <name> does not have any alpha characters
for _, pattern in ipairs (patterns) do -- spin through patterns table and
utilities.set_message ('err_numeric_names', name_alias); -- add an error message
if name:match (pattern) then
added_numeric_name_errs = true; -- set the flag so we emit only one error message
add_maint_cat ('extra_text_names', cfg.special_case_translation [list_name]); -- add a maint cat for this template
return; -- when here no point in further testing; abandon
break;
end
 
if not added_numeric_name_maint then -- if we have already set a maint message
for _, pattern in ipairs (patterns) do -- spin through list of patterns
if mw.ustring.match (name, pattern) then -- digits preceded or followed by anything but digits; %D+ includes punctuation
utilities.set_message ('maint_numeric_names', cfg.special_case_translation [list_name]); -- add a maint cat for this template
added_numeric_name_maint = true; -- set the flag so we emit only one maint message
return; -- when here no point in further testing; abandon
end
end
end
end
end
end
return name; -- and done
end
end




--[[--------------------------< N A M E _ H A S _ M U L T _ N A M E S >----------------------------------------
--[[-----------------< N A M E _ H A S _ M U L T _ N A M E S >------------------


Evaluates the content of author and editor (surnames only) parameters for multiple names. Multiple names are
Evaluates the content of last/surname (authors etc.) parameters for multiple names.
indicated if there is more than one comma and or semicolon. If found, the function adds the multiple name
Multiple names are indicated if there is more than one comma or any "unescaped"
(author or editor) maintenance category.
semicolons. Escaped semicolons are ones used as part of selected HTML entities.
If the condition is met, the function adds the multiple name maintenance category.
 
Same test for first except that commas should not appear in given names (MOS:JR says
that the generational suffix does not take a separator character). Titles, degrees,
postnominals, affiliations, all normally comma separated don't belong in a citation.
 
<name> – name parameter value
<list_name> – AuthorList, EditorList, etc
<limit> – number of allowed commas; 1 (default) for surnames; 0 for given names
 
returns nothing


]]
]]


local function name_has_mult_names (name, list_name)
local function name_has_mult_names (name, list_name, limit)
local count, _;
local _, commas, semicolons, nbsps;
if is_set (name) then
limit = limit and limit or 1;
_, count = name:gsub ('[;,]', ''); -- count the number of separator-like characters
if utilities.is_set (name) then
_, commas = name:gsub (',', ''); -- count the number of commas
_, semicolons = name:gsub (';', ''); -- count the number of semicolons
-- nbsps probably should be its own separate count rather than merged in
-- some way with semicolons because Lua patterns do not support the
-- grouping operator that regex does, which means there is no way to add
-- more entities to escape except by adding more counts with the new
-- entities
_, nbsps = name:gsub ('&nbsp;',''); -- count nbsps
if 1 < count then -- param could be |author= or |editor= so one separator character is acceptable
-- There is exactly 1 semicolon per &nbsp; entity, so subtract nbsps
add_maint_cat ('mult_names', cfg.special_case_translation [list_name]); -- more than one separator indicates multiple names so add a maint cat for this template
-- from semicolons to 'escape' them. If additional entities are added,
-- they also can be subtracted.
if limit < commas or 0 < (semicolons - nbsps) then
utilities.set_message ('maint_mult_names', cfg.special_case_translation [list_name]); -- add a maint message
end
end
end
end
return name; -- and done
end
end




--[[--------------------------< N A M E _ C H E C K S >--------------------------------------------------------
--[=[-------------------------< I S _ G E N E R I C >----------------------------------------------------------
This function calls various name checking functions used to validate the content of the various name-holding
 
parameters.
Compares values assigned to various parameters according to the string provided as <item> in the function call.
<item> can have on of two values:
'generic_names' – for name-holding parameters: |last=, |first=, |editor-last=, etc
'generic_titles' – for |title=
 
There are two types of generic tests.  The 'accept' tests look for a pattern that should not be rejected by the
'reject' test.  For example,
|author=[[John Smith (author)|Smith, John]]
would be rejected by the 'author' reject test.  But piped wikilinks with 'author' disambiguation should not be
rejected so the 'accept' test prevents that from happening.  Accept tests are always performed before reject
tests.
 
Each of the 'accept' and 'reject' sequence tables hold tables for en.wiki (['en']) and local.wiki (['local'])
that each can hold a test sequence table  The sequence table holds, at index [1], a test pattern, and, at index
[2], a boolean control value.  The control value tells string.find() or mw.ustring.find() to do plain-text search (true)
or a pattern search (false).  The intent of all this complexity is to make these searches as fast as possible so
that we don't run out of processing time on very large articles.
 
Returns
true when a reject test finds the pattern or string
false when an accept test finds the pattern or string
nil else
 
]=]


]]
local function is_generic (item, value, wiki)
local test_val;
local str_lower = { -- use string.lower() for en.wiki (['en']) and use mw.ustring.lower() or local.wiki (['local'])
['en'] = string.lower,
['local'] = mw.ustring.lower,
}
local str_find = { -- use string.find() for en.wiki (['en']) and use mw.ustring.find() or local.wiki (['local'])
['en'] = string.find,
['local'] = mw.ustring.find,
}


local function name_checks (last, first, list_name)
local function test (val, test_t, wiki) -- local function to do the testing; <wiki> selects lower() and find() functions
if is_set (last) then
val = test_t[2] and str_lower[wiki](value) or val; -- when <test_t[2]> set to 'true', plaintext search using lowercase value
if last:match ('^%(%(.*%)%)$') then -- if wrapped in doubled parentheses, accept as written
return str_find[wiki] (val, test_t[1], 1, test_t[2]); -- return nil when not found or matched
last = last:match ('^%(%((.*)%)%)$'); -- strip parens
else
last = name_has_mult_names (last, list_name); -- check for multiple names in the parameter (last only)
last = name_has_ed_markup (last, list_name); -- check for extraneous 'editor' annotation
end
end
end
if is_set (first) then
if first:match ('^%(%(.*%)%)$') then -- if wrapped in doubled parentheses, accept as written
local test_types_t = {'accept', 'reject'}; -- test accept patterns first, then reject patterns
first = first:match ('^%(%((.*)%)%)$'); -- strip parens
local wikis_t = {'en', 'local'}; -- do tests for each of these keys; en.wiki first, local.wiki second
else
 
first = name_has_ed_markup (first, list_name); -- check for extraneous 'editor' annotation
for _, test_type in ipairs (test_types_t) do -- for each test type
for _, generic_value in pairs (cfg.special_case_translation[item][test_type]) do -- spin through the list of generic value fragments to accept or reject
for _, wiki in ipairs (wikis_t) do
if generic_value[wiki] then
if test (value, generic_value[wiki], wiki) then -- go do the test
return ('reject' == test_type); -- param value rejected, return true; false else
end
end
end
end
end
end
end
return last, first; -- done
end
end




--[[--------------------------< E X T R A C T _ N A M E S >----------------------------------------------------
--[[--------------------------< N A M E _ I S _ G E N E R I C >------------------------------------------------
Gets name list from the input arguments


Searches through args in sequential order to find |lastn= and |firstn= parameters (or their aliases), and their matching link and mask parameters.
calls is_generic() to determine if <name> is a 'generic name' listed in cfg.generic_names; <name_alias> is the
Stops searching when both |lastn= and |firstn= are not found in args after two sequential attempts: found |last1=, |last2=, and |last3= but doesn't
parameter name used in error messaging
find |last4= and |last5= then the search is done.


This function emits an error message when there is a |firstn= without a matching |lastn=.  When there are 'holes' in the list of last names, |last1= and |last3=
]]
are present but |last2= is missing, an error message is emitted. |lastn= is not required to have a matching |firstn=.


When an author or editor parameter contains some form of 'et al.', the 'et al.' is stripped from the parameter and a flag (etal) returned
local function name_is_generic (name, name_alias)
that will cause list_people() to add the static 'et al.' text from Module:Citation/CS1/Configuration. This keeps 'et al.' out of the
if not added_generic_name_errs  and is_generic ('generic_names', name) then
template's metadata.  When this occurs, the page is added to a maintenance category.
utilities.set_message ('err_generic_name', name_alias); -- set an error message
added_generic_name_errs = true;
end
end


]]


local function extract_names(args, list_name)
--[[--------------------------< N A M E _ C H E C K S >--------------------------------------------------------
local names = {}; -- table of names
 
local last; -- individual name components
This function calls various name checking functions used to validate the content of the various name-holding parameters.
local first;
 
local link;
]]
local mask;
local i = 1; -- loop counter/indexer
local n = 1; -- output table indexer
local count = 0; -- used to count the number of times we haven't found a |last= (or alias for authors, |editor-last or alias for editors)
local etal=false; -- return value set to true when we find some form of et al. in an author parameter


local last_alias, first_alias; -- selected parameter aliases used in error messaging
local function name_checks (last, first, list_name, last_alias, first_alias)
while true do
local accept_name;
last, last_alias = select_one( args, cfg.aliases[list_name .. '-Last'], 'redundant_parameters', i ); -- search through args for name components beginning at 1
first, first_alias = select_one( args, cfg.aliases[list_name .. '-First'], 'redundant_parameters', i );
link = select_one( args, cfg.aliases[list_name .. '-Link'], 'redundant_parameters', i );
mask = select_one( args, cfg.aliases[list_name .. '-Mask'], 'redundant_parameters', i );


last, etal = name_has_etal (last, etal, false, last_alias); -- find and remove variations on et al.
if utilities.is_set (last) then
first, etal = name_has_etal (first, etal, false, first_alias); -- find and remove variations on et al.
last, accept_name = utilities.has_accept_as_written (last); -- remove accept-this-as-written markup when it wraps all of <last>
last, first= name_checks (last, first, list_name); -- multiple names, extraneous annotation, etc checks
if first and not last then -- if there is a firstn without a matching lastn
table.insert( z.message_tail, { set_error( 'first_missing_last', {first_alias, first_alias:gsub('first', 'last')}, true ) } ); -- add this error message
elseif not first and not last then -- if both firstn and lastn aren't found, are we done?
count = count + 1; -- number of times we haven't found last and first
if 2 <= count then -- two missing names and we give up
break; -- normal exit or there is a two-name hole in the list; can't tell which
end
else -- we have last with or without a first
link_title_ok (link, list_name:match ("(%w+)List"):lower() .. '-link' .. i, last, list_name:match ("(%w+)List"):lower() .. '-last' .. i); -- check for improper wikimarkup


names[n] = {last = last, first = first, link = link, mask = mask, corporate=false}; -- add this name to our names list (corporate for |vauthors= only)
if not accept_name then -- <last> not wrapped in accept-as-written markup
n = n + 1; -- point to next location in the names table
name_has_mult_names (last, list_name); -- check for multiple names in the parameter
if 1 == count then -- if the previous name was missing
name_is_numeric (last, last_alias, list_name); -- check for names that have no letters or are a mix of digits and other characters
table.insert( z.message_tail, { set_error( 'missing_name', {list_name:match ("(%w+)List"):lower(), i-1}, true ) } ); -- add this error message
name_is_generic (last, last_alias); -- check for names found in the generic names list
end
count = 0; -- reset the counter, we're looking for two consecutive missing names
end
end
i = i + 1; -- point to next args location
end
end
return names, etal; -- all done, return our list of names
end


if utilities.is_set (first) then
first, accept_name = utilities.has_accept_as_written (first); -- remove accept-this-as-written markup when it wraps all of <first>
if not accept_name then -- <first> not wrapped in accept-as-written markup
name_has_mult_names (first, list_name, 0); -- check for multiple names in the parameter; 0 is number of allowed commas in a given name
name_is_numeric (first, first_alias, list_name); -- check for names that have no letters or are a mix of digits and other characters
name_is_generic (first, first_alias); -- check for names found in the generic names list
end
local wl_type, D = utilities.is_wikilink (first);
if 0 ~= wl_type then
first = D;
utilities.set_message ('err_bad_paramlink', first_alias);
end
end


--[[--------------------------< G E T _ I S O 6 3 9 _ C O D E >------------------------------------------------
return last, first; -- done
end


Validates language names provided in |language= parameter if not an ISO639-1 or 639-2 code.


Returns the language name and associated two- or three-character code.  Because case of the source may be incorrect
--[[----------------------< E X T R A C T _ N A M E S >-------------------------
or different from the case that WikiMedia uses, the name comparisons are done in lower case and when a match is
found, the Wikimedia version (assumed to be correct) is returned along with the code.  When there is no match, we
return the original language name string.


mw.language.fetchLanguageNames(<local wiki language>, 'all') returns a list of languages that in some cases may include
Gets name list from the input arguments
extensions. For example, code 'cbk-zam' and its associated name 'Chavacano de Zamboanga' (MediaWiki does not support
code 'cbk' or name 'Chavacano'.  Most (all?) of these languages are not used a 'language' codes per se, rather they
are used as sub-domain names: cbk-zam.wikipedia.org.  A list of language names and codes supported by fetchLanguageNames()
can be found at Template:Citation Style documentation/language/doc


Names that are included in the list will be found if that name is provided in the |language= parameter.  For example,
Searches through args in sequential order to find |lastn= and |firstn= parameters
if |language=Chavacano de Zamboanga, that name will be found with the associated code 'cbk-zam'.  When names are found
(or their aliases), and their matching link and mask parameters. Stops searching
and the associated code is not two or three characters, this function returns only the WikiMedia language name.
when both |lastn= and |firstn= are not found in args after two sequential attempts:
found |last1=, |last2=, and |last3= but doesn't find |last4= and |last5= then the
search is done.


Some language names have multiple entries under different codes:
This function emits an error message when there is a |firstn= without a matching
Aromanian has code rup and code roa-rup
|lastn=.  When there are 'holes' in the list of last names, |last1= and |last3=
When this occurs, this function returns the language name and the 2- or 3-character code
are present but |last2= is missing, an error message is emitted. |lastn= is not
required to have a matching |firstn=.


Adapted from code taken from Module:Check ISO 639-1.
When an author or editor parameter contains some form of 'et al.', the 'et al.'
is stripped from the parameter and a flag (etal) returned that will cause list_people()
to add the static 'et al.' text from Module:Citation/CS1/Configuration.  This keeps
'et al.' out of the template's metadata.  When this occurs, an error is emitted.


]]
]]


local function get_iso639_code (lang, this_wiki_code)
local function extract_names(args, list_name)
if cfg.lang_name_remap[lang:lower()] then -- if there is a remapped name (because MediaWiki uses something that we don't think is correct)
local names = {}; -- table of names
return cfg.lang_name_remap[lang:lower()][1], cfg.lang_name_remap[lang:lower()][2]; -- for this language 'name', return a possibly new name and appropriate code
local last; -- individual name components
end
local first;
local link;
local mask;
local i = 1; -- loop counter/indexer
local n = 1; -- output table indexer
local count = 0; -- used to count the number of times we haven't found a |last= (or alias for authors, |editor-last or alias for editors)
local etal = false; -- return value set to true when we find some form of et al. in an author parameter


local ietf_code; -- because some languages have both ietf-like codes and iso 639-like codes
local last_alias, first_alias, link_alias; -- selected parameter aliases used in error messaging
local ietf_name;
while true do
last, last_alias = utilities.select_one ( args, cfg.aliases[list_name .. '-Last'], 'err_redundant_parameters', i ); -- search through args for name components beginning at 1
first, first_alias = utilities.select_one ( args, cfg.aliases[list_name .. '-First'], 'err_redundant_parameters', i );
link, link_alias = utilities.select_one ( args, cfg.aliases[list_name .. '-Link'], 'err_redundant_parameters', i );
mask = utilities.select_one ( args, cfg.aliases[list_name .. '-Mask'], 'err_redundant_parameters', i );
local languages = mw.language.fetchLanguageNames(this_wiki_code, 'all') -- get a list of language names known to Wikimedia
if last then -- error check |lastn= alias for unknown interwiki link prefix; done here because this is where we have the parameter name
-- ('all' is required for North Ndebele, South Ndebele, and Ojibwa)
local project, language = interwiki_prefixen_get (last, true); -- true because we expect interwiki links in |lastn= to be wikilinked
local langlc = mw.ustring.lower(lang); -- lower case version for comparisons
if nil == project and nil == language then -- when both are nil
utilities.set_message ('err_bad_paramlink', last_alias); -- not known, emit an error message -- TODO: err_bad_interwiki?
last = utilities.remove_wiki_link (last); -- remove wikilink markup; show display value only
end
end
if link then -- error check |linkn= alias for unknown interwiki link prefix
local project, language = interwiki_prefixen_get (link, false); -- false because wiki links in |author-linkn= is an error
if nil == project and nil == language then -- when both are nil
utilities.set_message ('err_bad_paramlink', link_alias); -- not known, emit an error message -- TODO: err_bad_interwiki?
link = nil; -- unset so we don't link
link_alias = nil;
end
end
last, etal = name_has_etal (last, etal, false, last_alias); -- find and remove variations on et al.
first, etal = name_has_etal (first, etal, false, first_alias); -- find and remove variations on et al.
last, first = name_checks (last, first, list_name, last_alias, first_alias); -- multiple names, extraneous annotation, etc. checks


for code, name in pairs(languages) do -- scan the list to see if we can find our language
if first and not last then -- if there is a firstn without a matching lastn
if langlc == mw.ustring.lower(name) then
local alias = first_alias:find ('given', 1, true) and 'given' or 'first'; -- get first or given form of the alias
if 2 == code:len() or 3 == code:len() then -- two- or three-character codes only; extensions not supported
utilities.set_message ('err_first_missing_last', {
return name, code; -- so return the name and the code
first_alias, -- param name of alias missing its mate
first_alias:gsub (alias, {['first'] = 'last', ['given'] = 'surname'}), -- make param name appropriate to the alias form
}); -- add this error message
elseif not first and not last then -- if both firstn and lastn aren't found, are we done?
count = count + 1; -- number of times we haven't found last and first
if 2 <= count then -- two missing names and we give up
break; -- normal exit or there is a two-name hole in the list; can't tell which
end
else -- we have last with or without a first
local result;
link = link_title_ok (link, link_alias, last, last_alias); -- check for improper wiki-markup
 
if first then
link = link_title_ok (link, link_alias, first, first_alias); -- check for improper wiki-markup
end
 
names[n] = {last = last, first = first, link = link, mask = mask, corporate = false}; -- add this name to our names list (corporate for |vauthors= only)
n = n + 1; -- point to next location in the names table
if 1 == count then -- if the previous name was missing
utilities.set_message ('err_missing_name', {list_name:match ("(%w+)List"):lower(), i - 1}); -- add this error message
end
end
ietf_code = code; -- remember that we found an ietf-like code and save its name
count = 0; -- reset the counter, we're looking for two consecutive missing names
ietf_name = name; -- but keep looking for a 2- or 3-char code
end
end
i = i + 1; -- point to next args location
end
end
-- didn't find name with 2- or 3-char code; if ietf-like code found return
return ietf_code and ietf_name or lang; -- associated name; return original language text else
return names, etal; -- all done, return our list of names and the etal flag
end
end




--[[--------------------------< L A N G U A G E _ P A R A M E T E R >------------------------------------------
--[[--------------------------< N A M E _ T A G _ G E T >------------------------------------------------------


Gets language name from a provided two- or three-character ISO 639 code.  If a code is recognized by MediaWiki,
attempt to decode |language=<lang_param> and return language name and matching tag; nil else.
use the returned name; if not, then use the value that was provided with the language parameter.


When |language= contains a recognized language (either code or name), the page is assigned to the category for
This function looks for:
that code: Category:Norwegian-language sources (no). For valid three-character code languages, the page is assigned
<lang_param> as a tag in cfg.lang_tag_remap{}
to the single category for '639-2' codes: Category:CS1 ISO 639-2 language sources.
<lang_param> as a name in cfg.lang_name_remap{}
 
Languages that are the same as the local wiki are not categorized. MediaWiki does not recognize three-character
<lang_param> as a name in cfg.mw_languages_by_name_t
equivalents of two-character codes: code 'ar' is recognized bit code 'ara' is not.
<lang_param> as a tag in cfg.mw_languages_by_tag_t
when those fail, presume that <lang_param> is an IETF-like tag that MediaWiki does not recognize.  Strip all
script, region, variant, whatever subtags from <lang_param> to leave just a two or three character language tag
and look for the new <lang_param> in cfg.mw_languages_by_tag_t{}


This function supports multiple languages in the form |language=nb, French, th where the language names or codes are
on success, returns name (in properly capitalized form) and matching tag (in lowercase); on failure returns nil
separated from each other by commas.


]]
]]


local function language_parameter (lang)
local function name_tag_get (lang_param)
local code; -- the two- or three-character language code
local lang_param_lc = mw.ustring.lower (lang_param); -- use lowercase as an index into the various tables
local name; -- the language name
local name;
local language_list = {}; -- table of language names to be rendered
local tag;
local names_table = {}; -- table made from the value assigned to |language=


local this_wiki = mw.getContentLanguage(); -- get a language object for this wiki
name = cfg.lang_tag_remap[lang_param_lc]; -- assume <lang_param_lc> is a tag; attempt to get remapped language name
local this_wiki_code = this_wiki:getCode() -- get this wiki's language code
if name then -- when <name>, <lang_param> is a tag for a remapped language name
local this_wiki_name = mw.language.fetchLanguageName(this_wiki_code, this_wiki_code); -- get this wiki's language name
if cfg.lang_name_remap[name:lower()][2] ~= lang_param_lc then
utilities.set_message ('maint_unknown_lang'); -- add maint category if not already added
return name, cfg.lang_name_remap[name:lower()][2]; -- so return name and tag from lang_name_remap[name]; special case to xlate sr-ec and sr-el to sr-cyrl and sr-latn
end
return name, lang_param_lc; -- so return <name> from remap and <lang_param_lc>
end


names_table = mw.text.split (lang, '%s*,%s*'); -- names should be a comma separated list
tag = lang_param_lc:match ('^(%a%a%a?)%-.*'); -- still assuming that <lang_param_lc> is a tag; strip script, region, variant subtags
name = cfg.lang_tag_remap[tag]; -- attempt to get remapped language name with language subtag only
if name then -- when <name>, <tag> is a tag for a remapped language name
return name, tag; -- so return <name> from remap and <tag>
end


for _, lang in ipairs (names_table) do -- reuse lang
if cfg.lang_name_remap[lang_param_lc] then -- not a remapped tag, assume <lang_param_lc> is a name; attempt to get remapped language tag
name = cfg.lang_code_remap[lang:lower()]; -- first see if this is a code that is not supported by MediaWiki but is in remap
return cfg.lang_name_remap[lang_param_lc][1], cfg.lang_name_remap[lang_param_lc][2]; -- for this <lang_param_lc>, return a (possibly) new name and appropriate tag
end


if name then -- there was a remapped code so
name = cfg.mw_languages_by_tag_t[lang_param_lc]; -- assume that <lang_param_lc> is a tag; attempt to get its matching language name
lang = lang:gsub ('^(%a%a%a?)%-.*', '%1'); -- strip ietf tags from code
else
if lang:match ('^%a%a%-') then -- strip ietf tags from code; TODO: is there a need to support 3-char with tag?
lang = lang:match ('(%a%a)%-') -- keep only 639-1 code portion to lang; TODO: do something with 3166 alpha 2 country code?
end
if 2 == lang:len() or 3 == lang:len() then -- if two-or three-character code
name = mw.language.fetchLanguageName (lang:lower(), this_wiki_code); -- get language name if |language= is a proper code
end
end
 
if is_set (name) then -- if |language= specified a valid code
code = lang:lower(); -- save it
else
name, code = get_iso639_code (lang, this_wiki_code); -- attempt to get code from name (assign name here so that we are sure of proper capitalization)
end
if is_set (code) then -- only 2- or 3-character codes
if name then
name = cfg.lang_code_remap[code] or name; -- override wikimedia when they misuse language codes/names
return name, lang_param_lc; -- <lang_param_lc> is a tag so return it and <name>
 
if this_wiki_code ~= code then -- when the language is not the same as this wiki's language
if 2 == code:len() then -- and is a two-character code
add_prop_cat ('foreign_lang_source' .. code, {name, code}) -- categorize it
else -- or is a recognized language (but has a three-character code)
add_prop_cat ('foreign_lang_source_2' .. code, {code}) -- categorize it differently TODO: support mutliple three-character code categories per cs1|2 template
end
end
else
add_maint_cat ('unknown_lang'); -- add maint category if not already added
end
table.insert (language_list, name);
name = ''; -- so we can reuse it
end
end
code = #language_list -- reuse code as number of languages in the list
tag = cfg.mw_languages_by_name_t[lang_param_lc]; -- assume that <lang_param_lc> is a language name; attempt to get its matching tag
if 2 >= code then
name = table.concat (language_list, cfg.messages['parameter-pair-separator']) -- insert '<space>and<space>' between two language names
if tag then
elseif 2 < code then
return cfg.mw_languages_by_tag_t[tag], tag; -- <lang_param_lc> is a name so return the name from the table and <tag>
name = table.concat (language_list, ', '); -- and concatenate with '<comma><space>' separators
name = name:gsub (', ([^,]+)$', cfg.messages['parameter-final-separator'] .. '%1'); -- replace last '<comma><space>' separator with '<comma><space>and<space>' separator
end
end
if this_wiki_name == name then
 
return ''; -- if one language and that language is this wiki's return an empty string (no annotation)
tag = lang_param_lc:match ('^(%a%a%a?)%-.*'); -- is <lang_param_lc> an IETF-like tag that MediaWiki doesn't recognize? <tag> gets the language subtag; nil else
 
if tag then
name = cfg.mw_languages_by_tag_t[tag]; -- attempt to get a language name using the shortened <tag>
if name then
return name, tag; -- <lang_param_lc> is an unrecognized IETF-like tag so return <name> and language subtag
end
end
end
return (" " .. wrap_msg ('language', name)); -- otherwise wrap with '(in ...)'
--[[ TODO: should only return blank or name rather than full list
so we can clean up the bunched parenthetical elements Language, Type, Format
]]
end
end




--[[--------------------------< S E T _ C S 1 _ S T Y L E >----------------------------------------------------
--[[-------------------< L A N G U A G E _ P A R A M E T E R >------------------


Set style settings for CS1 citation templates. Returns separator and postscript settings
Gets language name from a provided two- or three-character ISO 639 code. If a code
At en.wiki, for cs1:
is recognized by MediaWiki, use the returned name; if not, then use the value that
ps gets: '.'
was provided with the language parameter.
sep gets: '.'


]]
When |language= contains a recognized language (either code or name), the page is
assigned to the category for that code: Category:Norwegian-language sources (no).
For valid three-character code languages, the page is assigned to the single category
for '639-2' codes: Category:CS1 ISO 639-2 language sources.


local function set_cs1_style (ps)
Languages that are the same as the local wiki are not categorized.  MediaWiki does
if not is_set (ps) then -- unless explicitely set to something
not recognize three-character equivalents of two-character codes: code 'ar' is
ps = cfg.presentation['ps_cs1']; -- terminate the rendered citation
recognized but code 'ara' is not.
end
return cfg.presentation['sep_cs1'], ps; -- element separator
end


This function supports multiple languages in the form |language=nb, French, th
where the language names or codes are separated from each other by commas with
optional space characters.


--[[--------------------------< S E T _ C S 2 _ S T Y L E >----------------------------------------------------
]]


Set style settings for CS2 citation templates. Returns separator, postscript, ref settings
local function language_parameter (lang)
At en.wiki, for cs2:
local tag; -- some form of IETF-like language tag; language subtag with optional region, sript, vatiant, etc subtags
ps gets: '' (empty string - no terminal punctuation)
local lang_subtag; -- ve populates |language= with mostly unecessary region subtags the MediaWiki does not recognize; this is the base language subtag
sep gets: ','
local name; -- the language name
local language_list = {}; -- table of language names to be rendered
local names_t = {}; -- table made from the value assigned to |language=


]]
local this_wiki_name = mw.language.fetchLanguageName (cfg.this_wiki_code, cfg.this_wiki_code); -- get this wiki's language name
 
names_t = mw.text.split (lang, '%s*,%s*'); -- names should be a comma separated list
 
for _, lang in ipairs (names_t) do -- reuse lang here because we don't yet know if lang is a language name or a language tag
name, tag = name_tag_get (lang); -- attempt to get name/tag pair for <lang>; <name> has proper capitalization; <tag> is lowercase
 
if utilities.is_set (tag) then
lang_subtag = tag:gsub ('^(%a%a%a?)%-.*', '%1'); -- for categorization, strip any IETF-like tags from language tag


local function set_cs2_style (ps, ref)
if cfg.this_wiki_code ~= lang_subtag then -- when the language is not the same as this wiki's language
if not is_set (ps) then -- if |postscript= has not been set, set cs2 default
if 2 == lang_subtag:len() then -- and is a two-character tag
ps = cfg.presentation['ps_cs2']; -- terminate the rendered citation
utilities.add_prop_cat ('foreign-lang-source', {name, tag}, lang_subtag); -- categorize it; tag appended to allow for multiple language categorization
else -- or is a recognized language (but has a three-character tag)
utilities.add_prop_cat ('foreign-lang-source-2', {lang_subtag}, lang_subtag); -- categorize it differently TODO: support multiple three-character tag categories per cs1|2 template?
end
elseif cfg.local_lang_cat_enable then -- when the language and this wiki's language are the same and categorization is enabled
utilities.add_prop_cat ('local-lang-source', {name, lang_subtag}); -- categorize it
end
else
name = lang; -- return whatever <lang> has so that we show something
utilities.set_message ('maint_unknown_lang'); -- add maint category if not already added
end
table.insert (language_list, name);
name = ''; -- so we can reuse it
end
end
if not is_set (ref) then -- if |ref= is not set
ref = "harv"; -- set default |ref=harv
name = utilities.make_sep_list (#language_list, language_list);
if (1 == #language_list) and (lang_subtag == cfg.this_wiki_code) then -- when only one language, find lang name in this wiki lang name; for |language=en-us, 'English' in 'American English'
return ''; -- if one language and that language is this wiki's return an empty string (no annotation)
end
end
return cfg.presentation['sep_cs2'], ps, ref; -- element separator
return (" " .. wrap_msg ('language', name)); -- otherwise wrap with '(in ...)'
--[[ TODO: should only return blank or name rather than full list
so we can clean up the bunched parenthetical elements Language, Type, Format
]]
end
end




--[[--------------------------< G E T _ S E T T I N G S _ F R O M _ C I T E _ C L A S S >----------------------
--[[-----------------------< S E T _ C S _ S T Y L E >--------------------------


When |mode= is not set or when its value is invalid, use config.CitationClass and parameter values to establish
Gets the default CS style configuration for the given mode.
rendered style.
Returns default separator and either postscript as passed in or the default.
In CS1, the default postscript and separator are '.'.
In CS2, the default postscript is the empty string and the default separator is ','.


]]
]]


local function get_settings_from_cite_class (ps, ref, cite_class)
local function set_cs_style (postscript, mode)
local sep;
if utilities.is_set(postscript) then
if (cite_class == "citation") then -- for citation templates (CS2)
-- emit a maintenance message if user postscript is the default cs1 postscript
sep, ps, ref = set_cs2_style (ps, ref);
-- we catch the opposite case for cs2 in set_style
else -- not a citation template so CS1
if mode == 'cs1' and postscript == cfg.presentation['ps_' .. mode] then
sep, ps = set_cs1_style (ps);
utilities.set_message ('maint_postscript');
end
else
postscript = cfg.presentation['ps_' .. mode];
end
end
 
return cfg.presentation['sep_' .. mode], postscript;
return sep, ps, ref -- return them all
end
end




--[[--------------------------< S E T _ S T Y L E >------------------------------------------------------------
--[[--------------------------< S E T _ S T Y L E >-----------------------------


Establish basic style settings to be used when rendering the citation. Uses |mode= if set and valid or uses
Sets the separator and postscript styles. Checks the |mode= first and the
config.CitationClass from the template's #invoke: to establish style.
#invoke CitationClass second. Removes the postscript if postscript == none.


]]
]]


local function set_style (mode, ps, ref, cite_class)
local function set_style (mode, postscript, cite_class)
local sep;
local sep;
if 'cs2' == mode then -- if this template is to be rendered in CS2 (citation) style
if 'cs2' == mode then
sep, ps, ref = set_cs2_style (ps, ref);
sep, postscript = set_cs_style (postscript, 'cs2');
elseif 'cs1' == mode then -- if this template is to be rendered in CS1 (cite xxx) style
elseif 'cs1' == mode then
sep, ps = set_cs1_style (ps);
sep, postscript = set_cs_style (postscript, 'cs1');
else -- anything but cs1 or cs2
elseif 'citation' == cite_class then
sep, ps, ref = get_settings_from_cite_class (ps, ref, cite_class); -- get settings based on the template's CitationClass
sep, postscript = set_cs_style (postscript, 'cs2');
else
sep, postscript = set_cs_style (postscript, 'cs1');
end
end
if 'none' == ps:lower() then -- if assigned value is 'none' then
 
ps = ''; -- set to empty string
if cfg.keywords_xlate[postscript:lower()] == 'none' then
-- emit a maintenance message if user postscript is the default cs2 postscript
-- we catch the opposite case for cs1 in set_cs_style
if 'cs2' == mode or 'citation' == cite_class then
utilities.set_message ('maint_postscript');
end
postscript = '';
end
end
return sep, ps, ref
return sep, postscript
end
end




--[=[-------------------------< I S _ P D F >------------------------------------------------------------------
--[=[-------------------------< I S _ P D F >-----------------------------------


Determines if a url has the file extension that is one of the pdf file extensions used by [[MediaWiki:Common.css]] when
Determines if a URL has the file extension that is one of the PDF file extensions
applying the pdf icon to external links.
used by [[MediaWiki:Common.css]] when applying the PDF icon to external links.


returns true if file extension is one of the recognized extensions, else false
returns true if file extension is one of the recognized extensions, else false
Line 1,648: Line 1,873:




--[[--------------------------< S T Y L E _ F O R M A T >------------------------------------------------------
--[[--------------------------< S T Y L E _ F O R M A T >-----------------------


Applies css style to |format=, |chapter-format=, etc.  Also emits an error message if the format parameter does
Applies CSS style to |format=, |chapter-format=, etc.  Also emits an error message
not have a matching url parameter.  If the format parameter is not set and the url contains a file extension that
if the format parameter does not have a matching URL parameter.  If the format parameter
is recognized as a pdf document by MediaWiki's commons.css, this code will set the format parameter to (PDF) with
is not set and the URL contains a file extension that is recognized as a PDF document
by MediaWiki's commons.css, this code will set the format parameter to (PDF) with
the appropriate styling.
the appropriate styling.


Line 1,658: Line 1,884:


local function style_format (format, url, fmt_param, url_param)
local function style_format (format, url, fmt_param, url_param)
if is_set (format) then
if utilities.is_set (format) then
format = wrap_style ('format', format); -- add leading space, parentheses, resize
format = utilities.wrap_style ('format', format); -- add leading space, parentheses, resize
if not is_set (url) then
if not utilities.is_set (url) then
format = format .. set_error( 'format_missing_url', {fmt_param, url_param} ); -- add an error message
utilities.set_message ('err_format_missing_url', {fmt_param, url_param}); -- add an error message
end
end
elseif is_pdf (url) then -- format is not set so if url is a pdf file then
elseif is_pdf (url) then -- format is not set so if URL is a PDF file then
format = wrap_style ('format', 'PDF'); -- set format to pdf
format = utilities.wrap_style ('format', 'PDF'); -- set format to PDF
else
else
format = ''; -- empty string for concatenation
format = ''; -- empty string for concatenation
Line 1,672: Line 1,898:




--[[--------------------------< G E T _ D I S P L A Y _ N A M E S >--------------------------------------------
--[[---------------------< G E T _ D I S P L A Y _ N A M E S >------------------


Returns a number that defines the number of names displayed for author and editor name lists and a boolean flag
Returns a number that defines the number of names displayed for author and editor
to indicate when et al. should be appended to the name list.
name lists and a Boolean flag to indicate when et al. should be appended to the name list.


When the value assigned to |display-xxxxors= is a number greater than or equal to zero, return the number and
When the value assigned to |display-xxxxors= is a number greater than or equal to zero,
the previous state of the 'etal' flag (false by default but may have been set to true if the name list contains
return the number and the previous state of the 'etal' flag (false by default
some variant of the text 'et al.').
but may have been set to true if the name list contains some variant of the text 'et al.').


When the value assigned to |display-xxxxors= is the keyword 'etal', return a number that is one greater than the
When the value assigned to |display-xxxxors= is the keyword 'etal', return a number
number of authors in the list and set the 'etal' flag true. This will cause the list_people() to display all of
that is one greater than the number of authors in the list and set the 'etal' flag true.
the names in the name list followed by 'et al.'
This will cause the list_people() to display all of the names in the name list followed by 'et al.'


In all other cases, returns nil and the previous state of the 'etal' flag.
In all other cases, returns nil and the previous state of the 'etal' flag.


inputs:
inputs:
max: A['DisplayAuthors'] or A['DisplayEditors']; a number or some flavor of etal
max: A['DisplayAuthors'] or A['DisplayEditors'], etc; a number or some flavor of etal
count: #a or #e
count: #a or #e
list_name: 'authors' or 'editors'
list_name: 'authors' or 'editors'
etal: author_etal or editor_etal
etal: author_etal or editor_etal
This function sets an error message when |display-xxxxors= value greater than or equal to number of names but
not when <max> comes from {{cs1 config}} global settings.  When using global settings, <param> is set to the
keyword 'cs1 config' which is used to supress the normal error.  Error is suppressed because it is to be expected
that some citations in an article will have the same or fewer names that the limit specified in {{cs1 config}}.


]]
]]


local function get_display_names (max, count, list_name, etal)
local function get_display_names (max, count, list_name, etal, param)
if is_set (max) then
if utilities.is_set (max) then
if 'etal' == max:lower():gsub("[ '%.]", '') then -- the :gsub() portion makes 'etal' from a variety of 'et al.' spellings and stylings
if 'etal' == max:lower():gsub("[ '%.]", '') then -- the :gsub() portion makes 'etal' from a variety of 'et al.' spellings and stylings
max = count + 1; -- number of authors + 1 so display all author name plus et al.
max = count + 1; -- number of authors + 1 so display all author name plus et al.
Line 1,702: Line 1,933:
elseif max:match ('^%d+$') then -- if is a string of numbers
elseif max:match ('^%d+$') then -- if is a string of numbers
max = tonumber (max); -- make it a number
max = tonumber (max); -- make it a number
if max >= count then -- if |display-xxxxors= value greater than or equal to number of authors/editors
if (max >= count) and ('cs1 config' ~= param) then -- error when local |display-xxxxors= value greater than or equal to number of names; not an error when using global setting
add_maint_cat ('disp_name', cfg.special_case_translation [list_name]);
utilities.set_message ('err_disp_name', {param, max}); -- add error message
max = nil;
end
end
else -- not a valid keyword or number
else -- not a valid keyword or number
table.insert( z.message_tail, { set_error( 'invalid_param_val', {'display-' .. list_name, max}, true ) } ); -- add error message
utilities.set_message ('err_disp_name', {param, max}); -- add error message
max = nil; -- unset; as if |display-xxxxors= had not been set
max = nil; -- unset; as if |display-xxxxors= had not been set
end
end
Line 1,715: Line 1,947:




--[[--------------------------< E X T R A _ T E X T _ I N _ P A G E _ C H E C K >------------------------------
--[[----------< E X T R A _ T E X T _ I N _ P A G E _ C H E C K >---------------


Adds page to Category:CS1 maint: extra text if |page= or |pages= has what appears to be some form of p. or pp.  
Adds error if |page=, |pages=, |quote-page=, |quote-pages= has what appears to be
abbreviation in the first characters of the parameter content.
some form of p. or pp. abbreviation in the first characters of the parameter content.


check Page and Pages for extraneous p, p., pp, and pp. at start of parameter value:
check page for extraneous p, p., pp, pp., pg, pg. at start of parameter value:
good pattern: '^P[^%.P%l]' matches when |page(s)= begins PX or P# but not Px where x and X are letters and # is a dgiit
good pattern: '^P[^%.P%l]' matches when page begins PX or P# but not Px
bad pattern: '^[Pp][Pp]' matches matches when |page(s)= begins pp or pP or Pp or PP
      where x and X are letters and # is a digit
bad pattern: '^[Pp][PpGg]' matches when page begins pp, pP, Pp, PP, pg, pG, Pg, PG


]]
]]


local function extra_text_in_page_check (page)
local function extra_text_in_page_check (val, name)
local good_pattern = '^P[^%.Pp]'; -- ok to begin with uppercase P: P7 (pg 7 of section P) but not p123 (page 123) TODO: add Gg for PG or Pg?
if not val:match (cfg.vol_iss_pg_patterns.good_ppattern) then
local bad_pattern = '^[Pp]?[Pp]%.?[ %d]';
for _, pattern in ipairs (cfg.vol_iss_pg_patterns.bad_ppatterns) do -- spin through the selected sequence table of patterns
 
if val:match (pattern) then -- when a match, error so
if not page:match (good_pattern) and (page:match (bad_pattern) or  page:match ('^[Pp]ages?')) then
utilities.set_message ('err_extra_text_pages', name); -- add error message
add_maint_cat ('extra_text');
return; -- and done
end
end
end
end
end
 
 
--[[--------------------------< E X T R A _ T E X T _ I N _ V O L _ I S S _ C H E C K >------------------------
 
Adds error if |volume= or |issue= has what appears to be some form of redundant 'type' indicator.
 
For |volume=:
'V.', or 'Vol.' (with or without the dot) abbreviations or 'Volume' in the first characters of the parameter
content (all case insensitive). 'V' and 'v' (without the dot) are presumed to be roman numerals so
are allowed.
 
For |issue=:
'No.', 'I.', 'Iss.' (with or without the dot) abbreviations, or 'Issue' in the first characters of the
parameter content (all case insensitive).
Single character values ('v', 'i', 'n') allowed when not followed by separator character ('.', ':', '=', or
whitespace character) – param values are trimmed of whitespace by MediaWiki before delivered to the module.
<val> is |volume= or |issue= parameter value
<name> is |volume= or |issue= parameter name for error message
<selector> is 'v' for |volume=, 'i' for |issue=
 
sets error message on failure; returns nothing
 
]]
 
local function extra_text_in_vol_iss_check (val, name, selector)
if not utilities.is_set (val) then
return;
end
local patterns = 'v' == selector and cfg.vol_iss_pg_patterns.vpatterns or cfg.vol_iss_pg_patterns.ipatterns;
 
local handler = 'v' == selector and 'err_extra_text_volume' or 'err_extra_text_issue';
val = val:lower(); -- force parameter value to lower case
for _, pattern in ipairs (patterns) do -- spin through the selected sequence table of patterns
if val:match (pattern) then -- when a match, error so
utilities.set_message (handler, name); -- add error message
return; -- and done
end
end
end
end


Line 1,739: Line 2,016:


split apart a |vauthors= or |veditors= parameter.  This function allows for corporate names, wrapped in doubled
split apart a |vauthors= or |veditors= parameter.  This function allows for corporate names, wrapped in doubled
parentheses to also have commas; in the old version of the code, the doubled parnetheses were included in the
parentheses to also have commas; in the old version of the code, the doubled parentheses were included in the
rendered citation and in the metadata.  Individual author names may be wikilinked
rendered citation and in the metadata.  Individual author names may be wikilinked


Line 1,748: Line 2,025:
local function get_v_name_table (vparam, output_table, output_link_table)
local function get_v_name_table (vparam, output_table, output_link_table)
local name_table = mw.text.split(vparam, "%s*,%s*"); -- names are separated by commas
local name_table = mw.text.split(vparam, "%s*,%s*"); -- names are separated by commas
local wl_type, label, link; -- wl_type not used here; just a place holder
local wl_type, label, link; -- wl_type not used here; just a placeholder
local i = 1;
local i = 1;
while name_table[i] do
while name_table[i] do
if name_table[i]:match ('^%(%(.*[^%)][^%)]$') then -- first segment of corporate with one or more commas; this segment has the opening doubled parens
if name_table[i]:match ('^%(%(.*[^%)][^%)]$') then -- first segment of corporate with one or more commas; this segment has the opening doubled parentheses
local name = name_table[i];
local name = name_table[i];
i=i+1; -- bump indexer to next segment
i = i + 1; -- bump indexer to next segment
while name_table[i] do
while name_table[i] do
name = name .. ', ' .. name_table[i]; -- concatenate with previous segments
name = name .. ', ' .. name_table[i]; -- concatenate with previous segments
if name_table[i]:match ('^.*%)%)$') then -- if this table member has the closing doubled parens
if name_table[i]:match ('^.*%)%)$') then -- if this table member has the closing doubled parentheses
break; -- and done reassembling so
break; -- and done reassembling so
end
end
i=i+1; -- bump indexer
i = i + 1; -- bump indexer
end
end
table.insert (output_table, name); -- and add corporate name to the output table
table.insert (output_table, name); -- and add corporate name to the output table
table.insert (output_link_table, ''); -- no wikilink
table.insert (output_link_table, ''); -- no wikilink
else
else
wl_type, label, link = is_wikilink (name_table[i]); -- wl_type is: 0, no wl (text in label variable); 1, [[D]]; 2, [[L|D]]
wl_type, label, link = utilities.is_wikilink (name_table[i]); -- wl_type is: 0, no wl (text in label variable); 1, [[D]]; 2, [[L|D]]
table.insert (output_table, label); -- add this name
table.insert (output_table, label); -- add this name
if 1 == wl_type then
if 1 == wl_type then
Line 1,774: Line 2,051:
end
end
end
end
i = i+1;
i = i + 1;
end
end
return output_table;
return output_table;
Line 1,787: Line 2,064:
Author / editor names in |vauthors= or |veditors= must be in Vancouver system style. Corporate or institutional names
Author / editor names in |vauthors= or |veditors= must be in Vancouver system style. Corporate or institutional names
may sometimes be required and because such names will often fail the is_good_vanc_name() and other format compliance
may sometimes be required and because such names will often fail the is_good_vanc_name() and other format compliance
tests, are wrapped in doubled paranethese ((corporate name)) to suppress the format tests.
tests, are wrapped in doubled parentheses ((corporate name)) to suppress the format tests.


Supports generational suffixes Jr, 2nd, 3rd, 4th–6th.
Supports generational suffixes Jr, 2nd, 3rd, 4th–6th.


This function sets the vancouver error when a reqired comma is missing and when there is a space between an author's initials.
This function sets the Vancouver error when a required comma is missing and when there is a space between an author's initials.


]]
]]
Line 1,807: Line 2,084:


for i, v_name in ipairs(v_name_table) do
for i, v_name in ipairs(v_name_table) do
if v_name:match ('^%(%(.+%)%)$') then -- corporate authors are wrapped in doubled parentheses to supress vanc formatting and error detection
first = ''; -- set to empty string for concatenation and because it may have been set for previous author/editor
first = ''; -- set to empty string for concatenation and because it may have been set for previous author/editor
local accept_name;
last = v_name:match ('^%(%((.+)%)%)$') -- remove doubled parntheses
v_name, accept_name = utilities.has_accept_as_written (v_name); -- remove accept-this-as-written markup when it wraps all of <v_name>
 
if accept_name then
last = v_name;
corporate = true; -- flag used in list_people()
corporate = true; -- flag used in list_people()
elseif string.find(v_name, "%s") then
elseif string.find(v_name, "%s") then
if v_name:find('[;%.]') then -- look for commonly occurring punctuation characters;  
if v_name:find('[;%.]') then -- look for commonly occurring punctuation characters;  
add_vanc_error ('punctuation');
add_vanc_error (cfg.err_msg_supl.punctuation, i);
end
end
local lastfirstTable = {}
local lastfirstTable = {}
lastfirstTable = mw.text.split(v_name, "%s")
lastfirstTable = mw.text.split(v_name, "%s+")
first = table.remove(lastfirstTable); -- removes and returns value of last element in table which should be author intials
first = table.remove(lastfirstTable); -- removes and returns value of last element in table which should be initials or generational suffix
if is_suffix (first) then -- if a valid suffix
 
suffix = first -- save it as a suffix and
if not mw.ustring.match (first, '^%u+$') then -- mw.ustring here so that later we will catch non-Latin characters
suffix = first; -- not initials so assume that whatever we got is a generational suffix
first = table.remove(lastfirstTable); -- get what should be the initials from the table
first = table.remove(lastfirstTable); -- get what should be the initials from the table
end -- no suffix error message here because letter combination may be result of Romanization; check for digits?
end
last = table.concat(lastfirstTable, " ") -- returns a string that is the concatenation of all other names that are not initials
last = table.concat(lastfirstTable, ' ') -- returns a string that is the concatenation of all other names that are not initials and generational suffix
if not utilities.is_set (last) then
first = ''; -- unset
last = v_name; -- last empty because something wrong with first
add_vanc_error (cfg.err_msg_supl.name, i);
end
if mw.ustring.match (last, '%a+%s+%u+%s+%a+') then
if mw.ustring.match (last, '%a+%s+%u+%s+%a+') then
add_vanc_error ('missing comma'); -- matches last II last; the case when a comma is missing
add_vanc_error (cfg.err_msg_supl['missing comma'], i); -- matches last II last; the case when a comma is missing
end
end
if mw.ustring.match (v_name, ' %u %u$') then -- this test is in the wrong place TODO: move or replace with a more appropriate test
if mw.ustring.match (v_name, ' %u %u$') then -- this test is in the wrong place TODO: move or replace with a more appropriate test
add_vanc_error ('name'); -- matches a space between two intiials
add_vanc_error (cfg.err_msg_supl.initials, i); -- matches a space between two initials
end
end
else
else
first = ''; -- set to empty string for concatenation and because it may have been set for previous author/editor
last = v_name; -- last name or single corporate name?  Doesn't support multiword corporate names? do we need this?
last = v_name; -- last name or single corporate name?  Doesn't support multiword corporate names? do we need this?
end
end
if is_set (first) then
if utilities.is_set (first) then
if not mw.ustring.match (first, "^%u?%u$") then -- first shall contain one or two upper-case letters, nothing else
if not mw.ustring.match (first, "^%u?%u$") then -- first shall contain one or two upper-case letters, nothing else
add_vanc_error ('initials'); -- too many initials; mixed case initials (which may be ok Romanization); hyphenated initials
add_vanc_error (cfg.err_msg_supl.initials, i); -- too many initials; mixed case initials (which may be ok Romanization); hyphenated initials
end
end
is_good_vanc_name (last, first); -- check first and last before restoring the suffix which may have a non-Latin digit
is_good_vanc_name (last, first, suffix, i); -- check first and last before restoring the suffix which may have a non-Latin digit
if is_set (suffix) then
if utilities.is_set (suffix) then
first = first .. ' ' .. suffix; -- if there was a suffix concatenate with the initials
first = first .. ' ' .. suffix; -- if there was a suffix concatenate with the initials
suffix = ''; -- unset so we don't add this suffix to all subsequent names
suffix = ''; -- unset so we don't add this suffix to all subsequent names
Line 1,845: Line 2,130:
else
else
if not corporate then
if not corporate then
is_good_vanc_name (last, '');
is_good_vanc_name (last, '', nil, i);
end
end
end
end


link = select_one( args, cfg.aliases[list_name .. '-Link'], 'redundant_parameters', i ) or v_link_table[i];
link = utilities.select_one ( args, cfg.aliases[list_name .. '-Link'], 'err_redundant_parameters', i ) or v_link_table[i];
mask = select_one( args, cfg.aliases[list_name .. '-Mask'], 'redundant_parameters', i );
mask = utilities.select_one ( args, cfg.aliases[list_name .. '-Mask'], 'err_redundant_parameters', i );
names[i] = {last = last, first = first, link = link, mask = mask, corporate=corporate}; -- add this assembled name to our names list
names[i] = {last = last, first = first, link = link, mask = mask, corporate = corporate}; -- add this assembled name to our names list
end
end
return names, etal; -- all done, return our list of names
return names, etal; -- all done, return our list of names
Line 1,860: Line 2,145:


Select one of |authors=, |authorn= / |lastn / firstn=, or |vauthors= as the source of the author name list or
Select one of |authors=, |authorn= / |lastn / firstn=, or |vauthors= as the source of the author name list or
select one of |editors=, |editorn= / editor-lastn= / |editor-firstn= or |veditors= as the source of the editor name list.
select one of |editorn= / editor-lastn= / |editor-firstn= or |veditors= as the source of the editor name list.


Only one of these appropriate three will be used.  The hierarchy is: |authorn= (and aliases) highest and |authors= lowest and
Only one of these appropriate three will be used.  The hierarchy is: |authorn= (and aliases) highest and |authors= lowest;
similarly, |editorn= (and aliases) highest and |editors= lowest
|editorn= (and aliases) highest and |veditors= lowest (support for |editors= withdrawn)


When looking for |authorn= / |editorn= parameters, test |xxxxor1= and |xxxxor2= (and all of their aliases); stops after the second
When looking for |authorn= / |editorn= parameters, test |xxxxor1= and |xxxxor2= (and all of their aliases); stops after the second
Line 1,871: Line 2,156:
Emits an error message when more than one xxxxor name source is provided.
Emits an error message when more than one xxxxor name source is provided.


In this function, vxxxxors = vauthors or veditors; xxxxors = authors or editors as appropriate.
In this function, vxxxxors = vauthors or veditors; xxxxors = authors as appropriate.


]]
]]


local function select_author_editor_source (vxxxxors, xxxxors, args, list_name)
local function select_author_editor_source (vxxxxors, xxxxors, args, list_name)
local lastfirst = false;
local lastfirst = false;
if select_one( args, cfg.aliases[list_name .. '-Last'], 'none', 1 ) or -- do this twice incase we have a |first1= without a |last1=; this ...
if utilities.select_one ( args, cfg.aliases[list_name .. '-Last'], 'none', 1 ) or -- do this twice in case we have a |first1= without a |last1=; this ...
select_one( args, cfg.aliases[list_name .. '-First'], 'none', 1 ) or -- ... also catches the case where |first= is used with |vauthors=
utilities.select_one ( args, cfg.aliases[list_name .. '-First'], 'none', 1 ) or -- ... also catches the case where |first= is used with |vauthors=
select_one( args, cfg.aliases[list_name .. '-Last'], 'none', 2 ) or
utilities.select_one ( args, cfg.aliases[list_name .. '-Last'], 'none', 2 ) or
select_one( args, cfg.aliases[list_name .. '-First'], 'none', 2 ) then
utilities.select_one ( args, cfg.aliases[list_name .. '-First'], 'none', 2 ) then
lastfirst=true;
lastfirst = true;
end
end


if (is_set (vxxxxors) and true == lastfirst) or -- these are the three error conditions
if (utilities.is_set (vxxxxors) and true == lastfirst) or -- these are the three error conditions
(is_set (vxxxxors) and is_set (xxxxors)) or
(utilities.is_set (vxxxxors) and utilities.is_set (xxxxors)) or
(true == lastfirst and is_set (xxxxors)) then
(true == lastfirst and utilities.is_set (xxxxors)) then
local err_name;
local err_name;
if 'AuthorList' == list_name then -- figure out which name should be used in error message
if 'AuthorList' == list_name then -- figure out which name should be used in error message
Line 1,893: Line 2,178:
err_name = 'editor';
err_name = 'editor';
end
end
table.insert( z.message_tail, { set_error( 'redundant_parameters',
utilities.set_message ('err_redundant_parameters', err_name .. '-name-list parameters'); -- add error message
{err_name .. '-name-list parameters'}, true ) } ); -- add error message
end
end


if true == lastfirst then return 1 end; -- return a number indicating which author name source to use
if true == lastfirst then return 1 end; -- return a number indicating which author name source to use
if is_set (vxxxxors) then return 2 end;
if utilities.is_set (vxxxxors) then return 2 end;
if is_set (xxxxors) then return 3 end;
if utilities.is_set (xxxxors) then return 3 end;
return 1; -- no authors so return 1; this allows missing author name test to run in case there is a first without last  
return 1; -- no authors so return 1; this allows missing author name test to run in case there is a first without last  
end
end
Line 1,907: Line 2,191:


This function is used to validate a parameter's assigned value for those parameters that have only a limited number
This function is used to validate a parameter's assigned value for those parameters that have only a limited number
of allowable values (yes, y, true, no, etc).  When the parameter value has not been assigned a value (missing or empty
of allowable values (yes, y, true, live, dead, etc.).  When the parameter value has not been assigned a value (missing
in the source template) the function returns true.  If the parameter value is one of the list of allowed values returns
or empty in the source template) the function returns the value specified by ret_val.  If the parameter value is one
true; else, emits an error message and returns false.
of the list of allowed values returns the translated value; else, emits an error message and returns the value
specified by ret_val.
 
TODO: explain <invert>


]]
]]


local function is_valid_parameter_value (value, name, possible)
local function is_valid_parameter_value (value, name, possible, ret_val, invert)
if not is_set (value) then
if not utilities.is_set (value) then
return true; -- an empty parameter is ok
return ret_val; -- an empty parameter is ok
elseif in_array (value, possible) then
end
return true;
 
if (not invert and utilities.in_array (value, possible)) then -- normal; <value> is in <possible> table
return cfg.keywords_xlate[value]; -- return translation of parameter keyword
elseif invert and not utilities.in_array (value, possible) then -- invert; <value> is not in <possible> table
return value; -- return <value> as it is
else
else
table.insert( z.message_tail, { set_error( 'invalid_param_val', {name, value}, true ) } ); -- not an allowed value so add error message
utilities.set_message ('err_invalid_param_val', {name, value}); -- not an allowed value so add error message
return false
return ret_val;
end
end
end
end
Line 1,935: Line 2,226:


local function terminate_name_list (name_list, sepc)
local function terminate_name_list (name_list, sepc)
if (string.sub (name_list,-3,-1) == sepc .. '. ') then -- if already properly terminated
if (string.sub (name_list, -3, -1) == sepc .. '. ') then -- if already properly terminated
return name_list; -- just return the name list
return name_list; -- just return the name list
elseif (string.sub (name_list,-1,-1) == sepc) or (string.sub (name_list,-3,-1) == sepc .. ']]') then -- if last name in list ends with sepc char
elseif (string.sub (name_list, -1, -1) == sepc) or (string.sub (name_list, -3, -1) == sepc .. ']]') then -- if last name in list ends with sepc char
return name_list .. " "; -- don't add another
return name_list .. " "; -- don't add another
else
else
return name_list .. sepc .. ' '; -- otherwise terninate the name list
return name_list .. sepc .. ' '; -- otherwise terminate the name list
end
end
end
end




--[[-------------------------< F O R M A T _ V O L U M E _ I S S U E >----------------------------------------
--[[-------------------------< F O R M A T _ V O L U M E _ I S S U E >-----------------------------------------


returns the concatenation of the formatted volume and issue parameters as a single string; or formatted volume
returns the concatenation of the formatted volume and issue (or journal article number) parameters as a single
or formatted issue, or an empty string if neither are set.
string; or formatted volume or formatted issue, or an empty string if neither are set.


]]
]]
local function format_volume_issue (volume, issue, cite_class, origin, sepc, lower)
local function format_volume_issue (volume, issue, article, cite_class, origin, sepc, lower)
if not is_set (volume) and not is_set (issue) then
if not utilities.is_set (volume) and not utilities.is_set (issue) and not utilities.is_set (article) then
return '';
return '';
end
end
-- same condition as in format_pages_sheets()
local is_journal = 'journal' == cite_class or (utilities.in_array (cite_class, {'citation', 'map', 'interview'}) and 'journal' == origin);
local is_numeric_vol = volume and (volume:match ('^[MDCLXVI]+$') or volume:match ('^%d+$')); -- is only uppercase roman numerals or only digits?
local is_long_vol = volume and (4 < mw.ustring.len(volume)); -- is |volume= value longer than 4 characters?
if 'magazine' == cite_class or (in_array (cite_class, {'citation', 'map'}) and 'magazine' == origin) then
if volume and (not is_numeric_vol and is_long_vol) then -- when not all digits or Roman numerals, is |volume= longer than 4 characters?
if is_set (volume) and is_set (issue) then
utilities.add_prop_cat ('long-vol'); -- yes, add properties cat
return wrap_msg ('vol-no', {sepc, volume, issue}, lower);
elseif is_set (volume) then
return wrap_msg ('vol', {sepc, volume}, lower);
else
return wrap_msg ('issue', {sepc, issue}, lower);
end
end
end


local vol = ''; -- here for all cites except magazine
if is_journal then -- journal-style formatting
local vol = '';
if is_set (volume) then
 
if volume:match ('^[MDCLXVI]+$') or volume:match ('^%d+$')then -- volume value is all digits or all uppercase roman numerals
if utilities.is_set (volume) then
vol = substitute (cfg.presentation['vol-bold'], {sepc, hyphen_to_dash(volume)}); -- render in bold face
if is_numeric_vol then -- |volume= value all digits or all uppercase Roman numerals?
elseif (4 < mw.ustring.len(volume)) then -- not all digits or roman numerals and longer than 4 characters
vol = utilities.substitute (cfg.presentation['vol-bold'], {sepc, volume}); -- render in bold face
vol = substitute (cfg.messages['j-vol'], {sepc, volume}); -- not bold
elseif is_long_vol then -- not all digits or Roman numerals; longer than 4 characters?
add_prop_cat ('long_vol');
vol = utilities.substitute (cfg.messages['j-vol'], {sepc, utilities.hyphen_to_dash (volume)}); -- not bold
else -- four or less characters
else -- four or fewer characters
vol = substitute (cfg.presentation['vol-bold'], {sepc, hyphen_to_dash(volume)}); -- bold
vol = utilities.substitute (cfg.presentation['vol-bold'], {sepc, utilities.hyphen_to_dash (volume)}); -- bold
end
end
end
vol = vol .. (utilities.is_set (issue) and utilities.substitute (cfg.messages['j-issue'], issue) or '')
vol = vol .. (utilities.is_set (article) and utilities.substitute (cfg.messages['j-article-num'], article) or '')
return vol;
end
end
if is_set (issue) then
return vol .. substitute (cfg.messages['j-issue'], issue);
if 'podcast' == cite_class and utilities.is_set (issue) then
return wrap_msg ('issue', {sepc, issue}, lower);
end
if 'conference' == cite_class and utilities.is_set (article) then -- |article-number= supported only in journal and conference cites
if utilities.is_set (volume) and utilities.is_set (article) then -- both volume and article number
return wrap_msg ('vol-art', {sepc, utilities.hyphen_to_dash (volume), article}, lower);
elseif utilities.is_set (article) then -- article number alone; when volume alone, handled below
return wrap_msg ('art', {sepc, article}, lower);
end
end
 
-- all other types of citation
if utilities.is_set (volume) and utilities.is_set (issue) then
return wrap_msg ('vol-no', {sepc, utilities.hyphen_to_dash (volume), issue}, lower);
elseif utilities.is_set (volume) then
return wrap_msg ('vol', {sepc, utilities.hyphen_to_dash (volume)}, lower);
else
return wrap_msg ('issue', {sepc, issue}, lower);
end
end
return vol;
end
end


Line 1,998: Line 2,310:
local function format_pages_sheets (page, pages, sheet, sheets, cite_class, origin, sepc, nopp, lower)
local function format_pages_sheets (page, pages, sheet, sheets, cite_class, origin, sepc, nopp, lower)
if 'map' == cite_class then -- only cite map supports sheet(s) as in-source locators
if 'map' == cite_class then -- only cite map supports sheet(s) as in-source locators
if is_set (sheet) then
if utilities.is_set (sheet) then
if 'journal' == origin then
if 'journal' == origin then
return '', '', wrap_msg ('j-sheet', sheet, lower), '';
return '', '', wrap_msg ('j-sheet', sheet, lower), '';
Line 2,004: Line 2,316:
return '', '', wrap_msg ('sheet', {sepc, sheet}, lower), '';
return '', '', wrap_msg ('sheet', {sepc, sheet}, lower), '';
end
end
elseif is_set (sheets) then
elseif utilities.is_set (sheets) then
if 'journal' == origin then
if 'journal' == origin then
return '', '', '', wrap_msg ('j-sheets', sheets, lower);
return '', '', '', wrap_msg ('j-sheets', sheets, lower);
Line 2,013: Line 2,325:
end
end


local is_journal = 'journal' == cite_class or (in_array (cite_class, {'citation', 'map', 'interview'}) and 'journal' == origin);
local is_journal = 'journal' == cite_class or (utilities.in_array (cite_class, {'citation', 'map', 'interview'}) and 'journal' == origin);
if is_set (page) then
if utilities.is_set (page) then
if is_journal then
if is_journal then
return substitute (cfg.messages['j-page(s)'], page), '', '', '';
return utilities.substitute (cfg.messages['j-page(s)'], page), '', '', '';
elseif not nopp then
elseif not nopp then
return substitute (cfg.messages['p-prefix'], {sepc, page}), '', '', '';
return utilities.substitute (cfg.messages['p-prefix'], {sepc, page}), '', '', '';
else
else
return substitute (cfg.messages['nopp'], {sepc, page}), '', '', '';
return utilities.substitute (cfg.messages['nopp'], {sepc, page}), '', '', '';
end
end
elseif is_set(pages) then
elseif utilities.is_set (pages) then
if is_journal then
if is_journal then
return substitute (cfg.messages['j-page(s)'], pages), '', '', '';
return utilities.substitute (cfg.messages['j-page(s)'], pages), '', '', '';
elseif tonumber(pages) ~= nil and not nopp then -- if pages is only digits, assume a single page number
elseif tonumber(pages) ~= nil and not nopp then -- if pages is only digits, assume a single page number
return '', substitute (cfg.messages['p-prefix'], {sepc, pages}), '', '';
return '', utilities.substitute (cfg.messages['p-prefix'], {sepc, pages}), '', '';
elseif not nopp then
elseif not nopp then
return '', substitute (cfg.messages['pp-prefix'], {sepc, pages}), '', '';
return '', utilities.substitute (cfg.messages['pp-prefix'], {sepc, pages}), '', '';
else
else
return '', substitute (cfg.messages['nopp'], {sepc, pages}), '', '';
return '', utilities.substitute (cfg.messages['nopp'], {sepc, pages}), '', '';
end
end
end
end
Line 2,043: Line 2,355:
returns one of the in-source locators: page, pages, or at.  
returns one of the in-source locators: page, pages, or at.  


If any of these are interwiki links to wikisource, returns the label portion of the interwikilink as plain text
If any of these are interwiki links to Wikisource, returns the label portion of the interwiki-link as plain text
for use in COinS.  This COinS thing is done because here we convert an interwiki link to and external link and
for use in COinS.  This COinS thing is done because here we convert an interwiki-link to an external link and
add an icon span around that; get_coins_pages() doesn't know about the span.  TODO: should it?   
add an icon span around that; get_coins_pages() doesn't know about the span.  TODO: should it?   


Line 2,054: Line 2,366:
]]
]]


local function insource_loc_get (page, pages, at)
local function insource_loc_get (page, page_orig, pages, pages_orig, at)
local ws_url, ws_label, coins_pages, L; -- for wikisource interwikilinks; TODO: this corrupts page metadata (span remains in place after cleanup; fix there?)
local ws_url, ws_label, coins_pages, L; -- for Wikisource interwiki-links; TODO: this corrupts page metadata (span remains in place after cleanup; fix there?)


if is_set (page) then
if utilities.is_set (page) then
if is_set (pages) or is_set(at) then
if utilities.is_set (pages) or utilities.is_set (at) then
pages = ''; -- unset the others
pages = ''; -- unset the others
at = '';
at = '';
end
end
extra_text_in_page_check (page); -- add this page to maint cat if |page= value begins with what looks like p. or pp.
extra_text_in_page_check (page, page_orig); -- emit error message when |page= value begins with what looks like p., pp., etc.


ws_url, ws_label, L = wikisource_url_make (page); -- make ws url from |page= interwiki link; link portion L becomes tool tip label
ws_url, ws_label, L = wikisource_url_make (page); -- make ws URL from |page= interwiki link; link portion L becomes tooltip label
if ws_url then
if ws_url then
page = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in page'); -- space char after label to move icon away from in-source text; TODO: a better way to do this?
page = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in page'); -- space char after label to move icon away from in-source text; TODO: a better way to do this?
page = substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, page});
page = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, page});
coins_pages = ws_label;
coins_pages = ws_label;
end
end
elseif is_set (pages) then
elseif utilities.is_set (pages) then
if is_set (at) then
if utilities.is_set (at) then
at = ''; -- unset
at = ''; -- unset
end
end
extra_text_in_page_check (pages); -- add this page to maint cat if |pages= value begins with what looks like p. or pp.
extra_text_in_page_check (pages, pages_orig); -- emit error message when |page= value begins with what looks like p., pp., etc.


ws_url, ws_label, L = wikisource_url_make (pages); -- make ws url from |pages= interwiki link; link portion L becomes tool tip label
ws_url, ws_label, L = wikisource_url_make (pages); -- make ws URL from |pages= interwiki link; link portion L becomes tooltip label
if ws_url then
if ws_url then
pages = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in pages'); -- space char after label to move icon away from in-source text; TODO: a better way to do this?
pages = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in pages'); -- space char after label to move icon away from in-source text; TODO: a better way to do this?
pages = substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, pages});
pages = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, pages});
coins_pages = ws_label;
coins_pages = ws_label;
end
end
elseif is_set (at) then
elseif utilities.is_set (at) then
ws_url, ws_label, L = wikisource_url_make (at); -- make ws url from |at= interwiki link; link portion L becomes tool tip label
ws_url, ws_label, L = wikisource_url_make (at); -- make ws URL from |at= interwiki link; link portion L becomes tooltip label
if ws_url then
if ws_url then
at = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in at'); -- space char after label to move icon away from in-source text; TODO: a better way to do this?
at = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in at'); -- space char after label to move icon away from in-source text; TODO: a better way to do this?
at = substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, at});
at = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, at});
coins_pages = ws_label;
coins_pages = ws_label;
end
end
Line 2,094: Line 2,406:
end
end


--[[--------------------------< I S _ U N I Q U E _ A R C H I V E _ U R L >------------------------------------
add error message when |archive-url= value is same as |url= or chapter-url= (or alias...) value
]]
local function is_unique_archive_url (archive, url, c_url, source, date)
if utilities.is_set (archive) then
if archive == url or archive == c_url then
utilities.set_message ('err_bad_url', {utilities.wrap_style ('parameter', source)}); -- add error message
return '', ''; -- unset |archive-url= and |archive-date= because same as |url= or |chapter-url=
end
end
return archive, date;
end




--[=[-------------------------< A R C H I V E _ U R L _ C H E C K >--------------------------------------------
--[=[-------------------------< A R C H I V E _ U R L _ C H E C K >--------------------------------------------


Check archive.org urls to make sure they at least look like they are pointing at valid archives and not to the  
Check archive.org URLs to make sure they at least look like they are pointing at valid archives and not to the  
save snapshot url or to calendar pages.  When the archive url is 'https://web.archive.org/save/' (or http://...)
save snapshot URL or to calendar pages.  When the archive URL is 'https://web.archive.org/save/' (or http://...)
archive.org saves a snapshot of the target page in the url.  That is something that Wikipedia should not allow
archive.org saves a snapshot of the target page in the URL.  That is something that Wikipedia should not allow
unwitting readers to do.
unwitting readers to do.


When the archive.org url does not have a complete timestamp, archive.org chooses a snapshot according to its own
When the archive.org URL does not have a complete timestamp, archive.org chooses a snapshot according to its own
algorithm or provides a calendar 'search' result.  [[WP:ELNO]] discourages links to search results.
algorithm or provides a calendar 'search' result.  [[WP:ELNO]] discourages links to search results.


This function looks at the value assigned to |archive-url= and returns empty strings for |archive-url= and
This function looks at the value assigned to |archive-url= and returns empty strings for |archive-url= and
|archive-date= and an error message when:
|archive-date= and an error message when:
|archive-url= holds an archive.org save command url
|archive-url= holds an archive.org save command URL
|archive-url= is an archive.org url that does not have a complete timestamp (YYYYMMDDhhmmss 14 digits) in the
|archive-url= is an archive.org URL that does not have a complete timestamp (YYYYMMDDhhmmss 14 digits) in the
correct place
correct place
otherwise returns |archive-url= and |archive-date=
otherwise returns |archive-url= and |archive-date=


There are two mostly compatible archive.org urls:
There are two mostly compatible archive.org URLs:
//web.archive.org/<timestamp>... -- the old form
//web.archive.org/<timestamp>... -- the old form
//web.archive.org/web/<timestamp>... -- the new form
//web.archive.org/web/<timestamp>... -- the new form
Line 2,121: Line 2,449:
we don't check for these specific flags but we do check the form.
we don't check for these specific flags but we do check the form.


This function supports a preview mode.  When the article is rendered in preview mode, this funct may return a modified
This function supports a preview mode.  When the article is rendered in preview mode, this function may return a modified
archive url:
archive URL:
for save command errors, return undated wildcard (/*/)
for save command errors, return undated wildcard (/*/)
for timestamp errors when the timestamp has a wildcard, return the url unmodified
for timestamp errors when the timestamp has a wildcard, return the URL unmodified
for timestamp errors when the timestamp does not have a wildcard, return with timestamp limited to six digits plus wildcard (/yyyymm*/)
for timestamp errors when the timestamp does not have a wildcard, return with timestamp limited to six digits plus wildcard (/yyyymm*/)
A secondary function is to return an archive-url timestamp from those urls that have them (archive.org and
archive.today).  The timestamp is used by validation.archive_date_check() to see if the value in |archive-date=
matches the timestamp in the archive url.


]=]
]=]
Line 2,131: Line 2,463:
local function archive_url_check (url, date)
local function archive_url_check (url, date)
local err_msg = ''; -- start with the error message empty
local err_msg = ''; -- start with the error message empty
local path, timestamp, flag; -- portions of the archive.or url
local path, timestamp, flag; -- portions of the archive.org URL
if (not url:match('//web%.archive%.org/')) and (not url:match('//liveweb%.archive%.org/')) then -- also deprecated liveweb Wayback machine url
timestamp = url:match ('//archive.today/(%d%d%d%d%d%d%d%d%d%d%d%d%d%d)/'); -- get timestamp from archive.today urls
if timestamp then -- if this was an archive.today url ...
return url, date, timestamp; -- return ArchiveURL, ArchiveDate, and timestamp from |archive-url=, and done
end
-- here for archive.org urls
if (not url:match('//web%.archive%.org/')) and (not url:match('//liveweb%.archive%.org/')) then -- also deprecated liveweb Wayback machine URL
return url, date; -- not an archive.org archive, return ArchiveURL and ArchiveDate
return url, date; -- not an archive.org archive, return ArchiveURL and ArchiveDate
end
end


if url:match('//web%.archive%.org/save/') then -- if a save command url, we don't want to allow saving of the target page  
if url:match('//web%.archive%.org/save/') then -- if a save command URL, we don't want to allow saving of the target page  
err_msg = 'save command';
err_msg = cfg.err_msg_supl.save;
url = url:gsub ('(//web%.archive%.org)/save/', '%1/*/', 1); -- for preview mode: modify ArchiveURL
url = url:gsub ('(//web%.archive%.org)/save/', '%1/*/', 1); -- for preview mode: modify ArchiveURL
elseif url:match('//liveweb%.archive%.org/') then
elseif url:match('//liveweb%.archive%.org/') then
err_msg = 'liveweb';
err_msg = cfg.err_msg_supl.liveweb;
else
else
path, timestamp, flag = url:match('//web%.archive%.org/([^%d]*)(%d+)([^/]*)/'); -- split out some of the url parts for evaluation
path, timestamp, flag = url:match('//web%.archive%.org/([^%d]*)(%d+)([^/]*)/'); -- split out some of the URL parts for evaluation
if not path then -- malformed in some way; pattern did not match
if not is_set(timestamp) or 14 ~= timestamp:len() then -- path and flag optional, must have 14-digit timestamp here
err_msg = cfg.err_msg_supl.timestamp;
err_msg = 'timestamp';
elseif 14 ~= timestamp:len() then -- path and flag optional, must have 14-digit timestamp here
err_msg = cfg.err_msg_supl.timestamp;
if '*' ~= flag then
if '*' ~= flag then
url=url:gsub ('(//web%.archive%.org/[^%d]*%d?%d?%d?%d?%d?%d?)[^/]*', '%1*', 1) -- for preview, modify ts to be yearmo* max (0-6 digits plus splat)
local replacement = timestamp:match ('^%d%d%d%d%d%d') or timestamp:match ('^%d%d%d%d'); -- get the first 6 (YYYYMM) or first 4 digits (YYYY)
if replacement then -- nil if there aren't at least 4 digits (year)
replacement = replacement .. string.rep ('0', 14 - replacement:len()); -- year or yearmo (4 or 6 digits) zero-fill to make 14-digit timestamp
url=url:gsub ('(//web%.archive%.org/[^%d]*)%d[^/]*', '%1' .. replacement .. '*', 1) -- for preview, modify ts to 14 digits plus splat for calendar display
end
end
end
elseif is_set(path) and 'web/' ~= path then -- older archive urls do not have the extra 'web/' path element
elseif utilities.is_set (path) and 'web/' ~= path then -- older archive URLs do not have the extra 'web/' path element
err_msg = 'path';
err_msg = cfg.err_msg_supl.path;
elseif is_set (flag) and not is_set (path) then -- flag not allowed with the old form url (without the 'web/' path element)
elseif utilities.is_set (flag) and not utilities.is_set (path) then -- flag not allowed with the old form URL (without the 'web/' path element)
err_msg = 'flag';
err_msg = cfg.err_msg_supl.flag;
elseif is_set (flag) and not flag:match ('%a%a_') then -- flag if present must be two alpha characters and underscore (requires 'web/' path element)
elseif utilities.is_set (flag) and not flag:match ('%a%a_') then -- flag if present must be two alpha characters and underscore (requires 'web/' path element)
err_msg = 'flag';
err_msg = cfg.err_msg_supl.flag;
else
else
return url, date; -- return archiveURL and ArchiveDate
return url, date, timestamp; -- return ArchiveURL, ArchiveDate, and timestamp from |archive-url=
end
end
end
end
-- if here, something not right so
-- if here, something not right so
table.insert( z.message_tail, { set_error( 'archive_url', {err_msg}, true ) } ); -- add error message and
utilities.set_message ('err_archive_url', {err_msg}); -- add error message and
if is_set (Frame:preprocess('{{REVISIONID}}')) then
 
return '', ''; -- return empty strings for archiveURL and ArchiveDate
if is_preview_mode then
return url, date, timestamp; -- preview mode so return ArchiveURL, ArchiveDate, and timestamp from |archive-url=
else
else
return url, date; -- preview mode so return archiveURL and ArchiveDate
return '', ''; -- return empty strings for ArchiveURL and ArchiveDate
end
end
end
end




--[[--------------------------< C I T A T I O N 0 >------------------------------------------------------------
--[[--------------------------< P L A C E _ C H E C K >--------------------------------------------------------
 
check |place=, |publication-place=, |location= to see if these params include digits.  This function added because
many editors misuse location to specify the in-source location (|page(s)= and |at= are supposed to do that)


This is the main function doing the majority of the citation formatting.
returns the original parameter value without modification; added maint cat when parameter value contains digits


]]
]]


local function citation0( config, args)
local function place_check (param_val)
--[[
if not utilities.is_set (param_val) then -- parameter empty or omitted
Load Input Parameters
return param_val; -- return that empty state
The argument_wrapper facilitates the mapping of multiple aliases to single internal variable.
]]
local A = argument_wrapper( args );
local i
 
-- Pick out the relevant fields from the arguments.  Different citation templates
-- define different field names for the same underlying things.
 
-- set default parameter values defined by |mode= parameter.
local Mode = A['Mode'];
if not is_valid_parameter_value (Mode, 'mode', cfg.keywords['mode']) then
Mode = '';
end
end
if mw.ustring.find (param_val, '%d') then -- not empty, are there digits in the parameter value
utilities.set_message ('maint_location'); -- yep, add maint cat
end
return param_val; -- and done
end


local author_etal;
local a = {}; -- authors list from |lastn= / |firstn= pairs or |vauthors=
local Authors;
local NameListFormat = A['NameListFormat'];
local Collaboration = A['Collaboration'];


do -- to limit scope of selected
--[[--------------------------< I S _ A R C H I V E D _ C O P Y >----------------------------------------------
local selected = select_author_editor_source (A['Vauthors'], A['Authors'], args, 'AuthorList');
if 1 == selected then
a, author_etal = extract_names (args, 'AuthorList'); -- fetch author list from |authorn= / |lastn= / |firstn=, |author-linkn=, and |author-maskn=
elseif 2 == selected then
NameListFormat = 'vanc'; -- override whatever |name-list-format= might be
a, author_etal = parse_vauthors_veditors (args, args.vauthors, 'AuthorList'); -- fetch author list from |vauthors=, |author-linkn=, and |author-maskn=
elseif 3 == selected then
Authors = A['Authors']; -- use content of |authors=
if 'authors' == A:ORIGIN('Authors') then -- but add a maint cat if the parameter is |authors=
add_maint_cat ('authors'); -- because use of this parameter is discouraged; what to do about the aliases is a TODO:
end
end
if is_set (Collaboration) then
author_etal = true; -- so that |display-authors=etal not required
end
end


local Others = A['Others'];
compares |title= to 'Archived copy' (placeholder added by bots that can't find proper title); if matches, return true; nil else


local editor_etal;
]]
local e = {}; -- editors list from |editor-lastn= / |editor-firstn= pairs or |veditors=
local Editors;


do -- to limit scope of selected
local function is_archived_copy (title)
local selected = select_author_editor_source (A['Veditors'], A['Editors'], args, 'EditorList');
title = mw.ustring.lower(title); -- switch title to lower case
if 1 == selected then
if title:find (cfg.special_case_translation.archived_copy.en) then -- if title is 'Archived copy'
e, editor_etal = extract_names (args, 'EditorList'); -- fetch editor list from |editorn= / |editor-lastn= / |editor-firstn=, |editor-linkn=, and |editor-maskn=
return true;
elseif 2 == selected then
elseif cfg.special_case_translation.archived_copy['local'] then
NameListFormat = 'vanc'; -- override whatever |name-list-format= might be
if mw.ustring.find (title, cfg.special_case_translation.archived_copy['local']) then -- mw.ustring() because might not be Latin script
e, editor_etal = parse_vauthors_veditors (args, args.veditors, 'EditorList'); -- fetch editor list from |veditors=, |editor-linkn=, and |editor-maskn=
return true;
elseif 3 == selected then
Editors = A['Editors']; -- use content of |editors=
add_maint_cat ('editors'); -- but add a maint cat because use of this parameter is discouraged
end
end
end
end
end


local translator_etal;
local t = {}; -- translators list from |translator-lastn= / translator-firstn= pairs
local Translators; -- assembled translators name list
t = extract_names (args, 'TranslatorList'); -- fetch translator list from |translatorn= / |translator-lastn=, -firstn=, -linkn=, -maskn=


local interviewer_etal;
--[[--------------------------< D I S P L A Y _ N A M E S _ S E L E C T >--------------------------------------
local interviewers_list = {};
 
local Interviewers; -- used later
for any of the |display-authors=, |display-editors=, etc parameters, select either the local or global setting.
interviewers_list = extract_names (args, 'InterviewerList'); -- process preferred interviewers parameters
When both are present, look at <local_display_names> value.  When the value is some sort of 'et al.'string,
special handling is required.
 
When {{cs1 config}} has |display-<namelist>= AND this template has |display-<namelist>=etal AND:
the number of names specified by <number_of_names> is:
greater than the number specified in the global |display-<namelist>= parameter (<global_display_names>)
use global |display-<namelist>= parameter value
set overridden maint category
less than or equal to the number specified in the global |display-<namelist>=  parameter
use local |display-<namelist>= parameter value
 
The purpose of this function is to prevent categorizing a template that has fewer names than the global setting
to keep the etal annotation specified by <local_display_names>.
 
]]
 
local function display_names_select (global_display_names, local_display_names, param_name, number_of_names, test)
if global_display_names and utilities.is_set (local_display_names) then -- when both
if 'etal' == local_display_names:lower():gsub("[ '%.]", '') then -- the :gsub() portion makes 'etal' from a variety of 'et al.' spellings and stylings
number_of_names = tonumber (number_of_names); -- convert these to numbers for comparison
local global_display_names_num = tonumber (global_display_names); -- <global_display_names> not set when parameter value is not digits


local contributor_etal;
if number_of_names > global_display_names_num then -- template has more names than global config allows to be displayed?
local c = {}; -- contributors list from |contributor-lastn= / contributor-firstn= pairs
utilities.set_message ('maint_overridden_setting'); -- set a maint message because global is overriding local |display-<namelist>=etal
local Contributors; -- assembled contributors name list
return global_display_names, 'cs1 config'; -- return global with spoof parameter name (for get_display_names())
local Contribution = A['Contribution'];
else
if in_array(config.CitationClass, {"book","citation"}) and not is_set(A['Periodical']) then -- |contributor= and |contribution= only supported in book cites
return local_display_names, param_name; -- return local because fewer names so let <local_display_names> control
c = extract_names (args, 'ContributorList'); -- fetch contributor list from |contributorn= / |contributor-lastn=, -firstn=, -linkn=, -maskn=
if 0 < #c then
if not is_set (Contribution) then -- |contributor= requires |contribution=
table.insert( z.message_tail, { set_error( 'contributor_missing_required_param', 'contribution')}); -- add missing contribution error message
c = {}; -- blank the contributors' table; it is used as a flag later
end
if 0 == #a then -- |contributor= requires |author=
table.insert( z.message_tail, { set_error( 'contributor_missing_required_param', 'author')}); -- add missing author error message
c = {}; -- blank the contributors' table; it is used as a flag later
end
end
end
end
else -- if not a book cite
-- here when <global_display_names> and <local_display_names> both numbers; <global_display_names> controls
if select_one (args, cfg.aliases['ContributorList-Last'], 'redundant_parameters', 1 ) then -- are there contributor name list parameters?
utilities.set_message ('maint_overridden_setting'); -- set a maint message
table.insert( z.message_tail, { set_error( 'contributor_ignored')}); -- add contributor ignored error message
return global_display_names, 'cs1 config'; -- return global with spoof parameter name (for get_display_names())
end
Contribution = nil; -- unset
end
end
 
-- here when only one of <global_display_names> or <local_display_names> set
if not is_valid_parameter_value (NameListFormat, 'name-list-format', cfg.keywords['name-list-format']) then -- only accepted value for this parameter is 'vanc'
if global_display_names then
NameListFormat = ''; -- anything else, set to empty string
return global_display_names, 'cs1 config'; -- return global with spoof parameter name (for get_display_names())
else
return local_display_names, param_name; -- return local
end
end
end


if is_set (Others) then
if 0 == #a and 0 == #e then -- add maint cat when |others= has value and used without |author=, |editor=
add_maint_cat ('others');
end
end


local Year = A['Year'];
--[[--------------------------< C I T A T I O N 0 >------------------------------------------------------------
local PublicationDate = A['PublicationDate'];
 
local OrigYear = A['OrigYear'];
This is the main function doing the majority of the citation formatting.
local Date = A['Date'];
local LayDate = A['LayDate'];
------------------------------------------------- Get title data
local Title = A['Title'];
local ScriptTitle = A['ScriptTitle'];
local BookTitle = A['BookTitle'];
local Conference = A['Conference'];
local TransTitle = A['TransTitle'];
local TitleNote = A['TitleNote'];
local TitleLink = A['TitleLink'];
link_title_ok (TitleLink, A:ORIGIN ('TitleLink'), Title, 'title'); -- check for wikimarkup in |title-link= or wikimarkup in |title= when |title-link= is set


local Chapter = A['Chapter'];
]]
local ScriptChapter = A['ScriptChapter'];
local ScriptChapterOrigin = A:ORIGIN ('ScriptChapter');
local ChapterLink -- = A['ChapterLink']; -- deprecated as a parameter but still used internally by cite episode
local TransChapter = A['TransChapter'];
local TitleType = A['TitleType'];
local Degree = A['Degree'];
local Docket = A['Docket'];
local ArchiveFormat = A['ArchiveFormat'];


local ArchiveDate;
local function citation0( config, args )
local ArchiveURL;
--[[
Load Input Parameters
The argument_wrapper facilitates the mapping of multiple aliases to single internal variable.
]]
local A = argument_wrapper ( args );
local i


ArchiveURL, ArchiveDate = archive_url_check (A['ArchiveURL'], A['ArchiveDate'])
-- Pick out the relevant fields from the arguments.  Different citation templates
-- define different field names for the same underlying things.
local UrlStatus = A['UrlStatus'];
 
if 'url-status' == A:ORIGIN ('UrlStatus') then -- interim: TODO: this line goes away
local author_etal;
if not is_valid_parameter_value (UrlStatus, 'url-status', cfg.keywords ['url-status']) then -- set in config.defaults to 'dead'
local a = {}; -- authors list from |lastn= / |firstn= pairs or |vauthors=
UrlStatus = ''; -- anything else, set to empty string
local Authors;
end -- interim: TODO: this line goes away
local NameListStyle;
elseif is_set (A:ORIGIN ('UrlStatus')) then -- interim: while both |dead-url= and |url-status= allowed use separate keyword lists; TODO: remove this section
if cfg.global_cs1_config_t['NameListStyle'] then -- global setting in {{cs1 config}} overrides local |name-list-style= parameter value; nil when empty or assigned value invalid
if not is_valid_parameter_value (UrlStatus, A:ORIGIN ('UrlStatus'), cfg.keywords ['deadurl']) then -- interim: assume |dead-url=; use those keywords; TODO: this line goes away
NameListStyle = is_valid_parameter_value (cfg.global_cs1_config_t['NameListStyle'], 'cs1 config: name-list-style', cfg.keywords_lists['name-list-style'], ''); -- error messaging 'param' here is a hoax
UrlStatus = ''; -- anything else, set to empty string; interim: TODO: this line goes away
else
end -- interim: TODO: this line goes away
NameListStyle = is_valid_parameter_value (A['NameListStyle'], A:ORIGIN('NameListStyle'), cfg.keywords_lists['name-list-style'], '');
end -- interim: TODO: this line goes away
end
 
if in_array (UrlStatus, {'yes', 'true', 'y'}) then -- TODO: remove this whole thing when |dead-url=[yes|no] parameters removed from articles
if cfg.global_cs1_config_t['NameListStyle'] and utilities.is_set (A['NameListStyle']) then -- when template has |name-list-style=<something> which global setting has overridden
UrlStatus = 'dead';
utilities.set_message ('maint_overridden_setting'); -- set a maint message
elseif 'no' == UrlStatus then
UrlStatus = 'live';
end
local URL = A['URL']
local URLorigin = A:ORIGIN('URL'); -- get name of parameter that holds URL
local ChapterURL = A['ChapterURL'];
local ChapterURLorigin = A:ORIGIN('ChapterURL'); -- get name of parameter that holds ChapterURL
local ConferenceFormat = A['ConferenceFormat'];
local ConferenceURL = A['ConferenceURL'];
local ConferenceURLorigin = A:ORIGIN('ConferenceURL'); -- get name of parameter that holds ConferenceURL
local Periodical = A['Periodical'];
local Periodical_origin = '';
if is_set (Periodical) then
Periodical_origin = A:ORIGIN('Periodical'); -- get the name of the periodical parameter
local i;
Periodical, i = strip_apostrophe_markup (Periodical); -- strip appostrophe markup so that metadata isn't contaminated
if i then -- non-zero when markup was stripped so emit an error message
table.insert( z.message_tail, {set_error ('apostrophe_markup', {Periodical_origin}, true)});
end
end
end


local ScriptPeriodical = A['ScriptPeriodical'];
local Collaboration = A['Collaboration'];
local ScriptPeriodical_origin = A:ORIGIN('ScriptPeriodical');


if not (is_set (Periodical) or is_set (ScriptPeriodical)) then -- 'periodical' templates require periodical parameter
do -- to limit scope of selected
local p = {['journal'] = 'journal', ['magazine'] = 'magazine', ['news'] = 'newspaper', ['web'] = 'website'}; -- for error message
local selected = select_author_editor_source (A['Vauthors'], A['Authors'], args, 'AuthorList');
if p[config.CitationClass]  then
if 1 == selected then
table.insert( z.message_tail, {set_error ('missing_periodical', {config.CitationClass, p[config.CitationClass]}, true)});
a, author_etal = extract_names (args, 'AuthorList'); -- fetch author list from |authorn= / |lastn= / |firstn=, |author-linkn=, and |author-maskn=
end
elseif 2 == selected then
NameListStyle = 'vanc'; -- override whatever |name-list-style= might be
a, author_etal = parse_vauthors_veditors (args, A['Vauthors'], 'AuthorList'); -- fetch author list from |vauthors=, |author-linkn=, and |author-maskn=
elseif 3 == selected then
Authors = A['Authors']; -- use content of |people= or |credits=; |authors= is deprecated; TODO: constrain |people= and |credits= to cite av media, episode, serial?
end
if utilities.is_set (Collaboration) then
author_etal = true; -- so that |display-authors=etal not required
end
end
end


local TransPeriodical = A['TransPeriodical'];
local editor_etal;
local e = {}; -- editors list from |editor-lastn= / |editor-firstn= pairs or |veditors=


local Series = A['Series'];
do -- to limit scope of selected
local selected = select_author_editor_source (A['Veditors'], nil, args, 'EditorList'); -- support for |editors= withdrawn
local Volume;
if 1 == selected then
local Issue;
e, editor_etal = extract_names (args, 'EditorList'); -- fetch editor list from |editorn= / |editor-lastn= / |editor-firstn=, |editor-linkn=, and |editor-maskn=
local Page;
elseif 2 == selected then
local Pages;
NameListStyle = 'vanc'; -- override whatever |name-list-style= might be
local At;
e, editor_etal = parse_vauthors_veditors (args, args.veditors, 'EditorList'); -- fetch editor list from |veditors=, |editor-linkn=, and |editor-maskn=
 
end
if 'citation' == config.CitationClass then
end
if is_set (Periodical) then
if not in_array (Periodical_origin, {'website', 'mailinglist'}) then -- {{citation}} does not render volume for these 'periodicals'
local Chapter = A['Chapter']; -- done here so that we have access to |contribution= from |chapter= aliases
Volume = A['Volume']; -- but does for all other 'periodicals'
local Chapter_origin = A:ORIGIN ('Chapter');
local Contribution; -- because contribution is required for contributor(s)
if 'contribution' == Chapter_origin then
Contribution = Chapter; -- get the name of the contribution
end
local c = {}; -- contributors list from |contributor-lastn= / contributor-firstn= pairs
if utilities.in_array (config.CitationClass, {"book", "citation"}) and not utilities.is_set (A['Periodical']) then -- |contributor= and |contribution= only supported in book cites
c = extract_names (args, 'ContributorList'); -- fetch contributor list from |contributorn= / |contributor-lastn=, -firstn=, -linkn=, -maskn=
if 0 < #c then
if not utilities.is_set (Contribution) then -- |contributor= requires |contribution=
utilities.set_message ('err_contributor_missing_required_param', 'contribution'); -- add missing contribution error message
c = {}; -- blank the contributors' table; it is used as a flag later
end
end
elseif is_set (ScriptPeriodical) then
if 0 == #a then -- |contributor= requires |author=
if 'script-website' ~= ScriptPeriodical_origin then -- {{citation}} does not render volume for |script-website=
utilities.set_message ('err_contributor_missing_required_param', 'author'); -- add missing author error message
Volume = A['Volume']; -- but does for all other 'periodicals'
c = {}; -- blank the contributors' table; it is used as a flag later
end
end
else
Volume = A['Volume']; -- and does for non-'periodical' cites
end
end
elseif in_array (config.CitationClass, cfg.templates_using_volume) then -- render |volume= for cs1 according to the configuration settings
else -- if not a book cite
Volume = A['Volume'];
if utilities.select_one (args, cfg.aliases['ContributorList-Last'], 'err_redundant_parameters', 1 ) then -- are there contributor name list parameters?
end
utilities.set_message ('err_contributor_ignored'); -- add contributor ignored error message
 
if 'citation' == config.CitationClass then
if is_set (Periodical) and in_array (Periodical_origin, {'journal', 'magazine', 'newspaper', 'periodical', 'work'}) or -- {{citation}} renders issue for these 'periodicals'
is_set (ScriptPeriodical) and in_array (ScriptPeriodical_origin, {'script-journal', 'script-magazine', 'script-newspaper', 'script-periodical', 'script-work'}) then -- and these 'script-periodicals'
Issue = hyphen_to_dash (A['Issue']);
end
elseif in_array (config.CitationClass, cfg.templates_using_issue) then -- conference & map books do not support issue; {{citation}} listed here because included in settings table
if not (in_array (config.CitationClass, {'conference', 'map', 'citation'}) and not (is_set (Periodical) or is_set (ScriptPeriodical))) then
Issue = hyphen_to_dash (A['Issue']);
end
end
Contribution = nil; -- unset
end
end


local Position = '';
local Title = A['Title'];
if not in_array (config.CitationClass, cfg.templates_not_using_page) then
local TitleLink = A['TitleLink'];
Page = A['Page'];
 
Pages = hyphen_to_dash (A['Pages']);
local auto_select = ''; -- default is auto
At = A['At'];
local accept_link;
TitleLink, accept_link = utilities.has_accept_as_written (TitleLink, true); -- test for accept-this-as-written markup
if (not accept_link) and utilities.in_array (TitleLink, {'none', 'pmc', 'doi'}) then -- check for special keywords
auto_select = TitleLink; -- remember selection for later
TitleLink = ''; -- treat as if |title-link= would have been empty
end
 
TitleLink = link_title_ok (TitleLink, A:ORIGIN ('TitleLink'), Title, 'title'); -- check for wiki-markup in |title-link= or wiki-markup in |title= when |title-link= is set
 
local Section = ''; -- {{cite map}} only; preset to empty string for concatenation if not used
if 'map' == config.CitationClass and 'section' == Chapter_origin then
Section = A['Chapter']; -- get |section= from |chapter= alias list; |chapter= and the other aliases not supported in {{cite map}}
Chapter = ''; -- unset for now; will be reset later from |map= if present
end
end


local Edition = A['Edition'];
local Periodical = A['Periodical'];
local PublicationPlace = A['PublicationPlace']
local Periodical_origin = A:ORIGIN('Periodical');
local Place = A['Place'];
local ScriptPeriodical = A['ScriptPeriodical'];
local ScriptPeriodical_origin = A:ORIGIN('ScriptPeriodical');
local TransPeriodical =  A['TransPeriodical'];
local TransPeriodical_origin =  A:ORIGIN ('TransPeriodical');
local PublisherName = A['PublisherName'];
if (utilities.in_array (config.CitationClass, {'book', 'encyclopaedia'}) and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical) or utilities.is_set (TransPeriodical))) then
local PublisherName_origin = A:ORIGIN('PublisherName');
local param;
if is_set (PublisherName) then
if utilities.is_set (Periodical) then -- get a parameter name from one of these periodical related meta-parameters
local i=0;
Periodical = ''; -- unset because not valid {{cite book}} or {{cite encyclopedia}} parameters
PublisherName, i = strip_apostrophe_markup (PublisherName); -- strip appostrophe markup so that metadata isn't contaminated; publisher is never italicized
param = Periodical_origin -- get parameter name for error messaging
 
elseif utilities.is_set (TransPeriodical) then
if i then -- non-zero when markup was stripped so emit an error message
TransPeriodical = ''; -- unset because not valid {{cite book}} or {{cite encyclopedia}} parameters
table.insert( z.message_tail, {set_error ('apostrophe_markup', {PublisherName_origin}, true)});
param = TransPeriodical_origin; -- get parameter name for error messaging
end
elseif utilities.is_set (ScriptPeriodical) then
ScriptPeriodical = ''; -- unset because not valid {{cite book}} or {{cite encyclopedia}} parameters
param = ScriptPeriodical_origin; -- get parameter name for error messaging
end
end


local RegistrationRequired = A['RegistrationRequired'];
if utilities.is_set (param) then -- if we found one
if not is_valid_parameter_value (RegistrationRequired, 'registration', cfg.keywords ['yes_true_y']) then
utilities.set_message ('err_periodical_ignored', {param}); -- emit an error message
RegistrationRequired=nil;
end
end
end


local SubscriptionRequired = A['SubscriptionRequired'];
if utilities.is_set (Periodical) then
if not is_valid_parameter_value (SubscriptionRequired, 'subscription', cfg.keywords ['yes_true_y']) then
local i;
SubscriptionRequired=nil;
Periodical, i = utilities.strip_apostrophe_markup (Periodical); -- strip apostrophe markup so that metadata isn't contaminated
if i then -- non-zero when markup was stripped so emit an error message
utilities.set_message ('err_apostrophe_markup', {Periodical_origin});
end
end
end


local UrlAccess = A['UrlAccess'];
if 'mailinglist' == config.CitationClass then -- special case for {{cite mailing list}}
if not is_valid_parameter_value (UrlAccess, 'url-access', cfg.keywords ['url-access']) then
if utilities.is_set (Periodical) and utilities.is_set (A ['MailingList']) then -- both set emit an error TODO: make a function for this and similar?
UrlAccess = nil;
utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', Periodical_origin) .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'mailinglist')});
end
if not is_set(URL) and is_set(UrlAccess) then
UrlAccess = nil;
table.insert( z.message_tail, { set_error( 'param_access_requires_param', {'url'}, true ) } );
end
if is_set (UrlAccess) and is_set (SubscriptionRequired) then -- while not aliases, these are much the same so if both are set
table.insert( z.message_tail, { set_error( 'redundant_parameters', {wrap_style ('parameter', 'url-access') .. ' and ' .. wrap_style ('parameter', 'subscription')}, true ) } ); -- add error message
SubscriptionRequired = nil; -- unset; prefer |access= over |subscription=
end
if is_set (UrlAccess) and is_set (RegistrationRequired) then -- these are not the same but contradictory so if both are set
table.insert( z.message_tail, { set_error( 'redundant_parameters', {wrap_style ('parameter', 'url-access') .. ' and ' .. wrap_style ('parameter', 'registration')}, true ) } ); -- add error message
RegistrationRequired = nil; -- unset; prefer |access= over |registration=
end
end


local ChapterUrlAccess = A['ChapterUrlAccess'];
Periodical = A ['MailingList']; -- error or no, set Periodical to |mailinglist= value because this template is {{cite mailing list}}
if not is_valid_parameter_value (ChapterUrlAccess, A:ORIGIN('ChapterUrlAccess'), cfg.keywords ['url-access']) then -- same as url-access
Periodical_origin = A:ORIGIN('MailingList');
ChapterUrlAccess = nil;
end
end
if not is_set(ChapterURL) and is_set(ChapterUrlAccess) then
ChapterUrlAccess = nil;
table.insert( z.message_tail, { set_error( 'param_access_requires_param', {A:ORIGIN('ChapterUrlAccess'):gsub ('%-access', '')}, true ) } );
end


local MapUrlAccess = A['MapUrlAccess'];
-- web and news not tested for now because of
if not is_valid_parameter_value (MapUrlAccess, 'map-url-access', cfg.keywords ['url-access']) then
-- Wikipedia:Administrators%27_noticeboard#Is_there_a_semi-automated_tool_that_could_fix_these_annoying_"Cite_Web"_errors?
MapUrlAccess = nil;
if not (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) then -- 'periodical' templates require periodical parameter
-- local p = {['journal'] = 'journal', ['magazine'] = 'magazine', ['news'] = 'newspaper', ['web'] = 'website'}; -- for error message
local p = {['journal'] = 'journal', ['magazine'] = 'magazine'}; -- for error message
if p[config.CitationClass]  then
utilities.set_message ('err_missing_periodical', {config.CitationClass, p[config.CitationClass]});
end
end
if not is_set(A['MapURL']) and is_set(MapUrlAccess) then
end
MapUrlAccess = nil;
table.insert( z.message_tail, { set_error( 'param_access_requires_param', {'map-url'}, true ) } );
local Volume;
end
if 'citation' == config.CitationClass then
 
if utilities.is_set (Periodical) then
local Via = A['Via'];
if not utilities.in_array (Periodical_origin, cfg.citation_no_volume_t) then -- {{citation}} does not render |volume= when these parameters are used
local AccessDate = A['AccessDate'];
Volume = A['Volume']; -- but does for all other 'periodicals'
local Agency = A['Agency'];
end
 
elseif utilities.is_set (ScriptPeriodical) then
local Language = A['Language'];
if 'script-website' ~= ScriptPeriodical_origin then -- {{citation}} does not render volume for |script-website=
local Format = A['Format'];
Volume = A['Volume']; -- but does for all other 'periodicals'
local ChapterFormat = A['ChapterFormat'];
end
local DoiBroken = A['DoiBroken'];
else
local ID = A['ID'];
Volume = A['Volume']; -- and does for non-'periodical' cites
local ASINTLD = A['ASINTLD'];
local IgnoreISBN = A['IgnoreISBN'];
if not is_valid_parameter_value (IgnoreISBN, 'ignore-isbn-error', cfg.keywords ['yes_true_y']) then
IgnoreISBN = nil; -- anything else, set to empty string
end
end
local Embargo = A['Embargo'];
elseif utilities.in_array (config.CitationClass, cfg.templates_using_volume) then -- render |volume= for cs1 according to the configuration settings
local Class = A['Class']; -- arxiv class identifier
Volume = A['Volume'];
end
extra_text_in_vol_iss_check (Volume, A:ORIGIN ('Volume'), 'v');


local ID_list = extract_ids( args );
local Issue;
if is_set (DoiBroken) and not ID_list['DOI'] then
if 'citation' == config.CitationClass then
table.insert( z.message_tail, { set_error( 'doibroken_missing_doi', A:ORIGIN('DoiBroken'))});
if utilities.is_set (Periodical) and utilities.in_array (Periodical_origin, cfg.citation_issue_t) then -- {{citation}} may render |issue= when these parameters are used
Issue = utilities.hyphen_to_dash (A['Issue']);
end
end
local ID_access_levels = extract_id_access_levels( args, ID_list );
elseif utilities.in_array (config.CitationClass, cfg.templates_using_issue) then -- conference & map books do not support issue; {{citation}} listed here because included in settings table
 
if not (utilities.in_array (config.CitationClass, {'conference', 'map', 'citation'}) and not (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical))) then
local Quote = A['Quote'];
Issue = utilities.hyphen_to_dash (A['Issue']);
 
local LayFormat = A['LayFormat'];
local LayURL = A['LayURL'];
local LaySource = A['LaySource'];
local Transcript = A['Transcript'];
local TranscriptFormat = A['TranscriptFormat'];
local TranscriptURL = A['TranscriptURL']
local TranscriptURLorigin = A:ORIGIN('TranscriptURL'); -- get name of parameter that holds TranscriptURL
 
local LastAuthorAmp = A['LastAuthorAmp'];
if not is_valid_parameter_value (LastAuthorAmp, 'last-author-amp', cfg.keywords ['yes_true_y']) then
LastAuthorAmp = nil; -- set to empty string
end
end
end
local ArticleNumber;


local no_tracking_cats = A['NoTracking'];
if utilities.in_array (config.CitationClass, {'journal', 'conference'}) or ('citation' == config.CitationClass and utilities.is_set (Periodical) and 'journal' == Periodical_origin) then
if not is_valid_parameter_value (no_tracking_cats, 'no-tracking', cfg.keywords ['yes_true_y']) then
ArticleNumber = A['ArticleNumber'];
no_tracking_cats = nil; -- set to empty string
end
end


--local variables that are not cs1 parameters
extra_text_in_vol_iss_check (Issue, A:ORIGIN ('Issue'), 'i');
local use_lowercase; -- controls capitalization of certain static text
local this_page = mw.title.getCurrentTitle(); -- also used for COinS and for language
local anchor_year; -- used in the CITEREF identifier
local COinS_date = {}; -- holds date info extracted from |date= for the COinS metadata by Module:Date verification


local DF = A['DF']; -- date format set in cs1|2 template
local Page;
if not is_valid_parameter_value (DF, 'df', cfg.keywords['date-format']) then -- validate reformatting keyword
local Pages;
DF = ''; -- not valid, set to empty string
local At;
local QuotePage;
local QuotePages;
if not utilities.in_array (config.CitationClass, cfg.templates_not_using_page) then -- TODO: rewrite to emit ignored parameter error message?
Page = A['Page'];
Pages = utilities.hyphen_to_dash (A['Pages']);
At = A['At'];
QuotePage = A['QuotePage'];
QuotePages = utilities.hyphen_to_dash (A['QuotePages']);
end
end
if not is_set (DF) then
 
DF = cfg.global_df;
local Edition = A['Edition'];
local PublicationPlace = place_check (A['PublicationPlace'], A:ORIGIN('PublicationPlace'));
local Place = place_check (A['Place'], A:ORIGIN('Place'));
local PublisherName = A['PublisherName'];
local PublisherName_origin = A:ORIGIN('PublisherName');
if utilities.is_set (PublisherName) and (cfg.keywords_xlate['none'] ~= PublisherName) then
local i = 0;
PublisherName, i = utilities.strip_apostrophe_markup (PublisherName); -- strip apostrophe markup so that metadata isn't contaminated; publisher is never italicized
if i and (0 < i) then -- non-zero when markup was stripped so emit an error message
utilities.set_message ('err_apostrophe_markup', {PublisherName_origin});
end
end
if ('document' == config.CitationClass) and not utilities.is_set (PublisherName) then
utilities.set_message ('err_missing_publisher', {config.CitationClass, 'publisher'});
end
end


local sepc; -- separator between citation elements for CS1 a period, for CS2, a comma
local Newsgroup = A['Newsgroup']; -- TODO: strip apostrophe markup?
local PostScript;
local Newsgroup_origin = A:ORIGIN('Newsgroup');
local Ref;
 
sepc, PostScript, Ref = set_style (Mode:lower(), A['PostScript'], A['Ref'], config.CitationClass);
if 'newsgroup' == config.CitationClass then
use_lowercase = ( sepc == ',' ); -- used to control capitalization for certain static text
if utilities.is_set (PublisherName) and (cfg.keywords_xlate['none'] ~= PublisherName) then -- general use parameter |publisher= not allowed in cite newsgroup
utilities.set_message ('err_parameter_ignored', {PublisherName_origin});
end
 
PublisherName = nil; -- ensure that this parameter is unset for the time being; will be used again after COinS
end


--check this page to see if it is in one of the namespaces that cs1 is not supposed to add to the error categories
local URL = A['URL']; -- TODO: better way to do this for URL, ChapterURL, and MapURL?
if not is_set (no_tracking_cats) then -- ignore if we are already not going to categorize this page
local UrlAccess = is_valid_parameter_value (A['UrlAccess'], A:ORIGIN('UrlAccess'), cfg.keywords_lists['url-access'], nil);
if in_array (this_page.nsText, cfg.uncategorized_namespaces) then
if not utilities.is_set (URL) and utilities.is_set (UrlAccess) then
UrlAccess = nil;
utilities.set_message ('err_param_access_requires_param', 'url');
end
local ChapterURL = A['ChapterURL'];
local ChapterUrlAccess = is_valid_parameter_value (A['ChapterUrlAccess'], A:ORIGIN('ChapterUrlAccess'), cfg.keywords_lists['url-access'], nil);
if not utilities.is_set (ChapterURL) and utilities.is_set (ChapterUrlAccess) then
ChapterUrlAccess = nil;
utilities.set_message ('err_param_access_requires_param', {A:ORIGIN('ChapterUrlAccess'):gsub ('%-access', '')});
end
 
local MapUrlAccess = is_valid_parameter_value (A['MapUrlAccess'], A:ORIGIN('MapUrlAccess'), cfg.keywords_lists['url-access'], nil);
if not utilities.is_set (A['MapURL']) and utilities.is_set (MapUrlAccess) then
MapUrlAccess = nil;
utilities.set_message ('err_param_access_requires_param', {'map-url'});
end
 
local this_page = mw.title.getCurrentTitle(); -- also used for COinS and for language
local no_tracking_cats = is_valid_parameter_value (A['NoTracking'], A:ORIGIN('NoTracking'), cfg.keywords_lists['yes_true_y'], nil);
 
-- check this page to see if it is in one of the namespaces that cs1 is not supposed to add to the error categories
if not utilities.is_set (no_tracking_cats) then -- ignore if we are already not going to categorize this page
if cfg.uncategorized_namespaces[this_page.namespace] then -- is this page's namespace id one of the uncategorized namespace ids?
no_tracking_cats = "true"; -- set no_tracking_cats
no_tracking_cats = "true"; -- set no_tracking_cats
end
end
for _,v in ipairs (cfg.uncategorized_subpages) do -- cycle through page name patterns
for _, v in ipairs (cfg.uncategorized_subpages) do -- cycle through page name patterns
if this_page.text:match (v) then -- test page name against each pattern
if this_page.text:match (v) then -- test page name against each pattern
no_tracking_cats = "true"; -- set no_tracking_cats
no_tracking_cats = "true"; -- set no_tracking_cats
Line 2,535: Line 2,873:
end
end
end
end
 
-- check for extra |page=, |pages= or |at= parameters. (also sheet and sheets while we're at it)
-- check for extra |page=, |pages= or |at= parameters. (also sheet and sheets while we're at it)
utilities.select_one (args, {'page', 'p', 'pp', 'pages', 'at', 'sheet', 'sheets'}, 'err_redundant_parameters'); -- this is a dummy call simply to get the error message and category
select_one( args, {'page', 'p', 'pp', 'pages', 'at', 'sheet', 'sheets'}, 'redundant_parameters' ); -- this is a dummy call simply to get the error message and category


local coins_pages;
local coins_pages;
Page, Pages, At, coins_pages = insource_loc_get (Page, Pages, At);
Page, Pages, At, coins_pages = insource_loc_get (Page, A:ORIGIN('Page'), Pages, A:ORIGIN('Pages'), At);


local NoPP = A['NoPP']  
local NoPP = is_valid_parameter_value (A['NoPP'], A:ORIGIN('NoPP'), cfg.keywords_lists['yes_true_y'], nil);
if is_set (NoPP) and is_valid_parameter_value (NoPP, 'nopp', cfg.keywords ['yes_true_y']) then
NoPP = true;
else
NoPP = nil; -- unset, used as a flag later
end


if not is_set(PublicationPlace) and is_set(Place) then -- both |publication-place= and |place= (|location=) allowed if different
if utilities.is_set (PublicationPlace) and utilities.is_set (Place) then -- both |publication-place= and |place= (|location=) allowed if different
utilities.add_prop_cat ('location-test'); -- add property cat to evaluate how often PublicationPlace and Place are used together
if PublicationPlace == Place then
Place = ''; -- unset; don't need both if they are the same
end
elseif not utilities.is_set (PublicationPlace) and utilities.is_set (Place) then -- when only |place= (|location=) is set ...
PublicationPlace = Place; -- promote |place= (|location=) to |publication-place
PublicationPlace = Place; -- promote |place= (|location=) to |publication-place
end
end
 
if PublicationPlace == Place then Place = ''; end -- don't need both if they are the same
if PublicationPlace == Place then Place = ''; end -- don't need both if they are the same
local URL_origin = A:ORIGIN('URL'); -- get name of parameter that holds URL
local ChapterURL_origin = A:ORIGIN('ChapterURL'); -- get name of parameter that holds ChapterURL
local ScriptChapter = A['ScriptChapter'];
local ScriptChapter_origin = A:ORIGIN ('ScriptChapter');
local Format = A['Format'];
local ChapterFormat = A['ChapterFormat'];
local TransChapter = A['TransChapter'];
local TransChapter_origin = A:ORIGIN ('TransChapter');
local TransTitle = A['TransTitle'];
local ScriptTitle = A['ScriptTitle'];
--[[
--[[
Parameter remapping for cite encyclopedia:
Parameter remapping for cite encyclopedia:
When the citation has these parameters:
When the citation has these parameters:
|encyclopedia and |title then map |title to |article and |encyclopedia to |title
|encyclopedia= and |title= then map |title= to |article= and |encyclopedia= to |title= for rendering
|encyclopedia and |article then map |encyclopedia to |title
|encyclopedia= and |article= then map |encyclopedia= to |title= for rendering


|trans-title maps to |trans-chapter when |title is re-mapped
|trans-title= maps to |trans-chapter= when |title= is re-mapped
|url maps to |chapterurl when |title is remapped
|url= maps to |chapter-url= when |title= is remapped
All other combinations of |encyclopedia, |title, and |article are not modified
All other combinations of |encyclopedia=, |title=, and |article= are not modified
]]
]]


local Encyclopedia = A['Encyclopedia'];
local Encyclopedia = A['Encyclopedia']; -- used as a flag by this module and by ~/COinS


if ( config.CitationClass == "encyclopaedia" ) or ( config.CitationClass == "citation" and is_set (Encyclopedia)) then -- test code for citation
if utilities.is_set (Encyclopedia) then -- emit error message when Encyclopedia set but template is other than {{cite encyclopedia}} or {{citation}}
if is_set (Periodical) then -- Periodical is set when |encyclopedia is set
if 'encyclopaedia' ~= config.CitationClass and 'citation' ~= config.CitationClass then
if is_set(Title) or is_set (ScriptTitle) then
utilities.set_message ('err_parameter_ignored', {A:ORIGIN ('Encyclopedia')});
if not is_set(Chapter) then
Encyclopedia = nil; -- unset because not supported by this template
Chapter = Title; -- |encyclopedia and |title are set so map |title to |article and |encyclopedia to |title
end
end
 
if ('encyclopaedia' == config.CitationClass) or ('citation' == config.CitationClass and utilities.is_set (Encyclopedia)) then
if utilities.is_set (Periodical) and utilities.is_set (Encyclopedia) then -- when both parameters set emit an error message; {{citation}} only; Periodical not allowed in {{cite encyclopedia}}
utilities.set_message ('err_periodical_ignored', {Periodical_origin});
end
 
if utilities.is_set (Encyclopedia) then
Periodical = Encyclopedia; -- error or no, set Periodical to Encyclopedia for rendering; {{citation}} could (not legitimately) have both; use Encyclopedia
Periodical_origin = A:ORIGIN ('Encyclopedia');
 
if utilities.is_set (Title) or utilities.is_set (ScriptTitle) then
if not utilities.is_set (Chapter) then
Chapter = Title; -- |encyclopedia= and |title= are set so map |title= to |article= and |encyclopedia= to |title= for rendering
ScriptChapter = ScriptTitle;
ScriptChapter = ScriptTitle;
ScriptChapterOrigin = 'title';
ScriptChapter_origin = A:ORIGIN('ScriptTitle')
TransChapter = TransTitle;
TransChapter = TransTitle;
ChapterURL = URL;
ChapterURL = URL;
ChapterURL_origin = URL_origin;
ChapterUrlAccess = UrlAccess;
ChapterUrlAccess = UrlAccess;


if not is_set (ChapterURL) and is_set (TitleLink) then
if not utilities.is_set (ChapterURL) and utilities.is_set (TitleLink) then
Chapter = make_wikilink (TitleLink, Chapter);
Chapter = utilities.make_wikilink (TitleLink, Chapter);
end
end
Title = Periodical;
Title = Periodical;
Line 2,594: Line 2,958:
ScriptTitle = '';
ScriptTitle = '';
end
end
elseif is_set (Chapter) then -- |title not set
elseif utilities.is_set (Chapter) or utilities.is_set (ScriptChapter) then -- |title= not set
Title = Periodical; -- |encyclopedia set and |article set so map |encyclopedia to |title
Title = Periodical; -- |encyclopedia= set and |article= set so map |encyclopedia= to |title= for rendering
Periodical = ''; -- redundant so unset
Periodical = ''; -- redundant so unset
end
end
Line 2,601: Line 2,965:
end
end


-- Special case for cite techreport.
-- special case for cite techreport.
local ID = A['ID'];
if (config.CitationClass == "techreport") then -- special case for cite techreport
if (config.CitationClass == "techreport") then -- special case for cite techreport
if is_set(A['Number']) then -- cite techreport uses 'number', which other citations alias to 'issue'
if utilities.is_set (A['Number']) then -- cite techreport uses 'number', which other citations alias to 'issue'
if not is_set(ID) then -- can we use ID for the "number"?
if not utilities.is_set (ID) then -- can we use ID for the "number"?
ID = A['Number']; -- yes, use it
ID = A['Number']; -- yes, use it
else -- ID has a value so emit error message
else -- ID has a value so emit error message
table.insert( z.message_tail, { set_error('redundant_parameters', {wrap_style ('parameter', 'id') .. ' and ' .. wrap_style ('parameter', 'number')}, true )});
utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'id') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'number')});
end
end
end
end
end
-- special case for cite mailing list
if (config.CitationClass == "mailinglist") then
Periodical = A ['MailingList'];
elseif 'mailinglist' == Periodical_origin then
Periodical = ''; -- unset because mailing list is only used for cite mailing list
end
end


-- Account for the oddity that is {{cite conference}}, before generation of COinS data.
-- Account for the oddity that is {{cite conference}}, before generation of COinS data.
local ChapterLink -- = A['ChapterLink']; -- deprecated as a parameter but still used internally by cite episode
local Conference = A['Conference'];
local BookTitle = A['BookTitle'];
local TransTitle_origin = A:ORIGIN ('TransTitle');
if 'conference' == config.CitationClass then
if 'conference' == config.CitationClass then
if is_set(BookTitle) then
if utilities.is_set (BookTitle) then
Chapter = Title;
Chapter = Title;
-- ChapterLink = TitleLink; -- |chapterlink= is deprecated
Chapter_origin = 'title';
-- ChapterLink = TitleLink; -- |chapter-link= is deprecated
ChapterURL = URL;
ChapterURL = URL;
ChapterUrlAccess = UrlAccess;
ChapterUrlAccess = UrlAccess;
ChapterURLorigin = URLorigin;
ChapterURL_origin = URL_origin;
URLorigin = '';
URL_origin = '';
ChapterFormat = Format;
ChapterFormat = Format;
TransChapter = TransTitle;
TransChapter = TransTitle;
TransChapter_origin = TransTitle_origin;
Title = BookTitle;
Title = BookTitle;
Format = '';
Format = '';
-- TitleLink = '';
-- TitleLink = '';
TransTitle = '';
TransTitle = '';
URL = '';
URL = '';
Line 2,639: Line 3,003:
Conference = ''; -- not cite conference or cite speech so make sure this is empty string
Conference = ''; -- not cite conference or cite speech so make sure this is empty string
end
end
-- CS1/2 mode
local Mode;
if cfg.global_cs1_config_t['Mode'] then -- global setting in {{cs1 config}} overrides local |mode= parameter value; nil when empty or assigned value invalid
Mode = is_valid_parameter_value (cfg.global_cs1_config_t['Mode'], 'cs1 config: mode', cfg.keywords_lists['mode'], ''); -- error messaging 'param' here is a hoax
else
Mode = is_valid_parameter_value (A['Mode'], A:ORIGIN('Mode'), cfg.keywords_lists['mode'], '');
end
if cfg.global_cs1_config_t['Mode'] and utilities.is_set (A['Mode']) then -- when template has |mode=<something> which global setting has overridden
utilities.set_message ('maint_overridden_setting'); -- set a maint message
end


-- separator character and postscript
local sepc, PostScript = set_style (Mode:lower(), A['PostScript'], config.CitationClass);
-- controls capitalization of certain static text
local use_lowercase = ( sepc == ',' );
-- cite map oddities
-- cite map oddities
local Cartography = "";
local Cartography = "";
Line 2,646: Line 3,027:
local Sheets = A['Sheets'] or '';
local Sheets = A['Sheets'] or '';
if config.CitationClass == "map" then
if config.CitationClass == "map" then
if utilities.is_set (Chapter) then --TODO: make a function for this and similar?
utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'map') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', Chapter_origin)}); -- add error message
end
Chapter = A['Map'];
Chapter = A['Map'];
Chapter_origin = A:ORIGIN('Map');
ChapterURL = A['MapURL'];
ChapterURL = A['MapURL'];
ChapterURL_origin = A:ORIGIN('MapURL');
TransChapter = A['TransMap'];
ScriptChapter = A['ScriptMap']
ScriptChapter_origin = A:ORIGIN('ScriptMap')
ChapterUrlAccess = MapUrlAccess;
ChapterUrlAccess = MapUrlAccess;
TransChapter = A['TransMap'];
ChapterURLorigin = A:ORIGIN('MapURL');
ChapterFormat = A['MapFormat'];
ChapterFormat = A['MapFormat'];
 
Cartography = A['Cartography'];
Cartography = A['Cartography'];
if is_set( Cartography ) then
if utilities.is_set ( Cartography ) then
Cartography = sepc .. " " .. wrap_msg ('cartography', Cartography, use_lowercase);
Cartography = sepc .. " " .. wrap_msg ('cartography', Cartography, use_lowercase);
end
end
Scale = A['Scale'];
Scale = A['Scale'];
if is_set( Scale ) then
if utilities.is_set ( Scale ) then
Scale = sepc .. " " .. Scale;
Scale = sepc .. " " .. Scale;
end
end
Line 2,664: Line 3,052:


-- Account for the oddities that are {{cite episode}} and {{cite serial}}, before generation of COinS data.
-- Account for the oddities that are {{cite episode}} and {{cite serial}}, before generation of COinS data.
local Series = A['Series'];
if 'episode' == config.CitationClass or 'serial' == config.CitationClass then
if 'episode' == config.CitationClass or 'serial' == config.CitationClass then
local AirDate = A['AirDate'];
local SeriesLink = A['SeriesLink'];
local SeriesLink = A['SeriesLink'];


link_title_ok (SeriesLink, A:ORIGIN ('SeriesLink'), Series, 'series'); -- check for wikimarkup in |series-link= or wikimarkup in |series= when |series-link= is set
SeriesLink = link_title_ok (SeriesLink, A:ORIGIN ('SeriesLink'), Series, 'series'); -- check for wiki-markup in |series-link= or wiki-markup in |series= when |series-link= is set


local Network = A['Network'];
local Network = A['Network'];
Line 2,674: Line 3,062:
local s, n = {}, {};
local s, n = {}, {};
-- do common parameters first
-- do common parameters first
if is_set(Network) then table.insert(n, Network); end
if utilities.is_set (Network) then table.insert(n, Network); end
if is_set(Station) then table.insert(n, Station); end
if utilities.is_set (Station) then table.insert(n, Station); end
ID = table.concat(n, sepc .. ' ');
ID = table.concat(n, sepc .. ' ');
if not is_set (Date) and is_set (AirDate) then -- promote airdate to date
if 'episode' == config.CitationClass then -- handle the oddities that are strictly {{cite episode}}
Date = AirDate;
end
 
if 'episode' == config.CitationClass then -- handle the oddities that are strictly {{cite episode}}
local Season = A['Season'];
local Season = A['Season'];
local SeriesNumber = A['SeriesNumber'];
local SeriesNumber = A['SeriesNumber'];


if is_set (Season) and is_set (SeriesNumber) then -- these are mutually exclusive so if both are set
if utilities.is_set (Season) and utilities.is_set (SeriesNumber) then -- these are mutually exclusive so if both are set TODO: make a function for this and similar?
table.insert( z.message_tail, { set_error( 'redundant_parameters', {wrap_style ('parameter', 'season') .. ' and ' .. wrap_style ('parameter', 'seriesno')}, true ) } ); -- add error message
utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'season') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'seriesno')}); -- add error message
SeriesNumber = ''; -- unset; prefer |season= over |seriesno=
SeriesNumber = ''; -- unset; prefer |season= over |seriesno=
end
end
-- assemble a table of parts concatenated later into Series
-- assemble a table of parts concatenated later into Series
if is_set(Season) then table.insert(s, wrap_msg ('season', Season, use_lowercase)); end
if utilities.is_set (Season) then table.insert(s, wrap_msg ('season', Season, use_lowercase)); end
if is_set(SeriesNumber) then table.insert(s, wrap_msg ('seriesnum', SeriesNumber, use_lowercase)); end
if utilities.is_set (SeriesNumber) then table.insert(s, wrap_msg ('seriesnum', SeriesNumber, use_lowercase)); end
if is_set(Issue) then table.insert(s, wrap_msg ('episode', Issue, use_lowercase)); end
if utilities.is_set (Issue) then table.insert(s, wrap_msg ('episode', Issue, use_lowercase)); end
Issue = ''; -- unset because this is not a unique parameter
Issue = ''; -- unset because this is not a unique parameter
Chapter = Title; -- promote title parameters to chapter
Chapter = Title; -- promote title parameters to chapter
ScriptChapter = ScriptTitle;
ScriptChapter = ScriptTitle;
ScriptChapterOrigin = 'title';
ScriptChapter_origin = A:ORIGIN('ScriptTitle');
ChapterLink = TitleLink; -- alias episodelink
ChapterLink = TitleLink; -- alias |episode-link=
TransChapter = TransTitle;
TransChapter = TransTitle;
ChapterURL = URL;
ChapterURL = URL;
ChapterUrlAccess = UrlAccess;
ChapterUrlAccess = UrlAccess;
ChapterURLorigin = A:ORIGIN('URL');
ChapterURL_origin = URL_origin;
ChapterFormat = Format;
 
Title = Series; -- promote series to title
Title = Series; -- promote series to title
TitleLink = SeriesLink;
TitleLink = SeriesLink;
Series = table.concat(s, sepc .. ' '); -- this is concatenation of season, seriesno, episode number
Series = table.concat(s, sepc .. ' '); -- this is concatenation of season, seriesno, episode number


if is_set (ChapterLink) and not is_set (ChapterURL) then -- link but not URL
if utilities.is_set (ChapterLink) and not utilities.is_set (ChapterURL) then -- link but not URL
Chapter = make_wikilink (ChapterLink, Chapter);
Chapter = utilities.make_wikilink (ChapterLink, Chapter);
elseif is_set (ChapterLink) and is_set (ChapterURL) then -- if both are set, URL links episode;
elseif utilities.is_set (ChapterLink) and utilities.is_set (ChapterURL) then -- if both are set, URL links episode;
Series = make_wikilink (ChapterLink, Series);
Series = utilities.make_wikilink (ChapterLink, Series);
end
end
URL = ''; -- unset
URL = ''; -- unset
TransTitle = '';
TransTitle = '';
ScriptTitle = '';
ScriptTitle = '';
Format = '';
else -- now oddities that are cite serial
else -- now oddities that are cite serial
Issue = ''; -- unset because this parameter no longer supported by the citation/core version of cite serial
Issue = ''; -- unset because this parameter no longer supported by the citation/core version of cite serial
Chapter = A['Episode']; -- TODO: make |episode= available to cite episode someday?
Chapter = A['Episode']; -- TODO: make |episode= available to cite episode someday?
if is_set (Series) and is_set (SeriesLink) then
if utilities.is_set (Series) and utilities.is_set (SeriesLink) then
Series = make_wikilink (SeriesLink, Series);
Series = utilities.make_wikilink (SeriesLink, Series);
end
end
Series = wrap_style ('italic-title', Series); -- series is italicized
Series = utilities.wrap_style ('italic-title', Series); -- series is italicized
end
end
end
end
-- end of {{cite episode}} stuff
-- end of {{cite episode}} stuff
-- Account for the oddities that are {{cite arxiv}}, {{cite biorxiv}}, {{cite citeseerx}}, {{cite ssrn}}, before generation of COinS data.
do
if in_array (config.CitationClass, {'arxiv', 'biorxiv', 'citeseerx', 'ssrn'}) then
if not is_set (ID_list[config.CitationClass:upper()]) then -- |arxiv= or |eprint= required for cite arxiv; |biorxiv= & |citeseerx= required for their templates
table.insert( z.message_tail, { set_error( config.CitationClass .. '_missing', {}, true ) } ); -- add error message
end
Periodical = ({['arxiv'] = 'arXiv', ['biorxiv'] = bioRxiv, ['citeseerx'] = 'CiteSeerX', ['ssrn'] = 'Social Science Research Network'})[config.CitationClass];
end
end


-- handle type parameter for those CS1 citations that have default values
-- handle type parameter for those CS1 citations that have default values
if in_array(config.CitationClass, {"AV-media-notes", "interview", "mailinglist", "map", "podcast", "pressrelease", "report", "techreport", "thesis"}) then
local TitleType = A['TitleType'];
local Degree = A['Degree'];
if utilities.in_array (config.CitationClass, {'AV-media-notes', 'document', 'interview', 'mailinglist', 'map', 'podcast', 'pressrelease', 'report', 'speech', 'techreport', 'thesis'}) then
TitleType = set_titletype (config.CitationClass, TitleType);
TitleType = set_titletype (config.CitationClass, TitleType);
if is_set(Degree) and "Thesis" == TitleType then -- special case for cite thesis
if utilities.is_set (Degree) and "Thesis" == TitleType then -- special case for cite thesis
TitleType = Degree .. ' ' .. cfg.title_types ['thesis']:lower();
TitleType = Degree .. ' ' .. cfg.title_types ['thesis']:lower();
end
end
end
end


if is_set(TitleType) then -- if type parameter is specified
if utilities.is_set (TitleType) then -- if type parameter is specified
TitleType = substitute( cfg.messages['type'], TitleType); -- display it in parentheses
TitleType = utilities.substitute ( cfg.messages['type'], TitleType); -- display it in parentheses
-- TODO: Hack on TitleType to fix bunched parentheses problem
-- TODO: Hack on TitleType to fix bunched parentheses problem
end
end


-- legacy: promote PublicationDate to Date if neither Date nor Year are set.
-- legacy: promote PublicationDate to Date if neither Date nor Year are set.
local Date = A['Date'];
  local Date_origin; -- to hold the name of parameter promoted to Date; required for date error messaging
  local Date_origin; -- to hold the name of parameter promoted to Date; required for date error messaging
local PublicationDate = A['PublicationDate'];
local Year = A['Year'];


if not is_set (Date) then
if not utilities.is_set (Date) then
Date = Year; -- promote Year to Date
Date = Year; -- promote Year to Date
Year = nil; -- make nil so Year as empty string isn't used for CITEREF
Year = nil; -- make nil so Year as empty string isn't used for CITEREF
if not is_set (Date) and is_set(PublicationDate) then -- use PublicationDate when |date= and |year= are not set
if not utilities.is_set (Date) and utilities.is_set (PublicationDate) then -- use PublicationDate when |date= and |year= are not set
Date = PublicationDate; -- promote PublicationDate to Date
Date = PublicationDate; -- promote PublicationDate to Date
PublicationDate = ''; -- unset, no longer needed
PublicationDate = ''; -- unset, no longer needed
Line 2,778: Line 3,158:
Date validation supporting code is in Module:Citation/CS1/Date_validation
Date validation supporting code is in Module:Citation/CS1/Date_validation
]]
]]
local DF = is_valid_parameter_value (A['DF'], A:ORIGIN('DF'), cfg.keywords_lists['df'], '');
if not utilities.is_set (DF) then
DF = cfg.global_df; -- local |df= if present overrides global df set by {{use xxx date}} template
end
local ArchiveURL;
local ArchiveDate;
local ArchiveFormat = A['ArchiveFormat'];
local archive_url_timestamp; -- timestamp from wayback machine url
ArchiveURL, ArchiveDate, archive_url_timestamp = archive_url_check (A['ArchiveURL'], A['ArchiveDate'])
ArchiveFormat = style_format (ArchiveFormat, ArchiveURL, 'archive-format', 'archive-url');
ArchiveURL, ArchiveDate = is_unique_archive_url (ArchiveURL, URL, ChapterURL, A:ORIGIN('ArchiveURL'), ArchiveDate); -- add error message when URL or ChapterURL == ArchiveURL
local AccessDate = A['AccessDate'];
local COinS_date = {}; -- holds date info extracted from |date= for the COinS metadata by Module:Date verification
local DoiBroken = A['DoiBroken'];
local Embargo = A['Embargo'];
local anchor_year; -- used in the CITEREF identifier
do -- create defined block to contain local variables error_message, date_parameters_list, mismatch
do -- create defined block to contain local variables error_message, date_parameters_list, mismatch
local error_message = '';
local error_message = '';
-- AirDate has been promoted to Date so not necessary to check it
-- AirDate has been promoted to Date so not necessary to check it
local date_parameters_list = {
local date_parameters_list = {
['access-date'] = {val=AccessDate, name=A:ORIGIN ('AccessDate')},
['access-date'] = {val = AccessDate, name = A:ORIGIN ('AccessDate')},
['archive-date'] = {val=ArchiveDate, name=A:ORIGIN ('ArchiveDate')},
['archive-date'] = {val = ArchiveDate, name = A:ORIGIN ('ArchiveDate')},
['date'] = {val=Date, name=Date_origin},
['date'] = {val = Date, name = Date_origin},
['doi-broken-date'] = {val=DoiBroken, name=A:ORIGIN ('DoiBroken')},
['doi-broken-date'] = {val = DoiBroken, name = A:ORIGIN ('DoiBroken')},
['embargo'] = {val=Embargo, name=A:ORIGIN ('Embargo')},
['pmc-embargo-date'] = {val = Embargo, name = A:ORIGIN ('Embargo')},
['lay-date'] = {val=LayDate, name=A:ORIGIN ('LayDate')},
['publication-date'] = {val = PublicationDate, name = A:ORIGIN ('PublicationDate')},
['publication-date'] ={val=PublicationDate, name=A:ORIGIN ('PublicationDate')},
['year'] = {val = Year, name = A:ORIGIN ('Year')},
['year'] = {val=Year, name=A:ORIGIN ('Year')},
};
};
anchor_year, Embargo, error_message = dates(date_parameters_list, COinS_date);
 
local error_list = {};
anchor_year, Embargo = validation.dates(date_parameters_list, COinS_date, error_list);


-- start temporary Julian / Gregorian calendar uncertainty categorization
-- start temporary Julian / Gregorian calendar uncertainty categorization
if COinS_date.inter_cal_cat then
if COinS_date.inter_cal_cat then
add_prop_cat ('jul_greg_uncertainty');
utilities.add_prop_cat ('jul-greg-uncertainty');
end
end
-- end temporary Julian / Gregorian calendar uncertainty categorization
-- end temporary Julian / Gregorian calendar uncertainty categorization


if is_set (Year) and is_set (Date) then -- both |date= and |year= not normally needed;  
if utilities.is_set (Year) and utilities.is_set (Date) then -- both |date= and |year= not normally needed;  
local mismatch = year_date_check (Year, Date)
validation.year_date_check (Year, A:ORIGIN ('Year'), Date, A:ORIGIN ('Date'), error_list);
if 0 == mismatch then -- |year= does not match a year-value in |date=
if is_set (error_message) then -- if there is already an error message
error_message = error_message .. ', '; -- tack on this additional message
end
error_message = error_message .. '&#124;year= / &#124;date= mismatch';
elseif 1 == mismatch then -- |year= matches year-value in |date=
add_maint_cat ('date_year');
end
end
end
if not is_set(error_message) then -- error free dates only
if 0 == #error_list then -- error free dates only; 0 when error_list is empty
local modified = false; -- flag
local modified = false; -- flag
if is_set (DF) then -- if we need to reformat dates
if utilities.is_set (DF) then -- if we need to reformat dates
modified = reformat_dates (date_parameters_list, DF, false); -- reformat to DF format, use long month names if appropriate
modified = validation.reformat_dates (date_parameters_list, DF); -- reformat to DF format, use long month names if appropriate
end
end


if true == date_hyphen_to_dash (date_parameters_list) then -- convert hyphens to dashes where appropriate
if true == validation.date_hyphen_to_dash (date_parameters_list) then -- convert hyphens to dashes where appropriate
modified = true;
modified = true;
add_maint_cat ('date_format'); -- hyphens were converted so add maint category
utilities.set_message ('maint_date_format'); -- hyphens were converted so add maint category
end
end
-- for those wikis that can and want to have English date names translated to the local language,
-- for those wikis that can and want to have English date names translated to the local language; not supported at en.wiki
-- uncomment these three lines.  Not supported by en.wiki (for obvious reasons)
if cfg.date_name_auto_xlate_enable and validation.date_name_xlate (date_parameters_list, cfg.date_digit_auto_xlate_enable ) then
-- set date_name_xlate() second argument to true to translate English digits to local digits (will translate ymd dates)
utilities.set_message ('maint_date_auto_xlated'); -- add maint cat
-- if date_name_xlate (date_parameters_list, false) then
modified = true;
-- modified = true;
end
-- end


if modified then -- if the date_parameters_list values were modified
if modified then -- if the date_parameters_list values were modified
Line 2,835: Line 3,228:
Date = date_parameters_list['date'].val;
Date = date_parameters_list['date'].val;
DoiBroken = date_parameters_list['doi-broken-date'].val;
DoiBroken = date_parameters_list['doi-broken-date'].val;
LayDate = date_parameters_list['lay-date'].val;
PublicationDate = date_parameters_list['publication-date'].val;
PublicationDate = date_parameters_list['publication-date'].val;
end
if archive_url_timestamp and utilities.is_set (ArchiveDate) then
validation.archive_date_check (ArchiveDate, archive_url_timestamp, DF); -- does YYYYMMDD in archive_url_timestamp match date in ArchiveDate
end
end
else
else
table.insert( z.message_tail, { set_error( 'bad_date', {error_message}, true ) } ); -- add this error message
utilities.set_message ('err_bad_date', {utilities.make_sep_list (#error_list, error_list)}); -- add this error message
end
end
end -- end of do
end -- end of do


-- Account for the oddity that is {{cite journal}} with |pmc= set and |url= not set. Do this after date check but before COInS.
if utilities.in_array (config.CitationClass, {'book', 'encyclopaedia'}) or -- {{cite book}}, {{cite encyclopedia}}; TODO: {{cite conference}} and others?
-- Here we unset Embargo if PMC not embargoed (|embargo= not set in the citation) or if the embargo time has expired. Otherwise, holds embargo date
('citation' == config.CitationClass and utilities.is_set (Encyclopedia)) or -- {{citation}} as an encylopedia citation
Embargo = is_embargoed (Embargo);
('citation' == config.CitationClass and not utilities.is_set (Periodical)) then -- {{citation}} as a book citation
if utilities.is_set (PublicationPlace) then
if not utilities.is_set (PublisherName) then
local date = COinS_date.rftdate and tonumber (COinS_date.rftdate:match ('%d%d%d%d')); -- get year portion of COinS date (because in Arabic numerals); convert string to number
if date and (1850 <= date) then -- location has no publisher; if date is 1850 or later
utilities.set_message ('maint_location_no_publisher'); -- add maint cat
end
else -- PublisherName has a value
if cfg.keywords_xlate['none'] == PublisherName then -- if that value is 'none' (only for book and encyclopedia citations)
PublisherName = ''; -- unset
end
end
end
end
 
local ID_list = {}; -- sequence table of rendered identifiers
local ID_list_coins = {}; -- table of identifiers and their values from args; key is same as cfg.id_handlers's key
local Class = A['Class']; -- arxiv class identifier
local ID_support = {
{A['ASINTLD'], 'ASIN', 'err_asintld_missing_asin', A:ORIGIN ('ASINTLD')},
{DoiBroken, 'DOI', 'err_doibroken_missing_doi', A:ORIGIN ('DoiBroken')},
{Embargo, 'PMC', 'err_embargo_missing_pmc', A:ORIGIN ('Embargo')},
}


if config.CitationClass == "journal" and not is_set(URL) and is_set(ID_list['PMC']) then
ID_list, ID_list_coins = identifiers.identifier_lists_get (args, {DoiBroken = DoiBroken, ASINTLD = A['ASINTLD'], Embargo = Embargo, Class = Class}, ID_support);
if not is_set (Embargo) then -- if not embargoed or embargo has expired
 
URL=cfg.id_handlers['PMC'].prefix .. ID_list['PMC']; -- set url to be the same as the PMC external link if not embargoed
-- Account for the oddities that are {{cite arxiv}}, {{cite biorxiv}}, {{cite citeseerx}}, {{cite medrxiv}}, {{cite ssrn}}, before generation of COinS data.
URLorigin = cfg.id_handlers['PMC'].parameters[1]; -- set URLorigin to parameter name for use in error message if citation is missing a |title=
if utilities.in_array (config.CitationClass, whitelist.preprint_template_list_t) then -- |arxiv= or |eprint= required for cite arxiv; |biorxiv=, |citeseerx=, |medrxiv=, |ssrn= required for their templates
if is_set(AccessDate) then -- access date requires |url=; pmc created url is not |url=
if not (args[cfg.id_handlers[config.CitationClass:upper()].parameters[1]] or -- can't use ID_list_coins k/v table here because invalid parameters omitted
table.insert( z.message_tail, { set_error( 'accessdate_missing_url', {}, true ) } );
args[cfg.id_handlers[config.CitationClass:upper()].parameters[2]]) then -- which causes unexpected parameter missing error message
utilities.set_message ('err_' .. config.CitationClass .. '_missing'); -- add error message
end
 
Periodical = ({['arxiv'] = 'arXiv', ['biorxiv'] = 'bioRxiv', ['citeseerx'] = 'CiteSeerX', ['medrxiv'] = 'medRxiv', ['ssrn'] = 'Social Science Research Network'})[config.CitationClass];
end
 
-- Link the title of the work if no |url= was provided, but we have a |pmc= or a |doi= with |doi-access=free
 
if config.CitationClass == "journal" and not utilities.is_set (URL) and not utilities.is_set (TitleLink) and not utilities.in_array (cfg.keywords_xlate[Title], {'off', 'none'}) then -- TODO: remove 'none' once existing citations have been switched to 'off', so 'none' can be used as token for "no title" instead
if 'none' ~= cfg.keywords_xlate[auto_select] then -- if auto-linking not disabled
if identifiers.auto_link_urls[auto_select] then -- manual selection
URL = identifiers.auto_link_urls[auto_select]; -- set URL to be the same as identifier's external link
URL_origin = cfg.id_handlers[auto_select:upper()].parameters[1]; -- set URL_origin to parameter name for use in error message if citation is missing a |title=
elseif identifiers.auto_link_urls['pmc'] then -- auto-select PMC
URL = identifiers.auto_link_urls['pmc']; -- set URL to be the same as the PMC external link if not embargoed
URL_origin = cfg.id_handlers['PMC'].parameters[1]; -- set URL_origin to parameter name for use in error message if citation is missing a |title=
elseif identifiers.auto_link_urls['doi'] then -- auto-select DOI
URL = identifiers.auto_link_urls['doi'];
URL_origin = cfg.id_handlers['DOI'].parameters[1];
end
end
 
if utilities.is_set (URL) then -- set when using an identifier-created URL
if utilities.is_set (AccessDate) then -- |access-date= requires |url=; identifier-created URL is not |url=
utilities.set_message ('err_accessdate_missing_url'); -- add an error message
AccessDate = ''; -- unset
AccessDate = ''; -- unset
end
end


if utilities.is_set (ArchiveURL) then -- |archive-url= requires |url=; identifier-created URL is not |url=
utilities.set_message ('err_archive_missing_url'); -- add an error message
ArchiveURL = ''; -- unset
end
end
end
end
end
Line 2,861: Line 3,309:
-- At this point fields may be nil if they weren't specified in the template use.  We can use that fact.
-- At this point fields may be nil if they weren't specified in the template use.  We can use that fact.
-- Test if citation has no title
-- Test if citation has no title
if not is_set(Title) and
if not utilities.is_set (Title) and not utilities.is_set (TransTitle) and not utilities.is_set (ScriptTitle) then -- has special case for cite episode
not is_set(TransTitle) and
utilities.set_message ('err_citation_missing_title', {'episode' == config.CitationClass and 'series' or 'title'});
not is_set(ScriptTitle) then
if 'episode' == config.CitationClass then -- special case for cite episode; TODO: is there a better way to do this?
table.insert( z.message_tail, { set_error( 'citation_missing_title', {'series'}, true ) } );
else
table.insert( z.message_tail, { set_error( 'citation_missing_title', {'title'}, true ) } );
end
end
end


if 'none' == Title and
if utilities.in_array (cfg.keywords_xlate[Title], {'off', 'none'}) and
in_array (config.CitationClass, {'journal', 'citation'}) and
utilities.in_array (config.CitationClass, {'journal', 'citation'}) and
(is_set (Periodical) or is_set (ScriptPeriodical)) and
(utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and
('journal' == Periodical_origin or 'script-journal' == ScriptPeriodical_origin) then -- special case for journal cites
('journal' == Periodical_origin or 'script-journal' == ScriptPeriodical_origin) then -- special case for journal cites
Title = ''; -- set title to empty string
Title = ''; -- set title to empty string
add_maint_cat ('untitled');
utilities.set_message ('maint_untitled'); -- add maint cat
end
end
check_for_url ({ -- add error message when any of these parameters hold a URL
['title']=Title,
[A:ORIGIN('Chapter')]=Chapter,
[Periodical_origin] = Periodical,
[PublisherName_origin] = PublisherName
});


-- COinS metadata (see <http://ocoins.info/>) for automated parsing of citation information.
-- COinS metadata (see <http://ocoins.info/>) for automated parsing of citation information.
Line 2,893: Line 3,328:
local coins_chapter = Chapter; -- default assuming that remapping not required
local coins_chapter = Chapter; -- default assuming that remapping not required
local coins_title = Title; -- et tu
local coins_title = Title; -- et tu
if 'encyclopaedia' == config.CitationClass or ('citation' == config.CitationClass and is_set (Encyclopedia)) then
if 'encyclopaedia' == config.CitationClass or ('citation' == config.CitationClass and utilities.is_set (Encyclopedia)) then
if is_set (Chapter) and is_set (Title) and is_set (Periodical) then -- if all are used then
if utilities.is_set (Chapter) and utilities.is_set (Title) and utilities.is_set (Periodical) then -- if all are used then
coins_chapter = Title; -- remap
coins_chapter = Title; -- remap
coins_title = Periodical;
coins_title = Periodical;
Line 2,903: Line 3,338:
coins_author = c; -- use that instead
coins_author = c; -- use that instead
end
end
 
-- this is the function call to COinS()
-- this is the function call to COinS()
local OCinSoutput = COinS({
local OCinSoutput = metadata.COinS({
['Periodical'] = strip_apostrophe_markup (Periodical), -- no markup in the metadata
['Periodical'] = utilities.strip_apostrophe_markup (Periodical), -- no markup in the metadata
['Encyclopedia'] = strip_apostrophe_markup (Encyclopedia),
['Encyclopedia'] = Encyclopedia, -- just a flag; content ignored by ~/COinS
['Chapter'] = make_coins_title (coins_chapter, ScriptChapter), -- Chapter and ScriptChapter stripped of bold / italic wikimarkup
['Chapter'] = metadata.make_coins_title (coins_chapter, ScriptChapter), -- Chapter and ScriptChapter stripped of bold / italic / accept-as-written markup
['Degree'] = Degree; -- cite thesis only
['Degree'] = Degree; -- cite thesis only
['Title'] = make_coins_title (coins_title, ScriptTitle), -- Title and ScriptTitle stripped of bold / italic wikimarkup
['Title'] = metadata.make_coins_title (coins_title, ScriptTitle), -- Title and ScriptTitle stripped of bold / italic / accept-as-written markup
['PublicationPlace'] = PublicationPlace,
['PublicationPlace'] = PublicationPlace,
['Date'] = COinS_date.rftdate, -- COinS_date has correctly formatted date if Date is valid;
['Date'] = COinS_date.rftdate, -- COinS_date.* has correctly formatted date values if Date is valid;
['Season'] = COinS_date.rftssn,
['Season'] = COinS_date.rftssn,
['Chron'] =  COinS_date.rftchron or (not COinS_date.rftdate and Date) or '', -- chron but if not set and invalid date format use Date; keep this last bit?
['Quarter'] = COinS_date.rftquarter,
['Chron'] =  COinS_date.rftchron,
['Series'] = Series,
['Series'] = Series,
['Volume'] = Volume,
['Volume'] = Volume,
['Issue'] = Issue,
['Issue'] = Issue,
['Pages'] = coins_pages or get_coins_pages (first_set ({Sheet, Sheets, Page, Pages, At}, 5)), -- pages stripped of external links
['ArticleNumber'] = ArticleNumber,
['Pages'] = coins_pages or metadata.get_coins_pages (first_set ({Sheet, Sheets, Page, Pages, At, QuotePage, QuotePages}, 7)), -- pages stripped of external links
['Edition'] = Edition,
['Edition'] = Edition,
['PublisherName'] = PublisherName, -- any apostrophe markup already removed
['PublisherName'] = PublisherName or Newsgroup, -- any apostrophe markup already removed from PublisherName
['URL'] = first_set ({ChapterURL, URL}, 2),
['URL'] = first_set ({ChapterURL, URL}, 2),
['Authors'] = coins_author,
['Authors'] = coins_author,
['ID_list'] = ID_list,
['ID_list'] = ID_list_coins,
['RawPage'] = this_page.prefixedText,
['RawPage'] = this_page.prefixedText,
}, config.CitationClass);
}, config.CitationClass);


-- Account for the oddities that are {{cite arxiv}}, {{cite biorxiv}}, {{cite citeseerx}}, and {{cite ssrn}} AFTER generation of COinS data.
-- Account for the oddities that are {{cite arxiv}}, {{cite biorxiv}}, {{cite citeseerx}}, {{cite medrxiv}}, and {{cite ssrn}} AFTER generation of COinS data.
if in_array (config.CitationClass, {'arxiv', 'biorxiv', 'citeseerx', 'ssrn'}) then -- we have set rft.jtitle in COinS to arXiv, bioRxiv, CiteSeerX, or ssrn now unset so it isn't displayed
if utilities.in_array (config.CitationClass, whitelist.preprint_template_list_t) then -- we have set rft.jtitle in COinS to arXiv, bioRxiv, CiteSeerX, medRxiv, or ssrn now unset so it isn't displayed
Periodical = ''; -- periodical not allowed in these templates; if article has been published, use cite journal
Periodical = ''; -- periodical not allowed in these templates; if article has been published, use cite journal
end
end


-- special case for cite newsgroup.  Do this after COinS because we are modifying Publishername to include some static text
-- special case for cite newsgroup.  Do this after COinS because we are modifying Publishername to include some static text
if 'newsgroup' == config.CitationClass then
if 'newsgroup' == config.CitationClass and utilities.is_set (Newsgroup) then
if is_set (PublisherName) then
PublisherName = utilities.substitute (cfg.messages['newsgroup'], external_link( 'news:' .. Newsgroup, Newsgroup, Newsgroup_origin, nil ));
PublisherName = substitute (cfg.messages['newsgroup'], external_link( 'news:' .. PublisherName, PublisherName, PublisherName_origin, nil ));
end
end
end


local Editors;
local EditorCount; -- used only for choosing {ed.) or (eds.) annotation at end of editor name-list
local Contributors; -- assembled contributors name list
local contributor_etal;
local Translators; -- assembled translators name list
local translator_etal;
local t = {}; -- translators list from |translator-lastn= / translator-firstn= pairs
t = extract_names (args, 'TranslatorList'); -- fetch translator list from |translatorn= / |translator-lastn=, -firstn=, -linkn=, -maskn=
local Interviewers;
local interviewers_list = {};
interviewers_list = extract_names (args, 'InterviewerList'); -- process preferred interviewers parameters
local interviewer_etal;
-- Now perform various field substitutions.
-- Now perform various field substitutions.
-- We also add leading spaces and surrounding markup and punctuation to the
-- We also add leading spaces and surrounding markup and punctuation to the
-- various parts of the citation, but only when they are non-nil.
-- various parts of the citation, but only when they are non-nil.
local EditorCount; -- used only for choosing {ed.) or (eds.) annotation at end of editor name-list
do
do
local last_first_list;
local last_first_list;
local control = {  
local control = {  
format = NameListFormat, -- empty string or 'vanc'
format = NameListStyle, -- empty string, '&', 'amp', 'and', or 'vanc'
maximum = nil, -- as if display-authors or display-editors not set
maximum = nil, -- as if display-authors or display-editors not set
lastauthoramp = LastAuthorAmp,
page_name = this_page.text, -- get current page name so that we don't wikilink to it via editorlinkn
mode = Mode
mode = Mode
};
};


do -- do editor name list first because the now unsupported coauthors used to modify control table
do -- do editor name list first because the now unsupported coauthors used to modify control table
control.maximum , editor_etal = get_display_names (A['DisplayEditors'], #e, 'editors', editor_etal);
local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayEditors'], A['DisplayEditors'], A:ORIGIN ('DisplayEditors'), #e);
last_first_list, EditorCount = list_people(control, e, editor_etal);
control.maximum, editor_etal = get_display_names (display_names, #e, 'editors', editor_etal, param);


if is_set (Editors) then
Editors, EditorCount = list_people (control, e, editor_etal);
Editors, editor_etal = name_has_etal (Editors, editor_etal, false, 'editors'); -- find and remove variations on et al.
if editor_etal then
Editors = Editors .. ' ' .. cfg.messages['et al']; -- add et al. to editors parameter beause |display-editors=etal
EditorCount = 2; -- with et al., |editors= is multiple names; spoof to display (eds.) annotation
else
EditorCount = 2; -- we don't know but assume |editors= is multiple names; spoof to display (eds.) annotation
end
else
Editors = last_first_list; -- either an author name list or an empty string
end


if 1 == EditorCount and (true == editor_etal or 1 < #e) then -- only one editor displayed but includes etal then  
if 1 == EditorCount and (true == editor_etal or 1 < #e) then -- only one editor displayed but includes etal then  
Line 2,974: Line 3,409:
end
end
do -- now do interviewers
do -- now do interviewers
control.maximum , interviewer_etal = get_display_names (A['DisplayInterviewers'], #interviewers_list, 'interviewers', interviewer_etal);
local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayInterviewers'], A['DisplayInterviewers'], A:ORIGIN ('DisplayInterviewers'), #interviewers_list);
control.maximum, interviewer_etal = get_display_names (display_names, #interviewers_list, 'interviewers', interviewer_etal, param);
 
Interviewers = list_people (control, interviewers_list, interviewer_etal);
Interviewers = list_people (control, interviewers_list, interviewer_etal);
end
end
do -- now do translators
do -- now do translators
control.maximum , translator_etal = get_display_names (A['DisplayTranslators'], #t, 'translators', translator_etal);
local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayTranslators'], A['DisplayTranslators'], A:ORIGIN ('DisplayTranslators'), #t);
control.maximum, translator_etal = get_display_names (display_names, #t, 'translators', translator_etal, param);
 
Translators = list_people (control, t, translator_etal);
Translators = list_people (control, t, translator_etal);
end
end
do -- now do contributors
do -- now do contributors
control.maximum , contributor_etal = get_display_names (A['DisplayContributors'], #c, 'contributors', contributor_etal);
local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayContributors'], A['DisplayContributors'], A:ORIGIN ('DisplayContributors'), #c);
control.maximum, contributor_etal = get_display_names (display_names, #c, 'contributors', contributor_etal, param);
 
Contributors = list_people (control, c, contributor_etal);
Contributors = list_people (control, c, contributor_etal);
end
end
do -- now do authors
do -- now do authors
control.maximum , author_etal = get_display_names (A['DisplayAuthors'], #a, 'authors', author_etal);
local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayAuthors'], A['DisplayAuthors'], A:ORIGIN ('DisplayAuthors'), #a, author_etal);
control.maximum, author_etal = get_display_names (display_names, #a, 'authors', author_etal, param);


last_first_list = list_people(control, a, author_etal);
last_first_list = list_people (control, a, author_etal);


if is_set (Authors) then
if utilities.is_set (Authors) then
Authors, author_etal = name_has_etal (Authors, author_etal, false, 'authors'); -- find and remove variations on et al.
Authors, author_etal = name_has_etal (Authors, author_etal, false, 'authors'); -- find and remove variations on et al.
if author_etal then
if author_etal then
Line 3,000: Line 3,442:
end -- end of do
end -- end of do
if is_set (Authors) and is_set (Collaboration) then
if utilities.is_set (Authors) and utilities.is_set (Collaboration) then
Authors = Authors .. ' (' .. Collaboration .. ')'; -- add collaboration after et al.
Authors = Authors .. ' (' .. Collaboration .. ')'; -- add collaboration after et al.
end
end
Line 3,006: Line 3,448:
end
end


-- apply |[xx-]format= styling; at the end, these parameters hold correctly styled format annotation,
local ConferenceFormat = A['ConferenceFormat'];
-- an error message if the associated url is not set, or an empty string for concatenation
local ConferenceURL = A['ConferenceURL'];
ArchiveFormat = style_format (ArchiveFormat, ArchiveURL, 'archive-format', 'archive-url');
ConferenceFormat = style_format (ConferenceFormat, ConferenceURL, 'conference-format', 'conference-url');
ConferenceFormat = style_format (ConferenceFormat, ConferenceURL, 'conference-format', 'conference-url');
Format = style_format (Format, URL, 'format', 'url');
Format = style_format (Format, URL, 'format', 'url');
LayFormat = style_format (LayFormat, LayURL, 'lay-format', 'lay-url');
TranscriptFormat = style_format (TranscriptFormat, TranscriptURL, 'transcript-format', 'transcripturl');


-- special case for chapter format so no error message or cat when chapter not supported
-- special case for chapter format so no error message or cat when chapter not supported
if not (in_array(config.CitationClass, {'web', 'news', 'journal', 'magazine', 'pressrelease', 'podcast', 'newsgroup', 'arxiv', 'biorxiv', 'citeseerx', 'ssrn'}) or
if not (utilities.in_array (config.CitationClass, {'web', 'news', 'journal', 'magazine', 'pressrelease', 'podcast', 'newsgroup', 'arxiv', 'biorxiv', 'citeseerx', 'medrxiv', 'ssrn'}) or
('citation' == config.CitationClass and (is_set (Periodical) or is_set (ScriptPeriodical)) and not is_set (Encyclopedia))) then
('citation' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and not utilities.is_set (Encyclopedia))) then
ChapterFormat = style_format (ChapterFormat, ChapterURL, 'chapter-format', 'chapter-url');
ChapterFormat = style_format (ChapterFormat, ChapterURL, 'chapter-format', 'chapter-url');
end
end


if not is_set(URL) then
if not utilities.is_set (URL) then
if in_array(config.CitationClass, {"web","podcast", "mailinglist"}) or -- |url= required for cite web, cite podcast, and cite mailinglist
if utilities.in_array (config.CitationClass, {"web", "podcast", "mailinglist"}) or -- |url= required for cite web, cite podcast, and cite mailinglist
('citation' == config.CitationClass and ('website' == Periodical_origin or 'script-website' == ScriptPeriodical_origin)) then -- and required for {{citation}} with |website= or |script-website=
('citation' == config.CitationClass and ('website' == Periodical_origin or 'script-website' == ScriptPeriodical_origin)) then -- and required for {{citation}} with |website= or |script-website=
table.insert( z.message_tail, { set_error( 'cite_web_url', {}, true ) } );
utilities.set_message ('err_cite_web_url');
end
end
-- do we have |accessdate= without either |url= or |chapter-url=?
-- do we have |accessdate= without either |url= or |chapter-url=?
if is_set(AccessDate) and not is_set(ChapterURL)then -- ChapterURL may be set when URL is not set;
if utilities.is_set (AccessDate) and not utilities.is_set (ChapterURL) then -- ChapterURL may be set when URL is not set;
table.insert( z.message_tail, { set_error( 'accessdate_missing_url', {}, true ) } );
utilities.set_message ('err_accessdate_missing_url');
AccessDate = '';
AccessDate = '';
end
end
end
end


local OriginalURL, OriginalURLorigin, OriginalFormat, OriginalAccess;
local UrlStatus = is_valid_parameter_value (A['UrlStatus'], A:ORIGIN('UrlStatus'), cfg.keywords_lists['url-status'], '');
-- DeadURL = DeadURL:lower(); -- used later when assembling archived text
local OriginalURL
local OriginalURL_origin
local OriginalFormat
local OriginalAccess;
UrlStatus = UrlStatus:lower(); -- used later when assembling archived text
UrlStatus = UrlStatus:lower(); -- used later when assembling archived text
if is_set( ArchiveURL ) then
if utilities.is_set ( ArchiveURL ) then
if is_set (ChapterURL) then -- if chapter-url is set apply archive url to it
if utilities.is_set (ChapterURL) then -- if chapter-url= is set apply archive url to it
OriginalURL = ChapterURL; -- save copy of source chapter's url for archive text
OriginalURL = ChapterURL; -- save copy of source chapter's url for archive text
OriginalURLorigin = ChapterURLorigin; -- name of chapter-url parameter for error messages
OriginalURL_origin = ChapterURL_origin; -- name of |chapter-url= parameter for error messages
OriginalFormat = ChapterFormat; -- and original |chapter-format=
OriginalFormat = ChapterFormat; -- and original |chapter-format=
-- if 'no' ~= DeadURL then
 
if 'live' ~= UrlStatus then
if 'live' ~= UrlStatus then
ChapterURL = ArchiveURL -- swap-in the archive's url
ChapterURL = ArchiveURL -- swap-in the archive's URL
ChapterURLorigin = A:ORIGIN('ArchiveURL') -- name of archive-url parameter for error messages
ChapterURL_origin = A:ORIGIN('ArchiveURL') -- name of |archive-url= parameter for error messages
ChapterFormat = ArchiveFormat or ''; -- swap in archive's format
ChapterFormat = ArchiveFormat or ''; -- swap in archive's format
ChapterUrlAccess = nil; -- restricted access levels do not make sense for archived urls
ChapterUrlAccess = nil; -- restricted access levels do not make sense for archived URLs
end
end
elseif is_set (URL) then
elseif utilities.is_set (URL) then
OriginalURL = URL; -- save copy of original source URL
OriginalURL = URL; -- save copy of original source URL
OriginalURLorigin = URLorigin; -- name of url parameter for error messages
OriginalURL_origin = URL_origin; -- name of URL parameter for error messages
OriginalFormat = Format; -- and original |format=
OriginalFormat = Format; -- and original |format=
OriginalAccess = UrlAccess;
OriginalAccess = UrlAccess;
-- if 'no' ~= DeadURL then -- if URL set then archive-url applies to it
 
if 'live' ~= UrlStatus then -- if URL set then archive-url applies to it
if 'live' ~= UrlStatus then -- if URL set then |archive-url= applies to it
URL = ArchiveURL -- swap-in the archive's url
URL = ArchiveURL -- swap-in the archive's URL
URLorigin = A:ORIGIN('ArchiveURL') -- name of archive url parameter for error messages
URL_origin = A:ORIGIN('ArchiveURL') -- name of archive URL parameter for error messages
Format = ArchiveFormat or ''; -- swap in archive's format
Format = ArchiveFormat or ''; -- swap in archive's format
UrlAccess = nil; -- restricted access levels do not make sense for archived urls
UrlAccess = nil; -- restricted access levels do not make sense for archived URLs
end
end
  end
end
elseif utilities.is_set (UrlStatus) then -- if |url-status= is set when |archive-url= is not set
  utilities.set_message ('maint_url_status'); -- add maint cat
end
end


if in_array(config.CitationClass, {'web','news','journal', 'magazine', 'pressrelease', 'podcast', 'newsgroup', 'arxiv', 'biorxiv', 'citeseerx', 'ssrn'}) or -- if any of the 'periodical' cites except encyclopedia
if utilities.in_array (config.CitationClass, {'web', 'news', 'journal', 'magazine', 'pressrelease', 'podcast', 'newsgroup', 'arxiv', 'biorxiv', 'citeseerx', 'medrxiv', 'ssrn'}) or -- if any of the 'periodical' cites except encyclopedia
('citation' == config.CitationClass and (is_set (Periodical) or is_set (ScriptPeriodical)) and not is_set (Encyclopedia)) then
('citation' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and not utilities.is_set (Encyclopedia)) then
local chap_param;
local chap_param;
if is_set (Chapter) then -- get a parameter name from one of these chapter related meta-parameters
if utilities.is_set (Chapter) then -- get a parameter name from one of these chapter related meta-parameters
chap_param = A:ORIGIN ('Chapter')
chap_param = A:ORIGIN ('Chapter')
elseif is_set (TransChapter) then
elseif utilities.is_set (TransChapter) then
chap_param = A:ORIGIN ('TransChapter')
chap_param = A:ORIGIN ('TransChapter')
elseif is_set (ChapterURL) then
elseif utilities.is_set (ChapterURL) then
chap_param = A:ORIGIN ('ChapterURL')
chap_param = A:ORIGIN ('ChapterURL')
elseif is_set (ScriptChapter) then
elseif utilities.is_set (ScriptChapter) then
chap_param = ScriptChapterOrigin;
chap_param = ScriptChapter_origin;
else is_set (ChapterFormat)
else utilities.is_set (ChapterFormat)
chap_param = A:ORIGIN ('ChapterFormat')
chap_param = A:ORIGIN ('ChapterFormat')
end
end


if is_set (chap_param) then -- if we found one
if utilities.is_set (chap_param) then -- if we found one
table.insert( z.message_tail, { set_error( 'chapter_ignored', {chap_param}, true ) } ); -- add error message
utilities.set_message ('err_chapter_ignored', {chap_param}); -- add error message
Chapter = ''; -- and set them to empty string to be safe with concatenation
Chapter = ''; -- and set them to empty string to be safe with concatenation
TransChapter = '';
TransChapter = '';
Line 3,088: Line 3,532:
else -- otherwise, format chapter / article title
else -- otherwise, format chapter / article title
local no_quotes = false; -- default assume that we will be quoting the chapter parameter value
local no_quotes = false; -- default assume that we will be quoting the chapter parameter value
if is_set (Contribution) and 0 < #c then -- if this is a contribution with contributor(s)
if utilities.is_set (Contribution) and 0 < #c then -- if this is a contribution with contributor(s)
if in_array (Contribution:lower(), cfg.keywords.contribution) then -- and a generic contribution title
if utilities.in_array (Contribution:lower(), cfg.keywords_lists.contribution) then -- and a generic contribution title
no_quotes = true; -- then render it unquoted
no_quotes = true; -- then render it unquoted
end
end
end
end


Chapter = format_chapter_title (ScriptChapter, ScriptChapterOrigin, Chapter, TransChapter, ChapterURL, ChapterURLorigin, no_quotes, ChapterUrlAccess); -- Contribution is also in Chapter
Chapter = format_chapter_title (ScriptChapter, ScriptChapter_origin, Chapter, Chapter_origin, TransChapter, TransChapter_origin, ChapterURL, ChapterURL_origin, no_quotes, ChapterUrlAccess); -- Contribution is also in Chapter
if is_set (Chapter) then
if utilities.is_set (Chapter) then
Chapter = Chapter .. ChapterFormat ;
Chapter = Chapter .. ChapterFormat ;
if 'map' == config.CitationClass and is_set (TitleType) then
if 'map' == config.CitationClass and utilities.is_set (TitleType) then
Chapter = Chapter .. ' ' .. TitleType; -- map annotation here; not after title
Chapter = Chapter .. ' ' .. TitleType; -- map annotation here; not after title
end
end
Chapter = Chapter.. sepc .. ' ';
Chapter = Chapter .. sepc .. ' ';
elseif is_set (ChapterFormat) then -- |chapter= not set but |chapter-format= is so ...
elseif utilities.is_set (ChapterFormat) then -- |chapter= not set but |chapter-format= is so ...
Chapter = ChapterFormat .. sepc .. ' '; -- ... ChapterFormat has error message, we want to see it
Chapter = ChapterFormat .. sepc .. ' '; -- ... ChapterFormat has error message, we want to see it
end
end
end
end


-- Format main title.
-- Format main title
if is_set (ArchiveURL) and mw.ustring.match (mw.ustring.lower(Title), cfg.special_case_translation['archived_copy']) then -- if title is 'Archived copy' (place holder added by bots that can't find proper title)
local plain_title = false;
add_maint_cat ('archived_copy'); -- add maintenance category before we modify the content of Title
local accept_title;
Title, accept_title = utilities.has_accept_as_written (Title, true); -- remove accept-this-as-written markup when it wraps all of <Title>
if accept_title and ('' == Title) then -- only support forced empty for now "(())"
Title = cfg.messages['notitle']; -- replace by predefined "No title" message
-- TODO: utilities.set_message ( 'err_redundant_parameters', ...); -- issue proper error message instead of muting
ScriptTitle = ''; -- just mute for now
TransTitle = ''; -- just mute for now
plain_title = true; -- suppress text decoration for descriptive title
utilities.set_message ('maint_untitled'); -- add maint cat
end
end


if Title:match ('^%(%(.*%)%)$') then -- if keep as written markup:
if not accept_title then -- <Title> not wrapped in accept-as-written markup
Title= Title:gsub ('^%(%((.*)%)%)$', '%1') -- remove the markup
if '...' == Title:sub (-3) then -- if ellipsis is the last three characters of |title=
else
if '...' == Title:sub (-3) then -- if elipsis is the last three characters of |title=
Title = Title:gsub ('(%.%.%.)%.+$', '%1'); -- limit the number of dots to three
Title = Title:gsub ('(%.%.%.)%.+$', '%1'); -- limit the number of dots to three
elseif not mw.ustring.find (Title, '%.%s*%a%.$') and -- end of title is not a 'dot-(optional space-)letter-dot' initialism ...
elseif not mw.ustring.find (Title, '%.%s*%a%.$') and -- end of title is not a 'dot-(optional space-)letter-dot' initialism ...
not mw.ustring.find (Title, '%s+%a%.$') then -- ...and not a 'space-letter-dot' initial (''Allium canadense'' L.)
not mw.ustring.find (Title, '%s+%a%.$') then -- ...and not a 'space-letter-dot' initial (''Allium canadense'' L.)
Title = mw.ustring.gsub(Title, '%'..sepc..'$', ''); -- remove any trailing separator character; sepc and ms.ustring() here for languages that use multibyte separator characters
Title = mw.ustring.gsub(Title, '%' .. sepc .. '$', ''); -- remove any trailing separator character; sepc and ms.ustring() here for languages that use multibyte separator characters
end
 
if utilities.is_set (ArchiveURL) and is_archived_copy (Title) then
utilities.set_message ('maint_archived_copy'); -- add maintenance category before we modify the content of Title
end
 
if is_generic ('generic_titles', Title) then
utilities.set_message ('err_generic_title'); -- set an error message
end
end
end
end
 
if in_array(config.CitationClass, {'web', 'news', 'journal', 'magazine', 'pressrelease', 'podcast', 'newsgroup', 'mailinglist', 'interview', 'arxiv', 'biorxiv', 'citeseerx', 'ssrn'}) or
if (not plain_title) and (utilities.in_array (config.CitationClass, {'web', 'news', 'journal', 'magazine', 'document', 'pressrelease', 'podcast', 'newsgroup', 'mailinglist', 'interview', 'arxiv', 'biorxiv', 'citeseerx', 'medrxiv', 'ssrn'}) or
('citation' == config.CitationClass and (is_set (Periodical) or is_set (ScriptPeriodical)) and not is_set (Encyclopedia)) or
('citation' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and not utilities.is_set (Encyclopedia)) or
('map' == config.CitationClass and (is_set (Periodical) or is_set (ScriptPeriodical))) then -- special case for cite map when the map is in a periodical treat as an article
('map' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)))) then -- special case for cite map when the map is in a periodical treat as an article
Title = kern_quotes (Title); -- if necessary, separate title's leading and trailing quote marks from Module provided quote marks
Title = kern_quotes (Title); -- if necessary, separate title's leading and trailing quote marks from module provided quote marks
Title = wrap_style ('quoted-title', Title);
Title = utilities.wrap_style ('quoted-title', Title);
Title = script_concatenate (Title, ScriptTitle, 'script-title'); -- <bdi> tags, lang atribute, categorization, etc; must be done after title is wrapped
Title = script_concatenate (Title, ScriptTitle, 'script-title'); -- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped
TransTitle= wrap_style ('trans-quoted-title', TransTitle );
TransTitle = utilities.wrap_style ('trans-quoted-title', TransTitle );
elseif 'report' == config.CitationClass then -- no styling for cite report
elseif plain_title or ('report' == config.CitationClass) then -- no styling for cite report and descriptive titles (otherwise same as above)
Title = script_concatenate (Title, ScriptTitle, 'script-title'); -- <bdi> tags, lang atribute, categorization, etc; must be done after title is wrapped
Title = script_concatenate (Title, ScriptTitle, 'script-title'); -- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped
TransTitle= wrap_style ('trans-quoted-title', TransTitle ); -- for cite report, use this form for trans-title
TransTitle = utilities.wrap_style ('trans-quoted-title', TransTitle ); -- for cite report, use this form for trans-title
else
else
Title = wrap_style ('italic-title', Title);
Title = utilities.wrap_style ('italic-title', Title);
Title = script_concatenate (Title, ScriptTitle, 'script-title'); -- <bdi> tags, lang atribute, categorization, etc; must be done after title is wrapped
Title = script_concatenate (Title, ScriptTitle, 'script-title'); -- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped
TransTitle = wrap_style ('trans-italic-title', TransTitle);
TransTitle = utilities.wrap_style ('trans-italic-title', TransTitle);
end
end


local TransError = "";
if utilities.is_set (TransTitle) then
if is_set(TransTitle) then
if utilities.is_set (Title) then
if is_set(Title) then
TransTitle = " " .. TransTitle;
TransTitle = " " .. TransTitle;
else
else
TransError = " " .. set_error( 'trans_missing_title', {'title'} );
utilities.set_message ('err_trans_missing_title', {'title'});
end
end
end
end


if is_set(Title) then
if utilities.is_set (Title) then -- TODO: is this the right place to be making Wikisource URLs?
if not is_set (TitleLink) and is_set (URL) then
if utilities.is_set (TitleLink) and utilities.is_set (URL) then
Title = external_link (URL, Title, URLorigin, UrlAccess) .. TransTitle .. TransError .. Format;
utilities.set_message ('err_wikilink_in_url'); -- set an error message because we can't have both
TitleLink = ''; -- unset
end
if not utilities.is_set (TitleLink) and utilities.is_set (URL) then
Title = external_link (URL, Title, URL_origin, UrlAccess) .. TransTitle .. Format;
URL = ''; -- unset these because no longer needed
URL = ''; -- unset these because no longer needed
Format = "";
Format = "";
elseif is_set (TitleLink) and not is_set (URL) then
elseif utilities.is_set (TitleLink) and not utilities.is_set (URL) then
local ws_url;
local ws_url;
ws_url = wikisource_url_make (TitleLink); -- ignore ws_label return; not used here
ws_url = wikisource_url_make (TitleLink); -- ignore ws_label return; not used here
if ws_url then
if ws_url then
Title = external_link (ws_url, Title .. '&nbsp;', 'ws link in title-link'); -- space char after Title to move icon away from italic text; TODO: a better way to do this?
Title = external_link (ws_url, Title .. '&nbsp;', 'ws link in title-link'); -- space char after Title to move icon away from italic text; TODO: a better way to do this?
Title = substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], TitleLink, Title});
Title = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], TitleLink, Title});
Title = Title .. TransTitle .. TransError;
Title = Title .. TransTitle;
else
else
Title = make_wikilink (TitleLink, Title) .. TransTitle .. TransError;
Title = utilities.make_wikilink (TitleLink, Title) .. TransTitle;
end
end
else
else
local ws_url, ws_label;
local ws_url, ws_label, L; -- Title has italic or quote markup by the time we get here which causes is_wikilink() to return 0 (not a wikilink)
ws_url, ws_label, L = wikisource_url_make (Title); -- make ws url from |title= interwiki link; link portion L becomes tool tip label
ws_url, ws_label, L = wikisource_url_make (Title:gsub('^[\'"]*(.-)[\'"]*$', '%1')); -- make ws URL from |title= interwiki link (strip italic or quote markup); link portion L becomes tooltip label
if ws_url then
if ws_url then
Title = Title:gsub ('%b[]', ws_label); -- replace interwiki link with ws_label to retain markup
Title = Title:gsub ('%b[]', ws_label); -- replace interwiki link with ws_label to retain markup
Title = external_link (ws_url, Title .. '&nbsp;', 'ws link in title'); -- space char after Title to move icon away from italic text; TODO: a better way to do this?
Title = external_link (ws_url, Title .. '&nbsp;', 'ws link in title'); -- space char after Title to move icon away from italic text; TODO: a better way to do this?
Title = substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, Title});
Title = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, Title});
Title = Title .. TransTitle .. TransError;
Title = Title .. TransTitle;
else
else
Title = Title .. TransTitle .. TransError;
Title = Title .. TransTitle;
end
end
end
end
else
else
Title = TransTitle .. TransError;
Title = TransTitle;
end
end


if is_set(Place) then
if utilities.is_set (Place) then
Place = " " .. wrap_msg ('written', Place, use_lowercase) .. sepc .. " ";
Place = " " .. wrap_msg ('written', Place, use_lowercase) .. sepc .. " ";
end
end


if is_set (Conference) then
local ConferenceURL_origin = A:ORIGIN('ConferenceURL'); -- get name of parameter that holds ConferenceURL
if is_set (ConferenceURL) then
if utilities.is_set (Conference) then
Conference = external_link( ConferenceURL, Conference, ConferenceURLorigin, nil );
if utilities.is_set (ConferenceURL) then
Conference = external_link( ConferenceURL, Conference, ConferenceURL_origin, nil );
end
end
Conference = sepc .. " " .. Conference .. ConferenceFormat;
Conference = sepc .. " " .. Conference .. ConferenceFormat;
elseif is_set(ConferenceURL) then
elseif utilities.is_set (ConferenceURL) then
Conference = sepc .. " " .. external_link( ConferenceURL, nil, ConferenceURLorigin, nil );
Conference = sepc .. " " .. external_link( ConferenceURL, nil, ConferenceURL_origin, nil );
end
end


if not is_set(Position) then
local Position = '';
if not utilities.is_set (Position) then
local Minutes = A['Minutes'];
local Minutes = A['Minutes'];
local Time = A['Time'];
local Time = A['Time'];


if is_set(Minutes) then
if utilities.is_set (Minutes) then
if is_set (Time) then
if utilities.is_set (Time) then --TODO: make a function for this and similar?
table.insert( z.message_tail, { set_error( 'redundant_parameters', {wrap_style ('parameter', 'minutes') .. ' and ' .. wrap_style ('parameter', 'time')}, true ) } );
utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'minutes') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'time')});
end
end
Position = " " .. Minutes .. " " .. cfg.messages['minutes'];
Position = " " .. Minutes .. " " .. cfg.messages['minutes'];
else
else
if is_set(Time) then
if utilities.is_set (Time) then
local TimeCaption = A['TimeCaption']
local TimeCaption = A['TimeCaption']
if not is_set(TimeCaption) then
if not utilities.is_set (TimeCaption) then
TimeCaption = cfg.messages['event'];
TimeCaption = cfg.messages['event'];
if sepc ~= '.' then
if sepc ~= '.' then
Line 3,219: Line 3,683:
Page, Pages, Sheet, Sheets = format_pages_sheets (Page, Pages, Sheet, Sheets, config.CitationClass, Periodical_origin, sepc, NoPP, use_lowercase);
Page, Pages, Sheet, Sheets = format_pages_sheets (Page, Pages, Sheet, Sheets, config.CitationClass, Periodical_origin, sepc, NoPP, use_lowercase);


At = is_set(At) and (sepc .. " " .. At) or "";
At = utilities.is_set (At) and (sepc .. " " .. At) or "";
Position = is_set(Position) and (sepc .. " " .. Position) or "";
Position = utilities.is_set (Position) and (sepc .. " " .. Position) or "";
if config.CitationClass == 'map' then
if config.CitationClass == 'map' then
local Section = A['Section'];
local Sections = A['Sections']; -- Section (singular) is an alias of Chapter so set earlier
local Sections = A['Sections'];
local Inset = A['Inset'];
local Inset = A['Inset'];
if is_set( Inset ) then
if utilities.is_set ( Inset ) then
Inset = sepc .. " " .. wrap_msg ('inset', Inset, use_lowercase);
Inset = sepc .. " " .. wrap_msg ('inset', Inset, use_lowercase);
end
end


if is_set( Sections ) then
if utilities.is_set ( Sections ) then
Section = sepc .. " " .. wrap_msg ('sections', Sections, use_lowercase);
Section = sepc .. " " .. wrap_msg ('sections', Sections, use_lowercase);
elseif is_set( Section ) then
elseif utilities.is_set ( Section ) then
Section = sepc .. " " .. wrap_msg ('section', Section, use_lowercase);
Section = sepc .. " " .. wrap_msg ('section', Section, use_lowercase);
end
end
Line 3,238: Line 3,701:
end
end


if is_set (Language) then
local Others = A['Others'];
Language = language_parameter (Language); -- format, categories, name from ISO639-1, etc
if utilities.is_set (Others) and 0 == #a and 0 == #e then -- add maint cat when |others= has value and used without |author=, |editor=
else
if config.CitationClass == "AV-media-notes"
Language=""; -- language not specified so make sure this is an empty string;
or config.CitationClass == "audio-visual" then -- special maint for AV/M which has a lot of 'false' positives right now
--[[ TODO: need to extract the wrap_msg from language_parameter
utilities.set_message ('maint_others_avm')
so that we can solve parentheses bunching problem with Format/Language/TitleType
else
]]
utilities.set_message ('maint_others');
end
end
end
 
Others = utilities.is_set (Others) and (sepc .. " " .. Others) or "";
Others = is_set(Others) and (sepc .. " " .. Others) or "";
if is_set (Translators) then
if utilities.is_set (Translators) then
Others = safe_join ({sepc .. ' ', wrap_msg ('translated', Translators, use_lowercase), Others}, sepc);
Others = safe_join ({sepc .. ' ', wrap_msg ('translated', Translators, use_lowercase), Others}, sepc);
end
end
if is_set (Interviewers) then
if utilities.is_set (Interviewers) then
Others = safe_join ({sepc .. ' ', wrap_msg ('interview', Interviewers, use_lowercase), Others}, sepc);
Others = safe_join ({sepc .. ' ', wrap_msg ('interview', Interviewers, use_lowercase), Others}, sepc);
end
end
TitleNote = is_set(TitleNote) and (sepc .. " " .. TitleNote) or "";
local TitleNote = A['TitleNote'];
if is_set (Edition) then
TitleNote = utilities.is_set (TitleNote) and (sepc .. " " .. TitleNote) or "";
if Edition:match ('%f[%a][Ee]d%.?$') or Edition:match ('%f[%a][Ee]dition$') then
if utilities.is_set (Edition) then
add_maint_cat ('extra_text', 'edition');
if Edition:match ('%f[%a][Ee]d%n?%.?$') or Edition:match ('%f[%a][Ee]dition$') then -- Ed, ed, Ed., ed., Edn, edn, Edn., edn.
utilities.set_message ('err_extra_text_edition'); -- add error message
end
end
Edition = " " .. wrap_msg ('edition', Edition);
Edition = " " .. wrap_msg ('edition', Edition);
Line 3,266: Line 3,730:
end
end


Series = is_set (Series) and wrap_msg ('series', {sepc, Series}) or ""; -- not the same as SeriesNum
Series = utilities.is_set (Series) and wrap_msg ('series', {sepc, Series}) or ""; -- not the same as SeriesNum
OrigYear = is_set (OrigYear) and wrap_msg ('origyear', OrigYear) or '';
local Agency = A['Agency'];
Agency = is_set (Agency) and wrap_msg ('agency', {sepc, Agency}) or "";
Agency = utilities.is_set (Agency) and wrap_msg ('agency', {sepc, Agency}) or "";
Volume = format_volume_issue (Volume, Issue, config.CitationClass, Periodical_origin, sepc, use_lowercase);
Volume = format_volume_issue (Volume, Issue, ArticleNumber, config.CitationClass, Periodical_origin, sepc, use_lowercase);
 
------------------------------------ totally unrelated data
Via = is_set (Via) and  wrap_msg ('via', Via) or '';


--[[
if utilities.is_set (AccessDate) then
Subscription implies paywall; Registration does not.  If both are used in a citation, the subscription required link
note is displayed. There are no error messages for this condition.
]]
if is_set (SubscriptionRequired) then
SubscriptionRequired = sepc .. " " .. cfg.messages['subscription']; -- subscription required message
elseif is_set (RegistrationRequired) then
SubscriptionRequired = sepc .. " " .. cfg.messages['registration']; -- registration required message
else
SubscriptionRequired = ''; -- either or both might be set to something other than yes true y
end
 
if is_set(AccessDate) then
local retrv_text = " " .. cfg.messages['retrieved']
local retrv_text = " " .. cfg.messages['retrieved']


AccessDate = nowrap_date (AccessDate); -- wrap in nowrap span if date in appropriate format
AccessDate = nowrap_date (AccessDate); -- wrap in nowrap span if date in appropriate format
if (sepc ~= ".") then retrv_text = retrv_text:lower() end -- if mode is cs2, lower case
if (sepc ~= ".") then retrv_text = retrv_text:lower() end -- if mode is cs2, lower case
AccessDate = substitute (retrv_text, AccessDate); -- add retrieved text
AccessDate = utilities.substitute (retrv_text, AccessDate); -- add retrieved text


AccessDate = substitute (cfg.presentation['accessdate'], {sepc, AccessDate}); -- allow editors to hide accessdates
AccessDate = utilities.substitute (cfg.presentation['accessdate'], {sepc, AccessDate}); -- allow editors to hide accessdates
end
end
if is_set(ID) then ID = sepc .." ".. ID; end
if utilities.is_set (ID) then ID = sepc .. " " .. ID; end
   if "thesis" == config.CitationClass and is_set(Docket) then
ID = sepc .." Docket ".. Docket .. ID;
local Docket = A['Docket'];
   if "thesis" == config.CitationClass and utilities.is_set (Docket) then
ID = sepc .. " Docket " .. Docket .. ID;
end
end
   if "report" == config.CitationClass and is_set(Docket) then -- for cite report when |docket= is set
   if "report" == config.CitationClass and utilities.is_set (Docket) then -- for cite report when |docket= is set
ID = sepc .. ' ' .. Docket; -- overwrite ID even if |id= is set
ID = sepc .. ' ' .. Docket; -- overwrite ID even if |id= is set
end
end


ID_list = build_id_list( ID_list, {IdAccessLevels=ID_access_levels, DoiBroken = DoiBroken, ASINTLD = ASINTLD, IgnoreISBN = IgnoreISBN, Embargo=Embargo, Class = Class} );
if utilities.is_set (URL) then
URL = " " .. external_link( URL, nil, URL_origin, UrlAccess );
end


if is_set(URL) then
local Quote = A['Quote'];
URL = " " .. external_link( URL, nil, URLorigin, UrlAccess );
local TransQuote = A['TransQuote'];
end
local ScriptQuote = A['ScriptQuote'];
if utilities.is_set (Quote) or utilities.is_set (TransQuote) or utilities.is_set (ScriptQuote) then


if is_set(Quote) then
if utilities.is_set (Quote) then
if Quote:sub(1,1) == '"' and Quote:sub(-1,-1) == '"' then -- if first and last characters of quote are quote marks
if Quote:sub(1, 1) == '"' and Quote:sub(-1, -1) == '"' then -- if first and last characters of quote are quote marks
Quote = Quote:sub(2,-2); -- strip them off
Quote = Quote:sub(2, -2); -- strip them off
end
end
end
Quote = sepc .." " .. wrap_style ('quoted-text', Quote ); -- wrap in <q>...</q> tags
PostScript = ""; -- cs1|2 does not supply terminal punctuation when |quote= is set
Quote = kern_quotes (Quote); -- kern if needed
end
Quote = utilities.wrap_style ('quoted-text', Quote ); -- wrap in <q>...</q> tags
local Archived
if utilities.is_set (ScriptQuote) then
if is_set(ArchiveURL) then
Quote = script_concatenate (Quote, ScriptQuote, 'script-quote'); -- <bdi> tags, lang attribute, categorization, etc.; must be done after quote is wrapped
if not is_set(ArchiveDate) then
ArchiveDate = set_error('archive_missing_date');
end
end
if "live" == UrlStatus then
 
local arch_text = cfg.messages['archived'];
if utilities.is_set (TransQuote) then
if sepc ~= "." then arch_text = arch_text:lower() end
if TransQuote:sub(1, 1) == '"' and TransQuote:sub(-1, -1) == '"' then -- if first and last characters of |trans-quote are quote marks
Archived = sepc .. " " .. substitute( cfg.messages['archived-live'],
TransQuote = TransQuote:sub(2, -2); -- strip them off
{ external_link( ArchiveURL, arch_text, A:ORIGIN('ArchiveURL'), nil ) .. ArchiveFormat, ArchiveDate } );
if not is_set(OriginalURL) then
Archived = Archived .. " " .. set_error('archive_missing_url');  
end
end
elseif is_set(OriginalURL) then -- UrlStatus is empty, 'dead', 'unfit', 'usurped', 'bot: unknown'
Quote = Quote .. " " .. utilities.wrap_style ('trans-quoted-title', TransQuote );
local arch_text = cfg.messages['archived-dead'];
end
if sepc ~= "." then arch_text = arch_text:lower() end
 
if in_array (UrlStatus, {'unfit', 'usurped', 'bot: unknown'}) then
if utilities.is_set (QuotePage) or utilities.is_set (QuotePages) then -- add page prefix
Archived = sepc .. " " .. 'Archived from the original on ' .. ArchiveDate; -- format already styled
local quote_prefix = '';
if utilities.is_set (QuotePage) then
extra_text_in_page_check (QuotePage, 'quote-page'); -- add to maint cat if |quote-page= value begins with what looks like p., pp., etc.
if not NoPP then
quote_prefix = utilities.substitute (cfg.messages['p-prefix'], {sepc, QuotePage}), '', '', '';
else
quote_prefix = utilities.substitute (cfg.messages['nopp'], {sepc, QuotePage}), '', '', '';
end
elseif utilities.is_set (QuotePages) then
extra_text_in_page_check (QuotePages, 'quote-pages'); -- add to maint cat if |quote-pages= value begins with what looks like p., pp., etc.
if tonumber(QuotePages) ~= nil and not NoPP then -- if only digits, assume single page
quote_prefix = utilities.substitute (cfg.messages['p-prefix'], {sepc, QuotePages}), '', '';
elseif not NoPP then
quote_prefix = utilities.substitute (cfg.messages['pp-prefix'], {sepc, QuotePages}), '', '';
else
quote_prefix = utilities.substitute (cfg.messages['nopp'], {sepc, QuotePages}), '', '';
end
end
                       
Quote = quote_prefix .. ": " .. Quote;
else
Quote = sepc .. " " .. Quote;
end
 
PostScript = ""; -- cs1|2 does not supply terminal punctuation when |quote= is set
end
-- We check length of PostScript here because it will have been nuked by
-- the quote parameters. We'd otherwise emit a message even if there wasn't
-- a displayed postscript.
-- TODO: Should the max size (1) be configurable?
-- TODO: Should we check a specific pattern?
if utilities.is_set(PostScript) and mw.ustring.len(PostScript) > 1 then
utilities.set_message ('maint_postscript')
end
local Archived;
if utilities.is_set (ArchiveURL) then
if not utilities.is_set (ArchiveDate) then -- ArchiveURL set but ArchiveDate not set
utilities.set_message ('err_archive_missing_date'); -- emit an error message
ArchiveURL = ''; -- empty string for concatenation
ArchiveDate = ''; -- empty string for concatenation
end
else
if utilities.is_set (ArchiveDate) then -- ArchiveURL not set but ArchiveDate is set
utilities.set_message ('err_archive_date_missing_url'); -- emit an error message
ArchiveURL = ''; -- empty string for concatenation
ArchiveDate = ''; -- empty string for concatenation
end
end
 
if utilities.is_set (ArchiveURL) then
local arch_text;
if "live" == UrlStatus then
arch_text = cfg.messages['archived'];
if sepc ~= "." then arch_text = arch_text:lower() end
if utilities.is_set (ArchiveDate) then
Archived = sepc .. ' ' .. utilities.substitute ( cfg.messages['archived-live'],
{external_link( ArchiveURL, arch_text, A:ORIGIN('ArchiveURL'), nil) .. ArchiveFormat, ArchiveDate } );
else
Archived = '';
end
if not utilities.is_set (OriginalURL) then
utilities.set_message ('err_archive_missing_url');
Archived = ''; -- empty string for concatenation
end
elseif utilities.is_set (OriginalURL) then -- UrlStatus is empty, 'dead', 'unfit', 'usurped', 'bot: unknown'
if utilities.in_array (UrlStatus, {'unfit', 'usurped', 'bot: unknown'}) then
arch_text = cfg.messages['archived-unfit'];
if sepc ~= "." then arch_text = arch_text:lower() end
Archived = sepc .. ' ' .. arch_text .. ArchiveDate; -- format already styled
if 'bot: unknown' == UrlStatus then
if 'bot: unknown' == UrlStatus then
add_maint_cat ('bot:_unknown'); -- and add a category if not already added
utilities.set_message ('maint_bot_unknown'); -- and add a category if not already added
else
else
add_maint_cat ('unfit'); -- and add a category if not already added
utilities.set_message ('maint_unfit'); -- and add a category if not already added
end
end
else -- UrlStatus is empty, 'dead'
else -- UrlStatus is empty, 'dead'
Archived = sepc .. " " .. substitute( arch_text,
arch_text = cfg.messages['archived-dead'];
{ external_link( OriginalURL, cfg.messages['original'], OriginalURLorigin, OriginalAccess ) .. OriginalFormat, ArchiveDate } ); -- format already styled
if sepc ~= "." then arch_text = arch_text:lower() end
if utilities.is_set (ArchiveDate) then
Archived = sepc .. " " .. utilities.substitute ( arch_text,
{ external_link( OriginalURL, cfg.messages['original'], OriginalURL_origin, OriginalAccess ) .. OriginalFormat, ArchiveDate } ); -- format already styled
else
Archived = ''; -- unset for concatenation
end
end
end
else
else -- OriginalUrl not set
local arch_text = cfg.messages['archived-missing'];
utilities.set_message ('err_archive_missing_url');
if sepc ~= "." then arch_text = arch_text:lower() end
Archived = ''; -- empty string for concatenation
Archived = sepc .. " " .. substitute( arch_text,
{ set_error('archive_missing_url'), ArchiveDate } );
end
end
elseif is_set (ArchiveFormat) then
elseif utilities.is_set (ArchiveFormat) then
Archived = ArchiveFormat; -- if set and ArchiveURL not set ArchiveFormat has error message
Archived = ArchiveFormat; -- if set and ArchiveURL not set ArchiveFormat has error message
else
else
Archived = ""
Archived = '';
end
end
local Lay = '';
local TranscriptURL = A['TranscriptURL']
if is_set(LayURL) then
local TranscriptFormat = A['TranscriptFormat'];
if is_set(LayDate) then LayDate = " (" .. LayDate .. ")" end
TranscriptFormat = style_format (TranscriptFormat, TranscriptURL, 'transcript-format', 'transcripturl');
if is_set(LaySource) then  
local Transcript = A['Transcript'];
LaySource = " &ndash; ''" .. safe_for_italics(LaySource) .. "''";
local TranscriptURL_origin = A:ORIGIN('TranscriptURL'); -- get name of parameter that holds TranscriptURL
else
if utilities.is_set (Transcript) then
LaySource = "";
if utilities.is_set (TranscriptURL) then
Transcript = external_link( TranscriptURL, Transcript, TranscriptURL_origin, nil );
end
end
if sepc == '.' then
Transcript = sepc .. ' ' .. Transcript .. TranscriptFormat;
Lay = sepc .. " " .. external_link( LayURL, cfg.messages['lay summary'], A:ORIGIN('LayURL'), nil ) .. LayFormat .. LaySource .. LayDate
elseif utilities.is_set (TranscriptURL) then
else
Transcript = external_link( TranscriptURL, nil, TranscriptURL_origin, nil );
Lay = sepc .. " " .. external_link( LayURL, cfg.messages['lay summary']:lower(), A:ORIGIN('LayURL'), nil ) .. LayFormat .. LaySource .. LayDate
end
elseif is_set (LayFormat) then -- Test if |lay-format= is given without giving a |lay-url=
Lay = sepc .. LayFormat; -- if set and LayURL not set, then LayFormat has error message
end
 
if is_set(Transcript) then
if is_set(TranscriptURL) then
Transcript = external_link( TranscriptURL, Transcript, TranscriptURLorigin, nil );
end
Transcript = sepc .. ' ' .. Transcript .. TranscriptFormat;
elseif is_set(TranscriptURL) then
Transcript = external_link( TranscriptURL, nil, TranscriptURLorigin, nil );
end
end


local Publisher;
local Publisher;
if is_set(PublicationDate) then
if utilities.is_set (PublicationDate) then
PublicationDate = wrap_msg ('published', PublicationDate);
PublicationDate = wrap_msg ('published', PublicationDate);
end
end
if is_set(PublisherName) then
if utilities.is_set (PublisherName) then
if is_set(PublicationPlace) then
if utilities.is_set (PublicationPlace) then
Publisher = sepc .. " " .. PublicationPlace .. ": " .. PublisherName .. PublicationDate;
Publisher = sepc .. " " .. PublicationPlace .. ": " .. PublisherName .. PublicationDate;
else
else
Publisher = sepc .. " " .. PublisherName .. PublicationDate;   
Publisher = sepc .. " " .. PublisherName .. PublicationDate;   
end
end
elseif is_set(PublicationPlace) then  
elseif utilities.is_set (PublicationPlace) then  
Publisher= sepc .. " " .. PublicationPlace .. PublicationDate;
Publisher= sepc .. " " .. PublicationPlace .. PublicationDate;
else  
else  
Line 3,401: Line 3,912:
-- Several of the above rely upon detecting this as nil, so do it last.
-- Several of the above rely upon detecting this as nil, so do it last.
if (is_set (Periodical) or is_set (ScriptPeriodical) or is_set (TransPeriodical)) then
if (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical) or utilities.is_set (TransPeriodical)) then
if is_set(Title) or is_set(TitleNote) then  
if utilities.is_set (Title) or utilities.is_set (TitleNote) then  
Periodical = sepc .. " " .. format_periodical (ScriptPeriodical, ScriptPeriodical_origin, Periodical, TransPeriodical);
Periodical = sepc .. " " .. format_periodical (ScriptPeriodical, ScriptPeriodical_origin, Periodical, TransPeriodical, TransPeriodical_origin);
else  
else  
Periodical = format_periodical (ScriptPeriodical, ScriptPeriodical_origin, Periodical, TransPeriodical);
Periodical = format_periodical (ScriptPeriodical, ScriptPeriodical_origin, Periodical, TransPeriodical, TransPeriodical_origin);
end
end
end
local Language = A['Language'];
if utilities.is_set (Language) then
Language = language_parameter (Language); -- format, categories, name from ISO639-1, etc.
else
Language=''; -- language not specified so make sure this is an empty string;
--[[ TODO: need to extract the wrap_msg from language_parameter
so that we can solve parentheses bunching problem with Format/Language/TitleType
]]
end
end


Line 3,414: Line 3,935:
]]
]]
if "speech" == config.CitationClass then -- cite speech only
if "speech" == config.CitationClass then -- cite speech only
TitleNote = " (Speech)"; -- annotate the citation
TitleNote = TitleType; -- move TitleType to TitleNote so that it renders ahead of |event=
if is_set (Periodical) then -- if Periodical, perhaps because of an included |website= or |journal= parameter  
TitleType = ''; -- and unset
if is_set (Conference) then -- and if |event= is set
 
if utilities.is_set (Periodical) then -- if Periodical, perhaps because of an included |website= or |journal= parameter  
if utilities.is_set (Conference) then -- and if |event= is set
Conference = Conference .. sepc .. " "; -- then add appropriate punctuation to the end of the Conference variable before rendering
Conference = Conference .. sepc .. " "; -- then add appropriate punctuation to the end of the Conference variable before rendering
end
end
Line 3,429: Line 3,952:
local tcommon2; -- used for book cite when |contributor= is set
local tcommon2; -- used for book cite when |contributor= is set
if in_array(config.CitationClass, {"journal","citation"}) and is_set(Periodical) then
if utilities.in_array (config.CitationClass, {"journal", "citation"}) and utilities.is_set (Periodical) then
if is_set(Others) then Others = safe_join ({Others, sepc .. " "}, sepc) end -- add terminal punctuation & space; check for dup sepc; TODO why do we need to do this here?
if not (utilities.is_set (Authors) or utilities.is_set (Editors)) then
Others = Others:gsub ('^' .. sepc .. ' ', ''); -- when no authors and no editors, strip leading sepc and space
end
if utilities.is_set (Others) then Others = safe_join ({Others, sepc .. " "}, sepc) end -- add terminal punctuation & space; check for dup sepc; TODO why do we need to do this here?
tcommon = safe_join( {Others, Title, TitleNote, Conference, Periodical, Format, TitleType, Series, Language, Edition, Publisher, Agency, Volume}, sepc );
tcommon = safe_join( {Others, Title, TitleNote, Conference, Periodical, Format, TitleType, Series, Language, Edition, Publisher, Agency, Volume}, sepc );
elseif in_array(config.CitationClass, {"book","citation"}) and not is_set(Periodical) then -- special cases for book cites
elseif utilities.in_array (config.CitationClass, {"book", "citation"}) and not utilities.is_set (Periodical) then -- special cases for book cites
if is_set (Contributors) then -- when we are citing foreword, preface, introduction, etc
if utilities.is_set (Contributors) then -- when we are citing foreword, preface, introduction, etc.
tcommon = safe_join( {Title, TitleNote}, sepc ); -- author and other stuff will come after this and before tcommon2
tcommon = safe_join( {Title, TitleNote}, sepc ); -- author and other stuff will come after this and before tcommon2
tcommon2 = safe_join( {Conference, Periodical, Format, TitleType, Series, Language, Volume, Others, Edition, Publisher, Agency}, sepc );
tcommon2 = safe_join( {Conference, Periodical, Format, TitleType, Series, Language, Volume, Others, Edition, Publisher, Agency}, sepc );
Line 3,441: Line 3,967:


elseif 'map' == config.CitationClass then -- special cases for cite map
elseif 'map' == config.CitationClass then -- special cases for cite map
if is_set (Chapter) then -- map in a book; TitleType is part of Chapter
if utilities.is_set (Chapter) then -- map in a book; TitleType is part of Chapter
tcommon = safe_join( {Title, Format, Edition, Scale, Series, Language, Cartography, Others, Publisher, Volume}, sepc );
tcommon = safe_join( {Title, Format, Edition, Scale, Series, Language, Cartography, Others, Publisher, Volume}, sepc );
elseif is_set (Periodical) then -- map in a periodical
elseif utilities.is_set (Periodical) then -- map in a periodical
tcommon = safe_join( {Title, TitleType, Format, Periodical, Scale, Series, Language, Cartography, Others, Publisher, Volume}, sepc );
tcommon = safe_join( {Title, TitleType, Format, Periodical, Scale, Series, Language, Cartography, Others, Publisher, Volume}, sepc );
else -- a sheet or stand-alone map
else -- a sheet or stand-alone map
Line 3,463: Line 3,989:
end
end
local Via = A['Via'];
Via = utilities.is_set (Via) and  wrap_msg ('via', Via) or '';
local idcommon;
local idcommon;
if 'audio-visual' == config.CitationClass or 'episode' == config.CitationClass then -- special case for cite AV media & cite episode position transcript
if 'audio-visual' == config.CitationClass or 'episode' == config.CitationClass then -- special case for cite AV media & cite episode position transcript
idcommon = safe_join( { ID_list, URL, Archived, Transcript, AccessDate, Via, SubscriptionRequired, Lay, Quote }, sepc );
idcommon = safe_join( { ID_list, URL, Archived, Transcript, AccessDate, Via, Quote }, sepc );
else
else
idcommon = safe_join( { ID_list, URL, Archived, AccessDate, Via, SubscriptionRequired, Lay, Quote }, sepc );
idcommon = safe_join( { ID_list, URL, Archived, AccessDate, Via, Quote }, sepc );
end
end
Line 3,473: Line 4,001:
local pgtext = Position .. Sheet .. Sheets .. Page .. Pages .. At;
local pgtext = Position .. Sheet .. Sheets .. Page .. Pages .. At;


if is_set(Date) then
local OrigDate = A['OrigDate'];
if is_set (Authors) or is_set (Editors) then -- date follows authors or editors when authors not set
OrigDate = utilities.is_set (OrigDate) and wrap_msg ('origdate', OrigDate) or '';
Date = " (" .. Date ..")" .. OrigYear .. sepc .. " "; -- in paranetheses
if utilities.is_set (Date) then
if utilities.is_set (Authors) or utilities.is_set (Editors) then -- date follows authors or editors when authors not set
Date = " (" .. Date .. ")" .. OrigDate .. sepc .. " "; -- in parentheses
else -- neither of authors and editors set
else -- neither of authors and editors set
if (string.sub(tcommon,-1,-1) == sepc) then -- if the last character of tcommon is sepc
if (string.sub(tcommon, -1, -1) == sepc) then -- if the last character of tcommon is sepc
Date = " " .. Date .. OrigYear; -- Date does not begin with sepc
Date = " " .. Date .. OrigDate; -- Date does not begin with sepc
else
else
Date = sepc .. " " .. Date .. OrigYear; -- Date begins with sepc
Date = sepc .. " " .. Date .. OrigDate; -- Date begins with sepc
end
end
end
end
end
end
if is_set(Authors) then
if utilities.is_set (Authors) then
if (not is_set (Date)) then -- when date is set it's in parentheses; no Authors termination
if (not utilities.is_set (Date)) then -- when date is set it's in parentheses; no Authors termination
Authors = terminate_name_list (Authors, sepc); -- when no date, terminate with 0 or 1 sepc and a space
Authors = terminate_name_list (Authors, sepc); -- when no date, terminate with 0 or 1 sepc and a space
end
end
if is_set(Editors) then
if utilities.is_set (Editors) then
local in_text = " ";
local in_text = '';
local post_text = "";
local post_text = '';
if is_set(Chapter) and 0 == #c then
if utilities.is_set (Chapter) and 0 == #c then
in_text = in_text .. cfg.messages['in'] .. " "
in_text = cfg.messages['in'] .. ' ';
if (sepc ~= '.') then
if (sepc ~= '.') then
in_text = in_text:lower() -- lowercase for cs2
in_text = in_text:lower(); -- lowercase for cs2
end
end
end
end
if EditorCount <= 1 then
if EditorCount <= 1 then
post_text = " (" .. cfg.messages['editor'] .. ")"; -- be consistent with no-author, no-date case
post_text = ' (' .. cfg.messages['editor'] .. ')'; -- be consistent with no-author, no-date case
else
else
post_text = " (" .. cfg.messages['editors'] .. ")";
post_text = ' (' .. cfg.messages['editors'] .. ')';
end
end
Editors = terminate_name_list (in_text .. Editors .. post_text, sepc); -- terminate with 0 or 1 sepc and a space
Editors = terminate_name_list (in_text .. Editors .. post_text, sepc); -- terminate with 0 or 1 sepc and a space
end
end
if is_set (Contributors) then -- book cite and we're citing the intro, preface, etc
if utilities.is_set (Contributors) then -- book cite and we're citing the intro, preface, etc.
local by_text = sepc .. ' ' .. cfg.messages['by'] .. ' ';
local by_text = sepc .. ' ' .. cfg.messages['by'] .. ' ';
if (sepc ~= '.') then by_text = by_text:lower() end -- lowercase for cs2
if (sepc ~= '.') then by_text = by_text:lower() end -- lowercase for cs2
Authors = by_text .. Authors; -- author follows title so tweak it here
Authors = by_text .. Authors; -- author follows title so tweak it here
if is_set (Editors) and is_set (Date) then -- when Editors make sure that Authors gets terminated
if utilities.is_set (Editors) and utilities.is_set (Date) then -- when Editors make sure that Authors gets terminated
Authors = terminate_name_list (Authors, sepc); -- terminate with 0 or 1 sepc and a space
Authors = terminate_name_list (Authors, sepc); -- terminate with 0 or 1 sepc and a space
end
end
if (not is_set (Date)) then -- when date is set it's in parentheses; no Contributors termination
if (not utilities.is_set (Date)) then -- when date is set it's in parentheses; no Contributors termination
Contributors = terminate_name_list (Contributors, sepc); -- terminate with 0 or 1 sepc and a space
Contributors = terminate_name_list (Contributors, sepc); -- terminate with 0 or 1 sepc and a space
end
end
Line 3,518: Line 4,048:
text = safe_join( {Authors, Date, Chapter, Place, Editors, tcommon, pgtext, idcommon }, sepc );
text = safe_join( {Authors, Date, Chapter, Place, Editors, tcommon, pgtext, idcommon }, sepc );
end
end
elseif is_set(Editors) then
elseif utilities.is_set (Editors) then
if is_set(Date) then
if utilities.is_set (Date) then
if EditorCount <= 1 then
if EditorCount <= 1 then
Editors = Editors .. ", " .. cfg.messages['editor'];
Editors = Editors .. cfg.presentation['sep_name'] .. cfg.messages['editor'];
else
else
Editors = Editors .. ", " .. cfg.messages['editors'];
Editors = Editors .. cfg.presentation['sep_name'] .. cfg.messages['editors'];
end
end
else
else
Line 3,534: Line 4,064:
text = safe_join( {Editors, Date, Chapter, Place, tcommon, pgtext, idcommon}, sepc );
text = safe_join( {Editors, Date, Chapter, Place, tcommon, pgtext, idcommon}, sepc );
else
else
if in_array(config.CitationClass, {"journal","citation"}) and is_set(Periodical) then
if utilities.in_array (config.CitationClass, {"journal", "citation"}) and utilities.is_set (Periodical) then
text = safe_join( {Chapter, Place, tcommon, pgtext, Date, idcommon}, sepc );
text = safe_join( {Chapter, Place, tcommon, pgtext, Date, idcommon}, sepc );
else
else
Line 3,541: Line 4,071:
end
end
if is_set(PostScript) and PostScript ~= sepc then
if utilities.is_set (PostScript) and PostScript ~= sepc then
text = safe_join( {text, sepc}, sepc ); --Deals with italics, spaces, etc.
text = safe_join( {text, sepc}, sepc ); -- Deals with italics, spaces, etc.
text = text:sub(1,-sepc:len()-1);
text = text:sub(1, -sepc:len() - 1);
end
end
text = safe_join( {text, PostScript}, sepc );
text = safe_join( {text, PostScript}, sepc );


-- Now enclose the whole thing in a <cite/> element
-- Now enclose the whole thing in a <cite> element
local options = {};
local options_t = {};
options_t.class = cite_class_attribute_make (config.CitationClass, Mode);
if is_set(config.CitationClass) and config.CitationClass ~= "citation" then
 
options.class = config.CitationClass;
local Ref = is_valid_parameter_value (A['Ref'], A:ORIGIN('Ref'), cfg.keywords_lists['ref'], nil, true); -- nil when |ref=harv; A['Ref'] else
options.class = "citation " .. config.CitationClass; -- class=citation required for blue highlight when used with |ref=
 
else
if 'none' ~= cfg.keywords_xlate[(Ref and Ref:lower()) or ''] then
options.class = "citation";
local namelist_t = {}; -- holds selected contributor, author, editor name list
end
local year = first_set ({Year, anchor_year}, 2); -- Year first for legacy citations and for YMD dates that require disambiguation
if is_set(Ref) and Ref:lower() ~= "none" then -- set reference anchor if appropriate
local id = Ref
if ('harv' == Ref ) then
local namelist = {}; -- holds selected contributor, author, editor name list
local year = first_set ({Year, anchor_year}, 2); -- Year first for legacy citations and for YMD dates that require disambiguation


if #c > 0 then -- if there is a contributor list
if #c > 0 then -- if there is a contributor list
namelist = c; -- select it
namelist_t = c; -- select it
elseif #a > 0 then -- or an author list
elseif #a > 0 then -- or an author list
namelist = a;
namelist_t = a;
elseif #e > 0 then -- or an editor list
elseif #e > 0 then -- or an editor list
namelist = e;
namelist_t = e;
end
end
if #namelist > 0 then -- if there are names in namelist
local citeref_id;
id = anchor_id (namelist, year); -- go make the CITEREF anchor
if #namelist_t > 0 then -- if there are names in namelist_t
else
citeref_id = make_citeref_id (namelist_t, year); -- go make the CITEREF anchor
id = ''; -- unset
if mw.uri.anchorEncode (citeref_id) == ((Ref and mw.uri.anchorEncode (Ref)) or '') then -- Ref may already be encoded (by {{sfnref}}) so citeref_id must be encoded before comparison
utilities.set_message ('maint_ref_duplicates_default');
end
end
else
citeref_id = ''; -- unset
end
end
options.id = id;
options_t.id = Ref or citeref_id;
end
end
 
if string.len(text:gsub("<span[^>/]*>(.-)</span>", "%1"):gsub("%b<>","")) <= 2 then -- remove <span> tags and other html-like markup; then get length of what remains
if string.len (text:gsub('%b<>', '')) <= 2 then -- remove html and html-like tags; then get length of what remains;
z.error_categories = {};
z.error_cats_t = {}; -- blank the categories list
text = set_error('empty_citation');
z.error_msgs_t = {}; -- blank the error messages list
z.message_tail = {};
OCinSoutput = nil; -- blank the metadata string
text = ''; -- blank the the citation
utilities.set_message ('err_empty_citation'); -- set empty citation message and category
end
end
local render = {}; -- here we collect the final bits for concatenation into the rendered citation
local render_t = {}; -- here we collect the final bits for concatenation into the rendered citation


if is_set(options.id) then -- here we wrap the rendered citation in <cite ...>...</cite> tags
if utilities.is_set (options_t.id) then -- here we wrap the rendered citation in <cite ...>...</cite> tags
table.insert (render, substitute (cfg.presentation['cite-id'], {mw.uri.anchorEncode(options.id), mw.text.nowiki(options.class), text})); -- when |ref= is set
table.insert (render_t, utilities.substitute (cfg.presentation['cite-id'], {mw.uri.anchorEncode(options_t.id), mw.text.nowiki(options_t.class), text})); -- when |ref= is set or when there is a namelist
else
else
table.insert (render, substitute (cfg.presentation['cite'], {mw.text.nowiki(options.class), text})); -- all other cases
table.insert (render_t, utilities.substitute (cfg.presentation['cite'], {mw.text.nowiki(options_t.class), text})); -- when |ref=none or when namelist_t empty and |ref= is missing or is empty
end
end


table.insert (render, substitute (cfg.presentation['ocins'], {OCinSoutput})); -- append metadata to the citation
if OCinSoutput then -- blanked when citation is 'empty' so don't bother to add boilerplate metadata span
table.insert (render_t, utilities.substitute (cfg.presentation['ocins'], OCinSoutput)); -- format and append metadata to the citation
end
 
local template_name = ('citation' == config.CitationClass) and 'citation' or 'cite ' .. (cfg.citation_class_map_t[config.CitationClass] or config.CitationClass);
local template_link = '[[Template:' .. template_name .. '|' .. template_name .. ']]';
local msg_prefix = '<code class="cs1-code">{{' .. template_link .. '}}</code>: ';
 
if 0 ~= #z.error_msgs_t then
mw.addWarning (utilities.substitute (cfg.messages.warning_msg_e, template_link));
 
table.insert (render_t, ' '); -- insert a space between citation and its error messages
table.sort (z.error_msgs_t); -- sort the error messages list; sorting includes wrapping <span> and <code> tags; hidden-error sorts ahead of visible-error


if 0 ~= #z.message_tail then
local hidden = true; -- presume that the only error messages emited by this template are hidden
table.insert (render, ' ');
for _, v in ipairs (z.error_msgs_t) do -- spin through the list of error messages
for i,v in ipairs( z.message_tail ) do
if v:find ('cs1-visible-error', 1, true) then -- look for the visible error class name
if is_set(v[1]) then
hidden = false; -- found one; so don't hide the error message prefix
if i == #z.message_tail then
break; -- and done because no need to look further
table.insert (render, error_comment( v[1], v[2] ));
else
table.insert (render, error_comment( v[1] .. "; ", v[2] ));
end
end
end
end
end
z.error_msgs_t[1] = table.concat ({utilities.error_comment (msg_prefix, hidden), z.error_msgs_t[1]}); -- add error message prefix to first error message to prevent extraneous punctuation
table.insert (render_t, table.concat (z.error_msgs_t, '; ')); -- make a big string of error messages and add it to the rendering
end
end


if 0 ~= #z.maintenance_cats then
if 0 ~= #z.maint_cats_t then
local maint_msgs = {}; -- here we collect all of the maint messages
mw.addWarning (utilities.substitute (cfg.messages.warning_msg_m, template_link));
for _, v in ipairs( z.maintenance_cats ) do -- append maintenance categories
 
local maint = {}; -- here we assemble a maintenence message
table.sort (z.maint_cats_t); -- sort the maintenance messages list
table.insert (maint, v); -- maint msg is the category name
 
table.insert (maint, ' ('); -- open the link text
local maint_msgs_t = {}; -- here we collect all of the maint messages
table.insert (maint, make_wikilink (':Category:' .. v, 'link')); -- add the link
 
table.insert (maint, ')'); -- and close it
if 0 == #z.error_msgs_t then -- if no error messages
table.insert (maint_msgs, table.concat (maint)); -- assemble new maint message and add it to the maint_msgs table
table.insert (maint_msgs_t, msg_prefix); -- insert message prefix in maint message livery
end
end
table.insert (render, substitute (cfg.presentation['hidden-maint'], table.concat (maint_msgs, ' '))); -- wrap the group of maint message with proper presentation and save
for _, v in ipairs( z.maint_cats_t ) do -- append maintenance categories
table.insert (maint_msgs_t, -- assemble new maint message and add it to the maint_msgs_t table
table.concat ({v, ' (', utilities.substitute (cfg.messages[':cat wikilink'], v), ')'})
);
end
table.insert (render_t, utilities.substitute (cfg.presentation['hidden-maint'], table.concat (maint_msgs_t, ' '))); -- wrap the group of maint messages with proper presentation and save
end
end
 
no_tracking_cats = no_tracking_cats:lower();
if not no_tracking_cats then
if in_array(no_tracking_cats, {"", "no", "false", "n"}) then
local sort_key;
for _, v in ipairs( z.error_categories ) do
local cat_wikilink = 'cat wikilink';
table.insert (render, make_wikilink ('Category:' .. v));
if cfg.enable_sort_keys then -- when namespace sort keys enabled
local namespace_number = mw.title.getCurrentTitle().namespace; -- get namespace number for this wikitext
sort_key = (0 ~= namespace_number and (cfg.name_space_sort_keys[namespace_number] or cfg.name_space_sort_keys.other)) or nil; -- get sort key character; nil for mainspace
cat_wikilink = (not sort_key and 'cat wikilink') or 'cat wikilink sk'; -- make <cfg.messages> key
end
 
for _, v in ipairs (z.error_cats_t) do -- append error categories
table.insert (render_t, utilities.substitute (cfg.messages[cat_wikilink], {v, sort_key}));
end
end
for _, v in ipairs( z.maintenance_cats ) do -- append maintenance categories
for _, v in ipairs (z.maint_cats_t) do -- append maintenance categories
table.insert (render, make_wikilink ('Category:' .. v));
table.insert (render_t, utilities.substitute (cfg.messages[cat_wikilink], {v, sort_key}));
end
end
for _, v in ipairs( z.properties_cats ) do -- append properties categories
 
table.insert (render, make_wikilink ('Category:' .. v));
for _, v in ipairs (z.prop_cats_t) do -- append properties categories
table.insert (render_t, utilities.substitute (cfg.messages['cat wikilink'], v)); -- no sort keys
end
end
end
end


return table.concat (render);
return table.concat (render_t); -- make a big string and done
end
end


Line 3,650: Line 4,204:
]]
]]


local function validate (name, cite_class)
local function validate (name, cite_class, empty)
local name = tostring (name);
local name = tostring (name);
local enum_name; -- parameter name with enumerator (if any) replaced with '#'
local state;
local state;
local function state_test (state, name) -- local function to do testing of state values
if in_array (cite_class, {'arxiv', 'biorxiv', 'citeseerx', 'ssrn'}) then -- limited parameter sets allowed for these templates
state = whitelist.limited_basic_arguments[name];
if true == state then return true; end -- valid actively supported parameter
if true == state then return true; end -- valid actively supported parameter
if false == state then
if false == state then
if empty then return nil; end -- empty deprecated parameters are treated as unknowns
deprecated_parameter (name); -- parameter is deprecated but still supported
deprecated_parameter (name); -- parameter is deprecated but still supported
return true;
return true;
end
end
 
if 'tracked' == state then
state = whitelist[cite_class .. '_basic_arguments'][name]; -- look in the parameter-list for the template identified by cite_class
local base_name = name:gsub ('%d', ''); -- strip enumerators from parameter names that have them to get the base name
 
utilities.add_prop_cat ('tracked-param', {base_name}, base_name); -- add a properties category; <base_name> modifies <key>
if true == state then return true; end -- valid actively supported parameter
if false == state then
deprecated_parameter (name); -- parameter is deprecated but still supported
return true;
end
-- limited enumerated parameters list
name = name:gsub("%d+", "#" ); -- replace digit(s) with # (last25 becomes last#) (mw.ustring because non-Western 'local' digits)
state = whitelist.limited_numbered_arguments[name];
if true == state then return true; end -- valid actively supported parameter
if false == state then
deprecated_parameter (name); -- parameter is deprecated but still supported
return true;
return true;
end
end
return nil;
end


return false; -- not supported because not found or name is set to nil
if name:find ('#') then -- # is a cs1|2 reserved character so parameters with # not permitted
end -- end limited parameter-set templates
return nil;
state = whitelist.basic_arguments[name]; -- all other templates; all normal parameters allowed
if true == state then return true; end -- valid actively supported parameter
if false == state then
deprecated_parameter (name); -- parameter is deprecated but still supported
return true;
end
end
-- all enumerated parameters allowed
-- replace wnumerator digit(s) with # (|last25= becomes |last#=) (mw.ustring because non-Western 'local' digits)
name = name:gsub("%d+", "#" ); -- replace digit(s) with # (last25 becomes last#) (mw.ustring because non-Western 'local' digits)
enum_name = mw.ustring.gsub (name, '%d+$', '#'); -- where enumerator is last charaters in parameter name (these to protect |s2cid=)
state = whitelist.numbered_arguments[name];
enum_name = mw.ustring.gsub (enum_name, '%d+([%-l])', '#%1'); -- where enumerator is in the middle of the parameter name; |author#link= is the oddity


if true == state then return true; end -- valid actively supported parameter
if 'document' == cite_class then -- special case for {{cite document}}
if false == state then
state = whitelist.document_parameters_t[enum_name]; -- this list holds enumerated and nonenumerated parameters
deprecated_parameter (name); -- parameter is deprecated but still supported
if true == state_test (state, name) then return true; end
return true;
return false;
end
end
return false; -- not supported because not found or name is set to nil
end


if utilities.in_array (cite_class, whitelist.preprint_template_list_t) then -- limited parameter sets allowed for these templates
state = whitelist.limited_parameters_t[enum_name]; -- this list holds enumerated and nonenumerated parameters
if true == state_test (state, name) then return true; end


--[[--------------------------< M I S S I N G _ P I P E _ C H E C K >------------------------------------------
state = whitelist.preprint_arguments_t[cite_class][name]; -- look in the parameter-list for the template identified by cite_class
if true == state_test (state, name) then return true; end
 
return false; -- not supported because not found or name is set to nil
end -- end limited parameter-set templates
 
if utilities.in_array (cite_class, whitelist.unique_param_template_list_t) then -- template-specific parameters for templates that accept parameters from the basic argument list
state = whitelist.unique_arguments_t[cite_class][name]; -- look in the template-specific parameter-lists for the template identified by cite_class
if true == state_test (state, name) then return true; end
end -- if here, fall into general validation
 
state = whitelist.common_parameters_t[enum_name]; -- all other templates; all normal parameters allowed; this list holds enumerated and nonenumerated parameters
if true == state_test (state, name) then return true; end
 
return false; -- not supported because not found or name is set to nil
end
 
 
--[=[-------------------------< I N T E R _ W I K I _ C H E C K >----------------------------------------------
 
check <value> for inter-language interwiki-link markup.  <prefix> must be a MediaWiki-recognized language
code.  when these values have the form (without leading colon):
[[<prefix>:link|label]] return label as plain-text
[[<prefix>:link]] return <prefix>:link as plain-text
 
return value as is else
 
]=]
 
local function inter_wiki_check (parameter, value)
local prefix = value:match ('%[%[(%a+):'); -- get an interwiki prefix if one exists
local _;
if prefix and cfg.inter_wiki_map[prefix:lower()] then -- if prefix is in the map, needs preceding colon so
utilities.set_message ('err_bad_paramlink', parameter); -- emit an error message
_, value, _ = utilities.is_wikilink (value); -- extract label portion from wikilink
end
return value;
end
 
 
--[[--------------------------< M I S S I N G _ P I P E _ C H E C K >------------------------------------------


Look at the contents of a parameter. If the content has a string of characters and digits followed by an equal
Look at the contents of a parameter. If the content has a string of characters and digits followed by an equal
sign, compare the alphanumeric string to the list of cs1|2 parameters.  If found, then the string is possibly a
sign, compare the alphanumeric string to the list of cs1|2 parameters.  If found, then the string is possibly a
parameter that is missing its pipe:
parameter that is missing its pipe.  There are two tests made:
{{cite ... |title=Title access-date=2016-03-17}}
{{cite ... |title=Title access-date=2016-03-17}} -- the first parameter has a value and whitespace separates that value from the missing pipe parameter name
 
{{cite ... |title=access-date=2016-03-17}} -- the first parameter has no value (whitespace after the first = is trimmed by MediaWiki)
cs1|2 shares some parameter names with xml/html atributes: class=, title=, etc.  To prevent false positives xml/html
cs1|2 shares some parameter names with XML/HTML attributes: class=, title=, etc.  To prevent false positives XML/HTML
tags are removed before the search.
tags are removed before the search.


Line 3,718: Line 4,298:
local function missing_pipe_check (parameter, value)
local function missing_pipe_check (parameter, value)
local capture;
local capture;
value = value:gsub ('%b<>', ''); -- remove xml/html tags because attributes: class=, title=, etc  
value = value:gsub ('%b<>', ''); -- remove XML/HTML tags because attributes: class=, title=, etc.


capture = value:match ('%s+(%a[%w%-]+)%s*=') or value:match ('^(%a[%w%-]+)%s*='); -- find and categorize parameters with possible missing pipes
capture = value:match ('%s+(%a[%w%-]+)%s*=') or value:match ('^(%a[%w%-]+)%s*='); -- find and categorize parameters with possible missing pipes
if capture and validate (capture) then -- if the capture is a valid parameter name
if capture and validate (capture) then -- if the capture is a valid parameter name
table.insert( z.message_tail, {set_error ('missing_pipe',parameter)});
utilities.set_message ('err_missing_pipe', parameter);
end
end
end
end
Line 3,734: Line 4,314:


local function has_extraneous_punc (param, value)
local function has_extraneous_punc (param, value)
if 'number' == type (param) then
return;
end
param = param:gsub ('%d+', '#'); -- enumerated name-list mask params allow terminal punct; normalize
if cfg.punct_skip[param] then
if cfg.punct_skip[param] then
return; -- parameter name found in the skip table so done
return; -- parameter name found in the skip table so done
Line 3,739: Line 4,324:
if value:match ('[,;:]$') then
if value:match ('[,;:]$') then
add_maint_cat ('extra_punct'); -- has extraneous punctuation; add maint cat
utilities.set_message ('maint_extra_punct'); -- has extraneous punctuation; add maint cat
end
if value:match ('^=') then -- sometimes an extraneous '=' character appears ...
utilities.set_message ('maint_extra_punct'); -- has extraneous punctuation; add maint cat
end
end
 
 
--[[--------------------------< H A S _ E X T R A N E O U S _ U R L >------------------------------------------
 
look for extraneous url parameter values; parameters listed in skip table are not checked
 
]]
 
local function has_extraneous_url (url_param_t)
local url_error_t = {};
check_for_url (url_param_t, url_error_t); -- extraneous url check
if 0 ~= #url_error_t then -- non-zero when there are errors
table.sort (url_error_t);
utilities.set_message ('err_param_has_ext_link', {utilities.make_sep_list (#url_error_t, url_error_t)}); -- add this error message
end
end
end
end
Line 3,751: Line 4,356:


local function citation(frame)
local function citation(frame)
Frame = frame; -- save a copy incase we need to display an error message in preview mode
Frame = frame; -- save a copy in case we need to display an error message in preview mode
 
local config = {}; -- table to store parameters from the module {{#invoke:}}
for k, v in pairs( frame.args ) do -- get parameters from the {{#invoke}} frame
config[k] = v;
-- args[k] = v; -- crude debug support that allows us to render a citation from module {{#invoke:}}; skips parameter validation; TODO: keep?
end
-- i18n: set the name that your wiki uses to identify sandbox subpages from sandbox template invoke (or can be set here)
local sandbox = ((config.SandboxPath and '' ~= config.SandboxPath) and config.SandboxPath) or '/sandbox'; -- sandbox path from {{#invoke:Citation/CS1/sandbox|citation|SandboxPath=/...}}
is_sandbox = nil ~= string.find (frame:getTitle(), sandbox, 1, true); -- is this invoke the sandbox module?
sandbox = is_sandbox and sandbox or ''; -- use i18n sandbox to load sandbox modules when this module is the sandox; live modules else
 
local pframe = frame:getParent()
local pframe = frame:getParent()
local validation, utilities, identifiers, metadata, styles;
local styles;
if nil ~= string.find (frame:getTitle(), 'sandbox', 1, true) then -- did the {{#invoke:}} use sandbox version?
cfg = mw.loadData ('Module:Citation/CS1/Configuration' .. sandbox); -- load sandbox versions of support modules when {{#invoke:Citation/CS1/sandbox|...}}; live modules else
cfg = mw.loadData ('Module:Citation/CS1/Configuration/sandbox'); -- load sandbox versions of support modules
whitelist = mw.loadData ('Module:Citation/CS1/Whitelist' .. sandbox);
whitelist = mw.loadData ('Module:Citation/CS1/Whitelist/sandbox');
utilities = require ('Module:Citation/CS1/Utilities' .. sandbox);
utilities = require ('Module:Citation/CS1/Utilities/sandbox');
validation = require ('Module:Citation/CS1/Date_validation' .. sandbox);
validation = require ('Module:Citation/CS1/Date_validation/sandbox');
identifiers = require ('Module:Citation/CS1/Identifiers' .. sandbox);
identifiers = require ('Module:Citation/CS1/Identifiers/sandbox');
metadata = require ('Module:Citation/CS1/COinS' .. sandbox);
metadata = require ('Module:Citation/CS1/COinS/sandbox');
styles = 'Module:Citation/CS1' .. sandbox .. '/styles.css';
styles = 'Module:Citation/CS1/sandbox/styles.css';
else -- otherwise
cfg = mw.loadData ('Module:Citation/CS1/Configuration'); -- load live versions of support modules
whitelist = mw.loadData ('Module:Citation/CS1/Whitelist');
utilities = require ('Module:Citation/CS1/Utilities');
validation = require ('Module:Citation/CS1/Date_validation');
identifiers = require ('Module:Citation/CS1/Identifiers');
metadata = require ('Module:Citation/CS1/COinS');
styles = 'Module:Citation/CS1/styles.css';
 
end


utilities.set_selected_modules (cfg); -- so that functions in Utilities can see the cfg tables
utilities.set_selected_modules (cfg); -- so that functions in Utilities can see the selected cfg tables
identifiers.set_selected_modules (cfg, utilities); -- so that functions in Identifiers can see the selected cfg tables and selected Utilities module
identifiers.set_selected_modules (cfg, utilities); -- so that functions in Identifiers can see the selected cfg tables and selected Utilities module
validation.set_selected_modules (cfg, utilities); -- so that functions in Date validataion can see selected cfg tables and the selected Utilities module
validation.set_selected_modules (cfg, utilities); -- so that functions in Date validataion can see selected cfg tables and the selected Utilities module
metadata.set_selected_modules (cfg, utilities); -- so that functions in COinS can see the selected cfg tables and selected Utilities module
metadata.set_selected_modules (cfg, utilities); -- so that functions in COinS can see the selected cfg tables and selected Utilities module
dates = validation.dates; -- imported functions from Module:Citation/CS1/Date validation
year_date_check = validation.year_date_check;
reformat_dates = validation.reformat_dates;
date_hyphen_to_dash = validation.date_hyphen_to_dash;
date_name_xlate = validation.date_name_xlate;
is_set = utilities.is_set; -- imported functions from Module:Citation/CS1/Utilities
in_array = utilities.in_array;
substitute = utilities.substitute;
error_comment = utilities.error_comment;
set_error = utilities.set_error;
select_one = utilities.select_one;
add_maint_cat = utilities.add_maint_cat;
wrap_style = utilities.wrap_style;
safe_for_italics = utilities.safe_for_italics;
is_wikilink = utilities.is_wikilink;
make_wikilink = utilities.make_wikilink;
strip_apostrophe_markup = utilities.strip_apostrophe_markup;


z = utilities.z; -- table of error and category tables in Module:Citation/CS1/Utilities
z = utilities.z; -- table of error and category tables in Module:Citation/CS1/Utilities


extract_ids = identifiers.extract_ids; -- imported functions from Module:Citation/CS1/Identifiers
is_preview_mode = not utilities.is_set (frame:preprocess ('{{REVISIONID}}'));
build_id_list = identifiers.build_id_list;
is_embargoed = identifiers.is_embargoed;
extract_id_access_levels = identifiers.extract_id_access_levels;
make_coins_title = metadata.make_coins_title; -- imported functions from Module:Citation/CS1/COinS
get_coins_pages = metadata.get_coins_pages;
COinS = metadata.COinS;


local args = {}; -- table where we store all of the template's arguments
local args = {}; -- table where we store all of the template's arguments
local suggestions = {}; -- table where we store suggestions if we need to loadData them
local suggestions = {}; -- table where we store suggestions if we need to loadData them
local error_text, error_state;
local error_text; -- used as a flag
 
local config = {}; -- table to store parameters from the module {{#invoke:}}
for k, v in pairs( frame.args ) do
config[k] = v;
-- args[k] = v; -- debug tool that allows us to render a citation from module {{#invoke:}}
end


local capture; -- the single supported capture when matching unknown parameters using patterns
local capture; -- the single supported capture when matching unknown parameters using patterns
for k, v in pairs( pframe.args ) do
local empty_unknowns = {}; -- sequence table to hold empty unknown params for error message listing
for k, v in pairs( pframe.args ) do -- get parameters from the parent (template) frame
v = mw.ustring.gsub (v, '^%s*(.-)%s*$', '%1'); -- trim leading/trailing whitespace; when v is only whitespace, becomes empty string
if v ~= '' then
if v ~= '' then
if ('string' == type (k)) then
if ('string' == type (k)) then
Line 3,827: Line 4,401:
end
end
if not validate( k, config.CitationClass ) then
if not validate( k, config.CitationClass ) then
error_text = "";
if type (k) ~= 'string' then -- exclude empty numbered parameters
if type( k ) ~= 'string' then
-- Exclude empty numbered parameters
if v:match("%S+") ~= nil then
if v:match("%S+") ~= nil then
error_text, error_state = set_error( 'text_ignored', {v}, true );
error_text = utilities.set_message ('err_text_ignored', {v});
end
end
elseif validate( k:lower(), config.CitationClass ) then  
elseif validate (k:lower(), config.CitationClass) then  
error_text, error_state = set_error( 'parameter_ignored_suggest', {k, k:lower()}, true ); -- suggest the lowercase version of the parameter
error_text = utilities.set_message ('err_parameter_ignored_suggest', {k, k:lower()}); -- suggest the lowercase version of the parameter
else
else
if nil == suggestions.suggestions then -- if this table is nil then we need to load it
if nil == suggestions.suggestions then -- if this table is nil then we need to load it
if nil ~= string.find (frame:getTitle(), 'sandbox', 1, true) then -- did the {{#invoke:}} use sandbox version?
suggestions = mw.loadData ('Module:Citation/CS1/Suggestions' .. sandbox); --load sandbox version of suggestion module when {{#invoke:Citation/CS1/sandbox|...}}; live module else
suggestions = mw.loadData( 'Module:Citation/CS1/Suggestions/sandbox' ); -- use the sandbox version
else
suggestions = mw.loadData( 'Module:Citation/CS1/Suggestions' ); -- use the live version
end
end
end
for pattern, param in pairs (suggestions.patterns) do -- loop through the patterns to see if we can suggest a proper parameter
for pattern, param in pairs (suggestions.patterns) do -- loop through the patterns to see if we can suggest a proper parameter
capture = k:match (pattern); -- the whole match if no caputre in pattern else the capture if a match
capture = k:match (pattern); -- the whole match if no capture in pattern else the capture if a match
if capture then -- if the pattern matches  
if capture then -- if the pattern matches  
param = substitute (param, capture); -- add the capture to the suggested parameter (typically the enumerator)
param = utilities.substitute (param, capture); -- add the capture to the suggested parameter (typically the enumerator)
if validate (param, config.CitationClass) then -- validate the suggestion to make sure that the suggestion is supported by this template (necessary for limited parameter lists)
if validate (param, config.CitationClass) then -- validate the suggestion to make sure that the suggestion is supported by this template (necessary for limited parameter lists)
error_text, error_state = set_error ('parameter_ignored_suggest', {k, param}, true); -- set the suggestion error message
error_text = utilities.set_message ('err_parameter_ignored_suggest', {k, param}); -- set the suggestion error message
else
else
error_text, error_state = set_error( 'parameter_ignored', {param}, true ); -- suggested param not supported by this template
error_text = utilities.set_message ('err_parameter_ignored', {k}); -- suggested param not supported by this template
v = ''; -- unset
end
end
end
end
end
end
if not is_set (error_text) then -- couldn't match with a pattern, is there an expicit suggestion?
if not utilities.is_set (error_text) then -- couldn't match with a pattern, is there an explicit suggestion?
if suggestions.suggestions[ k:lower() ] ~= nil then
if (suggestions.suggestions[ k:lower() ] ~= nil) and validate (suggestions.suggestions[ k:lower() ], config.CitationClass) then
error_text, error_state = set_error( 'parameter_ignored_suggest', {k, suggestions.suggestions[ k:lower() ]}, true );
utilities.set_message ('err_parameter_ignored_suggest', {k, suggestions.suggestions[ k:lower() ]});
else
else
error_text, error_state = set_error( 'parameter_ignored', {k}, true );
utilities.set_message ('err_parameter_ignored', {k});
v = ''; -- unset value assigned to unrecognized parameters (this for the limited parameter lists)
v = ''; -- unset value assigned to unrecognized parameters (this for the limited parameter lists)
end
end
end
end
end    
end    
if error_text ~= '' then
table.insert( z.message_tail, {error_text, error_state} );
end
end
end
missing_pipe_check (k, v); -- do we think that there is a parameter that is missing a pipe?
 
-- TODO: is this the best place for this translation?
args[k] = v; -- save this parameter and its value
args[k] = v;
 
elseif args[k] ~= nil or (k == 'postscript') then -- here when v is empty string
elseif not utilities.is_set (v) then -- for empty parameters
args[k] = v; -- why do we do this?  we don't support 'empty' parameters
if not validate (k, config.CitationClass, true) then -- is this empty parameter a valid parameter
end
k = ('' == k) and '(empty string)' or k; -- when k is empty string (or was space(s) trimmed to empty string), replace with descriptive text
table.insert (empty_unknowns, utilities.wrap_style ('parameter', k)); -- format for error message and add to the list
end
-- crude debug support that allows us to render a citation from module {{#invoke:}} TODO: keep?
-- elseif args[k] ~= nil or (k == 'postscript') then -- when args[k] has a value from {{#invoke}} frame (we don't normally do that)
-- args[k] = v; -- overwrite args[k] with empty string from pframe.args[k] (template frame); v is empty string here
end -- not sure about the postscript bit; that gets handled in parameter validation; historical artifact?
end
end
if 0 ~= #empty_unknowns then -- create empty unknown error message
utilities.set_message ('err_param_unknown_empty', {
1 == #empty_unknowns and '' or 's',
utilities.make_sep_list (#empty_unknowns, empty_unknowns)
});
end
local url_param_t = {};


for k, v in pairs( args ) do
for k, v in pairs( args ) do
if 'string' == type (k) then -- don't evaluate positional parameters
if 'string' == type (k) then -- don't evaluate positional parameters
has_invisible_chars (k, v);
has_invisible_chars (k, v); -- look for invisible characters
has_extraneous_punc (k, v); -- look for extraneous terminal punctuation in parameter values
end
has_extraneous_punc (k, v); -- look for extraneous terminal punctuation in parameter values
missing_pipe_check (k, v); -- do we think that there is a parameter that is missing a pipe?
args[k] = inter_wiki_check (k, v); -- when language interwiki-linked parameter missing leading colon replace with wiki-link label
 
if 'string' == type (k) and not cfg.url_skip[k] then -- when parameter k is not positional and not in url skip table
url_param_t[k] = v; -- make a parameter/value list for extraneous url check
end
end
end
end
return table.concat ({citation0( config, args), frame:extensionTag ('templatestyles', '', {src=styles})});
 
has_extraneous_url (url_param_t); -- look for url in parameter values where a url does not belong
 
return table.concat ({
frame:extensionTag ('templatestyles', '', {src=styles}),
citation0( config, args)
});
end
end


--[[--------------------------< E X P O R T E D  F U N C T I O N S >------------------------------------------
--[[--------------------------< E X P O R T E D  F U N C T I O N S >------------------------------------------

Latest revision as of 20:03, 28 November 2023

Documentation for this module may be created at Module:Citation/CS1/doc

require ('strict');

--[[--------------------------< F O R W A R D   D E C L A R A T I O N S >--------------------------------------

each of these counts against the Lua upvalue limit

]]

local validation;																-- functions in Module:Citation/CS1/Date_validation

local utilities;																-- functions in Module:Citation/CS1/Utilities
local z = {};																	-- table of tables in Module:Citation/CS1/Utilities

local identifiers;																-- functions and tables in Module:Citation/CS1/Identifiers
local metadata;																	-- functions in Module:Citation/CS1/COinS
local cfg = {};																	-- table of configuration tables that are defined in Module:Citation/CS1/Configuration
local whitelist = {};															-- table of tables listing valid template parameter names; defined in Module:Citation/CS1/Whitelist


--[[------------------< P A G E   S C O P E   V A R I A B L E S >---------------

declare variables here that have page-wide scope that are not brought in from
other modules; that are created here and used here

]]

local added_deprecated_cat;														-- Boolean flag so that the category is added only once
local added_vanc_errs;															-- Boolean flag so we only emit one Vancouver error / category
local added_generic_name_errs;													-- Boolean flag so we only emit one generic name error / category and stop testing names once an error is encountered
local added_numeric_name_errs;													-- Boolean flag so we only emit one numeric name error / category and stop testing names once an error is encountered
local added_numeric_name_maint;													-- Boolean flag so we only emit one numeric name maint category and stop testing names once a category has been emitted
local Frame;																	-- holds the module's frame table
local is_preview_mode;															-- true when article is in preview mode; false when using 'Preview page with this template' (previewing the module)
local is_sandbox;																-- true when using sandbox modules to render citation


--[[--------------------------< F I R S T _ S E T >------------------------------------------------------------

Locates and returns the first set value in a table of values where the order established in the table,
left-to-right (or top-to-bottom), is the order in which the values are evaluated.  Returns nil if none are set.

This version replaces the original 'for _, val in pairs do' and a similar version that used ipairs.  With the pairs
version the order of evaluation could not be guaranteed.  With the ipairs version, a nil value would terminate
the for-loop before it reached the actual end of the list.

]]

local function first_set (list, count)
	local i = 1;
	while i <= count do															-- loop through all items in list
		if utilities.is_set( list[i] ) then
			return list[i];														-- return the first set list member
		end
		i = i + 1;																-- point to next
	end
end


--[[--------------------------< A D D _ V A N C _ E R R O R >----------------------------------------------------

Adds a single Vancouver system error message to the template's output regardless of how many error actually exist.
To prevent duplication, added_vanc_errs is nil until an error message is emitted.

added_vanc_errs is a Boolean declared in page scope variables above

]]

local function add_vanc_error (source, position)
	if added_vanc_errs then return end
		
	added_vanc_errs = true;														-- note that we've added this category
	utilities.set_message ('err_vancouver', {source, position});
end


--[[--------------------------< I S _ S C H E M E >------------------------------------------------------------

does this thing that purports to be a URI scheme seem to be a valid scheme?  The scheme is checked to see if it
is in agreement with http://tools.ietf.org/html/std66#section-3.1 which says:
	Scheme names consist of a sequence of characters beginning with a
   letter and followed by any combination of letters, digits, plus
   ("+"), period ("."), or hyphen ("-").

returns true if it does, else false

]]

local function is_scheme (scheme)
	return scheme and scheme:match ('^%a[%a%d%+%.%-]*:');						-- true if scheme is set and matches the pattern
end


--[=[-------------------------< I S _ D O M A I N _ N A M E >--------------------------------------------------

Does this thing that purports to be a domain name seem to be a valid domain name?

Syntax defined here: http://tools.ietf.org/html/rfc1034#section-3.5
BNF defined here: https://tools.ietf.org/html/rfc4234
Single character names are generally reserved; see https://tools.ietf.org/html/draft-ietf-dnsind-iana-dns-01#page-15;
	see also [[Single-letter second-level domain]]
list of TLDs: https://www.iana.org/domains/root/db

RFC 952 (modified by RFC 1123) requires the first and last character of a hostname to be a letter or a digit.  Between
the first and last characters the name may use letters, digits, and the hyphen.

Also allowed are IPv4 addresses. IPv6 not supported

domain is expected to be stripped of any path so that the last character in the last character of the TLD.  tld
is two or more alpha characters.  Any preceding '//' (from splitting a URL with a scheme) will be stripped
here.  Perhaps not necessary but retained in case it is necessary for IPv4 dot decimal.

There are several tests:
	the first character of the whole domain name including subdomains must be a letter or a digit
	internationalized domain name (ASCII characters with .xn-- ASCII Compatible Encoding (ACE) prefix xn-- in the TLD) see https://tools.ietf.org/html/rfc3490
	single-letter/digit second-level domains in the .org, .cash, and .today TLDs
	q, x, and z SL domains in the .com TLD
	i and q SL domains in the .net TLD
	single-letter SL domains in the ccTLDs (where the ccTLD is two letters)
	two-character SL domains in gTLDs (where the gTLD is two or more letters)
	three-plus-character SL domains in gTLDs (where the gTLD is two or more letters)
	IPv4 dot-decimal address format; TLD not allowed

returns true if domain appears to be a proper name and TLD or IPv4 address, else false

]=]

local function is_domain_name (domain)
	if not domain then
		return false;															-- if not set, abandon
	end
	
	domain = domain:gsub ('^//', '');											-- strip '//' from domain name if present; done here so we only have to do it once
	
	if not domain:match ('^[%w]') then											-- first character must be letter or digit
		return false;
	end

	if domain:match ('^%a+:') then												-- hack to detect things that look like s:Page:Title where Page: is namespace at Wikisource
		return false;
	end

	local patterns = {															-- patterns that look like URLs
		'%f[%w][%w][%w%-]+[%w]%.%a%a+$',										-- three or more character hostname.hostname or hostname.tld
		'%f[%w][%w][%w%-]+[%w]%.xn%-%-[%w]+$',									-- internationalized domain name with ACE prefix
		'%f[%a][qxz]%.com$',													-- assigned one character .com hostname (x.com times out 2015-12-10)
		'%f[%a][iq]%.net$',														-- assigned one character .net hostname (q.net registered but not active 2015-12-10)
		'%f[%w][%w]%.%a%a$',													-- one character hostname and ccTLD (2 chars)
		'%f[%w][%w][%w]%.%a%a+$',												-- two character hostname and TLD
		'^%d%d?%d?%.%d%d?%d?%.%d%d?%d?%.%d%d?%d?',								-- IPv4 address
		}

	for _, pattern in ipairs (patterns) do										-- loop through the patterns list
		if domain:match (pattern) then
			return true;														-- if a match then we think that this thing that purports to be a URL is a URL
		end
	end

	for _, d in ipairs (cfg.single_letter_2nd_lvl_domains_t) do					-- look for single letter second level domain names for these top level domains
		if domain:match ('%f[%w][%w]%.' .. d) then
			return true
		end
	end
	return false;																-- no matches, we don't know what this thing is
end


--[[--------------------------< I S _ U R L >------------------------------------------------------------------

returns true if the scheme and domain parts of a URL appear to be a valid URL; else false.

This function is the last step in the validation process.  This function is separate because there are cases that
are not covered by split_url(), for example is_parameter_ext_wikilink() which is looking for bracketted external
wikilinks.

]]

local function is_url (scheme, domain)
	if utilities.is_set (scheme) then											-- if scheme is set check it and domain
		return is_scheme (scheme) and is_domain_name (domain);
	else
		return is_domain_name (domain);											-- scheme not set when URL is protocol-relative
	end
end


--[[--------------------------< S P L I T _ U R L >------------------------------------------------------------

Split a URL into a scheme, authority indicator, and domain.

First remove Fully Qualified Domain Name terminator (a dot following TLD) (if any) and any path(/), query(?) or fragment(#).

If protocol-relative URL, return nil scheme and domain else return nil for both scheme and domain.

When not protocol-relative, get scheme, authority indicator, and domain.  If there is an authority indicator (one
or more '/' characters immediately following the scheme's colon), make sure that there are only 2.

Any URL that does not have news: scheme must have authority indicator (//).  TODO: are there other common schemes
like news: that don't use authority indicator?

Strip off any port and path;

]]

local function split_url (url_str)
	local scheme, authority, domain;
	
	url_str = url_str:gsub ('([%a%d])%.?[/%?#].*$', '%1');						-- strip FQDN terminator and path(/), query(?), fragment (#) (the capture prevents false replacement of '//')

	if url_str:match ('^//%S*') then											-- if there is what appears to be a protocol-relative URL
		domain = url_str:match ('^//(%S*)')
	elseif url_str:match ('%S-:/*%S+') then										-- if there is what appears to be a scheme, optional authority indicator, and domain name
		scheme, authority, domain = url_str:match ('(%S-:)(/*)(%S+)');			-- extract the scheme, authority indicator, and domain portions
		if utilities.is_set (authority) then
			authority = authority:gsub ('//', '', 1);							-- replace place 1 pair of '/' with nothing;
			if utilities.is_set(authority) then									-- if anything left (1 or 3+ '/' where authority should be) then
				return scheme;													-- return scheme only making domain nil which will cause an error message
			end
		else
			if not scheme:match ('^news:') then									-- except for news:..., MediaWiki won't link URLs that do not have authority indicator; TODO: a better way to do this test?
				return scheme;													-- return scheme only making domain nil which will cause an error message
			end
		end
		domain = domain:gsub ('(%a):%d+', '%1');								-- strip port number if present
	end
	
	return scheme, domain;
end


--[[--------------------------< L I N K _ P A R A M _ O K >---------------------------------------------------

checks the content of |title-link=, |series-link=, |author-link=, etc. for properly formatted content: no wikilinks, no URLs

Link parameters are to hold the title of a Wikipedia article, so none of the WP:TITLESPECIALCHARACTERS are allowed:
	# < > [ ] | { } _
except the underscore which is used as a space in wiki URLs and # which is used for section links

returns false when the value contains any of these characters.

When there are no illegal characters, this function returns TRUE if value DOES NOT appear to be a valid URL (the
|<param>-link= parameter is ok); else false when value appears to be a valid URL (the |<param>-link= parameter is NOT ok).

]]

local function link_param_ok (value)
	local scheme, domain;
	if value:find ('[<>%[%]|{}]') then											-- if any prohibited characters
		return false;
	end

	scheme, domain = split_url (value);											-- get scheme or nil and domain or nil from URL; 
	return not is_url (scheme, domain);											-- return true if value DOES NOT appear to be a valid URL
end


--[[--------------------------< L I N K _ T I T L E _ O K >---------------------------------------------------

Use link_param_ok() to validate |<param>-link= value and its matching |<title>= value.

|<title>= may be wiki-linked but not when |<param>-link= has a value.  This function emits an error message when
that condition exists

check <link> for inter-language interwiki-link prefix.  prefix must be a MediaWiki-recognized language
code and must begin with a colon.

]]

local function link_title_ok (link, lorig, title, torig)
	local orig;
	if utilities.is_set (link) then 											-- don't bother if <param>-link doesn't have a value
		if not link_param_ok (link) then										-- check |<param>-link= markup
			orig = lorig;														-- identify the failing link parameter
		elseif title:find ('%[%[') then											-- check |title= for wikilink markup
			orig = torig;														-- identify the failing |title= parameter
		elseif link:match ('^%a+:') then										-- if the link is what looks like an interwiki
			local prefix = link:match ('^(%a+):'):lower();						-- get the interwiki prefix

			if cfg.inter_wiki_map[prefix] then									-- if prefix is in the map, must have preceding colon
				orig = lorig;													-- flag as error
			end
		end
	end

	if utilities.is_set (orig) then
		link = '';																-- unset
		utilities.set_message ('err_bad_paramlink', orig);						-- URL or wikilink in |title= with |title-link=;
	end
	
	return link;																-- link if ok, empty string else
end


--[[--------------------------< C H E C K _ U R L >------------------------------------------------------------

Determines whether a URL string appears to be valid.

First we test for space characters.  If any are found, return false.  Then split the URL into scheme and domain
portions, or for protocol-relative (//example.com) URLs, just the domain.  Use is_url() to validate the two
portions of the URL.  If both are valid, or for protocol-relative if domain is valid, return true, else false.

Because it is different from a standard URL, and because this module used external_link() to make external links
that work for standard and news: links, we validate newsgroup names here.  The specification for a newsgroup name
is at https://tools.ietf.org/html/rfc5536#section-3.1.4

]]

local function check_url( url_str )
	if nil == url_str:match ("^%S+$") then										-- if there are any spaces in |url=value it can't be a proper URL
		return false;
	end
	local scheme, domain;

	scheme, domain = split_url (url_str);										-- get scheme or nil and domain or nil from URL;
	
	if 'news:' == scheme then													-- special case for newsgroups
		return domain:match('^[%a%d%+%-_]+%.[%a%d%+%-_%.]*[%a%d%+%-_]$');
	end
	
	return is_url (scheme, domain);												-- return true if value appears to be a valid URL
end


--[=[-------------------------< I S _ P A R A M E T E R _ E X T _ W I K I L I N K >----------------------------

Return true if a parameter value has a string that begins and ends with square brackets [ and ] and the first
non-space characters following the opening bracket appear to be a URL.  The test will also find external wikilinks
that use protocol-relative URLs. Also finds bare URLs.

The frontier pattern prevents a match on interwiki-links which are similar to scheme:path URLs.  The tests that
find bracketed URLs are required because the parameters that call this test (currently |title=, |chapter=, |work=,
and |publisher=) may have wikilinks and there are articles or redirects like '//Hus' so, while uncommon, |title=[[//Hus]]
is possible as might be [[en://Hus]].

]=]

local function is_parameter_ext_wikilink (value)
local scheme, domain;

	if value:match ('%f[%[]%[%a%S*:%S+.*%]') then								-- if ext. wikilink with scheme and domain: [xxxx://yyyyy.zzz]
		scheme, domain = split_url (value:match ('%f[%[]%[(%a%S*:%S+).*%]'));
	elseif value:match ('%f[%[]%[//%S+.*%]') then								-- if protocol-relative ext. wikilink: [//yyyyy.zzz]
		scheme, domain = split_url (value:match ('%f[%[]%[(//%S+).*%]'));
	elseif value:match ('%a%S*:%S+') then										-- if bare URL with scheme; may have leading or trailing plain text
		scheme, domain = split_url (value:match ('(%a%S*:%S+)'));
	elseif value:match ('//%S+') then											-- if protocol-relative bare URL: //yyyyy.zzz; may have leading or trailing plain text
		scheme, domain = split_url (value:match ('(//%S+)'));					-- what is left should be the domain
	else
		return false;															-- didn't find anything that is obviously a URL
	end

	return is_url (scheme, domain);												-- return true if value appears to be a valid URL
end


--[[-------------------------< C H E C K _ F O R _ U R L >-----------------------------------------------------

loop through a list of parameters and their values.  Look at the value and if it has an external link, emit an error message.

]]

local function check_for_url (parameter_list, error_list)
	for k, v in pairs (parameter_list) do										-- for each parameter in the list
		if is_parameter_ext_wikilink (v) then									-- look at the value; if there is a URL add an error message
			table.insert (error_list, utilities.wrap_style ('parameter', k));
		end
	end
end


--[[--------------------------< S A F E _ F O R _ U R L >------------------------------------------------------

Escape sequences for content that will be used for URL descriptions

]]

local function safe_for_url( str )
	if str:match( "%[%[.-%]%]" ) ~= nil then 
		utilities.set_message ('err_wikilink_in_url', {});
	end
	
	return str:gsub( '[%[%]\n]', {	
		['['] = '&#91;',
		[']'] = '&#93;',
		['\n'] = ' ' } );
end


--[[--------------------------< E X T E R N A L _ L I N K >----------------------------------------------------

Format an external link with error checking

]]

local function external_link (URL, label, source, access)
	local err_msg = '';
	local domain;
	local path;
	local base_url;

	if not utilities.is_set (label) then
		label = URL;
		if utilities.is_set (source) then
			utilities.set_message ('err_bare_url_missing_title', {utilities.wrap_style ('parameter', source)});
		else
			error (cfg.messages["bare_url_no_origin"]);							-- programmer error; valid parameter name does not have matching meta-parameter
		end			
	end
	if not check_url (URL) then
		utilities.set_message ('err_bad_url', {utilities.wrap_style ('parameter', source)});
	end
	
	domain, path = URL:match ('^([/%.%-%+:%a%d]+)([/%?#].*)$');					-- split the URL into scheme plus domain and path
	if path then																-- if there is a path portion
		path = path:gsub ('[%[%]]', {['['] = '%5b', [']'] = '%5d'});			-- replace '[' and ']' with their percent-encoded values
		URL = table.concat ({domain, path});									-- and reassemble
	end

	base_url = table.concat ({ "[", URL, " ", safe_for_url (label), "]" });		-- assemble a wiki-markup URL

	if utilities.is_set (access) then											-- access level (subscription, registration, limited)
		base_url = utilities.substitute (cfg.presentation['ext-link-access-signal'], {cfg.presentation[access].class, cfg.presentation[access].title, base_url});	-- add the appropriate icon
	end

	return base_url;
end


--[[--------------------------< D E P R E C A T E D _ P A R A M E T E R >--------------------------------------

Categorize and emit an error message when the citation contains one or more deprecated parameters.  The function includes the
offending parameter name to the error message.  Only one error message is emitted regardless of the number of deprecated
parameters in the citation.

added_deprecated_cat is a Boolean declared in page scope variables above

]]

local function deprecated_parameter(name)
	if not added_deprecated_cat then
		added_deprecated_cat = true;											-- note that we've added this category
		utilities.set_message ('err_deprecated_params', {name});				-- add error message
	end
end


--[=[-------------------------< K E R N _ Q U O T E S >--------------------------------------------------------

Apply kerning to open the space between the quote mark provided by the module and a leading or trailing quote
mark contained in a |title= or |chapter= parameter's value.

This function will positive kern either single or double quotes:
	"'Unkerned title with leading and trailing single quote marks'"
	" 'Kerned title with leading and trailing single quote marks' " (in real life the kerning isn't as wide as this example)
Double single quotes (italic or bold wiki-markup) are not kerned.

Replaces Unicode quote marks in plain text or in the label portion of a [[L|D]] style wikilink with typewriter
quote marks regardless of the need for kerning.  Unicode quote marks are not replaced in simple [[D]] wikilinks.

Call this function for chapter titles, for website titles, etc.; not for book titles.

]=]

local function kern_quotes (str)
	local cap = '';
	local wl_type, label, link;

	wl_type, label, link = utilities.is_wikilink (str);							-- wl_type is: 0, no wl (text in label variable); 1, [[D]]; 2, [[L|D]]
	
	if 1 == wl_type then														-- [[D]] simple wikilink with or without quote marks
		if mw.ustring.match (str, '%[%[[\"“”\'‘’].+[\"“”\'‘’]%]%]') then		-- leading and trailing quote marks
			str = utilities.substitute (cfg.presentation['kern-left'], str);
			str = utilities.substitute (cfg.presentation['kern-right'], str);
		elseif mw.ustring.match (str, '%[%[[\"“”\'‘’].+%]%]')	then			-- leading quote marks
			str = utilities.substitute (cfg.presentation['kern-left'], str);
		elseif mw.ustring.match (str, '%[%[.+[\"“”\'‘’]%]%]') then				-- trailing quote marks
			str = utilities.substitute (cfg.presentation['kern-right'], str);
		end

	else																		-- plain text or [[L|D]]; text in label variable
		label = mw.ustring.gsub (label, '[“”]', '\"');							-- replace “” (U+201C & U+201D) with " (typewriter double quote mark)
		label = mw.ustring.gsub (label, '[‘’]', '\'');							-- replace ‘’ (U+2018 & U+2019) with ' (typewriter single quote mark)

		cap = mw.ustring.match (label, "^([\"\'][^\'].+)");						-- match leading double or single quote but not doubled single quotes (italic markup)
		if utilities.is_set (cap) then
			label = utilities.substitute (cfg.presentation['kern-left'], cap);
		end
	
		cap = mw.ustring.match (label, "^(.+[^\'][\"\'])$")						-- match trailing double or single quote but not doubled single quotes (italic markup)
		if utilities.is_set (cap) then
			label = utilities.substitute (cfg.presentation['kern-right'], cap);
		end
		
		if 2 == wl_type then
			str = utilities.make_wikilink (link, label);						-- reassemble the wikilink
		else
			str = label;
		end
	end
	return str;
end


--[[--------------------------< F O R M A T _ S C R I P T _ V A L U E >----------------------------------------

|script-title= holds title parameters that are not written in Latin-based scripts: Chinese, Japanese, Arabic, Hebrew, etc. These scripts should
not be italicized and may be written right-to-left.  The value supplied by |script-title= is concatenated onto Title after Title has been wrapped
in italic markup.

Regardless of language, all values provided by |script-title= are wrapped in <bdi>...</bdi> tags to isolate RTL languages from the English left to right.

|script-title= provides a unique feature.  The value in |script-title= may be prefixed with a two-character ISO 639-1 language code and a colon:
	|script-title=ja:*** *** (where * represents a Japanese character)
Spaces between the two-character code and the colon and the colon and the first script character are allowed:
	|script-title=ja : *** ***
	|script-title=ja: *** ***
	|script-title=ja :*** ***
Spaces preceding the prefix are allowed: |script-title = ja:*** ***

The prefix is checked for validity.  If it is a valid ISO 639-1 language code, the lang attribute (lang="ja") is added to the <bdi> tag so that browsers can
know the language the tag contains.  This may help the browser render the script more correctly.  If the prefix is invalid, the lang attribute
is not added.  At this time there is no error message for this condition.

Supports |script-title=, |script-chapter=, |script-<periodical>=

]]

local function format_script_value (script_value, script_param)
	local lang='';																-- initialize to empty string
	local name;
	if script_value:match('^%l%l%l?%s*:') then									-- if first 3 or 4 non-space characters are script language prefix
		lang = script_value:match('^(%l%l%l?)%s*:%s*%S.*');						-- get the language prefix or nil if there is no script
		if not utilities.is_set (lang) then
			utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['missing title part']});		-- prefix without 'title'; add error message
			return '';															-- script_value was just the prefix so return empty string
		end
																				-- if we get this far we have prefix and script
		name = cfg.lang_tag_remap[lang] or mw.language.fetchLanguageName( lang, cfg.this_wiki_code );	-- get language name so that we can use it to categorize
		if utilities.is_set (name) then											-- is prefix a proper ISO 639-1 language code?
			script_value = script_value:gsub ('^%l+%s*:%s*', '');				-- strip prefix from script
																				-- is prefix one of these language codes?
			if utilities.in_array (lang, cfg.script_lang_codes) then
				utilities.add_prop_cat ('script', {name, lang})
			else
				utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['unknown language code']});	-- unknown script-language; add error message
			end
			lang = ' lang="' .. lang .. '" ';									-- convert prefix into a lang attribute
		else
			utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['invalid language code']});		-- invalid language code; add error message
			lang = '';															-- invalid so set lang to empty string
		end
	else
		utilities.set_message ('err_script_parameter', {script_param, cfg.err_msg_supl['missing prefix']});				-- no language code prefix; add error message
	end
	script_value = utilities.substitute (cfg.presentation['bdi'], {lang, script_value});	-- isolate in case script is RTL

	return script_value;
end


--[[--------------------------< S C R I P T _ C O N C A T E N A T E >------------------------------------------

Initially for |title= and |script-title=, this function concatenates those two parameter values after the script
value has been wrapped in <bdi> tags.

]]

local function script_concatenate (title, script, script_param)
	if utilities.is_set (script) then
		script = format_script_value (script, script_param);					-- <bdi> tags, lang attribute, categorization, etc.; returns empty string on error
		if utilities.is_set (script) then
			title = title .. ' ' .. script;										-- concatenate title and script title
		end
	end
	return title;
end


--[[--------------------------< W R A P _ M S G >--------------------------------------------------------------

Applies additional message text to various parameter values. Supplied string is wrapped using a message_list
configuration taking one argument.  Supports lower case text for {{citation}} templates.  Additional text taken
from citation_config.messages - the reason this function is similar to but separate from wrap_style().

]]

local function wrap_msg (key, str, lower)
	if not utilities.is_set ( str ) then
		return "";
	end
	if true == lower then
		local msg;
		msg = cfg.messages[key]:lower();										-- set the message to lower case before 
		return utilities.substitute ( msg, str );								-- including template text
	else
		return utilities.substitute ( cfg.messages[key], str );
	end		
end


--[[----------------< W I K I S O U R C E _ U R L _ M A K E >-------------------

Makes a Wikisource URL from Wikisource interwiki-link.  Returns the URL and appropriate
label; nil else.

str is the value assigned to |chapter= (or aliases) or |title= or |title-link=

]]

local function wikisource_url_make (str)
	local wl_type, D, L;
	local ws_url, ws_label;
	local wikisource_prefix = table.concat ({'https://', cfg.this_wiki_code, '.wikisource.org/wiki/'});

	wl_type, D, L = utilities.is_wikilink (str);								-- wl_type is 0 (not a wikilink), 1 (simple wikilink), 2 (complex wikilink)

	if 0 == wl_type then														-- not a wikilink; might be from |title-link=
		str = D:match ('^[Ww]ikisource:(.+)') or D:match ('^[Ss]:(.+)');		-- article title from interwiki link with long-form or short-form namespace
		if utilities.is_set (str) then
			ws_url = table.concat ({											-- build a Wikisource URL
				wikisource_prefix,												-- prefix
				str,															-- article title
				});
			ws_label = str;														-- label for the URL
		end
	elseif 1 == wl_type then													-- simple wikilink: [[Wikisource:ws article]]
		str = D:match ('^[Ww]ikisource:(.+)') or D:match ('^[Ss]:(.+)');		-- article title from interwiki link with long-form or short-form namespace
		if utilities.is_set (str) then
			ws_url = table.concat ({											-- build a Wikisource URL
				wikisource_prefix,												-- prefix
				str,															-- article title
				});
			ws_label = str;														-- label for the URL
		end
	elseif 2 == wl_type then													-- non-so-simple wikilink: [[Wikisource:ws article|displayed text]] ([[L|D]])
		str = L:match ('^[Ww]ikisource:(.+)') or L:match ('^[Ss]:(.+)');		-- article title from interwiki link with long-form or short-form namespace
		if utilities.is_set (str) then
			ws_label = D;														-- get ws article name from display portion of interwiki link
			ws_url = table.concat ({											-- build a Wikisource URL
				wikisource_prefix,												-- prefix
				str,															-- article title without namespace from link portion of wikilink
				});
		end
	end

	if ws_url then
		ws_url = mw.uri.encode (ws_url, 'WIKI');								-- make a usable URL
		ws_url = ws_url:gsub ('%%23', '#');										-- undo percent-encoding of fragment marker
	end

	return ws_url, ws_label, L or D;											-- return proper URL or nil and a label or nil
end


--[[----------------< F O R M A T _ P E R I O D I C A L >-----------------------

Format the three periodical parameters: |script-<periodical>=, |<periodical>=,
and |trans-<periodical>= into a single Periodical meta-parameter.

]]

local function format_periodical (script_periodical, script_periodical_source, periodical, trans_periodical)

	if not utilities.is_set (periodical) then
		periodical = '';														-- to be safe for concatenation
	else
		periodical = utilities.wrap_style ('italic-title', periodical);			-- style 
	end

	periodical = script_concatenate (periodical, script_periodical, script_periodical_source);	-- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped

	if utilities.is_set (trans_periodical) then
		trans_periodical = utilities.wrap_style ('trans-italic-title', trans_periodical);
		if utilities.is_set (periodical) then
			periodical = periodical .. ' ' .. trans_periodical;
		else																	-- here when trans-periodical without periodical or script-periodical
			periodical = trans_periodical;
			utilities.set_message ('err_trans_missing_title', {'periodical'});
		end
	end

	return periodical;
end


--[[------------------< F O R M A T _ C H A P T E R _ T I T L E >---------------

Format the four chapter parameters: |script-chapter=, |chapter=, |trans-chapter=,
and |chapter-url= into a single chapter meta- parameter (chapter_url_source used
for error messages).

]]

local function format_chapter_title (script_chapter, script_chapter_source, chapter, chapter_source, trans_chapter, trans_chapter_source, chapter_url, chapter_url_source, no_quotes, access)
	local ws_url, ws_label, L = wikisource_url_make (chapter);					-- make a wikisource URL and label from a wikisource interwiki link
	if ws_url then
		ws_label = ws_label:gsub ('_', ' ');									-- replace underscore separators with space characters
		chapter = ws_label;
	end

	if not utilities.is_set (chapter) then
		chapter = '';															-- to be safe for concatenation
	else
		if false == no_quotes then
			chapter = kern_quotes (chapter);									-- if necessary, separate chapter title's leading and trailing quote marks from module provided quote marks
			chapter = utilities.wrap_style ('quoted-title', chapter);
		end
	end

	chapter = script_concatenate (chapter, script_chapter, script_chapter_source);	-- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped

	if utilities.is_set (chapter_url) then
		chapter = external_link (chapter_url, chapter, chapter_url_source, access);	-- adds bare_url_missing_title error if appropriate
	elseif ws_url then
		chapter = external_link (ws_url, chapter .. '&nbsp;', 'ws link in chapter');	-- adds bare_url_missing_title error if appropriate; space char to move icon away from chap text; TODO: better way to do this?
		chapter = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, chapter});				
	end

	if utilities.is_set (trans_chapter) then
		trans_chapter = utilities.wrap_style ('trans-quoted-title', trans_chapter);
		if utilities.is_set (chapter) then
			chapter = chapter .. ' ' .. trans_chapter;
		else																	-- here when trans_chapter without chapter or script-chapter
			chapter = trans_chapter;
			chapter_source = trans_chapter_source:match ('trans%-?(.+)');		-- when no chapter, get matching name from trans-<param>
			utilities.set_message ('err_trans_missing_title', {chapter_source});
		end
	end

	return chapter;
end


--[[----------------< H A S _ I N V I S I B L E _ C H A R S >-------------------

This function searches a parameter's value for non-printable or invisible characters.
The search stops at the first match.

This function will detect the visible replacement character when it is part of the Wikisource.

Detects but ignores nowiki and math stripmarkers.  Also detects other named stripmarkers
(gallery, math, pre, ref) and identifies them with a slightly different error message.
See also coins_cleanup().

Output of this function is an error message that identifies the character or the
Unicode group, or the stripmarker that was detected along with its position (or,
for multi-byte characters, the position of its first byte) in the parameter value.

]]

local function has_invisible_chars (param, v)
	local position = '';														-- position of invisible char or starting position of stripmarker
	local capture;																-- used by stripmarker detection to hold name of the stripmarker
	local stripmarker;															-- boolean set true when a stripmarker is found

	capture = string.match (v, '[%w%p ]*');										-- test for values that are simple ASCII text and bypass other tests if true
	if capture == v then														-- if same there are no Unicode characters
		return;
	end

	for _, invisible_char in ipairs (cfg.invisible_chars) do
		local char_name = invisible_char[1];									-- the character or group name
		local pattern = invisible_char[2];										-- the pattern used to find it
		position, _, capture = mw.ustring.find (v, pattern);					-- see if the parameter value contains characters that match the pattern
		
		if position and (cfg.invisible_defs.zwj == capture) then				-- if we found a zero-width joiner character
			if mw.ustring.find (v, cfg.indic_script) then						-- it's ok if one of the Indic scripts
				position = nil;													-- unset position
			elseif cfg.emoji_t[mw.ustring.codepoint (v, position+1)] then			-- is zwj followed by a character listed in emoji{}?
				position = nil;													-- unset position
			end
		end
		
		if position then
			if 'nowiki' == capture or 'math' == capture or						-- nowiki and math stripmarkers (not an error condition)
				('templatestyles' == capture and utilities.in_array (param, {'id', 'quote'})) then	-- templatestyles stripmarker allowed in these parameters
					stripmarker = true;											-- set a flag
			elseif true == stripmarker and cfg.invisible_defs.del == capture then	-- because stripmakers begin and end with the delete char, assume that we've found one end of a stripmarker
				position = nil;													-- unset
			else
				local err_msg;
				if capture and not (cfg.invisible_defs.del == capture or cfg.invisible_defs.zwj == capture) then
					err_msg = capture .. ' ' .. char_name;
				else
					err_msg = char_name .. ' ' .. 'character';
				end

				utilities.set_message ('err_invisible_char', {err_msg, utilities.wrap_style ('parameter', param), position});	-- add error message
				return;															-- and done with this parameter
			end
		end
	end
end


--[[-------------------< A R G U M E N T _ W R A P P E R >----------------------

Argument wrapper.  This function provides support for argument mapping defined
in the configuration file so that multiple names can be transparently aliased to
single internal variable.

]]

local function argument_wrapper ( args )
	local origin = {};
	
	return setmetatable({
		ORIGIN = function ( self, k )
			local dummy = self[k];												-- force the variable to be loaded.
			return origin[k];
		end
	},
	{
		__index = function ( tbl, k )
			if origin[k] ~= nil then
				return nil;
			end
			
			local args, list, v = args, cfg.aliases[k];
			
			if type( list ) == 'table' then
				v, origin[k] = utilities.select_one ( args, list, 'err_redundant_parameters' );
				if origin[k] == nil then
					origin[k] = '';												-- Empty string, not nil
				end
			elseif list ~= nil then
				v, origin[k] = args[list], list;
			else
				-- maybe let through instead of raising an error?
				-- v, origin[k] = args[k], k;
				error( cfg.messages['unknown_argument_map'] .. ': ' .. k);
			end
			
			-- Empty strings, not nil;
			if v == nil then
				v = '';
				origin[k] = '';
			end
			
			tbl = rawset( tbl, k, v );
			return v;
		end,
	});
end


--[[--------------------------< N O W R A P _ D A T E >-------------------------

When date is YYYY-MM-DD format wrap in nowrap span: <span ...>YYYY-MM-DD</span>.
When date is DD MMMM YYYY or is MMMM DD, YYYY then wrap in nowrap span:
<span ...>DD MMMM</span> YYYY or <span ...>MMMM DD,</span> YYYY

DOES NOT yet support MMMM YYYY or any of the date ranges.

]]

local function nowrap_date (date)
	local cap = '';
	local cap2 = '';

	if date:match("^%d%d%d%d%-%d%d%-%d%d$") then
		date = utilities.substitute (cfg.presentation['nowrap1'], date);
	
	elseif date:match("^%a+%s*%d%d?,%s+%d%d%d%d$") or date:match ("^%d%d?%s*%a+%s+%d%d%d%d$") then
		cap, cap2 = string.match (date, "^(.*)%s+(%d%d%d%d)$");
		date = utilities.substitute (cfg.presentation['nowrap2'], {cap, cap2});
	end
	
	return date;
end


--[[--------------------------< S E T _ T I T L E T Y P E >---------------------

This function sets default title types (equivalent to the citation including
|type=<default value>) for those templates that have defaults. Also handles the
special case where it is desirable to omit the title type from the rendered citation
(|type=none).

]]

local function set_titletype (cite_class, title_type)
	if utilities.is_set (title_type) then
		if 'none' == cfg.keywords_xlate[title_type] then
			title_type = '';													-- if |type=none then type parameter not displayed
		end
		return title_type;														-- if |type= has been set to any other value use that value
	end

	return cfg.title_types [cite_class] or '';									-- set template's default title type; else empty string for concatenation
end


--[[--------------------------< S A F E _ J O I N >-----------------------------

Joins a sequence of strings together while checking for duplicate separation characters.

]]

local function safe_join( tbl, duplicate_char )
	local f = {};																-- create a function table appropriate to type of 'duplicate character'
		if 1 == #duplicate_char then											-- for single byte ASCII characters use the string library functions
			f.gsub = string.gsub
			f.match = string.match
			f.sub = string.sub
		else																	-- for multi-byte characters use the ustring library functions
			f.gsub = mw.ustring.gsub
			f.match = mw.ustring.match
			f.sub = mw.ustring.sub
		end

	local str = '';																-- the output string
	local comp = '';															-- what does 'comp' mean?
	local end_chr = '';
	local trim;
	for _, value in ipairs( tbl ) do
		if value == nil then value = ''; end
		
		if str == '' then														-- if output string is empty
			str = value;														-- assign value to it (first time through the loop)
		elseif value ~= '' then
			if value:sub(1, 1) == '<' then										-- special case of values enclosed in spans and other markup.
				comp = value:gsub( "%b<>", "" );								-- remove HTML markup (<span>string</span> -> string)
			else
				comp = value;
			end
																				-- typically duplicate_char is sepc
			if f.sub(comp, 1, 1) == duplicate_char then							-- is first character same as duplicate_char? why test first character?
																				--   Because individual string segments often (always?) begin with terminal punct for the
																				--   preceding segment: 'First element' .. 'sepc next element' .. etc.?
				trim = false;
				end_chr = f.sub(str, -1, -1);									-- get the last character of the output string
				-- str = str .. "<HERE(enchr=" .. end_chr .. ")"				-- debug stuff?
				if end_chr == duplicate_char then								-- if same as separator
					str = f.sub(str, 1, -2);									-- remove it
				elseif end_chr == "'" then										-- if it might be wiki-markup
					if f.sub(str, -3, -1) == duplicate_char .. "''" then		-- if last three chars of str are sepc'' 
						str = f.sub(str, 1, -4) .. "''";						-- remove them and add back ''
					elseif  f.sub(str, -5, -1) == duplicate_char .. "]]''" then	-- if last five chars of str are sepc]]'' 
						trim = true;											-- why? why do this and next differently from previous?
					elseif f.sub(str, -4, -1) == duplicate_char .. "]''" then	-- if last four chars of str are sepc]'' 
						trim = true;											-- same question
					end
				elseif end_chr == "]" then										-- if it might be wiki-markup
					if f.sub(str, -3, -1) == duplicate_char .. "]]" then		-- if last three chars of str are sepc]] wikilink 
						trim = true;
					elseif f.sub(str, -3, -1) == duplicate_char .. '"]' then	-- if last three chars of str are sepc"] quoted external link 
						trim = true;
					elseif  f.sub(str, -2, -1) == duplicate_char .. "]" then	-- if last two chars of str are sepc] external link
						trim = true;
					elseif f.sub(str, -4, -1) == duplicate_char .. "'']" then	-- normal case when |url=something & |title=Title.
						trim = true;
					end
				elseif end_chr == " " then										-- if last char of output string is a space
					if f.sub(str, -2, -1) == duplicate_char .. " " then			-- if last two chars of str are <sepc><space>
						str = f.sub(str, 1, -3);								-- remove them both
					end
				end

				if trim then
					if value ~= comp then 										-- value does not equal comp when value contains HTML markup
						local dup2 = duplicate_char;
						if f.match(dup2, "%A" ) then dup2 = "%" .. dup2; end	-- if duplicate_char not a letter then escape it
						
						value = f.gsub(value, "(%b<>)" .. dup2, "%1", 1 )		-- remove duplicate_char if it follows HTML markup
					else
						value = f.sub(value, 2, -1 );							-- remove duplicate_char when it is first character
					end
				end
			end
			str = str .. value; 												-- add it to the output string
		end
	end
	return str;
end


--[[--------------------------< I S _ S U F F I X >-----------------------------

returns true if suffix is properly formed Jr, Sr, or ordinal in the range 1–9.
Puncutation not allowed.

]]

local function is_suffix (suffix)
	if utilities.in_array (suffix, {'Jr', 'Sr', 'Jnr', 'Snr', '1st', '2nd', '3rd'}) or suffix:match ('^%dth$') then
		return true;
	end
	return false;
end


--[[--------------------< I S _ G O O D _ V A N C _ N A M E >-------------------

For Vancouver style, author/editor names are supposed to be rendered in Latin
(read ASCII) characters.  When a name uses characters that contain diacritical
marks, those characters are to be converted to the corresponding Latin
character. When a name is written using a non-Latin alphabet or logogram, that
name is to be transliterated into Latin characters. The module doesn't do this
so editors may/must.

This test allows |first= and |last= names to contain any of the letters defined
in the four Unicode Latin character sets
	[http://www.unicode.org/charts/PDF/U0000.pdf C0 Controls and Basic Latin] 0041–005A, 0061–007A
	[http://www.unicode.org/charts/PDF/U0080.pdf C1 Controls and Latin-1 Supplement] 00C0–00D6, 00D8–00F6, 00F8–00FF
	[http://www.unicode.org/charts/PDF/U0100.pdf Latin Extended-A] 0100–017F
	[http://www.unicode.org/charts/PDF/U0180.pdf Latin Extended-B] 0180–01BF, 01C4–024F

|lastn= also allowed to contain hyphens, spaces, and apostrophes.
	(http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35029/)
|firstn= also allowed to contain hyphens, spaces, apostrophes, and periods

This original test:
	if nil == mw.ustring.find (last, "^[A-Za-zÀ-ÖØ-öø-ƿDŽ-ɏ%-%s%']*$")
	or nil == mw.ustring.find (first, "^[A-Za-zÀ-ÖØ-öø-ƿDŽ-ɏ%-%s%'%.]+[2-6%a]*$") then
was written outside of the code editor and pasted here because the code editor
gets confused between character insertion point and cursor position. The test has
been rewritten to use decimal character escape sequence for the individual bytes
of the Unicode characters so that it is not necessary to use an external editor
to maintain this code.

	\195\128-\195\150 – À-Ö (U+00C0–U+00D6 – C0 controls)
	\195\152-\195\182 – Ø-ö (U+00D8-U+00F6 – C0 controls)
	\195\184-\198\191 – ø-ƿ (U+00F8-U+01BF – C0 controls, Latin extended A & B)
	\199\132-\201\143 – DŽ-ɏ (U+01C4-U+024F – Latin extended B)

]]

local function is_good_vanc_name (last, first, suffix, position)
	if not suffix then
		if first:find ('[,%s]') then											-- when there is a space or comma, might be first name/initials + generational suffix
			first = first:match ('(.-)[,%s]+');									-- get name/initials
			suffix = first:match ('[,%s]+(.+)$');								-- get generational suffix
		end
	end
	if utilities.is_set (suffix) then
		if not is_suffix (suffix) then
			add_vanc_error (cfg.err_msg_supl.suffix, position);
			return false;														-- not a name with an appropriate suffix
		end
	end
	if nil == mw.ustring.find (last, "^[A-Za-z\195\128-\195\150\195\152-\195\182\195\184-\198\191\199\132-\201\143%-%s%']*$") or
		nil == mw.ustring.find (first, "^[A-Za-z\195\128-\195\150\195\152-\195\182\195\184-\198\191\199\132-\201\143%-%s%'%.]*$") then
			add_vanc_error (cfg.err_msg_supl['non-Latin char'], position);
			return false;														-- not a string of Latin characters; Vancouver requires Romanization
	end;
	return true;
end


--[[--------------------------< R E D U C E _ T O _ I N I T I A L S >------------------------------------------

Attempts to convert names to initials in support of |name-list-style=vanc.  

Names in |firstn= may be separated by spaces or hyphens, or for initials, a period.
See http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35062/.

Vancouver style requires family rank designations (Jr, II, III, etc.) to be rendered
as Jr, 2nd, 3rd, etc.  See http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35085/.
This code only accepts and understands generational suffix in the Vancouver format
because Roman numerals look like, and can be mistaken for, initials.

This function uses ustring functions because firstname initials may be any of the
Unicode Latin characters accepted by is_good_vanc_name ().

]]

local function reduce_to_initials(first, position)
	local name, suffix = mw.ustring.match(first, "^(%u+) ([%dJS][%drndth]+)$");

	if not name then															-- if not initials and a suffix
		name = mw.ustring.match(first, "^(%u+)$");								-- is it just initials?
	end

	if name then																-- if first is initials with or without suffix
		if 3 > mw.ustring.len (name) then										-- if one or two initials
			if suffix then														-- if there is a suffix
				if is_suffix (suffix) then										-- is it legitimate?
					return first;												-- one or two initials and a valid suffix so nothing to do
				else
					add_vanc_error (cfg.err_msg_supl.suffix, position);			-- one or two initials with invalid suffix so error message
					return first;												-- and return first unmolested
				end
			else
				return first;													-- one or two initials without suffix; nothing to do
			end
		end
	end																			-- if here then name has 3 or more uppercase letters so treat them as a word

	local initials, names = {}, {};												-- tables to hold name parts and initials
	local i = 1;																-- counter for number of initials

	names = mw.text.split (first, '[%s,]+');									-- split into a table of names and possible suffix

	while names[i] do															-- loop through the table
		if 1 < i and names[i]:match ('[%dJS][%drndth]+%.?$') then				-- if not the first name, and looks like a suffix (may have trailing dot)
			names[i] = names[i]:gsub ('%.', '');								-- remove terminal dot if present
			if is_suffix (names[i]) then										-- if a legitimate suffix
				table.insert (initials, ' ' .. names[i]);						-- add a separator space, insert at end of initials table
				break;															-- and done because suffix must fall at the end of a name
			end																	-- no error message if not a suffix; possibly because of Romanization
		end
		if 3 > i then
			table.insert (initials, mw.ustring.sub(names[i], 1, 1));			-- insert the initial at end of initials table
		end
		i = i + 1;																-- bump the counter
	end
			
	return table.concat(initials)												-- Vancouver format does not include spaces.
end


--[[--------------------------< I N T E R W I K I _ P R E F I X E N _ G E T >----------------------------------

extract interwiki prefixen from <value>.  Returns two one or two values:
	false – no prefixen
	nil – prefix exists but not recognized
	project prefix, language prefix – when value has either of:
		:<project>:<language>:<article>
		:<language>:<project>:<article>
	project prefix, nil – when <value> has only a known single-letter prefix
	nil, language prefix – when <value> has only a known language prefix

accepts single-letter project prefixen: 'd' (wikidata), 's' (wikisource), and 'w' (wikipedia) prefixes; at this
writing, the other single-letter prefixen (b (wikibook), c (commons), m (meta), n (wikinews), q (wikiquote), and
v (wikiversity)) are not supported.

]]

local function interwiki_prefixen_get (value, is_link)
	if not value:find (':%l+:') then											-- if no prefix
		return false;															-- abandon; boolean here to distinguish from nil fail returns later
	end

	local prefix_patterns_linked_t = {											-- sequence of valid interwiki and inter project prefixen
		'^%[%[:([dsw]):(%l%l+):',												-- wikilinked; project and language prefixes
		'^%[%[:(%l%l+):([dsw]):',												-- wikilinked; language and project prefixes
		'^%[%[:([dsw]):',														-- wikilinked; project prefix
		'^%[%[:(%l%l+):',														-- wikilinked; language prefix
		}
		
	local prefix_patterns_unlinked_t = {										-- sequence of valid interwiki and inter project prefixen
		'^:([dsw]):(%l%l+):',													-- project and language prefixes
		'^:(%l%l+):([dsw]):',													-- language and project prefixes
		'^:([dsw]):',															-- project prefix
		'^:(%l%l+):',															-- language prefix
		}
	
	local cap1, cap2;
	for _, pattern in ipairs ((is_link and prefix_patterns_linked_t) or prefix_patterns_unlinked_t) do
		cap1, cap2 = value:match (pattern);
		if cap1 then
			break;																-- found a match so stop looking
		end
	end
	
	if cap1 and cap2 then														-- when both then :project:language: or :language:project: (both forms allowed)
		if 1 == #cap1 then														-- length == 1 then :project:language:
			if cfg.inter_wiki_map[cap2] then									-- is language prefix in the interwiki map?
				return cap1, cap2;												-- return interwiki project and interwiki language
			end
		else																	-- here when :language:project:
			if cfg.inter_wiki_map[cap1] then									-- is language prefix in the interwiki map?
				return cap2, cap1;												-- return interwiki project and interwiki language
			end
		end
		return nil;																-- unknown interwiki language
	elseif not (cap1 or cap2) then												-- both are nil?
		return nil;																-- we got something that looks like a project prefix but isn't; return fail
	elseif 1 == #cap1 then														-- here when one capture
		return cap1, nil;														-- length is 1 so return project, nil language
	else																		-- here when one capture and its length it more than 1
		if cfg.inter_wiki_map[cap1] then										-- is language prefix in the interwiki map?
			return nil, cap1;													-- return nil project, language
		end
	end
end


--[[--------------------------< L I S T _ P E O P L E >--------------------------

Formats a list of people (authors, contributors, editors, interviewers, translators) 

names in the list will be linked when
	|<name>-link= has a value
	|<name>-mask- does NOT have a value; masked names are presumed to have been
		rendered previously so should have been linked there

when |<name>-mask=0, the associated name is not rendered

]]

local function list_people (control, people, etal)
	local sep;
	local namesep;
	local format = control.format;
	local maximum = control.maximum;
	local name_list = {};

	if 'vanc' == format then													-- Vancouver-like name styling?
		sep = cfg.presentation['sep_nl_vanc'];									-- name-list separator between names is a comma
		namesep = cfg.presentation['sep_name_vanc'];							-- last/first separator is a space
	else
		sep = cfg.presentation['sep_nl'];										-- name-list separator between names is a semicolon
		namesep = cfg.presentation['sep_name'];									-- last/first separator is <comma><space>
	end
	
	if sep:sub (-1, -1) ~= " " then sep = sep .. " " end
	if utilities.is_set (maximum) and maximum < 1 then return "", 0; end		-- returned 0 is for EditorCount; not used for other names
	
	for i, person in ipairs (people) do
		if utilities.is_set (person.last) then
			local mask = person.mask;
			local one;
			local sep_one = sep;

			if utilities.is_set (maximum) and i > maximum then
				etal = true;
				break;
			end
			
			if mask then
				local n = tonumber (mask);										-- convert to a number if it can be converted; nil else
				if n then
					one = 0 ~= n and string.rep("&mdash;", n) or nil;			-- make a string of (n > 0) mdashes, nil else, to replace name
					person.link = nil;											-- don't create link to name if name is replaces with mdash string or has been set nil
				else
					one = mask;													-- replace name with mask text (must include name-list separator)
					sep_one = " ";												-- modify name-list separator
				end
			else
				one = person.last;												-- get surname
				local first = person.first										-- get given name
				if utilities.is_set (first) then
					if ("vanc" == format) then									-- if Vancouver format
						one = one:gsub ('%.', '');								-- remove periods from surnames (http://www.ncbi.nlm.nih.gov/books/NBK7271/box/A35029/)
						if not person.corporate and is_good_vanc_name (one, first, nil, i) then		-- and name is all Latin characters; corporate authors not tested
							first = reduce_to_initials (first, i);				-- attempt to convert first name(s) to initials
						end
					end
					one = one .. namesep .. first;
				end
			end
			if utilities.is_set (person.link) then
				one = utilities.make_wikilink (person.link, one);				-- link author/editor
			end

			if one then															-- if <one> has a value (name, mdash replacement, or mask text replacement)
				local proj, tag = interwiki_prefixen_get (one, true);			-- get the interwiki prefixen if present

				if 'w' == proj and ('Wikipedia' == mw.site.namespaces.Project['name']) then
					proj = nil;													-- for stuff like :w:de:<article>, :w is unnecessary TODO: maint cat?
				end
				if proj then
					proj = ({['d'] = 'Wikidata', ['s'] = 'Wikisource', ['w'] = 'Wikipedia'})[proj];	-- :w (wikipedia) for linking from a non-wikipedia project
					if proj then 
						one = one .. utilities.wrap_style ('interproj', proj);	-- add resized leading space, brackets, static text, language name
						tag = nil;												-- unset; don't do both project and language
					end
				end
				if tag == cfg.this_wiki_code then
					tag = nil;													-- stuff like :en:<article> at en.wiki is pointless TODO: maint cat?
				end
				if tag then
					local lang = cfg.lang_tag_remap[tag] or cfg.mw_languages_by_tag_t[tag];
					if lang then												-- error messaging done in extract_names() where we know parameter names
						one = one .. utilities.wrap_style ('interwiki', lang);	-- add resized leading space, brackets, static text, language name
					end
				end

				table.insert (name_list, one);									-- add it to the list of names
				table.insert (name_list, sep_one);								-- add the proper name-list separator
			end
		end
	end

	local count = #name_list / 2;												-- (number of names + number of separators) divided by 2
	if 0 < count then 
		if 1 < count and not etal then
			if 'amp' == format then
				name_list[#name_list-2] = " & ";								-- replace last separator with ampersand text
			elseif 'and' == format then
				if 2 == count then
					name_list[#name_list-2] = cfg.presentation.sep_nl_and;		-- replace last separator with 'and' text
				else
					name_list[#name_list-2] = cfg.presentation.sep_nl_end;		-- replace last separator with '(sep) and' text
				end
			end
		end
		name_list[#name_list] = nil;											-- erase the last separator
	end

	local result = table.concat (name_list);									-- construct list
	if etal and utilities.is_set (result) then									-- etal may be set by |display-authors=etal but we might not have a last-first list
		result = result .. sep .. cfg.messages['et al'];						-- we've got a last-first list and etal so add et al.
	end
	
	return result, count;														-- return name-list string and count of number of names (count used for editor names only)
end


--[[--------------------< M A K E _ C I T E R E F _ I D >-----------------------

Generates a CITEREF anchor ID if we have at least one name or a date.  Otherwise
returns an empty string.

namelist is one of the contributor-, author-, or editor-name lists chosen in that
order.  year is Year or anchor_year.

]]

local function make_citeref_id (namelist, year)
	local names={};							-- a table for the one to four names and year
	for i,v in ipairs (namelist) do			-- loop through the list and take up to the first four last names
		names[i] = v.last
		if i == 4 then break end			-- if four then done
	end
	table.insert (names, year);				-- add the year at the end
	local id = table.concat(names);			-- concatenate names and year for CITEREF id
	if utilities.is_set (id) then			-- if concatenation is not an empty string
		return "CITEREF" .. id;				-- add the CITEREF portion
	else
		return '';							-- return an empty string; no reason to include CITEREF id in this citation
	end
end


--[[--------------------------< C I T E _ C L A S S _A T T R I B U T E _M A K E >------------------------------

construct <cite> tag class attribute for this citation.

<cite_class> – config.CitationClass from calling template
<mode> – value from |mode= parameter

]]

local function cite_class_attribute_make (cite_class, mode)
	local class_t = {};
	table.insert (class_t, 'citation');											-- required for blue highlight
	if 'citation' ~= cite_class then
		table.insert (class_t, cite_class);										-- identify this template for user css
		table.insert (class_t, utilities.is_set (mode) and mode or 'cs1');		-- identify the citation style for user css or javascript
	else
		table.insert (class_t, utilities.is_set (mode) and mode or 'cs2');		-- identify the citation style for user css or javascript
	end
	for _, prop_key in ipairs (z.prop_keys_t) do
		table.insert (class_t, prop_key);										-- identify various properties for user css or javascript
	end

	return table.concat (class_t, ' ');											-- make a big string and done
end


--[[---------------------< N A M E _ H A S _ E T A L >--------------------------

Evaluates the content of name parameters (author, editor, etc.) for variations on
the theme of et al.  If found, the et al. is removed, a flag is set to true and
the function returns the modified name and the flag.

This function never sets the flag to false but returns its previous state because
it may have been set by previous passes through this function or by the associated
|display-<names>=etal parameter

]]

local function name_has_etal (name, etal, nocat, param)

	if utilities.is_set (name) then												-- name can be nil in which case just return
		local patterns = cfg.et_al_patterns; 									-- get patterns from configuration
		
		for _, pattern in ipairs (patterns) do									-- loop through all of the patterns
			if name:match (pattern) then										-- if this 'et al' pattern is found in name
				name = name:gsub (pattern, '');									-- remove the offending text
				etal = true;													-- set flag (may have been set previously here or by |display-<names>=etal)
				if not nocat then												-- no categorization for |vauthors=
					utilities.set_message ('err_etal', {param});				-- and set an error if not added
				end
			end
		end
	end

	return name, etal;
end


--[[---------------------< N A M E _ I S _ N U M E R I C >----------------------

Add an error message and category when <name> parameter value does not contain letters.  

Add a maintenance category when <name> parameter value has numeric characters mixed with characters that are 
not numeric characters; could be letters and/or punctuation characters.

This function will only emit one error and one maint message for the current template.  Does not emit both error
and maint messages/categories for the same parameter value.

returns nothing

]]

local function name_is_numeric (name, name_alias, list_name)
	local patterns = {
		'^%D+%d',																-- <name> must have digits preceded by other characters
		'^%D*%d+%D+',															-- <name> must have digits followed by other characters
		}

	if not added_numeric_name_errs and mw.ustring.match (name, '^[%A]+$') then	-- if we have not already set an error message and <name> does not have any alpha characters
		utilities.set_message ('err_numeric_names', name_alias);				-- add an error message
		added_numeric_name_errs = true;											-- set the flag so we emit only one error message
		return;																	-- when here no point in further testing; abandon
	end

	if not added_numeric_name_maint then										-- if we have already set a maint message
		for _, pattern in ipairs (patterns) do									-- spin through list of patterns
			if mw.ustring.match (name, pattern) then							-- digits preceded or followed by anything but digits; %D+ includes punctuation
				utilities.set_message ('maint_numeric_names', cfg.special_case_translation [list_name]);	-- add a maint cat for this template
				added_numeric_name_maint = true;								-- set the flag so we emit only one maint message
				return;															-- when here no point in further testing; abandon
			end
		end
	end
end


--[[-----------------< N A M E _ H A S _ M U L T _ N A M E S >------------------

Evaluates the content of last/surname (authors etc.) parameters for multiple names.
Multiple names are indicated if there is more than one comma or any "unescaped"
semicolons. Escaped semicolons are ones used as part of selected HTML entities.
If the condition is met, the function adds the multiple name maintenance category.

Same test for first except that commas should not appear in given names (MOS:JR says
that the generational suffix does not take a separator character).  Titles, degrees,
postnominals, affiliations, all normally comma separated don't belong in a citation.

<name> – name parameter value
<list_name> – AuthorList, EditorList, etc
<limit> – number of allowed commas; 1 (default) for surnames; 0 for given names

returns nothing

]]

local function name_has_mult_names (name, list_name, limit)
	local _, commas, semicolons, nbsps;
	limit = limit and limit or 1;
	if utilities.is_set (name) then
		_, commas = name:gsub (',', '');										-- count the number of commas
		_, semicolons = name:gsub (';', '');									-- count the number of semicolons
		-- nbsps probably should be its own separate count rather than merged in
		-- some way with semicolons because Lua patterns do not support the
		-- grouping operator that regex does, which means there is no way to add
		-- more entities to escape except by adding more counts with the new
		-- entities
		_, nbsps = name:gsub ('&nbsp;','');										-- count nbsps
		
		-- There is exactly 1 semicolon per &nbsp; entity, so subtract nbsps
		-- from semicolons to 'escape' them. If additional entities are added,
		-- they also can be subtracted.
		if limit < commas or 0 < (semicolons - nbsps) then
			utilities.set_message ('maint_mult_names', cfg.special_case_translation [list_name]);	-- add a maint message
		end
	end
end


--[=[-------------------------< I S _ G E N E R I C >----------------------------------------------------------

Compares values assigned to various parameters according to the string provided as <item> in the function call.
<item> can have on of two values:
	'generic_names' – for name-holding parameters: |last=, |first=, |editor-last=, etc
	'generic_titles' – for |title=

There are two types of generic tests.  The 'accept' tests look for a pattern that should not be rejected by the
'reject' test.  For example,
	|author=[[John Smith (author)|Smith, John]]
would be rejected by the 'author' reject test.  But piped wikilinks with 'author' disambiguation should not be
rejected so the 'accept' test prevents that from happening.  Accept tests are always performed before reject
tests.

Each of the 'accept' and 'reject' sequence tables hold tables for en.wiki (['en']) and local.wiki (['local'])
that each can hold a test sequence table  The sequence table holds, at index [1], a test pattern, and, at index
[2], a boolean control value.  The control value tells string.find() or mw.ustring.find() to do plain-text search (true)
or a pattern search (false).  The intent of all this complexity is to make these searches as fast as possible so
that we don't run out of processing time on very large articles.

Returns
	true when a reject test finds the pattern or string
	false when an accept test finds the pattern or string
	nil else

]=]

local function is_generic (item, value, wiki)
	local test_val;
	local str_lower = {															-- use string.lower() for en.wiki (['en']) and use mw.ustring.lower() or local.wiki (['local'])
		['en'] = string.lower,
		['local'] = mw.ustring.lower,
		}
	local str_find = {															-- use string.find() for en.wiki (['en']) and use mw.ustring.find() or local.wiki (['local'])
		['en'] = string.find,
		['local'] = mw.ustring.find,
		}

	local function test (val, test_t, wiki)										-- local function to do the testing; <wiki> selects lower() and find() functions
		val = test_t[2] and str_lower[wiki](value) or val;						-- when <test_t[2]> set to 'true', plaintext search using lowercase value
		return str_find[wiki] (val, test_t[1], 1, test_t[2]);					-- return nil when not found or matched
	end
		
	local test_types_t = {'accept', 'reject'};									-- test accept patterns first, then reject patterns
	local wikis_t = {'en', 'local'};											-- do tests for each of these keys; en.wiki first, local.wiki second

	for _, test_type in ipairs (test_types_t) do								-- for each test type
		for _, generic_value in pairs (cfg.special_case_translation[item][test_type]) do	-- spin through the list of generic value fragments to accept or reject
			for _, wiki in ipairs (wikis_t) do
				if generic_value[wiki] then
					if test (value, generic_value[wiki], wiki) then				-- go do the test
						return ('reject' == test_type);							-- param value rejected, return true; false else
					end
				end
			end
		end
	end
end


--[[--------------------------< N A M E _ I S _ G E N E R I C >------------------------------------------------

calls is_generic() to determine if <name> is a 'generic name' listed in cfg.generic_names; <name_alias> is the
parameter name used in error messaging

]]

local function name_is_generic (name, name_alias)
	if not added_generic_name_errs  and is_generic ('generic_names', name) then
		utilities.set_message ('err_generic_name', name_alias);					-- set an error message
		added_generic_name_errs = true;
	end
end


--[[--------------------------< N A M E _ C H E C K S >--------------------------------------------------------

This function calls various name checking functions used to validate the content of the various name-holding parameters.

]]

local function name_checks (last, first, list_name, last_alias, first_alias)
	local accept_name;

	if utilities.is_set (last) then
		last, accept_name = utilities.has_accept_as_written (last);				-- remove accept-this-as-written markup when it wraps all of <last>

		if not accept_name then													-- <last> not wrapped in accept-as-written markup
			name_has_mult_names (last, list_name);								-- check for multiple names in the parameter
			name_is_numeric (last, last_alias, list_name);						-- check for names that have no letters or are a mix of digits and other characters
			name_is_generic (last, last_alias);									-- check for names found in the generic names list
		end
	end

	if utilities.is_set (first) then
		first, accept_name = utilities.has_accept_as_written (first);			-- remove accept-this-as-written markup when it wraps all of <first>

		if not accept_name then													-- <first> not wrapped in accept-as-written markup
			name_has_mult_names (first, list_name, 0);							-- check for multiple names in the parameter; 0 is number of allowed commas in a given name
			name_is_numeric (first, first_alias, list_name);					-- check for names that have no letters or are a mix of digits and other characters
			name_is_generic (first, first_alias);								-- check for names found in the generic names list
		end
		local wl_type, D = utilities.is_wikilink (first);
		if 0 ~= wl_type then
			first = D;
			utilities.set_message ('err_bad_paramlink', first_alias);
		end
	end

	return last, first;															-- done
end


--[[----------------------< E X T R A C T _ N A M E S >-------------------------

Gets name list from the input arguments

Searches through args in sequential order to find |lastn= and |firstn= parameters
(or their aliases), and their matching link and mask parameters. Stops searching
when both |lastn= and |firstn= are not found in args after two sequential attempts:
found |last1=, |last2=, and |last3= but doesn't find |last4= and |last5= then the
search is done.

This function emits an error message when there is a |firstn= without a matching
|lastn=.  When there are 'holes' in the list of last names, |last1= and |last3=
are present but |last2= is missing, an error message is emitted. |lastn= is not
required to have a matching |firstn=.

When an author or editor parameter contains some form of 'et al.', the 'et al.'
is stripped from the parameter and a flag (etal) returned that will cause list_people()
to add the static 'et al.' text from Module:Citation/CS1/Configuration.  This keeps
'et al.' out of the template's metadata.  When this occurs, an error is emitted.

]]

local function extract_names(args, list_name)
	local names = {};															-- table of names
	local last;																	-- individual name components
	local first;
	local link;
	local mask;
	local i = 1;																-- loop counter/indexer
	local n = 1;																-- output table indexer
	local count = 0;															-- used to count the number of times we haven't found a |last= (or alias for authors, |editor-last or alias for editors)
	local etal = false;															-- return value set to true when we find some form of et al. in an author parameter

	local last_alias, first_alias, link_alias;									-- selected parameter aliases used in error messaging
	while true do
		last, last_alias = utilities.select_one ( args, cfg.aliases[list_name .. '-Last'], 'err_redundant_parameters', i );		-- search through args for name components beginning at 1
		first, first_alias = utilities.select_one ( args, cfg.aliases[list_name .. '-First'], 'err_redundant_parameters', i );
		link, link_alias = utilities.select_one ( args, cfg.aliases[list_name .. '-Link'], 'err_redundant_parameters', i );
		mask = utilities.select_one ( args, cfg.aliases[list_name .. '-Mask'], 'err_redundant_parameters', i );
	
		if last then															-- error check |lastn= alias for unknown interwiki link prefix; done here because this is where we have the parameter name
			local project, language = interwiki_prefixen_get (last, true);		-- true because we expect interwiki links in |lastn= to be wikilinked
			if nil == project and nil == language then							-- when both are nil
				utilities.set_message ('err_bad_paramlink', last_alias);		-- not known, emit an error message	-- TODO: err_bad_interwiki?
				last = utilities.remove_wiki_link (last);						-- remove wikilink markup; show display value only
			end
		end
		
		if link then															-- error check |linkn= alias for unknown interwiki link prefix
			local project, language = interwiki_prefixen_get (link, false);		-- false because wiki links in |author-linkn= is an error
			if nil == project and nil == language then							-- when both are nil
				utilities.set_message ('err_bad_paramlink', link_alias);		-- not known, emit an error message	-- TODO: err_bad_interwiki?
				link = nil;														-- unset so we don't link
				link_alias = nil;
			end
		end
		
		last, etal = name_has_etal (last, etal, false, last_alias);				-- find and remove variations on et al.
		first, etal = name_has_etal (first, etal, false, first_alias);			-- find and remove variations on et al.
		last, first = name_checks (last, first, list_name, last_alias, first_alias);						-- multiple names, extraneous annotation, etc. checks

		if first and not last then												-- if there is a firstn without a matching lastn
			local alias = first_alias:find ('given', 1, true) and 'given' or 'first';	-- get first or given form of the alias
			utilities.set_message ('err_first_missing_last', {
				first_alias,													-- param name of alias missing its mate
				first_alias:gsub (alias, {['first'] = 'last', ['given'] = 'surname'}),	-- make param name appropriate to the alias form
				});																-- add this error message
		elseif not first and not last then										-- if both firstn and lastn aren't found, are we done?
			count = count + 1;													-- number of times we haven't found last and first
			if 2 <= count then													-- two missing names and we give up
				break;															-- normal exit or there is a two-name hole in the list; can't tell which
			end
		else																	-- we have last with or without a first
			local result;
			link = link_title_ok (link, link_alias, last, last_alias);			-- check for improper wiki-markup

			if first then
				link = link_title_ok (link, link_alias, first, first_alias);	-- check for improper wiki-markup
			end

			names[n] = {last = last, first = first, link = link, mask = mask, corporate = false};	-- add this name to our names list (corporate for |vauthors= only)
			n = n + 1;															-- point to next location in the names table
			if 1 == count then													-- if the previous name was missing
				utilities.set_message ('err_missing_name', {list_name:match ("(%w+)List"):lower(), i - 1});	-- add this error message
			end
			count = 0;															-- reset the counter, we're looking for two consecutive missing names
		end
		i = i + 1;																-- point to next args location
	end
	
	return names, etal;															-- all done, return our list of names and the etal flag
end


--[[--------------------------< N A M E _ T A G _ G E T >------------------------------------------------------

attempt to decode |language=<lang_param> and return language name and matching tag; nil else.

This function looks for:
	<lang_param> as a tag in cfg.lang_tag_remap{}
	<lang_param> as a name in cfg.lang_name_remap{}
	
	<lang_param> as a name in cfg.mw_languages_by_name_t
	<lang_param> as a tag in cfg.mw_languages_by_tag_t
when those fail, presume that <lang_param> is an IETF-like tag that MediaWiki does not recognize.  Strip all
script, region, variant, whatever subtags from <lang_param> to leave just a two or three character language tag
and look for the new <lang_param> in cfg.mw_languages_by_tag_t{}

on success, returns name (in properly capitalized form) and matching tag (in lowercase); on failure returns nil

]]

local function name_tag_get (lang_param)
	local lang_param_lc = mw.ustring.lower (lang_param);						-- use lowercase as an index into the various tables
	local name;
	local tag;

	name = cfg.lang_tag_remap[lang_param_lc];									-- assume <lang_param_lc> is a tag; attempt to get remapped language name 
	if name then																-- when <name>, <lang_param> is a tag for a remapped language name
		if cfg.lang_name_remap[name:lower()][2] ~= lang_param_lc then
			utilities.set_message ('maint_unknown_lang');						-- add maint category if not already added
			return name, cfg.lang_name_remap[name:lower()][2];					-- so return name and tag from lang_name_remap[name]; special case to xlate sr-ec and sr-el to sr-cyrl and sr-latn
		end
		return name, lang_param_lc;												-- so return <name> from remap and <lang_param_lc>
	end

	tag = lang_param_lc:match ('^(%a%a%a?)%-.*');								-- still assuming that <lang_param_lc> is a tag; strip script, region, variant subtags
	name = cfg.lang_tag_remap[tag];												-- attempt to get remapped language name with language subtag only
	if name then																-- when <name>, <tag> is a tag for a remapped language name
		return name, tag;														-- so return <name> from remap and <tag>
	end

	if cfg.lang_name_remap[lang_param_lc] then									-- not a remapped tag, assume <lang_param_lc> is a name; attempt to get remapped language tag 
		return cfg.lang_name_remap[lang_param_lc][1], cfg.lang_name_remap[lang_param_lc][2];	-- for this <lang_param_lc>, return a (possibly) new name and appropriate tag
	end

	name = cfg.mw_languages_by_tag_t[lang_param_lc];							-- assume that <lang_param_lc> is a tag; attempt to get its matching language name
	
	if name then
		return name, lang_param_lc;												-- <lang_param_lc> is a tag so return it and <name>
	end
	
	tag = cfg.mw_languages_by_name_t[lang_param_lc];							-- assume that <lang_param_lc> is a language name; attempt to get its matching tag
	
	if tag then
		return cfg.mw_languages_by_tag_t[tag], tag;								-- <lang_param_lc> is a name so return the name from the table and <tag>
	end

	tag = lang_param_lc:match ('^(%a%a%a?)%-.*');								-- is <lang_param_lc> an IETF-like tag that MediaWiki doesn't recognize? <tag> gets the language subtag; nil else

	if tag then
		name = cfg.mw_languages_by_tag_t[tag];									-- attempt to get a language name using the shortened <tag>
		if name then
			return name, tag;													-- <lang_param_lc> is an unrecognized IETF-like tag so return <name> and language subtag
		end
	end
end


--[[-------------------< L A N G U A G E _ P A R A M E T E R >------------------

Gets language name from a provided two- or three-character ISO 639 code.  If a code
is recognized by MediaWiki, use the returned name; if not, then use the value that
was provided with the language parameter.

When |language= contains a recognized language (either code or name), the page is
assigned to the category for that code: Category:Norwegian-language sources (no).
For valid three-character code languages, the page is assigned to the single category
for '639-2' codes: Category:CS1 ISO 639-2 language sources.

Languages that are the same as the local wiki are not categorized.  MediaWiki does
not recognize three-character equivalents of two-character codes: code 'ar' is
recognized but code 'ara' is not.

This function supports multiple languages in the form |language=nb, French, th
where the language names or codes are separated from each other by commas with
optional space characters.

]]

local function language_parameter (lang)
	local tag;																	-- some form of IETF-like language tag; language subtag with optional region, sript, vatiant, etc subtags
	local lang_subtag;															-- ve populates |language= with mostly unecessary region subtags the MediaWiki does not recognize; this is the base language subtag
	local name;																	-- the language name
	local language_list = {};													-- table of language names to be rendered
	local names_t = {};															-- table made from the value assigned to |language=

	local this_wiki_name = mw.language.fetchLanguageName (cfg.this_wiki_code, cfg.this_wiki_code);	-- get this wiki's language name

	names_t = mw.text.split (lang, '%s*,%s*');									-- names should be a comma separated list

	for _, lang in ipairs (names_t) do											-- reuse lang here because we don't yet know if lang is a language name or a language tag
		name, tag = name_tag_get (lang);										-- attempt to get name/tag pair for <lang>; <name> has proper capitalization; <tag> is lowercase

		if utilities.is_set (tag) then
			lang_subtag = tag:gsub ('^(%a%a%a?)%-.*', '%1');					-- for categorization, strip any IETF-like tags from language tag

			if cfg.this_wiki_code ~= lang_subtag then							-- when the language is not the same as this wiki's language
				if 2 == lang_subtag:len() then									-- and is a two-character tag
					utilities.add_prop_cat ('foreign-lang-source', {name, tag}, lang_subtag);		-- categorize it; tag appended to allow for multiple language categorization
				else															-- or is a recognized language (but has a three-character tag)
					utilities.add_prop_cat ('foreign-lang-source-2', {lang_subtag}, lang_subtag);			-- categorize it differently TODO: support multiple three-character tag categories per cs1|2 template?
				end
			elseif cfg.local_lang_cat_enable then								-- when the language and this wiki's language are the same and categorization is enabled
				utilities.add_prop_cat ('local-lang-source', {name, lang_subtag});		-- categorize it
			end
		else
			name = lang;														-- return whatever <lang> has so that we show something
			utilities.set_message ('maint_unknown_lang');						-- add maint category if not already added
		end
		
		table.insert (language_list, name);
		name = '';																-- so we can reuse it
	end
 
	name = utilities.make_sep_list (#language_list, language_list);
	if (1 == #language_list) and (lang_subtag == cfg.this_wiki_code) then		-- when only one language, find lang name in this wiki lang name; for |language=en-us, 'English' in 'American English'
		return '';																-- if one language and that language is this wiki's return an empty string (no annotation)
	end
	return (" " .. wrap_msg ('language', name));								-- otherwise wrap with '(in ...)'
	--[[ TODO: should only return blank or name rather than full list
	so we can clean up the bunched parenthetical elements Language, Type, Format
	]]
end


--[[-----------------------< S E T _ C S _ S T Y L E >--------------------------

Gets the default CS style configuration for the given mode.
Returns default separator and either postscript as passed in or the default.
In CS1, the default postscript and separator are '.'.
In CS2, the default postscript is the empty string and the default separator is ','.

]]

local function set_cs_style (postscript, mode)
	if utilities.is_set(postscript) then
		-- emit a maintenance message if user postscript is the default cs1 postscript
		-- we catch the opposite case for cs2 in set_style
		if mode == 'cs1' and postscript == cfg.presentation['ps_' .. mode] then
			utilities.set_message ('maint_postscript');
		end
	else
		postscript = cfg.presentation['ps_' .. mode];
	end
	return cfg.presentation['sep_' .. mode], postscript;
end


--[[--------------------------< S E T _ S T Y L E >-----------------------------

Sets the separator and postscript styles. Checks the |mode= first and the
#invoke CitationClass second. Removes the postscript if postscript == none.

]]

local function set_style (mode, postscript, cite_class)
	local sep;
	if 'cs2' == mode then
		sep, postscript = set_cs_style (postscript, 'cs2');
	elseif 'cs1' == mode then
		sep, postscript = set_cs_style (postscript, 'cs1');
	elseif 'citation' == cite_class	then
		sep, postscript = set_cs_style (postscript, 'cs2');
	else
		sep, postscript = set_cs_style (postscript, 'cs1');
	end

	if cfg.keywords_xlate[postscript:lower()] == 'none' then
		-- emit a maintenance message if user postscript is the default cs2 postscript
		-- we catch the opposite case for cs1 in set_cs_style
		if 'cs2' == mode or 'citation' == cite_class then
			utilities.set_message ('maint_postscript');
		end
		postscript = '';
	end
	
	return sep, postscript
end


--[=[-------------------------< I S _ P D F >-----------------------------------

Determines if a URL has the file extension that is one of the PDF file extensions
used by [[MediaWiki:Common.css]] when applying the PDF icon to external links.

returns true if file extension is one of the recognized extensions, else false

]=]

local function is_pdf (url)
	return url:match ('%.pdf$') or url:match ('%.PDF$') or
		url:match ('%.pdf[%?#]') or url:match ('%.PDF[%?#]') or
		url:match ('%.PDF&#035') or url:match ('%.pdf&#035');
end


--[[--------------------------< S T Y L E _ F O R M A T >-----------------------

Applies CSS style to |format=, |chapter-format=, etc.  Also emits an error message
if the format parameter does not have a matching URL parameter.  If the format parameter
is not set and the URL contains a file extension that is recognized as a PDF document
by MediaWiki's commons.css, this code will set the format parameter to (PDF) with
the appropriate styling.

]]

local function style_format (format, url, fmt_param, url_param)
	if utilities.is_set (format) then
		format = utilities.wrap_style ('format', format);						-- add leading space, parentheses, resize
		if not utilities.is_set (url) then
			utilities.set_message ('err_format_missing_url', {fmt_param, url_param});	-- add an error message
		end
	elseif is_pdf (url) then													-- format is not set so if URL is a PDF file then
		format = utilities.wrap_style ('format', 'PDF');						-- set format to PDF
	else
		format = '';															-- empty string for concatenation
	end
	return format;
end


--[[---------------------< G E T _ D I S P L A Y _ N A M E S >------------------

Returns a number that defines the number of names displayed for author and editor
name lists and a Boolean flag to indicate when et al. should be appended to the name list.

When the value assigned to |display-xxxxors= is a number greater than or equal to zero,
return the number and the previous state of the 'etal' flag (false by default
but may have been set to true if the name list contains some variant of the text 'et al.').

When the value assigned to |display-xxxxors= is the keyword 'etal', return a number
that is one greater than the number of authors in the list and set the 'etal' flag true.
This will cause the list_people() to display all of the names in the name list followed by 'et al.'

In all other cases, returns nil and the previous state of the 'etal' flag.

inputs:
	max: A['DisplayAuthors'] or A['DisplayEditors'], etc; a number or some flavor of etal
	count: #a or #e
	list_name: 'authors' or 'editors'
	etal: author_etal or editor_etal

This function sets an error message when |display-xxxxors= value greater than or equal to number of names but
not when <max> comes from {{cs1 config}} global settings.  When using global settings, <param> is set to the
keyword 'cs1 config' which is used to supress the normal error.  Error is suppressed because it is to be expected
that some citations in an article will have the same or fewer names that the limit specified in {{cs1 config}}.

]]

local function get_display_names (max, count, list_name, etal, param)
	if utilities.is_set (max) then
		if 'etal' == max:lower():gsub("[ '%.]", '') then						-- the :gsub() portion makes 'etal' from a variety of 'et al.' spellings and stylings
			max = count + 1;													-- number of authors + 1 so display all author name plus et al.
			etal = true;														-- overrides value set by extract_names()
		elseif max:match ('^%d+$') then											-- if is a string of numbers
			max = tonumber (max);												-- make it a number
			if (max >= count) and ('cs1 config' ~= param) then					-- error when local |display-xxxxors= value greater than or equal to number of names; not an error when using global setting
				utilities.set_message ('err_disp_name', {param, max});			-- add error message
				max = nil;
			end
		else																	-- not a valid keyword or number
			utilities.set_message ('err_disp_name', {param, max});				-- add error message
			max = nil;															-- unset; as if |display-xxxxors= had not been set
		end
	end
	
	return max, etal;
end


--[[----------< E X T R A _ T E X T _ I N _ P A G E _ C H E C K >---------------

Adds error if |page=, |pages=, |quote-page=, |quote-pages= has what appears to be
some form of p. or pp. abbreviation in the first characters of the parameter content.

check page for extraneous p, p., pp, pp., pg, pg. at start of parameter value:
	good pattern: '^P[^%.P%l]' matches when page begins PX or P# but not Px
		      where x and X are letters and # is a digit
	bad pattern:  '^[Pp][PpGg]' matches when page begins pp, pP, Pp, PP, pg, pG, Pg, PG

]]

local function extra_text_in_page_check (val, name)
	if not val:match (cfg.vol_iss_pg_patterns.good_ppattern) then
		for _, pattern in ipairs (cfg.vol_iss_pg_patterns.bad_ppatterns) do		-- spin through the selected sequence table of patterns
			if val:match (pattern) then											-- when a match, error so
				utilities.set_message ('err_extra_text_pages', name);	 		-- add error message
				return;															-- and done
			end
		end
	end		
end


--[[--------------------------< E X T R A _ T E X T _ I N _ V O L _ I S S _ C H E C K >------------------------

Adds error if |volume= or |issue= has what appears to be some form of redundant 'type' indicator.

For |volume=:
	'V.', or 'Vol.' (with or without the dot) abbreviations or 'Volume' in the first characters of the parameter
	content (all case insensitive). 'V' and 'v' (without the dot) are presumed to be roman numerals so
	are allowed.

For |issue=:
	'No.', 'I.', 'Iss.' (with or without the dot) abbreviations, or 'Issue' in the first characters of the
	parameter content (all case insensitive).
	
Single character values ('v', 'i', 'n') allowed when not followed by separator character ('.', ':', '=', or
whitespace character) – param values are trimmed of whitespace by MediaWiki before delivered to the module.
	
<val> is |volume= or |issue= parameter value
<name> is |volume= or |issue= parameter name for error message
<selector> is 'v' for |volume=, 'i' for |issue=

sets error message on failure; returns nothing

]]

local function extra_text_in_vol_iss_check (val, name, selector)
	if not utilities.is_set (val) then
		return;
	end
	
	local patterns = 'v' == selector and cfg.vol_iss_pg_patterns.vpatterns or cfg.vol_iss_pg_patterns.ipatterns;

	local handler = 'v' == selector and 'err_extra_text_volume' or 'err_extra_text_issue';
	val = val:lower();															-- force parameter value to lower case
	for _, pattern in ipairs (patterns) do										-- spin through the selected sequence table of patterns
		if val:match (pattern) then												-- when a match, error so
			utilities.set_message (handler, name);								-- add error message
			return;																-- and done
		end
	end
end


--[=[-------------------------< G E T _ V _ N A M E _ T A B L E >----------------------------------------------

split apart a |vauthors= or |veditors= parameter.  This function allows for corporate names, wrapped in doubled
parentheses to also have commas; in the old version of the code, the doubled parentheses were included in the
rendered citation and in the metadata.  Individual author names may be wikilinked

	|vauthors=Jones AB, [[E. B. White|White EB]], ((Black, Brown, and Co.))

]=]

local function get_v_name_table (vparam, output_table, output_link_table)
	local name_table = mw.text.split(vparam, "%s*,%s*");						-- names are separated by commas
	local wl_type, label, link;													-- wl_type not used here; just a placeholder
	
	local i = 1;
	
	while name_table[i] do
		if name_table[i]:match ('^%(%(.*[^%)][^%)]$') then						-- first segment of corporate with one or more commas; this segment has the opening doubled parentheses
			local name = name_table[i];
			i = i + 1;															-- bump indexer to next segment
			while name_table[i] do
				name = name .. ', ' .. name_table[i];							-- concatenate with previous segments
				if name_table[i]:match ('^.*%)%)$') then						-- if this table member has the closing doubled parentheses
					break;														-- and done reassembling so
				end
				i = i + 1;														-- bump indexer
			end
			table.insert (output_table, name);									-- and add corporate name to the output table
			table.insert (output_link_table, '');								-- no wikilink
		else
			wl_type, label, link = utilities.is_wikilink (name_table[i]);		-- wl_type is: 0, no wl (text in label variable); 1, [[D]]; 2, [[L|D]]
			table.insert (output_table, label);									-- add this name
			if 1 == wl_type then
				table.insert (output_link_table, label);						-- simple wikilink [[D]]
			else
				table.insert (output_link_table, link);							-- no wikilink or [[L|D]]; add this link if there is one, else empty string
			end
		end
		i = i + 1;
	end	
	return output_table;
end


--[[--------------------------< P A R S E _ V A U T H O R S _ V E D I T O R S >--------------------------------

This function extracts author / editor names from |vauthors= or |veditors= and finds matching |xxxxor-maskn= and
|xxxxor-linkn= in args.  It then returns a table of assembled names just as extract_names() does.

Author / editor names in |vauthors= or |veditors= must be in Vancouver system style. Corporate or institutional names
may sometimes be required and because such names will often fail the is_good_vanc_name() and other format compliance
tests, are wrapped in doubled parentheses ((corporate name)) to suppress the format tests.

Supports generational suffixes Jr, 2nd, 3rd, 4th–6th.

This function sets the Vancouver error when a required comma is missing and when there is a space between an author's initials.

]]

local function parse_vauthors_veditors (args, vparam, list_name)
	local names = {};															-- table of names assembled from |vauthors=, |author-maskn=, |author-linkn=
	local v_name_table = {};
	local v_link_table = {};													-- when name is wikilinked, targets go in this table
	local etal = false;															-- return value set to true when we find some form of et al. vauthors parameter
	local last, first, link, mask, suffix;
	local corporate = false;

	vparam, etal = name_has_etal (vparam, etal, true);							-- find and remove variations on et al. do not categorize (do it here because et al. might have a period)
	v_name_table = get_v_name_table (vparam, v_name_table, v_link_table);		-- names are separated by commas

	for i, v_name in ipairs(v_name_table) do
		first = '';																-- set to empty string for concatenation and because it may have been set for previous author/editor
		local accept_name;
		v_name, accept_name = utilities.has_accept_as_written (v_name);			-- remove accept-this-as-written markup when it wraps all of <v_name>

		if accept_name then
			last = v_name;
			corporate = true;													-- flag used in list_people()
		elseif string.find(v_name, "%s") then
			if v_name:find('[;%.]') then										-- look for commonly occurring punctuation characters; 
				add_vanc_error (cfg.err_msg_supl.punctuation, i);
			end
			local lastfirstTable = {}
			lastfirstTable = mw.text.split(v_name, "%s+")
			first = table.remove(lastfirstTable);								-- removes and returns value of last element in table which should be initials or generational suffix

			if not mw.ustring.match (first, '^%u+$') then						-- mw.ustring here so that later we will catch non-Latin characters
				suffix = first;													-- not initials so assume that whatever we got is a generational suffix
				first = table.remove(lastfirstTable);							-- get what should be the initials from the table
			end
			last = table.concat(lastfirstTable, ' ')							-- returns a string that is the concatenation of all other names that are not initials and generational suffix
			if not utilities.is_set (last) then
				first = '';														-- unset
				last = v_name;													-- last empty because something wrong with first
				add_vanc_error (cfg.err_msg_supl.name, i);
			end
			if mw.ustring.match (last, '%a+%s+%u+%s+%a+') then
				add_vanc_error (cfg.err_msg_supl['missing comma'], i);			-- matches last II last; the case when a comma is missing
			end
			if mw.ustring.match (v_name, ' %u %u$') then						-- this test is in the wrong place TODO: move or replace with a more appropriate test
				add_vanc_error (cfg.err_msg_supl.initials, i);					-- matches a space between two initials
			end
		else
			last = v_name;														-- last name or single corporate name?  Doesn't support multiword corporate names? do we need this?
		end
		
		if utilities.is_set (first) then
			if not mw.ustring.match (first, "^%u?%u$") then						-- first shall contain one or two upper-case letters, nothing else
				add_vanc_error (cfg.err_msg_supl.initials, i);					-- too many initials; mixed case initials (which may be ok Romanization); hyphenated initials
			end
			is_good_vanc_name (last, first, suffix, i);							-- check first and last before restoring the suffix which may have a non-Latin digit
			if utilities.is_set (suffix) then
				first = first .. ' ' .. suffix;									-- if there was a suffix concatenate with the initials
				suffix = '';													-- unset so we don't add this suffix to all subsequent names
			end
		else
			if not corporate then
				is_good_vanc_name (last, '', nil, i);
			end
		end

		link = utilities.select_one ( args, cfg.aliases[list_name .. '-Link'], 'err_redundant_parameters', i ) or v_link_table[i];
		mask = utilities.select_one ( args, cfg.aliases[list_name .. '-Mask'], 'err_redundant_parameters', i );
		names[i] = {last = last, first = first, link = link, mask = mask, corporate = corporate};		-- add this assembled name to our names list
	end
	return names, etal;															-- all done, return our list of names
end


--[[--------------------------< S E L E C T _ A U T H O R _ E D I T O R _ S O U R C E >------------------------

Select one of |authors=, |authorn= / |lastn / firstn=, or |vauthors= as the source of the author name list or
select one of |editorn= / editor-lastn= / |editor-firstn= or |veditors= as the source of the editor name list.

Only one of these appropriate three will be used.  The hierarchy is: |authorn= (and aliases) highest and |authors= lowest;
|editorn= (and aliases) highest and |veditors= lowest (support for |editors= withdrawn)

When looking for |authorn= / |editorn= parameters, test |xxxxor1= and |xxxxor2= (and all of their aliases); stops after the second
test which mimicks the test used in extract_names() when looking for a hole in the author name list.  There may be a better
way to do this, I just haven't discovered what that way is.

Emits an error message when more than one xxxxor name source is provided.

In this function, vxxxxors = vauthors or veditors; xxxxors = authors as appropriate.

]]

local function select_author_editor_source (vxxxxors, xxxxors, args, list_name)
	local lastfirst = false;
	if utilities.select_one ( args, cfg.aliases[list_name .. '-Last'], 'none', 1 ) or		-- do this twice in case we have a |first1= without a |last1=; this ...
		utilities.select_one ( args, cfg.aliases[list_name .. '-First'], 'none', 1 ) or		-- ... also catches the case where |first= is used with |vauthors=
		utilities.select_one ( args, cfg.aliases[list_name .. '-Last'], 'none', 2 ) or
		utilities.select_one ( args, cfg.aliases[list_name .. '-First'], 'none', 2 ) then
			lastfirst = true;
	end

	if (utilities.is_set (vxxxxors) and true == lastfirst) or					-- these are the three error conditions
		(utilities.is_set (vxxxxors) and utilities.is_set (xxxxors)) or
		(true == lastfirst and utilities.is_set (xxxxors)) then
			local err_name;
			if 'AuthorList' == list_name then									-- figure out which name should be used in error message
				err_name = 'author';
			else
				err_name = 'editor';
			end
			utilities.set_message ('err_redundant_parameters', err_name .. '-name-list parameters');	-- add error message
	end

	if true == lastfirst then return 1 end;										-- return a number indicating which author name source to use
	if utilities.is_set (vxxxxors) then return 2 end;
	if utilities.is_set (xxxxors) then return 3 end;
	return 1;																	-- no authors so return 1; this allows missing author name test to run in case there is a first without last 
end


--[[--------------------------< I S _ V A L I D _ P A R A M E T E R _ V A L U E >------------------------------

This function is used to validate a parameter's assigned value for those parameters that have only a limited number
of allowable values (yes, y, true, live, dead, etc.).  When the parameter value has not been assigned a value (missing
or empty in the source template) the function returns the value specified by ret_val.  If the parameter value is one
of the list of allowed values returns the translated value; else, emits an error message and returns the value
specified by ret_val.

TODO: explain <invert>

]]

local function is_valid_parameter_value (value, name, possible, ret_val, invert)
	if not utilities.is_set (value) then
		return ret_val;															-- an empty parameter is ok
	end

	if (not invert and utilities.in_array (value, possible)) then				-- normal; <value> is in <possible> table
		return cfg.keywords_xlate[value];										-- return translation of parameter keyword
	elseif invert and not utilities.in_array (value, possible) then				-- invert; <value> is not in <possible> table
		return value;															-- return <value> as it is
	else
		utilities.set_message ('err_invalid_param_val', {name, value});			-- not an allowed value so add error message
		return ret_val;
	end
end


--[[--------------------------< T E R M I N A T E _ N A M E _ L I S T >----------------------------------------

This function terminates a name list (author, contributor, editor) with a separator character (sepc) and a space
when the last character is not a sepc character or when the last three characters are not sepc followed by two
closing square brackets (close of a wikilink).  When either of these is true, the name_list is terminated with a
single space character.

]]

local function terminate_name_list (name_list, sepc)
	if (string.sub (name_list, -3, -1) == sepc .. '. ') then					-- if already properly terminated
		return name_list;														-- just return the name list
	elseif (string.sub (name_list, -1, -1) == sepc) or (string.sub (name_list, -3, -1) == sepc .. ']]') then	-- if last name in list ends with sepc char
		return name_list .. " ";												-- don't add another
	else
		return name_list .. sepc .. ' ';										-- otherwise terminate the name list
	end
end


--[[-------------------------< F O R M A T _ V O L U M E _ I S S U E >-----------------------------------------

returns the concatenation of the formatted volume and issue (or journal article number) parameters as a single
string; or formatted volume or formatted issue, or an empty string if neither are set.

]]
	
local function format_volume_issue (volume, issue, article, cite_class, origin, sepc, lower)
	if not utilities.is_set (volume) and not utilities.is_set (issue) and not utilities.is_set (article) then
		return '';
	end

	-- same condition as in format_pages_sheets()
	local is_journal = 'journal' == cite_class or (utilities.in_array (cite_class, {'citation', 'map', 'interview'}) and 'journal' == origin);

	local is_numeric_vol = volume and (volume:match ('^[MDCLXVI]+$') or volume:match ('^%d+$'));	-- is only uppercase roman numerals or only digits?
	local is_long_vol = volume and (4 < mw.ustring.len(volume));				-- is |volume= value longer than 4 characters?
	
	if volume and (not is_numeric_vol and is_long_vol) then						-- when not all digits or Roman numerals, is |volume= longer than 4 characters?
		utilities.add_prop_cat ('long-vol');									-- yes, add properties cat
	end

	if is_journal then															-- journal-style formatting
		local vol = '';

		if utilities.is_set (volume) then
			if is_numeric_vol then												-- |volume= value all digits or all uppercase Roman numerals?
				vol = utilities.substitute (cfg.presentation['vol-bold'], {sepc, volume});	-- render in bold face
			elseif is_long_vol then												-- not all digits or Roman numerals; longer than 4 characters?
				vol = utilities.substitute (cfg.messages['j-vol'], {sepc, utilities.hyphen_to_dash (volume)});	-- not bold
			else																-- four or fewer characters
				vol = utilities.substitute (cfg.presentation['vol-bold'], {sepc, utilities.hyphen_to_dash (volume)});	-- bold
			end
		end
		vol = vol .. (utilities.is_set (issue) and utilities.substitute (cfg.messages['j-issue'], issue) or '')
		vol = vol .. (utilities.is_set (article) and utilities.substitute (cfg.messages['j-article-num'], article) or '')
		return vol;
	end
	
	if 'podcast' == cite_class and utilities.is_set (issue) then
		return wrap_msg ('issue', {sepc, issue}, lower);
	end
	
	if 'conference' == cite_class and utilities.is_set (article) then			-- |article-number= supported only in journal and conference cites
		if utilities.is_set (volume) and utilities.is_set (article) then		-- both volume and article number
			return wrap_msg ('vol-art', {sepc, utilities.hyphen_to_dash (volume), article}, lower);
		elseif utilities.is_set (article) then									-- article number alone; when volume alone, handled below
			return wrap_msg ('art', {sepc, article}, lower);
		end
	end

	-- all other types of citation
	if utilities.is_set (volume) and utilities.is_set (issue) then
		return wrap_msg ('vol-no', {sepc, utilities.hyphen_to_dash (volume), issue}, lower);
	elseif utilities.is_set (volume) then
		return wrap_msg ('vol', {sepc, utilities.hyphen_to_dash (volume)}, lower);
	else
		return wrap_msg ('issue', {sepc, issue}, lower);
	end
end


--[[-------------------------< F O R M A T _ P A G E S _ S H E E T S >-----------------------------------------

adds static text to one of |page(s)= or |sheet(s)= values and returns it with all of the others set to empty strings.
The return order is:
	page, pages, sheet, sheets

Singular has priority over plural when both are provided.

]]

local function format_pages_sheets (page, pages, sheet, sheets, cite_class, origin, sepc, nopp, lower)
	if 'map' == cite_class then													-- only cite map supports sheet(s) as in-source locators
		if utilities.is_set (sheet) then
			if 'journal' == origin then
				return '', '', wrap_msg ('j-sheet', sheet, lower), '';
			else
				return '', '', wrap_msg ('sheet', {sepc, sheet}, lower), '';
			end
		elseif utilities.is_set (sheets) then
			if 'journal' == origin then
				return '', '', '', wrap_msg ('j-sheets', sheets, lower);
			else
				return '', '', '', wrap_msg ('sheets', {sepc, sheets}, lower);
			end
		end
	end

	local is_journal = 'journal' == cite_class or (utilities.in_array (cite_class, {'citation', 'map', 'interview'}) and 'journal' == origin);
	
	if utilities.is_set (page) then
		if is_journal then
			return utilities.substitute (cfg.messages['j-page(s)'], page), '', '', '';
		elseif not nopp then
			return utilities.substitute (cfg.messages['p-prefix'], {sepc, page}), '', '', '';
		else
			return utilities.substitute (cfg.messages['nopp'], {sepc, page}), '', '', '';
		end
	elseif utilities.is_set (pages) then
		if is_journal then
			return utilities.substitute (cfg.messages['j-page(s)'], pages), '', '', '';
		elseif tonumber(pages) ~= nil and not nopp then							-- if pages is only digits, assume a single page number
			return '', utilities.substitute (cfg.messages['p-prefix'], {sepc, pages}), '', '';
		elseif not nopp then
			return '', utilities.substitute (cfg.messages['pp-prefix'], {sepc, pages}), '', '';
		else
			return '', utilities.substitute (cfg.messages['nopp'], {sepc, pages}), '', '';
		end
	end
	
	return '', '', '', '';														-- return empty strings
end


--[[--------------------------< I N S O U R C E _ L O C _ G E T >----------------------------------------------

returns one of the in-source locators: page, pages, or at. 

If any of these are interwiki links to Wikisource, returns the label portion of the interwiki-link as plain text
for use in COinS.  This COinS thing is done because here we convert an interwiki-link to an external link and
add an icon span around that; get_coins_pages() doesn't know about the span.  TODO: should it?  

TODO: add support for sheet and sheets?; streamline;

TODO: make it so that this function returns only one of the three as the single in-source (the return value assigned
to a new name)?

]]

local function insource_loc_get (page, page_orig, pages, pages_orig, at)
	local ws_url, ws_label, coins_pages, L;										-- for Wikisource interwiki-links; TODO: this corrupts page metadata (span remains in place after cleanup; fix there?)

	if utilities.is_set (page) then
		if utilities.is_set (pages) or utilities.is_set (at) then
			pages = '';															-- unset the others
			at = '';
		end
		extra_text_in_page_check (page, page_orig);								-- emit error message when |page= value begins with what looks like p., pp., etc.

		ws_url, ws_label, L = wikisource_url_make (page);						-- make ws URL from |page= interwiki link; link portion L becomes tooltip label
		if ws_url then
			page = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in page');	-- space char after label to move icon away from in-source text; TODO: a better way to do this?
			page = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, page});
			coins_pages = ws_label;
		end
	elseif utilities.is_set (pages) then
		if utilities.is_set (at) then
			at = '';															-- unset
		end
		extra_text_in_page_check (pages, pages_orig);							-- emit error message when |page= value begins with what looks like p., pp., etc.

		ws_url, ws_label, L = wikisource_url_make (pages);						-- make ws URL from |pages= interwiki link; link portion L becomes tooltip label
		if ws_url then
			pages = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in pages');	-- space char after label to move icon away from in-source text; TODO: a better way to do this?
			pages = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, pages});
			coins_pages = ws_label;
		end
	elseif utilities.is_set (at) then
		ws_url, ws_label, L = wikisource_url_make (at);							-- make ws URL from |at= interwiki link; link portion L becomes tooltip label
		if ws_url then
			at = external_link (ws_url, ws_label .. '&nbsp;', 'ws link in at');	-- space char after label to move icon away from in-source text; TODO: a better way to do this?
			at = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, at});
			coins_pages = ws_label;
		end
	end
	
	return page, pages, at, coins_pages;
end

--[[--------------------------< I S _ U N I Q U E _ A R C H I V E _ U R L >------------------------------------

add error message when |archive-url= value is same as |url= or chapter-url= (or alias...) value

]]

local function is_unique_archive_url (archive, url, c_url, source, date)
	if utilities.is_set (archive) then
		if archive == url or archive == c_url then
			utilities.set_message ('err_bad_url', {utilities.wrap_style ('parameter', source)});	-- add error message
			return '', '';														-- unset |archive-url= and |archive-date= because same as |url= or |chapter-url=
		end
	end

	return archive, date;
end


--[=[-------------------------< A R C H I V E _ U R L _ C H E C K >--------------------------------------------

Check archive.org URLs to make sure they at least look like they are pointing at valid archives and not to the 
save snapshot URL or to calendar pages.  When the archive URL is 'https://web.archive.org/save/' (or http://...)
archive.org saves a snapshot of the target page in the URL.  That is something that Wikipedia should not allow
unwitting readers to do.

When the archive.org URL does not have a complete timestamp, archive.org chooses a snapshot according to its own
algorithm or provides a calendar 'search' result.  [[WP:ELNO]] discourages links to search results.

This function looks at the value assigned to |archive-url= and returns empty strings for |archive-url= and
|archive-date= and an error message when:
	|archive-url= holds an archive.org save command URL
	|archive-url= is an archive.org URL that does not have a complete timestamp (YYYYMMDDhhmmss 14 digits) in the
		correct place
otherwise returns |archive-url= and |archive-date=

There are two mostly compatible archive.org URLs:
	//web.archive.org/<timestamp>...		-- the old form
	//web.archive.org/web/<timestamp>...	-- the new form

The old form does not support or map to the new form when it contains a display flag.  There are four identified flags
('id_', 'js_', 'cs_', 'im_') but since archive.org ignores others following the same form (two letters and an underscore)
we don't check for these specific flags but we do check the form.

This function supports a preview mode.  When the article is rendered in preview mode, this function may return a modified
archive URL:
	for save command errors, return undated wildcard (/*/)
	for timestamp errors when the timestamp has a wildcard, return the URL unmodified
	for timestamp errors when the timestamp does not have a wildcard, return with timestamp limited to six digits plus wildcard (/yyyymm*/)

A secondary function is to return an archive-url timestamp from those urls that have them (archive.org and
archive.today).  The timestamp is used by validation.archive_date_check() to see if the value in |archive-date=
matches the timestamp in the archive url.

]=]

local function archive_url_check (url, date)
	local err_msg = '';															-- start with the error message empty
	local path, timestamp, flag;												-- portions of the archive.org URL
	
	timestamp = url:match ('//archive.today/(%d%d%d%d%d%d%d%d%d%d%d%d%d%d)/');	-- get timestamp from archive.today urls
	if timestamp then															-- if this was an archive.today url ...
		return url, date, timestamp;											-- return ArchiveURL, ArchiveDate, and timestamp from |archive-url=, and done
	end
																				-- here for archive.org urls
	if (not url:match('//web%.archive%.org/')) and (not url:match('//liveweb%.archive%.org/')) then		-- also deprecated liveweb Wayback machine URL
		return url, date;														-- not an archive.org archive, return ArchiveURL and ArchiveDate
	end

	if url:match('//web%.archive%.org/save/') then								-- if a save command URL, we don't want to allow saving of the target page 
		err_msg = cfg.err_msg_supl.save;
		url = url:gsub ('(//web%.archive%.org)/save/', '%1/*/', 1);				-- for preview mode: modify ArchiveURL
	elseif url:match('//liveweb%.archive%.org/') then
		err_msg = cfg.err_msg_supl.liveweb;
	else
		path, timestamp, flag = url:match('//web%.archive%.org/([^%d]*)(%d+)([^/]*)/');	-- split out some of the URL parts for evaluation
		if not path then														-- malformed in some way; pattern did not match
			err_msg = cfg.err_msg_supl.timestamp;
		elseif 14 ~= timestamp:len() then										-- path and flag optional, must have 14-digit timestamp here
			err_msg = cfg.err_msg_supl.timestamp;
			if '*' ~= flag then
				local replacement = timestamp:match ('^%d%d%d%d%d%d') or timestamp:match ('^%d%d%d%d');	-- get the first 6 (YYYYMM) or first 4 digits (YYYY)
				if replacement then												-- nil if there aren't at least 4 digits (year)
					replacement = replacement .. string.rep ('0', 14 - replacement:len());	-- year or yearmo (4 or 6 digits) zero-fill to make 14-digit timestamp
					url=url:gsub ('(//web%.archive%.org/[^%d]*)%d[^/]*', '%1' .. replacement .. '*', 1)	-- for preview, modify ts to 14 digits plus splat for calendar display
				end
			end
		elseif utilities.is_set (path) and 'web/' ~= path then					-- older archive URLs do not have the extra 'web/' path element
			err_msg = cfg.err_msg_supl.path;
		elseif utilities.is_set (flag) and not utilities.is_set (path) then		-- flag not allowed with the old form URL (without the 'web/' path element)
			err_msg = cfg.err_msg_supl.flag;
		elseif utilities.is_set (flag) and not flag:match ('%a%a_') then		-- flag if present must be two alpha characters and underscore (requires 'web/' path element)
			err_msg = cfg.err_msg_supl.flag;
		else
			return url, date, timestamp;										-- return ArchiveURL, ArchiveDate, and timestamp from |archive-url=
		end
	end
																				-- if here, something not right so
	utilities.set_message ('err_archive_url', {err_msg});						-- add error message and

	if is_preview_mode then
		return url, date, timestamp;											-- preview mode so return ArchiveURL, ArchiveDate, and timestamp from |archive-url=
	else
		return '', '';															-- return empty strings for ArchiveURL and ArchiveDate
	end
end


--[[--------------------------< P L A C E _ C H E C K >--------------------------------------------------------

check |place=, |publication-place=, |location= to see if these params include digits.  This function added because
many editors misuse location to specify the in-source location (|page(s)= and |at= are supposed to do that)

returns the original parameter value without modification; added maint cat when parameter value contains digits

]]

local function place_check (param_val)
	if not utilities.is_set (param_val) then									-- parameter empty or omitted
		return param_val;														-- return that empty state
	end
	
	if mw.ustring.find (param_val, '%d') then									-- not empty, are there digits in the parameter value
		utilities.set_message ('maint_location');								-- yep, add maint cat
	end
	
	return param_val;															-- and done
end


--[[--------------------------< I S _ A R C H I V E D _ C O P Y >----------------------------------------------

compares |title= to 'Archived copy' (placeholder added by bots that can't find proper title); if matches, return true; nil else

]]

local function is_archived_copy (title)
	title = mw.ustring.lower(title);											-- switch title to lower case
	if title:find (cfg.special_case_translation.archived_copy.en) then			-- if title is 'Archived copy'
		return true;
	elseif cfg.special_case_translation.archived_copy['local'] then
		if mw.ustring.find (title, cfg.special_case_translation.archived_copy['local']) then	-- mw.ustring() because might not be Latin script
			return true;
		end
	end
end


--[[--------------------------< D I S P L A Y _ N A M E S _ S E L E C T >--------------------------------------

for any of the |display-authors=, |display-editors=, etc parameters, select either the local or global setting.
When both are present, look at <local_display_names> value.  When the value is some sort of 'et al.'string,
special handling is required.

When {{cs1 config}} has |display-<namelist>= AND this template has |display-<namelist>=etal AND:
	the number of names specified by <number_of_names> is:
		greater than the number specified in the global |display-<namelist>= parameter (<global_display_names>)
			use global |display-<namelist>= parameter value
			set overridden maint category
		less than or equal to the number specified in the global |display-<namelist>=  parameter
			use local |display-<namelist>= parameter value

The purpose of this function is to prevent categorizing a template that has fewer names than the global setting
to keep the etal annotation specified by <local_display_names>.

]]

local function display_names_select (global_display_names, local_display_names, param_name, number_of_names, test)
	if global_display_names and utilities.is_set (local_display_names) then		-- when both
		if 'etal' == local_display_names:lower():gsub("[ '%.]", '') then		-- the :gsub() portion makes 'etal' from a variety of 'et al.' spellings and stylings
			number_of_names = tonumber (number_of_names);						-- convert these to numbers for comparison
			local global_display_names_num = tonumber (global_display_names);	-- <global_display_names> not set when parameter value is not digits

			if number_of_names > global_display_names_num then					-- template has more names than global config allows to be displayed?
				utilities.set_message ('maint_overridden_setting');				-- set a maint message because global is overriding local |display-<namelist>=etal
				return global_display_names, 'cs1 config';						-- return global with spoof parameter name (for get_display_names())
			else
				return local_display_names, param_name;							-- return local because fewer names so let <local_display_names> control
			end
		end
																				-- here when <global_display_names> and <local_display_names> both numbers; <global_display_names> controls
		utilities.set_message ('maint_overridden_setting');						-- set a maint message
		return global_display_names, 'cs1 config';								-- return global with spoof parameter name (for get_display_names())
	end
																				-- here when only one of <global_display_names> or <local_display_names> set
	if global_display_names then
		return global_display_names, 'cs1 config';								-- return global with spoof parameter name (for get_display_names())
	else
		return local_display_names, param_name;									-- return local
	end
end


--[[--------------------------< C I T A T I O N 0 >------------------------------------------------------------

This is the main function doing the majority of the citation formatting.

]]

local function citation0( config, args )
	--[[ 
	Load Input Parameters
	The argument_wrapper facilitates the mapping of multiple aliases to single internal variable.
	]]
	local A = argument_wrapper ( args );
	local i 

	-- Pick out the relevant fields from the arguments.  Different citation templates
	-- define different field names for the same underlying things.	

	local author_etal;
	local a	= {};																-- authors list from |lastn= / |firstn= pairs or |vauthors=
	local Authors;
	local NameListStyle;
		if cfg.global_cs1_config_t['NameListStyle'] then						-- global setting in {{cs1 config}} overrides local |name-list-style= parameter value; nil when empty or assigned value invalid
			NameListStyle = is_valid_parameter_value (cfg.global_cs1_config_t['NameListStyle'], 'cs1 config: name-list-style', cfg.keywords_lists['name-list-style'], '');	-- error messaging 'param' here is a hoax
		else
			NameListStyle = is_valid_parameter_value (A['NameListStyle'], A:ORIGIN('NameListStyle'), cfg.keywords_lists['name-list-style'], '');
		end

		if cfg.global_cs1_config_t['NameListStyle'] and utilities.is_set (A['NameListStyle']) then	-- when template has |name-list-style=<something> which global setting has overridden
			utilities.set_message ('maint_overridden_setting');					-- set a maint message
		end

	local Collaboration = A['Collaboration'];

	do																			-- to limit scope of selected
		local selected = select_author_editor_source (A['Vauthors'], A['Authors'], args, 'AuthorList');
		if 1 == selected then
			a, author_etal = extract_names (args, 'AuthorList');				-- fetch author list from |authorn= / |lastn= / |firstn=, |author-linkn=, and |author-maskn=
		elseif 2 == selected then
			NameListStyle = 'vanc';												-- override whatever |name-list-style= might be
			a, author_etal = parse_vauthors_veditors (args, A['Vauthors'], 'AuthorList');	-- fetch author list from |vauthors=, |author-linkn=, and |author-maskn=
		elseif 3 == selected then
			Authors = A['Authors'];												-- use content of |people= or |credits=; |authors= is deprecated; TODO: constrain |people= and |credits= to cite av media, episode, serial?
		end
		if utilities.is_set (Collaboration) then
			author_etal = true;													-- so that |display-authors=etal not required
		end
	end

	local editor_etal;
	local e	= {};																-- editors list from |editor-lastn= / |editor-firstn= pairs or |veditors=

	do																			-- to limit scope of selected
		local selected = select_author_editor_source (A['Veditors'], nil, args, 'EditorList');	-- support for |editors= withdrawn
		if 1 == selected then
			e, editor_etal = extract_names (args, 'EditorList');				-- fetch editor list from |editorn= / |editor-lastn= / |editor-firstn=, |editor-linkn=, and |editor-maskn=
		elseif 2 == selected then
			NameListStyle = 'vanc';												-- override whatever |name-list-style= might be
			e, editor_etal = parse_vauthors_veditors (args, args.veditors, 'EditorList');	-- fetch editor list from |veditors=, |editor-linkn=, and |editor-maskn=
		end
	end
							
	local Chapter = A['Chapter'];												-- done here so that we have access to |contribution= from |chapter= aliases
	local Chapter_origin = A:ORIGIN ('Chapter');
	local Contribution;															-- because contribution is required for contributor(s)
		if 'contribution' == Chapter_origin then
			Contribution = Chapter;												-- get the name of the contribution
		end
	local c = {};																-- contributors list from |contributor-lastn= / contributor-firstn= pairs
	
	if utilities.in_array (config.CitationClass, {"book", "citation"}) and not utilities.is_set (A['Periodical']) then	-- |contributor= and |contribution= only supported in book cites
		c = extract_names (args, 'ContributorList');							-- fetch contributor list from |contributorn= / |contributor-lastn=, -firstn=, -linkn=, -maskn=
		
		if 0 < #c then
			if not utilities.is_set (Contribution) then							-- |contributor= requires |contribution=
				utilities.set_message ('err_contributor_missing_required_param', 'contribution');	-- add missing contribution error message
				c = {};															-- blank the contributors' table; it is used as a flag later
			end
			if 0 == #a then														-- |contributor= requires |author=
				utilities.set_message ('err_contributor_missing_required_param', 'author');	-- add missing author error message
				c = {};															-- blank the contributors' table; it is used as a flag later
			end
		end
	else																		-- if not a book cite
		if utilities.select_one (args, cfg.aliases['ContributorList-Last'], 'err_redundant_parameters', 1 ) then	-- are there contributor name list parameters?
			utilities.set_message ('err_contributor_ignored');					-- add contributor ignored error message
		end
		Contribution = nil;														-- unset
	end

	local Title = A['Title'];
	local TitleLink = A['TitleLink'];

	local auto_select = '';														-- default is auto
	local accept_link;
	TitleLink, accept_link = utilities.has_accept_as_written (TitleLink, true);	-- test for accept-this-as-written markup
	if (not accept_link) and utilities.in_array (TitleLink, {'none', 'pmc', 'doi'}) then -- check for special keywords
		auto_select = TitleLink;												-- remember selection for later
		TitleLink = '';															-- treat as if |title-link= would have been empty
	end

	TitleLink = link_title_ok (TitleLink, A:ORIGIN ('TitleLink'), Title, 'title');	-- check for wiki-markup in |title-link= or wiki-markup in |title= when |title-link= is set

	local Section = '';															-- {{cite map}} only; preset to empty string for concatenation if not used
	if 'map' == config.CitationClass and 'section' == Chapter_origin then
		Section = A['Chapter'];													-- get |section= from |chapter= alias list; |chapter= and the other aliases not supported in {{cite map}}
		Chapter = '';															-- unset for now; will be reset later from |map= if present
	end

	local Periodical = A['Periodical'];
	local Periodical_origin = A:ORIGIN('Periodical');
	local ScriptPeriodical = A['ScriptPeriodical'];
	local ScriptPeriodical_origin = A:ORIGIN('ScriptPeriodical');
	local TransPeriodical =  A['TransPeriodical'];
	local TransPeriodical_origin =  A:ORIGIN ('TransPeriodical');
	
	if (utilities.in_array (config.CitationClass, {'book', 'encyclopaedia'}) and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical) or utilities.is_set (TransPeriodical))) then
		local param;
		if utilities.is_set (Periodical) then									-- get a parameter name from one of these periodical related meta-parameters
			Periodical = '';													-- unset because not valid {{cite book}} or {{cite encyclopedia}} parameters 
			param = Periodical_origin											-- get parameter name for error messaging
		elseif utilities.is_set (TransPeriodical) then
			TransPeriodical = '';												-- unset because not valid {{cite book}} or {{cite encyclopedia}} parameters 
			param = TransPeriodical_origin;										-- get parameter name for error messaging
		elseif utilities.is_set (ScriptPeriodical) then
			ScriptPeriodical = '';												-- unset because not valid {{cite book}} or {{cite encyclopedia}} parameters 
			param = ScriptPeriodical_origin;									-- get parameter name for error messaging
		end

		if utilities.is_set (param) then										-- if we found one
			utilities.set_message ('err_periodical_ignored', {param});			-- emit an error message
		end
	end

	if utilities.is_set (Periodical) then
		local i;
		Periodical, i = utilities.strip_apostrophe_markup (Periodical);			-- strip apostrophe markup so that metadata isn't contaminated 
		if i then																-- non-zero when markup was stripped so emit an error message
			utilities.set_message ('err_apostrophe_markup', {Periodical_origin});
		end
	end

	if 'mailinglist' == config.CitationClass then								-- special case for {{cite mailing list}}
		if utilities.is_set (Periodical) and utilities.is_set (A ['MailingList']) then	-- both set emit an error TODO: make a function for this and similar?
			utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', Periodical_origin) .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'mailinglist')});
		end

		Periodical = A ['MailingList'];											-- error or no, set Periodical to |mailinglist= value because this template is {{cite mailing list}}
		Periodical_origin = A:ORIGIN('MailingList');
	end

	-- web and news not tested for now because of 
	-- Wikipedia:Administrators%27_noticeboard#Is_there_a_semi-automated_tool_that_could_fix_these_annoying_"Cite_Web"_errors?
	if not (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) then	-- 'periodical' templates require periodical parameter
	--	local p = {['journal'] = 'journal', ['magazine'] = 'magazine', ['news'] = 'newspaper', ['web'] = 'website'};	-- for error message
		local p = {['journal'] = 'journal', ['magazine'] = 'magazine'};			-- for error message
		if p[config.CitationClass]  then
			utilities.set_message ('err_missing_periodical', {config.CitationClass, p[config.CitationClass]});
		end
	end
	
	local Volume;
	if 'citation' == config.CitationClass then
		if utilities.is_set (Periodical) then
			if not utilities.in_array (Periodical_origin, cfg.citation_no_volume_t) then	-- {{citation}} does not render |volume= when these parameters are used
				Volume = A['Volume'];											-- but does for all other 'periodicals'
			end
		elseif utilities.is_set (ScriptPeriodical) then
			if 'script-website' ~= ScriptPeriodical_origin then					-- {{citation}} does not render volume for |script-website=
				Volume = A['Volume'];											-- but does for all other 'periodicals'
			end
		else
			Volume = A['Volume'];												-- and does for non-'periodical' cites
		end
	elseif utilities.in_array (config.CitationClass, cfg.templates_using_volume) then	-- render |volume= for cs1 according to the configuration settings
		Volume = A['Volume'];
	end	
	extra_text_in_vol_iss_check (Volume, A:ORIGIN ('Volume'), 'v');	

	local Issue;
	if 'citation' == config.CitationClass then
		if utilities.is_set (Periodical) and utilities.in_array (Periodical_origin, cfg.citation_issue_t) then	-- {{citation}} may render |issue= when these parameters are used
			Issue = utilities.hyphen_to_dash (A['Issue']);
		end
	elseif utilities.in_array (config.CitationClass, cfg.templates_using_issue) then	-- conference & map books do not support issue; {{citation}} listed here because included in settings table
		if not (utilities.in_array (config.CitationClass, {'conference', 'map', 'citation'}) and not (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical))) then
			Issue = utilities.hyphen_to_dash (A['Issue']);
		end
	end
	
	local ArticleNumber;

	if utilities.in_array (config.CitationClass, {'journal', 'conference'}) or ('citation' == config.CitationClass and utilities.is_set (Periodical) and 'journal' == Periodical_origin) then
		ArticleNumber = A['ArticleNumber'];
	end

	extra_text_in_vol_iss_check (Issue, A:ORIGIN ('Issue'), 'i');	

	local Page;
	local Pages;
	local At;
	local QuotePage;
	local QuotePages;
	if not utilities.in_array (config.CitationClass, cfg.templates_not_using_page) then		-- TODO: rewrite to emit ignored parameter error message?
		Page = A['Page'];
		Pages = utilities.hyphen_to_dash (A['Pages']);	
		At = A['At'];
		QuotePage = A['QuotePage'];
		QuotePages = utilities.hyphen_to_dash (A['QuotePages']);
	end

	local Edition = A['Edition'];
	local PublicationPlace = place_check (A['PublicationPlace'], A:ORIGIN('PublicationPlace'));
	local Place = place_check (A['Place'], A:ORIGIN('Place'));
	
	local PublisherName = A['PublisherName'];
	local PublisherName_origin = A:ORIGIN('PublisherName');
	if utilities.is_set (PublisherName) and (cfg.keywords_xlate['none'] ~= PublisherName) then
		local i = 0;
		PublisherName, i = utilities.strip_apostrophe_markup (PublisherName);	-- strip apostrophe markup so that metadata isn't contaminated; publisher is never italicized
		if i and (0 < i) then													-- non-zero when markup was stripped so emit an error message
			utilities.set_message ('err_apostrophe_markup', {PublisherName_origin});
		end
	end
	
	if ('document' == config.CitationClass) and not utilities.is_set (PublisherName) then
		utilities.set_message ('err_missing_publisher', {config.CitationClass, 'publisher'});
	end

	local Newsgroup = A['Newsgroup'];											-- TODO: strip apostrophe markup?
	local Newsgroup_origin = A:ORIGIN('Newsgroup');	

	if 'newsgroup' == config.CitationClass then
		if utilities.is_set (PublisherName) and (cfg.keywords_xlate['none'] ~= PublisherName) then	-- general use parameter |publisher= not allowed in cite newsgroup
			utilities.set_message ('err_parameter_ignored', {PublisherName_origin});
		end

		PublisherName = nil;													-- ensure that this parameter is unset for the time being; will be used again after COinS
	end

	local URL = A['URL'];														-- TODO: better way to do this for URL, ChapterURL, and MapURL?
	local UrlAccess = is_valid_parameter_value (A['UrlAccess'], A:ORIGIN('UrlAccess'), cfg.keywords_lists['url-access'], nil);
	
	if not utilities.is_set (URL) and utilities.is_set (UrlAccess) then
		UrlAccess = nil;
		utilities.set_message ('err_param_access_requires_param', 'url');
	end
	
	local ChapterURL = A['ChapterURL'];
	local ChapterUrlAccess = is_valid_parameter_value (A['ChapterUrlAccess'], A:ORIGIN('ChapterUrlAccess'), cfg.keywords_lists['url-access'], nil);
	if not utilities.is_set (ChapterURL) and utilities.is_set (ChapterUrlAccess) then
		ChapterUrlAccess = nil;
		utilities.set_message ('err_param_access_requires_param', {A:ORIGIN('ChapterUrlAccess'):gsub ('%-access', '')});
	end

	local MapUrlAccess = is_valid_parameter_value (A['MapUrlAccess'], A:ORIGIN('MapUrlAccess'), cfg.keywords_lists['url-access'], nil);
	if not utilities.is_set (A['MapURL']) and utilities.is_set (MapUrlAccess) then
		MapUrlAccess = nil;
		utilities.set_message ('err_param_access_requires_param', {'map-url'});
	end

	local this_page = mw.title.getCurrentTitle();								-- also used for COinS and for language
	local no_tracking_cats = is_valid_parameter_value (A['NoTracking'], A:ORIGIN('NoTracking'), cfg.keywords_lists['yes_true_y'], nil);

	-- check this page to see if it is in one of the namespaces that cs1 is not supposed to add to the error categories
	if not utilities.is_set (no_tracking_cats) then								-- ignore if we are already not going to categorize this page
		if cfg.uncategorized_namespaces[this_page.namespace] then				-- is this page's namespace id one of the uncategorized namespace ids?
			no_tracking_cats = "true";											-- set no_tracking_cats
		end
		for _, v in ipairs (cfg.uncategorized_subpages) do						-- cycle through page name patterns
			if this_page.text:match (v) then									-- test page name against each pattern
				no_tracking_cats = "true";										-- set no_tracking_cats
				break;															-- bail out if one is found
			end
		end
	end
																				-- check for extra |page=, |pages= or |at= parameters. (also sheet and sheets while we're at it)
	utilities.select_one (args, {'page', 'p', 'pp', 'pages', 'at', 'sheet', 'sheets'}, 'err_redundant_parameters');	-- this is a dummy call simply to get the error message and category

	local coins_pages;
	
	Page, Pages, At, coins_pages = insource_loc_get (Page, A:ORIGIN('Page'), Pages, A:ORIGIN('Pages'), At);

	local NoPP = is_valid_parameter_value (A['NoPP'], A:ORIGIN('NoPP'), cfg.keywords_lists['yes_true_y'], nil);

	if utilities.is_set (PublicationPlace) and utilities.is_set (Place) then	-- both |publication-place= and |place= (|location=) allowed if different
		utilities.add_prop_cat ('location-test');								-- add property cat to evaluate how often PublicationPlace and Place are used together
		if PublicationPlace == Place then
			Place = '';															-- unset; don't need both if they are the same
		end
	elseif not utilities.is_set (PublicationPlace) and utilities.is_set (Place) then	-- when only |place= (|location=) is set ...
		PublicationPlace = Place;												-- promote |place= (|location=) to |publication-place
	end

	if PublicationPlace == Place then Place = ''; end							-- don't need both if they are the same

	local URL_origin = A:ORIGIN('URL');											-- get name of parameter that holds URL
	local ChapterURL_origin = A:ORIGIN('ChapterURL');							-- get name of parameter that holds ChapterURL
	local ScriptChapter = A['ScriptChapter'];
	local ScriptChapter_origin = A:ORIGIN ('ScriptChapter');
	local Format = A['Format'];
	local ChapterFormat = A['ChapterFormat'];
	local TransChapter = A['TransChapter'];
	local TransChapter_origin = A:ORIGIN ('TransChapter');
	local TransTitle = A['TransTitle'];
	local ScriptTitle = A['ScriptTitle'];
	
	--[[
	Parameter remapping for cite encyclopedia:
	When the citation has these parameters:
		|encyclopedia= and |title= then map |title= to |article= and |encyclopedia= to |title= for rendering
		|encyclopedia= and |article= then map |encyclopedia= to |title= for rendering

		|trans-title= maps to |trans-chapter= when |title= is re-mapped
		|url= maps to |chapter-url= when |title= is remapped
	
	All other combinations of |encyclopedia=, |title=, and |article= are not modified
	
	]]

	local Encyclopedia = A['Encyclopedia'];										-- used as a flag by this module and by ~/COinS

	if utilities.is_set (Encyclopedia) then										-- emit error message when Encyclopedia set but template is other than {{cite encyclopedia}} or {{citation}}
		if 'encyclopaedia' ~= config.CitationClass and 'citation' ~= config.CitationClass then
			utilities.set_message ('err_parameter_ignored', {A:ORIGIN ('Encyclopedia')});
			Encyclopedia = nil;													-- unset because not supported by this template
		end
	end

	if ('encyclopaedia' == config.CitationClass) or ('citation' == config.CitationClass and utilities.is_set (Encyclopedia)) then
		if utilities.is_set (Periodical) and utilities.is_set (Encyclopedia) then	-- when both parameters set emit an error message; {{citation}} only; Periodical not allowed in {{cite encyclopedia}}
			utilities.set_message ('err_periodical_ignored', {Periodical_origin});
		end

		if utilities.is_set (Encyclopedia) then
			Periodical = Encyclopedia;											-- error or no, set Periodical to Encyclopedia for rendering; {{citation}} could (not legitimately) have both; use Encyclopedia
			Periodical_origin = A:ORIGIN ('Encyclopedia');

			if utilities.is_set (Title) or utilities.is_set (ScriptTitle) then
				if not utilities.is_set (Chapter) then
					Chapter = Title;											-- |encyclopedia= and |title= are set so map |title= to |article= and |encyclopedia= to |title= for rendering
					ScriptChapter = ScriptTitle;
					ScriptChapter_origin = A:ORIGIN('ScriptTitle')
					TransChapter = TransTitle;
					ChapterURL = URL;
					ChapterURL_origin = URL_origin;

					ChapterUrlAccess = UrlAccess;

					if not utilities.is_set (ChapterURL) and utilities.is_set (TitleLink) then
						Chapter = utilities.make_wikilink (TitleLink, Chapter);
					end
					Title = Periodical;
					ChapterFormat = Format;
					Periodical = '';											-- redundant so unset
					TransTitle = '';
					URL = '';
					Format = '';
					TitleLink = '';
					ScriptTitle = '';
				end
			elseif utilities.is_set (Chapter) or utilities.is_set (ScriptChapter) then								-- |title= not set
				Title = Periodical;												-- |encyclopedia= set and |article= set so map |encyclopedia= to |title= for rendering
				Periodical = '';												-- redundant so unset
			end
		end
	end

	-- special case for cite techreport.
	local ID = A['ID'];
	if (config.CitationClass == "techreport") then								-- special case for cite techreport
		if utilities.is_set (A['Number']) then									-- cite techreport uses 'number', which other citations alias to 'issue'
			if not utilities.is_set (ID) then									-- can we use ID for the "number"?
				ID = A['Number'];												-- yes, use it
			else																-- ID has a value so emit error message
				utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'id') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'number')});
			end
		end	
	end

	-- Account for the oddity that is {{cite conference}}, before generation of COinS data.
	local ChapterLink -- = A['ChapterLink'];									-- deprecated as a parameter but still used internally by cite episode
	local Conference = A['Conference'];
	local BookTitle = A['BookTitle'];
	local TransTitle_origin = A:ORIGIN ('TransTitle');
	if 'conference' == config.CitationClass then
		if utilities.is_set (BookTitle) then
			Chapter = Title;
			Chapter_origin = 'title';
	--		ChapterLink = TitleLink;											-- |chapter-link= is deprecated
			ChapterURL = URL;
			ChapterUrlAccess = UrlAccess;
			ChapterURL_origin = URL_origin;
			URL_origin = '';
			ChapterFormat = Format;
			TransChapter = TransTitle;
			TransChapter_origin = TransTitle_origin;
			Title = BookTitle;
			Format = '';
	--		TitleLink = '';
			TransTitle = '';
			URL = '';
		end
	elseif 'speech' ~= config.CitationClass then
		Conference = '';														-- not cite conference or cite speech so make sure this is empty string
	end
	
	-- CS1/2 mode
	local Mode;
		if cfg.global_cs1_config_t['Mode'] then									-- global setting in {{cs1 config}} overrides local |mode= parameter value; nil when empty or assigned value invalid
			Mode = is_valid_parameter_value (cfg.global_cs1_config_t['Mode'], 'cs1 config: mode', cfg.keywords_lists['mode'], '');	-- error messaging 'param' here is a hoax
		else
			Mode = is_valid_parameter_value (A['Mode'], A:ORIGIN('Mode'), cfg.keywords_lists['mode'], '');
		end

		if cfg.global_cs1_config_t['Mode'] and utilities.is_set (A['Mode']) then	-- when template has |mode=<something> which global setting has overridden
			utilities.set_message ('maint_overridden_setting');					-- set a maint message
		end

	-- separator character and postscript
	local sepc, PostScript = set_style (Mode:lower(), A['PostScript'], config.CitationClass);
	-- controls capitalization of certain static text
	local use_lowercase = ( sepc == ',' );
	
	-- cite map oddities
	local Cartography = "";
	local Scale = "";
	local Sheet = A['Sheet'] or '';
	local Sheets = A['Sheets'] or '';
	if config.CitationClass == "map" then
		if utilities.is_set (Chapter) then										--TODO: make a function for this and similar?
			utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'map') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', Chapter_origin)});	-- add error message
		end
		Chapter = A['Map'];
		Chapter_origin = A:ORIGIN('Map');
		ChapterURL = A['MapURL'];
		ChapterURL_origin = A:ORIGIN('MapURL');
		TransChapter = A['TransMap'];
		ScriptChapter = A['ScriptMap']
		ScriptChapter_origin = A:ORIGIN('ScriptMap')

		ChapterUrlAccess = MapUrlAccess;
		ChapterFormat = A['MapFormat'];

		Cartography = A['Cartography'];
		if utilities.is_set ( Cartography ) then
			Cartography = sepc .. " " .. wrap_msg ('cartography', Cartography, use_lowercase);
		end		
		Scale = A['Scale'];
		if utilities.is_set ( Scale ) then
			Scale = sepc .. " " .. Scale;
		end
	end

	-- Account for the oddities that are {{cite episode}} and {{cite serial}}, before generation of COinS data.
	local Series = A['Series'];
	if 'episode' == config.CitationClass or 'serial' == config.CitationClass then
		local SeriesLink = A['SeriesLink'];

		SeriesLink = link_title_ok (SeriesLink, A:ORIGIN ('SeriesLink'), Series, 'series');	-- check for wiki-markup in |series-link= or wiki-markup in |series= when |series-link= is set

		local Network = A['Network'];
		local Station = A['Station'];
		local s, n = {}, {};
																				-- do common parameters first
		if utilities.is_set (Network) then table.insert(n, Network); end
		if utilities.is_set (Station) then table.insert(n, Station); end
		ID = table.concat(n, sepc .. ' ');
		
		if 'episode' == config.CitationClass then								-- handle the oddities that are strictly {{cite episode}}
			local Season = A['Season'];
			local SeriesNumber = A['SeriesNumber'];

			if utilities.is_set (Season) and utilities.is_set (SeriesNumber) then	-- these are mutually exclusive so if both are set TODO: make a function for this and similar?
				utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'season') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'seriesno')});		-- add error message
				SeriesNumber = '';												-- unset; prefer |season= over |seriesno=
			end
																				-- assemble a table of parts concatenated later into Series
			if utilities.is_set (Season) then table.insert(s, wrap_msg ('season', Season, use_lowercase)); end
			if utilities.is_set (SeriesNumber) then table.insert(s, wrap_msg ('seriesnum', SeriesNumber, use_lowercase)); end
			if utilities.is_set (Issue) then table.insert(s, wrap_msg ('episode', Issue, use_lowercase)); end
			Issue = '';															-- unset because this is not a unique parameter
	
			Chapter = Title;													-- promote title parameters to chapter
			ScriptChapter = ScriptTitle;
			ScriptChapter_origin = A:ORIGIN('ScriptTitle');
			ChapterLink = TitleLink;											-- alias |episode-link=
			TransChapter = TransTitle;
			ChapterURL = URL;
			ChapterUrlAccess = UrlAccess;
			ChapterURL_origin = URL_origin;
			ChapterFormat = Format;

			Title = Series;														-- promote series to title
			TitleLink = SeriesLink;
			Series = table.concat(s, sepc .. ' ');								-- this is concatenation of season, seriesno, episode number

			if utilities.is_set (ChapterLink) and not utilities.is_set (ChapterURL) then	-- link but not URL
				Chapter = utilities.make_wikilink (ChapterLink, Chapter);
			elseif utilities.is_set (ChapterLink) and utilities.is_set (ChapterURL) then	-- if both are set, URL links episode;
				Series = utilities.make_wikilink (ChapterLink, Series);
			end
			URL = '';															-- unset
			TransTitle = '';
			ScriptTitle = '';
			Format = '';
			
		else																	-- now oddities that are cite serial
			Issue = '';															-- unset because this parameter no longer supported by the citation/core version of cite serial
			Chapter = A['Episode'];												-- TODO: make |episode= available to cite episode someday?
			if utilities.is_set (Series) and utilities.is_set (SeriesLink) then
				Series = utilities.make_wikilink (SeriesLink, Series);
			end
			Series = utilities.wrap_style ('italic-title', Series);				-- series is italicized
		end	
	end
	-- end of {{cite episode}} stuff

	-- handle type parameter for those CS1 citations that have default values
	local TitleType = A['TitleType'];
	local Degree = A['Degree'];
	if utilities.in_array (config.CitationClass, {'AV-media-notes', 'document', 'interview', 'mailinglist', 'map', 'podcast', 'pressrelease', 'report', 'speech', 'techreport', 'thesis'}) then
		TitleType = set_titletype (config.CitationClass, TitleType);
		if utilities.is_set (Degree) and "Thesis" == TitleType then				-- special case for cite thesis
			TitleType = Degree .. ' ' .. cfg.title_types ['thesis']:lower();
		end
	end

	if utilities.is_set (TitleType) then										-- if type parameter is specified
		TitleType = utilities.substitute ( cfg.messages['type'], TitleType);	-- display it in parentheses
	-- TODO: Hack on TitleType to fix bunched parentheses problem
	end

	-- legacy: promote PublicationDate to Date if neither Date nor Year are set.
	local Date = A['Date'];
 	local Date_origin;															-- to hold the name of parameter promoted to Date; required for date error messaging
	local PublicationDate = A['PublicationDate'];
	local Year = A['Year'];

	if not utilities.is_set (Date) then
		Date = Year;															-- promote Year to Date
		Year = nil;																-- make nil so Year as empty string isn't used for CITEREF
		if not utilities.is_set (Date) and utilities.is_set (PublicationDate) then	-- use PublicationDate when |date= and |year= are not set
			Date = PublicationDate;												-- promote PublicationDate to Date
			PublicationDate = '';												-- unset, no longer needed
			Date_origin = A:ORIGIN('PublicationDate');							-- save the name of the promoted parameter
		else
			Date_origin = A:ORIGIN('Year');										-- save the name of the promoted parameter
		end
	else
		Date_origin = A:ORIGIN('Date');											-- not a promotion; name required for error messaging
	end

	if PublicationDate == Date then PublicationDate = ''; end					-- if PublicationDate is same as Date, don't display in rendered citation

	--[[
	Go test all of the date-holding parameters for valid MOS:DATE format and make sure that dates are real dates. This must be done before we do COinS because here is where
	we get the date used in the metadata.
	
	Date validation supporting code is in Module:Citation/CS1/Date_validation
	]]

	local DF = is_valid_parameter_value (A['DF'], A:ORIGIN('DF'), cfg.keywords_lists['df'], '');
	if not utilities.is_set (DF) then
		DF = cfg.global_df;														-- local |df= if present overrides global df set by {{use xxx date}} template
	end

	local ArchiveURL;
	local ArchiveDate;
	local ArchiveFormat = A['ArchiveFormat'];
	local archive_url_timestamp;												-- timestamp from wayback machine url
	
	ArchiveURL, ArchiveDate, archive_url_timestamp = archive_url_check (A['ArchiveURL'], A['ArchiveDate'])
	ArchiveFormat = style_format (ArchiveFormat, ArchiveURL, 'archive-format', 'archive-url');
	
	ArchiveURL, ArchiveDate = is_unique_archive_url (ArchiveURL, URL, ChapterURL, A:ORIGIN('ArchiveURL'), ArchiveDate);		-- add error message when URL or ChapterURL == ArchiveURL

	local AccessDate = A['AccessDate'];
	local COinS_date = {};														-- holds date info extracted from |date= for the COinS metadata by Module:Date verification
	local DoiBroken = A['DoiBroken'];
	local Embargo = A['Embargo'];
	local anchor_year;															-- used in the CITEREF identifier
	do	-- create defined block to contain local variables error_message, date_parameters_list, mismatch
		local error_message = '';
																				-- AirDate has been promoted to Date so not necessary to check it
		local date_parameters_list = {
			['access-date'] = {val = AccessDate, name = A:ORIGIN ('AccessDate')},
			['archive-date'] = {val = ArchiveDate, name = A:ORIGIN ('ArchiveDate')},
			['date'] = {val = Date, name = Date_origin},
			['doi-broken-date'] = {val = DoiBroken, name = A:ORIGIN ('DoiBroken')},
			['pmc-embargo-date'] = {val = Embargo, name = A:ORIGIN ('Embargo')},
			['publication-date'] = {val = PublicationDate, name = A:ORIGIN ('PublicationDate')},
			['year'] = {val = Year, name = A:ORIGIN ('Year')},
			};

		local error_list = {};
		anchor_year, Embargo = validation.dates(date_parameters_list, COinS_date, error_list);

-- start temporary Julian / Gregorian calendar uncertainty categorization
		if COinS_date.inter_cal_cat then
			utilities.add_prop_cat ('jul-greg-uncertainty');
		end
-- end temporary Julian / Gregorian calendar uncertainty categorization

		if utilities.is_set (Year) and utilities.is_set (Date) then				-- both |date= and |year= not normally needed; 
			validation.year_date_check (Year, A:ORIGIN ('Year'), Date, A:ORIGIN ('Date'), error_list);
		end
		
		if 0 == #error_list then												-- error free dates only; 0 when error_list is empty
			local modified = false;												-- flag
			
			if utilities.is_set (DF) then										-- if we need to reformat dates
				modified = validation.reformat_dates (date_parameters_list, DF);	-- reformat to DF format, use long month names if appropriate
			end

			if true == validation.date_hyphen_to_dash (date_parameters_list) then	-- convert hyphens to dashes where appropriate
				modified = true;
				utilities.set_message ('maint_date_format');					-- hyphens were converted so add maint category
			end
			
	-- for those wikis that can and want to have English date names translated to the local language; not supported at en.wiki
			if cfg.date_name_auto_xlate_enable and validation.date_name_xlate (date_parameters_list, cfg.date_digit_auto_xlate_enable ) then
				utilities.set_message ('maint_date_auto_xlated');				-- add maint cat
				modified = true;
			end

			if modified then													-- if the date_parameters_list values were modified
				AccessDate = date_parameters_list['access-date'].val;			-- overwrite date holding parameters with modified values
				ArchiveDate = date_parameters_list['archive-date'].val;
				Date = date_parameters_list['date'].val;
				DoiBroken = date_parameters_list['doi-broken-date'].val;
				PublicationDate = date_parameters_list['publication-date'].val;
			end

			if archive_url_timestamp and utilities.is_set (ArchiveDate) then
				validation.archive_date_check (ArchiveDate, archive_url_timestamp, DF);	-- does YYYYMMDD in archive_url_timestamp match date in ArchiveDate
			end
		else
			utilities.set_message ('err_bad_date', {utilities.make_sep_list (#error_list, error_list)});	-- add this error message
		end
	end	-- end of do

	if utilities.in_array (config.CitationClass, {'book', 'encyclopaedia'}) or	-- {{cite book}}, {{cite encyclopedia}}; TODO: {{cite conference}} and others?
		('citation' == config.CitationClass and utilities.is_set (Encyclopedia)) or			-- {{citation}} as an encylopedia citation
		('citation' == config.CitationClass and not utilities.is_set (Periodical)) then		-- {{citation}} as a book citation
			if utilities.is_set (PublicationPlace) then
				if not utilities.is_set (PublisherName) then
					local date = COinS_date.rftdate and tonumber (COinS_date.rftdate:match ('%d%d%d%d'));	-- get year portion of COinS date (because in Arabic numerals); convert string to number
					if date and (1850 <= date) then								-- location has no publisher; if date is 1850 or later
						utilities.set_message ('maint_location_no_publisher');	-- add maint cat
					end
				else															-- PublisherName has a value
					if cfg.keywords_xlate['none'] == PublisherName then			-- if that value is 'none' (only for book and encyclopedia citations)
						PublisherName = '';										-- unset
					end
				end
			end
	end

	local ID_list = {};															-- sequence table of rendered identifiers
	local ID_list_coins = {};													-- table of identifiers and their values from args; key is same as cfg.id_handlers's key
	local Class = A['Class'];													-- arxiv class identifier
	
	local ID_support = {
		{A['ASINTLD'], 'ASIN', 'err_asintld_missing_asin', A:ORIGIN ('ASINTLD')},				
		{DoiBroken, 'DOI', 'err_doibroken_missing_doi', A:ORIGIN ('DoiBroken')},
		{Embargo, 'PMC', 'err_embargo_missing_pmc', A:ORIGIN ('Embargo')},
		}

	ID_list, ID_list_coins = identifiers.identifier_lists_get (args, {DoiBroken = DoiBroken, ASINTLD = A['ASINTLD'], Embargo = Embargo, Class = Class}, ID_support);

	-- Account for the oddities that are {{cite arxiv}}, {{cite biorxiv}}, {{cite citeseerx}}, {{cite medrxiv}}, {{cite ssrn}}, before generation of COinS data.
	if utilities.in_array (config.CitationClass, whitelist.preprint_template_list_t) then	-- |arxiv= or |eprint= required for cite arxiv; |biorxiv=, |citeseerx=, |medrxiv=, |ssrn= required for their templates
		if not (args[cfg.id_handlers[config.CitationClass:upper()].parameters[1]] or 		-- can't use ID_list_coins k/v table here because invalid parameters omitted
			args[cfg.id_handlers[config.CitationClass:upper()].parameters[2]]) then			-- which causes unexpected parameter missing error message
				utilities.set_message ('err_' .. config.CitationClass .. '_missing');		-- add error message
		end

		Periodical = ({['arxiv'] = 'arXiv', ['biorxiv'] = 'bioRxiv', ['citeseerx'] = 'CiteSeerX', ['medrxiv'] = 'medRxiv', ['ssrn'] = 'Social Science Research Network'})[config.CitationClass];
	end

	-- Link the title of the work if no |url= was provided, but we have a |pmc= or a |doi= with |doi-access=free

	if config.CitationClass == "journal" and not utilities.is_set (URL) and not utilities.is_set (TitleLink) and not utilities.in_array (cfg.keywords_xlate[Title], {'off', 'none'}) then -- TODO: remove 'none' once existing citations have been switched to 'off', so 'none' can be used as token for "no title" instead
		if 'none' ~= cfg.keywords_xlate[auto_select] then						-- if auto-linking not disabled
 	 		if identifiers.auto_link_urls[auto_select] then						-- manual selection
		 		URL = identifiers.auto_link_urls[auto_select];					-- set URL to be the same as identifier's external link
 				URL_origin = cfg.id_handlers[auto_select:upper()].parameters[1];	-- set URL_origin to parameter name for use in error message if citation is missing a |title=
			elseif identifiers.auto_link_urls['pmc'] then						-- auto-select PMC
				URL = identifiers.auto_link_urls['pmc'];						-- set URL to be the same as the PMC external link if not embargoed
				URL_origin = cfg.id_handlers['PMC'].parameters[1];				-- set URL_origin to parameter name for use in error message if citation is missing a |title=
			elseif identifiers.auto_link_urls['doi'] then						-- auto-select DOI
				URL = identifiers.auto_link_urls['doi'];
				URL_origin = cfg.id_handlers['DOI'].parameters[1];
			end
 	 	end

		if utilities.is_set (URL) then											-- set when using an identifier-created URL
			if utilities.is_set (AccessDate) then								-- |access-date= requires |url=; identifier-created URL is not |url=
				utilities.set_message ('err_accessdate_missing_url');			-- add an error message
				AccessDate = '';												-- unset
			end

			if utilities.is_set (ArchiveURL) then								-- |archive-url= requires |url=; identifier-created URL is not |url=
				utilities.set_message ('err_archive_missing_url');				-- add an error message
				ArchiveURL = '';												-- unset
			end
		end
	end

	-- At this point fields may be nil if they weren't specified in the template use.  We can use that fact.
	-- Test if citation has no title
	if	not utilities.is_set (Title) and not utilities.is_set (TransTitle) and not utilities.is_set (ScriptTitle) then	-- has special case for cite episode
		utilities.set_message ('err_citation_missing_title', {'episode' == config.CitationClass and 'series' or 'title'});
	end

	if utilities.in_array (cfg.keywords_xlate[Title], {'off', 'none'}) and
			utilities.in_array (config.CitationClass, {'journal', 'citation'}) and
			(utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and
			('journal' == Periodical_origin or 'script-journal' == ScriptPeriodical_origin) then	-- special case for journal cites
				Title = '';														-- set title to empty string
				utilities.set_message ('maint_untitled');						-- add maint cat
	end

	-- COinS metadata (see <http://ocoins.info/>) for automated parsing of citation information.
	-- handle the oddity that is cite encyclopedia and {{citation |encyclopedia=something}}. Here we presume that
	-- when Periodical, Title, and Chapter are all set, then Periodical is the book (encyclopedia) title, Title
	-- is the article title, and Chapter is a section within the article.  So, we remap 
	
	local coins_chapter = Chapter;												-- default assuming that remapping not required
	local coins_title = Title;													-- et tu
	if 'encyclopaedia' == config.CitationClass or ('citation' == config.CitationClass and utilities.is_set (Encyclopedia)) then
		if utilities.is_set (Chapter) and utilities.is_set (Title) and utilities.is_set (Periodical) then		-- if all are used then
			coins_chapter = Title;												-- remap
			coins_title = Periodical;
		end
	end
	local coins_author = a;														-- default for coins rft.au 
	if 0 < #c then																-- but if contributor list
		coins_author = c;														-- use that instead
	end
	
	-- this is the function call to COinS()
	local OCinSoutput = metadata.COinS({
		['Periodical'] = utilities.strip_apostrophe_markup (Periodical),		-- no markup in the metadata
		['Encyclopedia'] = Encyclopedia,										-- just a flag; content ignored by ~/COinS
		['Chapter'] = metadata.make_coins_title (coins_chapter, ScriptChapter),	-- Chapter and ScriptChapter stripped of bold / italic / accept-as-written markup
		['Degree'] = Degree;													-- cite thesis only
		['Title'] = metadata.make_coins_title (coins_title, ScriptTitle),		-- Title and ScriptTitle stripped of bold / italic / accept-as-written markup
		['PublicationPlace'] = PublicationPlace,
		['Date'] = COinS_date.rftdate,											-- COinS_date.* has correctly formatted date values if Date is valid;
		['Season'] = COinS_date.rftssn,
		['Quarter'] = COinS_date.rftquarter,
		['Chron'] =  COinS_date.rftchron,
		['Series'] = Series,
		['Volume'] = Volume,
		['Issue'] = Issue,
		['ArticleNumber'] = ArticleNumber,
		['Pages'] = coins_pages or metadata.get_coins_pages (first_set ({Sheet, Sheets, Page, Pages, At, QuotePage, QuotePages}, 7)),	-- pages stripped of external links
		['Edition'] = Edition,
		['PublisherName'] = PublisherName or Newsgroup,							-- any apostrophe markup already removed from PublisherName
		['URL'] = first_set ({ChapterURL, URL}, 2),
		['Authors'] = coins_author,
		['ID_list'] = ID_list_coins,
		['RawPage'] = this_page.prefixedText,
	}, config.CitationClass);

	-- Account for the oddities that are {{cite arxiv}}, {{cite biorxiv}}, {{cite citeseerx}}, {{cite medrxiv}}, and {{cite ssrn}} AFTER generation of COinS data.
	if utilities.in_array (config.CitationClass, whitelist.preprint_template_list_t) then	-- we have set rft.jtitle in COinS to arXiv, bioRxiv, CiteSeerX, medRxiv, or ssrn now unset so it isn't displayed
		Periodical = '';														-- periodical not allowed in these templates; if article has been published, use cite journal
	end

	-- special case for cite newsgroup.  Do this after COinS because we are modifying Publishername to include some static text
	if 'newsgroup' == config.CitationClass and utilities.is_set (Newsgroup) then
		PublisherName = utilities.substitute (cfg.messages['newsgroup'], external_link( 'news:' .. Newsgroup, Newsgroup, Newsgroup_origin, nil ));
	end

	local Editors;
	local EditorCount;															-- used only for choosing {ed.) or (eds.) annotation at end of editor name-list
	local Contributors;															-- assembled contributors name list
	local contributor_etal;
	local Translators;															-- assembled translators name list
	local translator_etal;
	local t = {};																-- translators list from |translator-lastn= / translator-firstn= pairs
	t = extract_names (args, 'TranslatorList');									-- fetch translator list from |translatorn= / |translator-lastn=, -firstn=, -linkn=, -maskn=
	local Interviewers;															
	local interviewers_list = {};					
	interviewers_list = extract_names (args, 'InterviewerList');				-- process preferred interviewers parameters
	local interviewer_etal;
	
	-- Now perform various field substitutions.
	-- We also add leading spaces and surrounding markup and punctuation to the
	-- various parts of the citation, but only when they are non-nil.
	do
		local last_first_list;
		local control = { 
			format = NameListStyle,												-- empty string, '&', 'amp', 'and', or 'vanc'
			maximum = nil,														-- as if display-authors or display-editors not set
			mode = Mode
		};

		do																		-- do editor name list first because the now unsupported coauthors used to modify control table
			local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayEditors'], A['DisplayEditors'], A:ORIGIN ('DisplayEditors'), #e);
			control.maximum, editor_etal = get_display_names (display_names, #e, 'editors', editor_etal, param);

			Editors, EditorCount = list_people (control, e, editor_etal);

			if 1 == EditorCount and (true == editor_etal or 1 < #e) then		-- only one editor displayed but includes etal then 
				EditorCount = 2;												-- spoof to display (eds.) annotation
			end
		end
		do																		-- now do interviewers
			local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayInterviewers'], A['DisplayInterviewers'], A:ORIGIN ('DisplayInterviewers'), #interviewers_list);
			control.maximum, interviewer_etal = get_display_names (display_names, #interviewers_list, 'interviewers', interviewer_etal, param);

			Interviewers = list_people (control, interviewers_list, interviewer_etal);
		end
		do																		-- now do translators
			local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayTranslators'], A['DisplayTranslators'], A:ORIGIN ('DisplayTranslators'), #t);
			control.maximum, translator_etal = get_display_names (display_names, #t, 'translators', translator_etal, param);

			Translators = list_people (control, t, translator_etal);
		end
		do																		-- now do contributors
			local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayContributors'], A['DisplayContributors'], A:ORIGIN ('DisplayContributors'), #c);
			control.maximum, contributor_etal = get_display_names (display_names, #c, 'contributors', contributor_etal, param);

			Contributors = list_people (control, c, contributor_etal);
		end
		do																		-- now do authors
			local display_names, param = display_names_select (cfg.global_cs1_config_t['DisplayAuthors'], A['DisplayAuthors'], A:ORIGIN ('DisplayAuthors'), #a, author_etal);
			control.maximum, author_etal = get_display_names (display_names, #a, 'authors', author_etal, param);

			last_first_list = list_people (control, a, author_etal);

			if utilities.is_set (Authors) then
				Authors, author_etal = name_has_etal (Authors, author_etal, false, 'authors');	-- find and remove variations on et al.
				if author_etal then
					Authors = Authors .. ' ' .. cfg.messages['et al'];			-- add et al. to authors parameter
				end
			else
				Authors = last_first_list;										-- either an author name list or an empty string
			end
		end																		-- end of do
	
		if utilities.is_set (Authors) and utilities.is_set (Collaboration) then
			Authors = Authors .. ' (' .. Collaboration .. ')';					-- add collaboration after et al.
		end

	end

	local ConferenceFormat = A['ConferenceFormat'];
	local ConferenceURL = A['ConferenceURL'];
	ConferenceFormat = style_format (ConferenceFormat, ConferenceURL, 'conference-format', 'conference-url');
	Format = style_format (Format, URL, 'format', 'url');

	-- special case for chapter format so no error message or cat when chapter not supported
	if not (utilities.in_array (config.CitationClass, {'web', 'news', 'journal', 'magazine', 'pressrelease', 'podcast', 'newsgroup', 'arxiv', 'biorxiv', 'citeseerx', 'medrxiv', 'ssrn'}) or
		('citation' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and not utilities.is_set (Encyclopedia))) then
			ChapterFormat = style_format (ChapterFormat, ChapterURL, 'chapter-format', 'chapter-url');
	end

	if not utilities.is_set (URL) then
		if utilities.in_array (config.CitationClass, {"web", "podcast", "mailinglist"}) or		-- |url= required for cite web, cite podcast, and cite mailinglist
			('citation' == config.CitationClass and ('website' == Periodical_origin or 'script-website' == ScriptPeriodical_origin)) then	-- and required for {{citation}} with |website= or |script-website=
				utilities.set_message ('err_cite_web_url');
		end
		
		-- do we have |accessdate= without either |url= or |chapter-url=?
		if utilities.is_set (AccessDate) and not utilities.is_set (ChapterURL) then		-- ChapterURL may be set when URL is not set;
			utilities.set_message ('err_accessdate_missing_url');
			AccessDate = '';
		end
	end

	local UrlStatus = is_valid_parameter_value (A['UrlStatus'], A:ORIGIN('UrlStatus'), cfg.keywords_lists['url-status'], '');
	local OriginalURL
	local OriginalURL_origin
	local OriginalFormat
	local OriginalAccess;
	UrlStatus = UrlStatus:lower();												-- used later when assembling archived text
	if utilities.is_set ( ArchiveURL ) then
		if utilities.is_set (ChapterURL) then 									-- if chapter-url= is set apply archive url to it
			OriginalURL = ChapterURL;											-- save copy of source chapter's url for archive text
			OriginalURL_origin = ChapterURL_origin;								-- name of |chapter-url= parameter for error messages
			OriginalFormat = ChapterFormat;										-- and original |chapter-format=

			if 'live' ~= UrlStatus then
				ChapterURL = ArchiveURL											-- swap-in the archive's URL
				ChapterURL_origin = A:ORIGIN('ArchiveURL')						-- name of |archive-url= parameter for error messages
				ChapterFormat = ArchiveFormat or '';							-- swap in archive's format
				ChapterUrlAccess = nil;											-- restricted access levels do not make sense for archived URLs
			end
		elseif utilities.is_set (URL) then
			OriginalURL = URL;													-- save copy of original source URL
			OriginalURL_origin = URL_origin;									-- name of URL parameter for error messages
			OriginalFormat = Format; 											-- and original |format=
			OriginalAccess = UrlAccess;

			if 'live' ~= UrlStatus then											-- if URL set then |archive-url= applies to it
				URL = ArchiveURL												-- swap-in the archive's URL
				URL_origin = A:ORIGIN('ArchiveURL')								-- name of archive URL parameter for error messages
				Format = ArchiveFormat or '';									-- swap in archive's format
				UrlAccess = nil;												-- restricted access levels do not make sense for archived URLs
			end
		end
	elseif utilities.is_set (UrlStatus) then									-- if |url-status= is set when |archive-url= is not set
 		utilities.set_message ('maint_url_status');								-- add maint cat
	end

	if utilities.in_array (config.CitationClass, {'web', 'news', 'journal', 'magazine', 'pressrelease', 'podcast', 'newsgroup', 'arxiv', 'biorxiv', 'citeseerx', 'medrxiv', 'ssrn'}) or	-- if any of the 'periodical' cites except encyclopedia
		('citation' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and not utilities.is_set (Encyclopedia)) then
			local chap_param;
			if utilities.is_set (Chapter) then									-- get a parameter name from one of these chapter related meta-parameters
				chap_param = A:ORIGIN ('Chapter')
			elseif utilities.is_set (TransChapter) then
				chap_param = A:ORIGIN ('TransChapter')
			elseif utilities.is_set (ChapterURL) then
				chap_param = A:ORIGIN ('ChapterURL')
			elseif utilities.is_set (ScriptChapter) then
				chap_param = ScriptChapter_origin;
			else utilities.is_set (ChapterFormat)
				chap_param = A:ORIGIN ('ChapterFormat')
			end

			if utilities.is_set (chap_param) then								-- if we found one
				utilities.set_message ('err_chapter_ignored', {chap_param});	-- add error message
				Chapter = '';													-- and set them to empty string to be safe with concatenation
				TransChapter = '';
				ChapterURL = '';
				ScriptChapter = '';
				ChapterFormat = '';
			end
	else																		-- otherwise, format chapter / article title
		local no_quotes = false;												-- default assume that we will be quoting the chapter parameter value
		if utilities.is_set (Contribution) and 0 < #c then						-- if this is a contribution with contributor(s)
			if utilities.in_array (Contribution:lower(), cfg.keywords_lists.contribution) then	-- and a generic contribution title
				no_quotes = true;												-- then render it unquoted
			end
		end

		Chapter = format_chapter_title (ScriptChapter, ScriptChapter_origin, Chapter, Chapter_origin, TransChapter, TransChapter_origin, ChapterURL, ChapterURL_origin, no_quotes, ChapterUrlAccess);		-- Contribution is also in Chapter
		if utilities.is_set (Chapter) then
			Chapter = Chapter .. ChapterFormat ;
			if 'map' == config.CitationClass and utilities.is_set (TitleType) then
				Chapter = Chapter .. ' ' .. TitleType;							-- map annotation here; not after title
			end
			Chapter = Chapter .. sepc .. ' ';
		elseif utilities.is_set (ChapterFormat) then							-- |chapter= not set but |chapter-format= is so ...
			Chapter = ChapterFormat .. sepc .. ' ';								-- ... ChapterFormat has error message, we want to see it
		end
	end

	-- Format main title
	local plain_title = false;
	local accept_title;
	Title, accept_title = utilities.has_accept_as_written (Title, true);		-- remove accept-this-as-written markup when it wraps all of <Title>
	if accept_title and ('' == Title) then										-- only support forced empty for now "(())"
		Title = cfg.messages['notitle'];										-- replace by predefined "No title" message
			-- TODO: utilities.set_message ( 'err_redundant_parameters', ...);	-- issue proper error message instead of muting	 
			ScriptTitle = '';													-- just mute for now	 
			TransTitle = '';													-- just mute for now
 		plain_title = true;														-- suppress text decoration for descriptive title
		utilities.set_message ('maint_untitled');								-- add maint cat
	end

	if not accept_title then													-- <Title> not wrapped in accept-as-written markup
		if '...' == Title:sub (-3) then											-- if ellipsis is the last three characters of |title=
			Title = Title:gsub ('(%.%.%.)%.+$', '%1');							-- limit the number of dots to three
		elseif not mw.ustring.find (Title, '%.%s*%a%.$') and					-- end of title is not a 'dot-(optional space-)letter-dot' initialism ...
			not mw.ustring.find (Title, '%s+%a%.$') then						-- ...and not a 'space-letter-dot' initial (''Allium canadense'' L.)
				Title = mw.ustring.gsub(Title, '%' .. sepc .. '$', '');			-- remove any trailing separator character; sepc and ms.ustring() here for languages that use multibyte separator characters
		end

		if utilities.is_set (ArchiveURL) and is_archived_copy (Title) then
			utilities.set_message ('maint_archived_copy');						-- add maintenance category before we modify the content of Title
		end

		if is_generic ('generic_titles', Title) then
			utilities.set_message ('err_generic_title');						-- set an error message
		end
	end

	if (not plain_title) and (utilities.in_array (config.CitationClass, {'web', 'news', 'journal', 'magazine', 'document', 'pressrelease', 'podcast', 'newsgroup', 'mailinglist', 'interview', 'arxiv', 'biorxiv', 'citeseerx', 'medrxiv', 'ssrn'}) or
		('citation' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)) and not utilities.is_set (Encyclopedia)) or
		('map' == config.CitationClass and (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical)))) then		-- special case for cite map when the map is in a periodical treat as an article
			Title = kern_quotes (Title);										-- if necessary, separate title's leading and trailing quote marks from module provided quote marks
			Title = utilities.wrap_style ('quoted-title', Title);
			Title = script_concatenate (Title, ScriptTitle, 'script-title');	-- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped
			TransTitle = utilities.wrap_style ('trans-quoted-title', TransTitle );
	elseif plain_title or ('report' == config.CitationClass) then				-- no styling for cite report and descriptive titles (otherwise same as above)
		Title = script_concatenate (Title, ScriptTitle, 'script-title');		-- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped
		TransTitle = utilities.wrap_style ('trans-quoted-title', TransTitle );	-- for cite report, use this form for trans-title
	else
		Title = utilities.wrap_style ('italic-title', Title);
		Title = script_concatenate (Title, ScriptTitle, 'script-title');		-- <bdi> tags, lang attribute, categorization, etc.; must be done after title is wrapped
		TransTitle = utilities.wrap_style ('trans-italic-title', TransTitle);
	end

	if utilities.is_set (TransTitle) then
		if utilities.is_set (Title) then
			TransTitle = " " .. TransTitle;
		else
			utilities.set_message ('err_trans_missing_title', {'title'});
		end
	end

	if utilities.is_set (Title) then											-- TODO: is this the right place to be making Wikisource URLs?
		if utilities.is_set (TitleLink) and utilities.is_set (URL) then
			utilities.set_message ('err_wikilink_in_url');						-- set an error message because we can't have both
			TitleLink = '';														-- unset
		end
	
		if not utilities.is_set (TitleLink) and utilities.is_set (URL) then
			Title = external_link (URL, Title, URL_origin, UrlAccess) .. TransTitle .. Format;
			URL = '';															-- unset these because no longer needed
			Format = "";
		elseif utilities.is_set (TitleLink) and not utilities.is_set (URL) then
			local ws_url;
			ws_url = wikisource_url_make (TitleLink);							-- ignore ws_label return; not used here
			if ws_url then
				Title = external_link (ws_url, Title .. '&nbsp;', 'ws link in title-link');	-- space char after Title to move icon away from italic text; TODO: a better way to do this?
				Title = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], TitleLink, Title});				
				Title = Title .. TransTitle;
			else
				Title = utilities.make_wikilink (TitleLink, Title) .. TransTitle;
			end
		else
			local ws_url, ws_label, L;											-- Title has italic or quote markup by the time we get here which causes is_wikilink() to return 0 (not a wikilink)
			ws_url, ws_label, L = wikisource_url_make (Title:gsub('^[\'"]*(.-)[\'"]*$', '%1'));	-- make ws URL from |title= interwiki link (strip italic or quote markup); link portion L becomes tooltip label
			if ws_url then
				Title = Title:gsub ('%b[]', ws_label);							-- replace interwiki link with ws_label to retain markup
				Title = external_link (ws_url, Title .. '&nbsp;', 'ws link in title');	-- space char after Title to move icon away from italic text; TODO: a better way to do this?
				Title = utilities.substitute (cfg.presentation['interwiki-icon'], {cfg.presentation['class-wikisource'], L, Title});				
				Title = Title .. TransTitle;
			else
				Title = Title .. TransTitle;
			end
		end
	else
		Title = TransTitle;
	end

	if utilities.is_set (Place) then
		Place = " " .. wrap_msg ('written', Place, use_lowercase) .. sepc .. " ";
	end

	local ConferenceURL_origin = A:ORIGIN('ConferenceURL');						-- get name of parameter that holds ConferenceURL
	if utilities.is_set (Conference) then
		if utilities.is_set (ConferenceURL) then
			Conference = external_link( ConferenceURL, Conference, ConferenceURL_origin, nil );
		end
		Conference = sepc .. " " .. Conference .. ConferenceFormat;
	elseif utilities.is_set (ConferenceURL) then
		Conference = sepc .. " " .. external_link( ConferenceURL, nil, ConferenceURL_origin, nil );
	end

	local Position = '';
	if not utilities.is_set (Position) then
		local Minutes = A['Minutes'];
		local Time = A['Time'];

		if utilities.is_set (Minutes) then
			if utilities.is_set (Time) then		--TODO: make a function for this and similar?
				utilities.set_message ('err_redundant_parameters', {utilities.wrap_style ('parameter', 'minutes') .. cfg.presentation['sep_list_pair'] .. utilities.wrap_style ('parameter', 'time')});
			end
			Position = " " .. Minutes .. " " .. cfg.messages['minutes'];
		else
			if utilities.is_set (Time) then
				local TimeCaption = A['TimeCaption']
				if not utilities.is_set (TimeCaption) then
					TimeCaption = cfg.messages['event'];
					if sepc ~= '.' then
						TimeCaption = TimeCaption:lower();
					end
				end
				Position = " " .. TimeCaption .. " " .. Time;
			end
		end
	else
		Position = " " .. Position;
		At = '';
	end

	Page, Pages, Sheet, Sheets = format_pages_sheets (Page, Pages, Sheet, Sheets, config.CitationClass, Periodical_origin, sepc, NoPP, use_lowercase);

	At = utilities.is_set (At) and (sepc .. " " .. At) or "";
	Position = utilities.is_set (Position) and (sepc .. " " .. Position) or "";
	if config.CitationClass == 'map' then
		local Sections = A['Sections'];											-- Section (singular) is an alias of Chapter so set earlier
		local Inset = A['Inset'];
		
		if utilities.is_set ( Inset ) then
			Inset = sepc .. " " .. wrap_msg ('inset', Inset, use_lowercase);
		end			

		if utilities.is_set ( Sections ) then
			Section = sepc .. " " .. wrap_msg ('sections', Sections, use_lowercase);
		elseif utilities.is_set ( Section ) then
			Section = sepc .. " " .. wrap_msg ('section', Section, use_lowercase);
		end
		At = At .. Inset .. Section;		
	end	

	local Others = A['Others'];
	if utilities.is_set (Others) and 0 == #a and 0 == #e then					-- add maint cat when |others= has value and used without |author=, |editor=
		if config.CitationClass == "AV-media-notes"
		or config.CitationClass == "audio-visual" then							-- special maint for AV/M which has a lot of 'false' positives right now
			utilities.set_message ('maint_others_avm')
		else
			utilities.set_message ('maint_others');
		end
	end
	Others = utilities.is_set (Others) and (sepc .. " " .. Others) or "";
	
	if utilities.is_set (Translators) then
		Others = safe_join ({sepc .. ' ', wrap_msg ('translated', Translators, use_lowercase), Others}, sepc);
	end
	if utilities.is_set (Interviewers) then
		Others = safe_join ({sepc .. ' ', wrap_msg ('interview', Interviewers, use_lowercase), Others}, sepc);
	end
	
	local TitleNote = A['TitleNote'];
	TitleNote = utilities.is_set (TitleNote) and (sepc .. " " .. TitleNote) or "";
	if utilities.is_set (Edition) then
		if Edition:match ('%f[%a][Ee]d%n?%.?$') or Edition:match ('%f[%a][Ee]dition$') then -- Ed, ed, Ed., ed., Edn, edn, Edn., edn.
			utilities.set_message ('err_extra_text_edition');					 -- add error message
		end
		Edition = " " .. wrap_msg ('edition', Edition);
	else
		Edition = '';
	end

	Series = utilities.is_set (Series) and wrap_msg ('series', {sepc, Series}) or "";	-- not the same as SeriesNum
	local Agency = A['Agency'];
	Agency = utilities.is_set (Agency) and wrap_msg ('agency', {sepc, Agency}) or "";
	Volume = format_volume_issue (Volume, Issue, ArticleNumber, config.CitationClass, Periodical_origin, sepc, use_lowercase);

	if utilities.is_set (AccessDate) then
		local retrv_text = " " .. cfg.messages['retrieved']

		AccessDate = nowrap_date (AccessDate);									-- wrap in nowrap span if date in appropriate format
		if (sepc ~= ".") then retrv_text = retrv_text:lower() end				-- if mode is cs2, lower case
		AccessDate = utilities.substitute (retrv_text, AccessDate);				-- add retrieved text

		AccessDate = utilities.substitute (cfg.presentation['accessdate'], {sepc, AccessDate});	-- allow editors to hide accessdates
	end
	
	if utilities.is_set (ID) then ID = sepc .. " " .. ID; end
	
	local Docket = A['Docket'];
   	if "thesis" == config.CitationClass and utilities.is_set (Docket) then
		ID = sepc .. " Docket " .. Docket .. ID;
	end
   	if "report" == config.CitationClass and utilities.is_set (Docket) then		-- for cite report when |docket= is set
		ID = sepc .. ' ' .. Docket;												-- overwrite ID even if |id= is set
	end

	if utilities.is_set (URL) then
		URL = " " .. external_link( URL, nil, URL_origin, UrlAccess );
	end

	local Quote = A['Quote'];
	local TransQuote = A['TransQuote'];
	local ScriptQuote = A['ScriptQuote'];
	if utilities.is_set (Quote) or utilities.is_set (TransQuote) or utilities.is_set (ScriptQuote) then

		if utilities.is_set (Quote) then
			if Quote:sub(1, 1) == '"' and Quote:sub(-1, -1) == '"' then			-- if first and last characters of quote are quote marks
				Quote = Quote:sub(2, -2);										-- strip them off
			end
		end
		
		Quote = kern_quotes (Quote);											-- kern if needed
		Quote = utilities.wrap_style ('quoted-text', Quote );					-- wrap in <q>...</q> tags
	
		if utilities.is_set (ScriptQuote) then
			Quote = script_concatenate (Quote, ScriptQuote, 'script-quote');	-- <bdi> tags, lang attribute, categorization, etc.; must be done after quote is wrapped
		end

		if utilities.is_set (TransQuote) then
			if TransQuote:sub(1, 1) == '"' and TransQuote:sub(-1, -1) == '"' then -- if first and last characters of |trans-quote are quote marks
				TransQuote = TransQuote:sub(2, -2); -- strip them off
			end
			Quote = Quote .. " " .. utilities.wrap_style ('trans-quoted-title', TransQuote );
		end

		if utilities.is_set (QuotePage) or utilities.is_set (QuotePages) then	-- add page prefix
			local quote_prefix = '';
			if utilities.is_set (QuotePage) then
				extra_text_in_page_check (QuotePage, 'quote-page');				-- add to maint cat if |quote-page= value begins with what looks like p., pp., etc.
				if not NoPP then
					quote_prefix = utilities.substitute (cfg.messages['p-prefix'], {sepc, QuotePage}), '', '', '';
				else
					quote_prefix = utilities.substitute (cfg.messages['nopp'], {sepc, QuotePage}), '', '', '';
				end
			elseif utilities.is_set (QuotePages) then
				extra_text_in_page_check (QuotePages, 'quote-pages');			-- add to maint cat if |quote-pages= value begins with what looks like p., pp., etc.
				if tonumber(QuotePages) ~= nil and not NoPP then				-- if only digits, assume single page
					quote_prefix = utilities.substitute (cfg.messages['p-prefix'], {sepc, QuotePages}), '', '';
				elseif not NoPP then
					quote_prefix = utilities.substitute (cfg.messages['pp-prefix'], {sepc, QuotePages}), '', '';
				else
					quote_prefix = utilities.substitute (cfg.messages['nopp'], {sepc, QuotePages}), '', '';
				end
			end
                        
			Quote = quote_prefix .. ": " .. Quote;
		else
			Quote = sepc .. " " .. Quote;
		end

		PostScript = "";														-- cs1|2 does not supply terminal punctuation when |quote= is set
	end
	
	-- We check length of PostScript here because it will have been nuked by
	-- the quote parameters. We'd otherwise emit a message even if there wasn't
	-- a displayed postscript.
	-- TODO: Should the max size (1) be configurable?
	-- TODO: Should we check a specific pattern?
	if utilities.is_set(PostScript) and mw.ustring.len(PostScript) > 1 then
		utilities.set_message ('maint_postscript')
	end
	
	local Archived;
	if utilities.is_set (ArchiveURL) then
		if not utilities.is_set (ArchiveDate) then								-- ArchiveURL set but ArchiveDate not set
			utilities.set_message ('err_archive_missing_date');					-- emit an error message
			ArchiveURL = '';													-- empty string for concatenation
			ArchiveDate = '';													-- empty string for concatenation
		end
	else														
		if utilities.is_set (ArchiveDate) then									-- ArchiveURL not set but ArchiveDate is set
			utilities.set_message ('err_archive_date_missing_url');				-- emit an error message
			ArchiveURL = '';													-- empty string for concatenation
			ArchiveDate = '';													-- empty string for concatenation
		end
	end

	if utilities.is_set (ArchiveURL) then
		local arch_text;
		if "live" == UrlStatus then
			arch_text = cfg.messages['archived'];
			if sepc ~= "." then arch_text = arch_text:lower() end
			if utilities.is_set (ArchiveDate) then
				Archived = sepc .. ' ' .. utilities.substitute ( cfg.messages['archived-live'],
					{external_link( ArchiveURL, arch_text, A:ORIGIN('ArchiveURL'), nil) .. ArchiveFormat, ArchiveDate } );
			else
				Archived = '';
			end
			if not utilities.is_set (OriginalURL) then
				utilities.set_message ('err_archive_missing_url');
				Archived = '';													-- empty string for concatenation
			end
		elseif utilities.is_set (OriginalURL) then								-- UrlStatus is empty, 'dead', 'unfit', 'usurped', 'bot: unknown'
			if utilities.in_array (UrlStatus, {'unfit', 'usurped', 'bot: unknown'}) then
				arch_text = cfg.messages['archived-unfit'];
				if sepc ~= "." then arch_text = arch_text:lower() end
				Archived = sepc .. ' ' .. arch_text .. ArchiveDate;				-- format already styled
				if 'bot: unknown' == UrlStatus then
					utilities.set_message ('maint_bot_unknown');				-- and add a category if not already added
				else
					utilities.set_message ('maint_unfit');						-- and add a category if not already added
				end
			else																-- UrlStatus is empty, 'dead'
				arch_text = cfg.messages['archived-dead'];
				if sepc ~= "." then arch_text = arch_text:lower() end
				if utilities.is_set (ArchiveDate) then
					Archived = sepc .. " " .. utilities.substitute ( arch_text,
						{ external_link( OriginalURL, cfg.messages['original'], OriginalURL_origin, OriginalAccess ) .. OriginalFormat, ArchiveDate } );	-- format already styled
				else
					Archived = '';												-- unset for concatenation
				end
			end	
		else																	-- OriginalUrl not set
			utilities.set_message ('err_archive_missing_url');
			Archived = '';														-- empty string for concatenation
		end
	elseif utilities.is_set (ArchiveFormat) then
		Archived = ArchiveFormat;												-- if set and ArchiveURL not set ArchiveFormat has error message
	else
		Archived = '';
	end
	
	local TranscriptURL = A['TranscriptURL']
	local TranscriptFormat = A['TranscriptFormat'];
	TranscriptFormat = style_format (TranscriptFormat, TranscriptURL, 'transcript-format', 'transcripturl');
	local Transcript = A['Transcript'];
	local TranscriptURL_origin = A:ORIGIN('TranscriptURL');						-- get name of parameter that holds TranscriptURL
	if utilities.is_set (Transcript) then
		if utilities.is_set (TranscriptURL) then
			Transcript = external_link( TranscriptURL, Transcript, TranscriptURL_origin, nil );
		end
		Transcript = sepc .. ' ' .. Transcript .. TranscriptFormat;
	elseif utilities.is_set (TranscriptURL) then
		Transcript = external_link( TranscriptURL, nil, TranscriptURL_origin, nil );
	end

	local Publisher;
	if utilities.is_set (PublicationDate) then
		PublicationDate = wrap_msg ('published', PublicationDate);
	end
	if utilities.is_set (PublisherName) then
		if utilities.is_set (PublicationPlace) then
			Publisher = sepc .. " " .. PublicationPlace .. ": " .. PublisherName .. PublicationDate;
		else
			Publisher = sepc .. " " .. PublisherName .. PublicationDate;  
		end			
	elseif utilities.is_set (PublicationPlace) then 
		Publisher= sepc .. " " .. PublicationPlace .. PublicationDate;
	else 
		Publisher = PublicationDate;
	end
	
	-- Several of the above rely upon detecting this as nil, so do it last.
	if (utilities.is_set (Periodical) or utilities.is_set (ScriptPeriodical) or utilities.is_set (TransPeriodical)) then
		if utilities.is_set (Title) or utilities.is_set (TitleNote) then 
			Periodical = sepc .. " " .. format_periodical (ScriptPeriodical, ScriptPeriodical_origin, Periodical, TransPeriodical, TransPeriodical_origin);
		else 
			Periodical = format_periodical (ScriptPeriodical, ScriptPeriodical_origin, Periodical, TransPeriodical, TransPeriodical_origin);
		end
	end
	
	local Language = A['Language'];
	if utilities.is_set (Language) then
		Language = language_parameter (Language);								-- format, categories, name from ISO639-1, etc.
	else
		Language='';															-- language not specified so make sure this is an empty string;
	--[[ TODO: need to extract the wrap_msg from language_parameter
	so that we can solve parentheses bunching problem with Format/Language/TitleType
	]]
	end

	--[[
	Handle the oddity that is cite speech.  This code overrides whatever may be the value assigned to TitleNote (through |department=) and forces it to be " (Speech)" so that
	the annotation directly follows the |title= parameter value in the citation rather than the |event= parameter value (if provided).
	]]
	if "speech" == config.CitationClass then									-- cite speech only
		TitleNote = TitleType;													-- move TitleType to TitleNote so that it renders ahead of |event=
		TitleType = '';															-- and unset

		if utilities.is_set (Periodical) then									-- if Periodical, perhaps because of an included |website= or |journal= parameter 
			if utilities.is_set (Conference) then								-- and if |event= is set
				Conference = Conference .. sepc .. " ";							-- then add appropriate punctuation to the end of the Conference variable before rendering
			end
		end
	end

	-- Piece all bits together at last.  Here, all should be non-nil.
	-- We build things this way because it is more efficient in LUA
	-- not to keep reassigning to the same string variable over and over.

	local tcommon;
	local tcommon2;																-- used for book cite when |contributor= is set
	
	if utilities.in_array (config.CitationClass, {"journal", "citation"}) and utilities.is_set (Periodical) then
		if not (utilities.is_set (Authors) or utilities.is_set (Editors)) then
			Others = Others:gsub ('^' .. sepc .. ' ', '');						-- when no authors and no editors, strip leading sepc and space
		end
		if utilities.is_set (Others) then Others = safe_join ({Others, sepc .. " "}, sepc) end		-- add terminal punctuation & space; check for dup sepc; TODO why do we need to do this here?
		tcommon = safe_join( {Others, Title, TitleNote, Conference, Periodical, Format, TitleType, Series, Language, Edition, Publisher, Agency, Volume}, sepc );
	elseif utilities.in_array (config.CitationClass, {"book", "citation"}) and not utilities.is_set (Periodical) then		-- special cases for book cites
		if utilities.is_set (Contributors) then									-- when we are citing foreword, preface, introduction, etc.
			tcommon = safe_join( {Title, TitleNote}, sepc );					-- author and other stuff will come after this and before tcommon2
			tcommon2 = safe_join( {Conference, Periodical, Format, TitleType, Series, Language, Volume, Others, Edition, Publisher, Agency}, sepc );
		else
			tcommon = safe_join( {Title, TitleNote, Conference, Periodical, Format, TitleType, Series, Language, Volume, Others, Edition, Publisher, Agency}, sepc );
		end

	elseif 'map' == config.CitationClass then									-- special cases for cite map
		if utilities.is_set (Chapter) then										-- map in a book; TitleType is part of Chapter
			tcommon = safe_join( {Title, Format, Edition, Scale, Series, Language, Cartography, Others, Publisher, Volume}, sepc );
		elseif utilities.is_set (Periodical) then								-- map in a periodical
			tcommon = safe_join( {Title, TitleType, Format, Periodical, Scale, Series, Language, Cartography, Others, Publisher, Volume}, sepc );
		else																	-- a sheet or stand-alone map
			tcommon = safe_join( {Title, TitleType, Format, Edition, Scale, Series, Language, Cartography, Others, Publisher}, sepc );
		end
		
	elseif 'episode' == config.CitationClass then								-- special case for cite episode
		tcommon = safe_join( {Title, TitleNote, TitleType, Series, Language, Edition, Publisher}, sepc );

	else																		-- all other CS1 templates
		tcommon = safe_join( {Title, TitleNote, Conference, Periodical, Format, TitleType, Series, Language, 
			Volume, Others, Edition, Publisher, Agency}, sepc );
	end
	
	if #ID_list > 0 then
		ID_list = safe_join( { sepc .. " ",  table.concat( ID_list, sepc .. " " ), ID }, sepc );
	else
		ID_list = ID;
	end
	
	local Via = A['Via'];
	Via = utilities.is_set (Via) and  wrap_msg ('via', Via) or '';
	local idcommon;
	if 'audio-visual' == config.CitationClass or 'episode' == config.CitationClass then	-- special case for cite AV media & cite episode position transcript
		idcommon = safe_join( { ID_list, URL, Archived, Transcript, AccessDate, Via, Quote }, sepc );
	else
		idcommon = safe_join( { ID_list, URL, Archived, AccessDate, Via, Quote }, sepc );
	end
	
	local text;
	local pgtext = Position .. Sheet .. Sheets .. Page .. Pages .. At;

	local OrigDate = A['OrigDate'];
	OrigDate = utilities.is_set (OrigDate) and wrap_msg ('origdate', OrigDate) or '';
	if utilities.is_set (Date) then
		if utilities.is_set (Authors) or utilities.is_set (Editors) then		-- date follows authors or editors when authors not set
			Date = " (" .. Date .. ")" .. OrigDate .. sepc .. " ";				-- in parentheses
		else																	-- neither of authors and editors set
			if (string.sub(tcommon, -1, -1) == sepc) then						-- if the last character of tcommon is sepc
				Date = " " .. Date .. OrigDate;									-- Date does not begin with sepc
			else
				Date = sepc .. " " .. Date .. OrigDate;							-- Date begins with sepc
			end
		end
	end	
	if utilities.is_set (Authors) then
		if (not utilities.is_set (Date)) then									-- when date is set it's in parentheses; no Authors termination
			Authors = terminate_name_list (Authors, sepc);						-- when no date, terminate with 0 or 1 sepc and a space
		end
		if utilities.is_set (Editors) then
			local in_text = '';
			local post_text = '';
			if utilities.is_set (Chapter) and 0 == #c then
				in_text = cfg.messages['in'] .. ' ';
				if (sepc ~= '.') then
					in_text = in_text:lower();									-- lowercase for cs2
				end
			end
			if EditorCount <= 1 then
				post_text = ' (' .. cfg.messages['editor'] .. ')';				-- be consistent with no-author, no-date case
			else
				post_text = ' (' .. cfg.messages['editors'] .. ')';
			end
			Editors = terminate_name_list (in_text .. Editors .. post_text, sepc);	-- terminate with 0 or 1 sepc and a space
		end
		if utilities.is_set (Contributors) then									-- book cite and we're citing the intro, preface, etc.
			local by_text = sepc .. ' ' .. cfg.messages['by'] .. ' ';
			if (sepc ~= '.') then by_text = by_text:lower() end					-- lowercase for cs2
			Authors = by_text .. Authors;										-- author follows title so tweak it here
			if utilities.is_set (Editors) and utilities.is_set (Date) then		-- when Editors make sure that Authors gets terminated
				Authors = terminate_name_list (Authors, sepc);					-- terminate with 0 or 1 sepc and a space
			end
			if (not utilities.is_set (Date)) then								-- when date is set it's in parentheses; no Contributors termination
				Contributors = terminate_name_list (Contributors, sepc);		-- terminate with 0 or 1 sepc and a space
			end
			text = safe_join( {Contributors, Date, Chapter, tcommon, Authors, Place, Editors, tcommon2, pgtext, idcommon }, sepc );
		else
			text = safe_join( {Authors, Date, Chapter, Place, Editors, tcommon, pgtext, idcommon }, sepc );
		end
	elseif utilities.is_set (Editors) then
		if utilities.is_set (Date) then
			if EditorCount <= 1 then
				Editors = Editors .. cfg.presentation['sep_name'] .. cfg.messages['editor'];
			else
				Editors = Editors .. cfg.presentation['sep_name'] .. cfg.messages['editors'];
			end
		else
			if EditorCount <= 1 then
				Editors = Editors .. " (" .. cfg.messages['editor'] .. ")" .. sepc .. " "
			else
				Editors = Editors .. " (" .. cfg.messages['editors'] .. ")" .. sepc .. " "
			end
		end
		text = safe_join( {Editors, Date, Chapter, Place, tcommon, pgtext, idcommon}, sepc );
	else
		if utilities.in_array (config.CitationClass, {"journal", "citation"}) and utilities.is_set (Periodical) then
			text = safe_join( {Chapter, Place, tcommon, pgtext, Date, idcommon}, sepc );
		else
			text = safe_join( {Chapter, Place, tcommon, Date, pgtext, idcommon}, sepc );
		end
	end
	
	if utilities.is_set (PostScript) and PostScript ~= sepc then
		text = safe_join( {text, sepc}, sepc ); 								-- Deals with italics, spaces, etc.
		text = text:sub(1, -sepc:len() - 1);
	end	
	
	text = safe_join( {text, PostScript}, sepc );

	-- Now enclose the whole thing in a <cite> element
	local options_t = {};
	options_t.class = cite_class_attribute_make (config.CitationClass, Mode);

	local Ref = is_valid_parameter_value (A['Ref'], A:ORIGIN('Ref'), cfg.keywords_lists['ref'], nil, true);	-- nil when |ref=harv; A['Ref'] else

	if 'none' ~= cfg.keywords_xlate[(Ref and Ref:lower()) or ''] then
		local namelist_t = {};													-- holds selected contributor, author, editor name list
		local year = first_set ({Year, anchor_year}, 2);						-- Year first for legacy citations and for YMD dates that require disambiguation

		if #c > 0 then															-- if there is a contributor list
			namelist_t = c;														-- select it
		elseif #a > 0 then														-- or an author list
			namelist_t = a;
		elseif #e > 0 then														-- or an editor list
			namelist_t = e;
		end
		local citeref_id;
		if #namelist_t > 0 then													-- if there are names in namelist_t
			citeref_id = make_citeref_id (namelist_t, year);					-- go make the CITEREF anchor
			if mw.uri.anchorEncode (citeref_id) == ((Ref and mw.uri.anchorEncode (Ref)) or '') then	-- Ref may already be encoded (by {{sfnref}}) so citeref_id must be encoded before comparison
				utilities.set_message ('maint_ref_duplicates_default');
			end
		else
			citeref_id = '';													-- unset
		end
		options_t.id = Ref or citeref_id;
	end

	if string.len (text:gsub('%b<>', '')) <= 2 then								-- remove html and html-like tags; then get length of what remains; 
		z.error_cats_t = {};													-- blank the categories list
		z.error_msgs_t = {};													-- blank the error messages list
		OCinSoutput = nil;														-- blank the metadata string
		text = '';																-- blank the the citation
		utilities.set_message ('err_empty_citation');							-- set empty citation message and category
	end
	
	local render_t = {};														-- here we collect the final bits for concatenation into the rendered citation

	if utilities.is_set (options_t.id) then										-- here we wrap the rendered citation in <cite ...>...</cite> tags
		table.insert (render_t, utilities.substitute (cfg.presentation['cite-id'], {mw.uri.anchorEncode(options_t.id), mw.text.nowiki(options_t.class), text}));	-- when |ref= is set or when there is a namelist
	else
		table.insert (render_t, utilities.substitute (cfg.presentation['cite'], {mw.text.nowiki(options_t.class), text}));	-- when |ref=none or when namelist_t empty and |ref= is missing or is empty
	end		

	if OCinSoutput then															-- blanked when citation is 'empty' so don't bother to add boilerplate metadata span
		table.insert (render_t, utilities.substitute (cfg.presentation['ocins'], OCinSoutput));	-- format and append metadata to the citation
	end

	local template_name = ('citation' == config.CitationClass) and 'citation' or 'cite ' .. (cfg.citation_class_map_t[config.CitationClass] or config.CitationClass);
	local template_link = '[[Template:' .. template_name .. '|' .. template_name .. ']]';
	local msg_prefix = '<code class="cs1-code">{{' .. template_link .. '}}</code>: ';

	if 0 ~= #z.error_msgs_t then
		mw.addWarning (utilities.substitute (cfg.messages.warning_msg_e, template_link));

		table.insert (render_t, ' ');											-- insert a space between citation and its error messages
		table.sort (z.error_msgs_t);											-- sort the error messages list; sorting includes wrapping <span> and <code> tags; hidden-error sorts ahead of visible-error

		local hidden = true;													-- presume that the only error messages emited by this template are hidden
		for _, v in ipairs (z.error_msgs_t) do									-- spin through the list of error messages
			if v:find ('cs1-visible-error', 1, true) then						-- look for the visible error class name
				hidden = false;													-- found one; so don't hide the error message prefix
				break;															-- and done because no need to look further
			end
		end

		z.error_msgs_t[1] = table.concat ({utilities.error_comment (msg_prefix, hidden), z.error_msgs_t[1]});	-- add error message prefix to first error message to prevent extraneous punctuation
		table.insert (render_t, table.concat (z.error_msgs_t, '; '));			-- make a big string of error messages and add it to the rendering
	end

	if 0 ~= #z.maint_cats_t then
		mw.addWarning (utilities.substitute (cfg.messages.warning_msg_m, template_link));

		table.sort (z.maint_cats_t);											-- sort the maintenance messages list

		local maint_msgs_t = {};												-- here we collect all of the maint messages

		if 0 == #z.error_msgs_t then											-- if no error messages
			table.insert (maint_msgs_t, msg_prefix);							-- insert message prefix in maint message livery
		end
		
		for _, v in ipairs( z.maint_cats_t ) do									-- append maintenance categories
			table.insert (maint_msgs_t, 										-- assemble new maint message and add it to the maint_msgs_t table
				table.concat ({v, ' (', utilities.substitute (cfg.messages[':cat wikilink'], v), ')'})
				);
		end
		table.insert (render_t, utilities.substitute (cfg.presentation['hidden-maint'], table.concat (maint_msgs_t, ' ')));	-- wrap the group of maint messages with proper presentation and save
	end

	if not no_tracking_cats then
		local sort_key;
		local cat_wikilink = 'cat wikilink';
		if cfg.enable_sort_keys then											-- when namespace sort keys enabled
			local namespace_number = mw.title.getCurrentTitle().namespace;		-- get namespace number for this wikitext
			sort_key = (0 ~= namespace_number and (cfg.name_space_sort_keys[namespace_number] or cfg.name_space_sort_keys.other)) or nil;	-- get sort key character; nil for mainspace
			cat_wikilink = (not sort_key and 'cat wikilink') or 'cat wikilink sk';	-- make <cfg.messages> key
		end				

		for _, v in ipairs (z.error_cats_t) do									-- append error categories
			table.insert (render_t, utilities.substitute (cfg.messages[cat_wikilink], {v, sort_key}));
		end
		for _, v in ipairs (z.maint_cats_t) do									-- append maintenance categories
			table.insert (render_t, utilities.substitute (cfg.messages[cat_wikilink], {v, sort_key}));
		end

		for _, v in ipairs (z.prop_cats_t) do									-- append properties categories
			table.insert (render_t, utilities.substitute (cfg.messages['cat wikilink'], v));	-- no sort keys
		end
	end

	return table.concat (render_t);												-- make a big string and done
end


--[[--------------------------< V A L I D A T E >--------------------------------------------------------------

Looks for a parameter's name in one of several whitelists.

Parameters in the whitelist can have three values:
	true - active, supported parameters
	false - deprecated, supported parameters
	nil - unsupported parameters
	
]]

local function validate (name, cite_class, empty)
	local name = tostring (name);
	local enum_name;															-- parameter name with enumerator (if any) replaced with '#'
	local state;
	local function state_test (state, name)										-- local function to do testing of state values
		if true == state then return true; end									-- valid actively supported parameter
		if false == state then
			if empty then return nil; end										-- empty deprecated parameters are treated as unknowns
			deprecated_parameter (name);										-- parameter is deprecated but still supported
			return true;
		end
		if 'tracked' == state then
			local base_name = name:gsub ('%d', '');								-- strip enumerators from parameter names that have them to get the base name
			utilities.add_prop_cat ('tracked-param', {base_name}, base_name);	-- add a properties category; <base_name> modifies <key>
			return true;
		end
		return nil;
	end		

	if name:find ('#') then														-- # is a cs1|2 reserved character so parameters with # not permitted
		return nil;
	end
																				-- replace wnumerator digit(s) with # (|last25= becomes |last#=) (mw.ustring because non-Western 'local' digits)
	enum_name = mw.ustring.gsub (name, '%d+$', '#');							-- where enumerator is last charaters in parameter name (these to protect |s2cid=)
	enum_name = mw.ustring.gsub (enum_name, '%d+([%-l])', '#%1');				-- where enumerator is in the middle of the parameter name; |author#link= is the oddity

	if 'document' == cite_class then											-- special case for {{cite document}}
		state = whitelist.document_parameters_t[enum_name];						-- this list holds enumerated and nonenumerated parameters
		if true == state_test (state, name) then return true; end
		
		return false;
	end

	if utilities.in_array (cite_class, whitelist.preprint_template_list_t) then	-- limited parameter sets allowed for these templates
		state = whitelist.limited_parameters_t[enum_name];						-- this list holds enumerated and nonenumerated parameters
		if true == state_test (state, name) then return true; end

		state = whitelist.preprint_arguments_t[cite_class][name];				-- look in the parameter-list for the template identified by cite_class
		if true == state_test (state, name) then return true; end

		return false;															-- not supported because not found or name is set to nil
	end																			-- end limited parameter-set templates

	if utilities.in_array (cite_class, whitelist.unique_param_template_list_t) then 	-- template-specific parameters for templates that accept parameters from the basic argument list
		state = whitelist.unique_arguments_t[cite_class][name];					-- look in the template-specific parameter-lists for the template identified by cite_class
		if true == state_test (state, name) then return true; end
	end																			-- if here, fall into general validation

	state = whitelist.common_parameters_t[enum_name];							-- all other templates; all normal parameters allowed; this list holds enumerated and nonenumerated parameters
	if true == state_test (state, name) then return true; end

	return false;																-- not supported because not found or name is set to nil
end


--[=[-------------------------< I N T E R _ W I K I _ C H E C K >----------------------------------------------

check <value> for inter-language interwiki-link markup.  <prefix> must be a MediaWiki-recognized language
code.  when these values have the form (without leading colon):
	[[<prefix>:link|label]] return label as plain-text
	[[<prefix>:link]] return <prefix>:link as plain-text

return value as is else

]=]

local function inter_wiki_check (parameter, value)
	local prefix = value:match ('%[%[(%a+):');									-- get an interwiki prefix if one exists
	local _;
	
	if prefix and cfg.inter_wiki_map[prefix:lower()] then						-- if prefix is in the map, needs preceding colon so
		utilities.set_message ('err_bad_paramlink', parameter);					-- emit an error message
		_, value, _ = utilities.is_wikilink (value);							-- extract label portion from wikilink
	end
	return value;
end


--[[--------------------------< M I S S I N G _ P I P E _ C H E C K >------------------------------------------

Look at the contents of a parameter. If the content has a string of characters and digits followed by an equal
sign, compare the alphanumeric string to the list of cs1|2 parameters.  If found, then the string is possibly a
parameter that is missing its pipe.  There are two tests made:
	{{cite ... |title=Title access-date=2016-03-17}}	-- the first parameter has a value and whitespace separates that value from the missing pipe parameter name
	{{cite ... |title=access-date=2016-03-17}}			-- the first parameter has no value (whitespace after the first = is trimmed by MediaWiki)
cs1|2 shares some parameter names with XML/HTML attributes: class=, title=, etc.  To prevent false positives XML/HTML
tags are removed before the search.

If a missing pipe is detected, this function adds the missing pipe maintenance category.

]]

local function missing_pipe_check (parameter, value)
	local capture;
	value = value:gsub ('%b<>', '');											-- remove XML/HTML tags because attributes: class=, title=, etc.

	capture = value:match ('%s+(%a[%w%-]+)%s*=') or value:match ('^(%a[%w%-]+)%s*=');	-- find and categorize parameters with possible missing pipes
	if capture and validate (capture) then										-- if the capture is a valid parameter name
		utilities.set_message ('err_missing_pipe', parameter);
	end
end


--[[--------------------------< H A S _ E X T R A N E O U S _ P U N C T >--------------------------------------

look for extraneous terminal punctuation in most parameter values; parameters listed in skip table are not checked

]]

local function has_extraneous_punc (param, value)
	if 'number' == type (param) then
		return;
	end
	
	param = param:gsub ('%d+', '#');											-- enumerated name-list mask params allow terminal punct; normalize 
	if cfg.punct_skip[param] then
		return;																	-- parameter name found in the skip table so done
	end
	
	if value:match ('[,;:]$') then
		utilities.set_message ('maint_extra_punct');							-- has extraneous punctuation; add maint cat
	end
	if value:match ('^=') then													-- sometimes an extraneous '=' character appears ...
		utilities.set_message ('maint_extra_punct');							-- has extraneous punctuation; add maint cat
	end
end


--[[--------------------------< H A S _ E X T R A N E O U S _ U R L >------------------------------------------

look for extraneous url parameter values; parameters listed in skip table are not checked

]]

local function has_extraneous_url (url_param_t)
	local url_error_t = {};
	
	check_for_url (url_param_t, url_error_t);									-- extraneous url check
	if 0 ~= #url_error_t then													-- non-zero when there are errors
		table.sort (url_error_t);
		utilities.set_message ('err_param_has_ext_link', {utilities.make_sep_list (#url_error_t, url_error_t)});	-- add this error message
	end
end


--[[--------------------------< C I T A T I O N >--------------------------------------------------------------

This is used by templates such as {{cite book}} to create the actual citation text.

]]

local function citation(frame)
	Frame = frame;																-- save a copy in case we need to display an error message in preview mode

	local config = {};															-- table to store parameters from the module {{#invoke:}}
	for k, v in pairs( frame.args ) do											-- get parameters from the {{#invoke}} frame
		config[k] = v;
	--	args[k] = v;															-- crude debug support that allows us to render a citation from module {{#invoke:}}; skips parameter validation; TODO: keep?
	end	
																				-- i18n: set the name that your wiki uses to identify sandbox subpages from sandbox template invoke (or can be set here)
	local sandbox = ((config.SandboxPath and '' ~= config.SandboxPath) and config.SandboxPath) or '/sandbox';	-- sandbox path from {{#invoke:Citation/CS1/sandbox|citation|SandboxPath=/...}}
	is_sandbox = nil ~= string.find (frame:getTitle(), sandbox, 1, true);		-- is this invoke the sandbox module?
	sandbox = is_sandbox and sandbox or '';										-- use i18n sandbox to load sandbox modules when this module is the sandox; live modules else

	local pframe = frame:getParent()
	local styles;
	
	cfg = mw.loadData ('Module:Citation/CS1/Configuration' .. sandbox);			-- load sandbox versions of support modules when {{#invoke:Citation/CS1/sandbox|...}}; live modules else
	whitelist = mw.loadData ('Module:Citation/CS1/Whitelist' .. sandbox);
	utilities = require ('Module:Citation/CS1/Utilities' .. sandbox);
	validation = require ('Module:Citation/CS1/Date_validation' .. sandbox);
	identifiers = require ('Module:Citation/CS1/Identifiers' .. sandbox);
	metadata = require ('Module:Citation/CS1/COinS' .. sandbox);
	styles = 'Module:Citation/CS1' .. sandbox .. '/styles.css';

	utilities.set_selected_modules (cfg);										-- so that functions in Utilities can see the selected cfg tables
	identifiers.set_selected_modules (cfg, utilities);							-- so that functions in Identifiers can see the selected cfg tables and selected Utilities module
	validation.set_selected_modules (cfg, utilities);							-- so that functions in Date validataion can see selected cfg tables and the selected Utilities module
	metadata.set_selected_modules (cfg, utilities);								-- so that functions in COinS can see the selected cfg tables and selected Utilities module

	z = utilities.z;															-- table of error and category tables in Module:Citation/CS1/Utilities

	is_preview_mode = not utilities.is_set (frame:preprocess ('{{REVISIONID}}'));

	local args = {};															-- table where we store all of the template's arguments
	local suggestions = {};														-- table where we store suggestions if we need to loadData them
	local error_text;															-- used as a flag

	local capture;																-- the single supported capture when matching unknown parameters using patterns
	local empty_unknowns = {};													-- sequence table to hold empty unknown params for error message listing
	for k, v in pairs( pframe.args ) do											-- get parameters from the parent (template) frame
		v = mw.ustring.gsub (v, '^%s*(.-)%s*$', '%1');							-- trim leading/trailing whitespace; when v is only whitespace, becomes empty string
		if v ~= '' then
			if ('string' == type (k)) then
				k = mw.ustring.gsub (k, '%d', cfg.date_names.local_digits);		-- for enumerated parameters, translate 'local' digits to Western 0-9
			end
			if not validate( k, config.CitationClass ) then			
				if type (k) ~= 'string' then									-- exclude empty numbered parameters
					if v:match("%S+") ~= nil then
						error_text = utilities.set_message ('err_text_ignored', {v});
					end
				elseif validate (k:lower(), config.CitationClass) then 
					error_text = utilities.set_message ('err_parameter_ignored_suggest', {k, k:lower()});	-- suggest the lowercase version of the parameter
				else
					if nil == suggestions.suggestions then						-- if this table is nil then we need to load it
						suggestions = mw.loadData ('Module:Citation/CS1/Suggestions' .. sandbox);	--load sandbox version of suggestion module when {{#invoke:Citation/CS1/sandbox|...}}; live module else
					end
					for pattern, param in pairs (suggestions.patterns) do		-- loop through the patterns to see if we can suggest a proper parameter
						capture = k:match (pattern);							-- the whole match if no capture in pattern else the capture if a match
						if capture then											-- if the pattern matches 
							param = utilities.substitute (param, capture);		-- add the capture to the suggested parameter (typically the enumerator)
							if validate (param, config.CitationClass) then		-- validate the suggestion to make sure that the suggestion is supported by this template (necessary for limited parameter lists)
								error_text = utilities.set_message ('err_parameter_ignored_suggest', {k, param});	-- set the suggestion error message
							else
								error_text = utilities.set_message ('err_parameter_ignored', {k});	-- suggested param not supported by this template
								v = '';											-- unset
							end
						end
					end
					if not utilities.is_set (error_text) then					-- couldn't match with a pattern, is there an explicit suggestion?						
						if (suggestions.suggestions[ k:lower() ] ~= nil) and validate (suggestions.suggestions[ k:lower() ], config.CitationClass) then
							utilities.set_message ('err_parameter_ignored_suggest', {k, suggestions.suggestions[ k:lower() ]});
						else
							utilities.set_message ('err_parameter_ignored', {k});
							v = '';												-- unset value assigned to unrecognized parameters (this for the limited parameter lists)
						end
					end
				end				  
			end

			args[k] = v;														-- save this parameter and its value

		elseif not utilities.is_set (v) then									-- for empty parameters
			if not validate (k, config.CitationClass, true) then				-- is this empty parameter a valid parameter
				k = ('' == k) and '(empty string)' or k;						-- when k is empty string (or was space(s) trimmed to empty string), replace with descriptive text
				table.insert (empty_unknowns, utilities.wrap_style ('parameter', k));	-- format for error message and add to the list
			end
																				-- crude debug support that allows us to render a citation from module {{#invoke:}} TODO: keep?
	--	elseif args[k] ~= nil or (k == 'postscript') then						-- when args[k] has a value from {{#invoke}} frame (we don't normally do that)
	--		args[k] = v;														-- overwrite args[k] with empty string from pframe.args[k] (template frame); v is empty string here
		end																		-- not sure about the postscript bit; that gets handled in parameter validation; historical artifact?
	end	

	if 0 ~= #empty_unknowns then												-- create empty unknown error message
		utilities.set_message ('err_param_unknown_empty', {
			1 == #empty_unknowns and '' or 's',
			utilities.make_sep_list (#empty_unknowns, empty_unknowns)
			});
	end

	local url_param_t = {};

	for k, v in pairs( args ) do
		if 'string' == type (k) then											-- don't evaluate positional parameters
			has_invisible_chars (k, v);											-- look for invisible characters
		end
		has_extraneous_punc (k, v);												-- look for extraneous terminal punctuation in parameter values
		missing_pipe_check (k, v);												-- do we think that there is a parameter that is missing a pipe?
		args[k] = inter_wiki_check (k, v);										-- when language interwiki-linked parameter missing leading colon replace with wiki-link label

		if 'string' == type (k) and not cfg.url_skip[k] then					-- when parameter k is not positional and not in url skip table
			url_param_t[k] = v;													-- make a parameter/value list for extraneous url check
		end
	end

	has_extraneous_url (url_param_t);											-- look for url in parameter values where a url does not belong

	return table.concat ({
		frame:extensionTag ('templatestyles', '', {src=styles}),
		citation0( config, args)
	});
end


--[[--------------------------< E X P O R T E D   F U N C T I O N S >------------------------------------------
]]

return {citation = citation};