Mudanças entre as edições de "Módulo:Citação/CS1/Identificadores"

De WikiINRICRISTOpedia
Ir para navegação Ir para pesquisar
(atualizando citeseedx a partir de en:Module:Citation/CS1/Identifiers)
 
m (uma edição)
 
(Sem diferença)

Edição atual tal como às 11h35min de 8 de agosto de 2020

{{#lst:Módulo:Citação/CS1/doc|header}}

Esta página mostra e processa a verificação de erros em vários identificadores nomeados nos identificadores suportados em Módulo:Citação/CS1.

{{#lst:Módulo:Citação/CS1/doc|module_components_table}}


local identifiers = {};


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

local is_set, in_array, set_error, select_one, add_maint_cat, substitute;                    -- functions in Module:Citation/CS1/Utilities

local z;                                                                        -- table of tables defined in Module:Citation/CS1/Utilities

local cfg;                                                                        -- table of configuration tables that are defined in Module:Citation/CS1/Configuration


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

Formats a wiki style external link

]]

local function external_link_id(options)
    local url_string = options.id;
    local ext_link;
   
    if options.encode == true or options.encode == nil then
        url_string = mw.uri.encode( url_string );
    end
   
    ext_link = mw.ustring.format ('[%s%s%s %s]', options.prefix, url_string, options.suffix or "", mw.text.nowiki(options.id));
    if is_set(options.access) then
        ext_link = substitute (cfg.presentation[options.access], ext_link);    -- add the free-to-read / paywall lock
    end
   
    return mw.ustring.format( '[[%s|%s]]%s%s', options.link, options.label, options.separator or "&nbsp;", ext_link);

--    return mw.ustring.format( '[[%s|%s]]%s[%s%s%s %s]',
--        options.link, options.label, options.separator or "&nbsp;",
--        options.prefix, url_string, options.suffix or "",
--        mw.text.nowiki(options.id)
--    );
end


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

Formats a wiki style internal link

]]

local function internal_link_id(options)
    return mw.ustring.format( '[[%s|%s]]%s[[%s%s%s|%s]]',
        options.link, options.label, options.separator or "&nbsp;",
        options.prefix, options.id, options.suffix or "",
        mw.text.nowiki(options.id)
    );
end


--[[--------------------------< IS _ V A L I D _ I S X N >-----------------------------------------------------

ISBN-10 and ISSN validator code calculates checksum across all isbn/issn digits including the check digit.
ISBN-13 is checked in check_isbn().

If the number is valid the result will be 0. Before calling this function, issbn/issn must be checked for length
and stripped of dashes, spaces and other non-isxn characters.

]]

local function is_valid_isxn (isxn_str, len)
    local temp = 0;
    isxn_str = { isxn_str:byte(1, len) };    -- make a table of byte values '0' → 0x30 .. '9'  → 0x39, 'X' → 0x58
    len = len+1;                            -- adjust to be a loop counter
    for i, v in ipairs( isxn_str ) do        -- loop through all of the bytes and calculate the checksum
        if v == string.byte( "X" ) then        -- if checkdigit is X (compares the byte value of 'X' which is 0x58)
            temp = temp + 10*( len - i );    -- it represents 10 decimal
        else
            temp = temp + tonumber( string.char(v) )*(len-i);
        end
    end
    return temp % 11 == 0;                    -- returns true if calculation result is zero
end


--[[--------------------------< IS _ V A L I D _ I S X N  _ 1 3 >----------------------------------------------

ISBN-13 and ISMN validator code calculates checksum across all 13 isbn/ismn digits including the check digit.
If the number is valid, the result will be 0. Before calling this function, isbn-13/ismn must be checked for length
and stripped of dashes, spaces and other non-isxn-13 characters.

]]

local function is_valid_isxn_13 (isxn_str)
    local temp=0;
   
    isxn_str = { isxn_str:byte(1, 13) };                                        -- make a table of byte values '0' → 0x30 .. '9'  → 0x39
    for i, v in ipairs( isxn_str ) do
        temp = temp + (3 - 2*(i % 2)) * tonumber( string.char(v) );                -- multiply odd index digits by 1, even index digits by 3 and sum; includes check digit
    end
    return temp % 10 == 0;                                                        -- sum modulo 10 is zero when isbn-13/ismn is correct
end


--[[--------------------------< C H E C K _ I S B N >------------------------------------------------------------

Determines whether an ISBN string is valid

]]

local function check_isbn( isbn_str )
    if nil ~= isbn_str:match("[^%s-0-9X]") then
        return false, 'invalid character';                                        -- fail if isbn_str contains anything but digits, hyphens, or the uppercase X
    end
    isbn_str = isbn_str:gsub( "-", "" ):gsub( " ", "" );                        -- remove hyphens and spaces
    local len = isbn_str:len();
 
    if len ~= 10 and len ~= 13 then
        return false, 'length';                                                    -- fail if incorrect length
    end

    if len == 10 then
        if isbn_str:match( "^%d*X?$" ) == nil then                                -- fail if isbn_str has 'X' anywhere but last position
            return false, 'invalid form';                                   
        end
        return is_valid_isxn(isbn_str, 10), 'checksum';
    else
        if isbn_str:match( "^%d+$" ) == nil then
            return false, 'invalid character';                                    -- fail if isbn13 is not all digits
        end
        if isbn_str:match( "^97[89]%d*$" ) == nil then
            return false, 'invalid prefix';                                        -- fail when isbn13 does not begin with 978 or 979
        end
        return is_valid_isxn_13 (isbn_str), 'checksum';
    end
end


--[[--------------------------< I S M N >----------------------------------------------------------------------

Determines whether an ISMN string is valid.  Similar to isbn-13, ismn is 13 digits begining 979-0-... and uses the
same check digit calculations.  See http://www.ismn-international.org/download/Web_ISMN_Users_Manual_2008-6.pdf
section 2, pages 9–12.

]]

local function ismn (id)
    local handler = cfg.id_handlers['ISMN'];
    local text;
    local valid_ismn = true;
    local id_copy;

    id_copy = id;                                                                -- save a copy because this testing is destructive
    id=id:gsub( "[%s-–]", "" );                                                    -- strip spaces, hyphens, and endashes from the ismn

    if 13 ~= id:len() or id:match( "^9790%d*$" ) == nil then                    -- ismn must be 13 digits and begin 9790
        valid_ismn = false;
    else
        valid_ismn=is_valid_isxn_13 (id);                                        -- validate ismn
    end

--    text = internal_link_id({link = handler.link, label = handler.label,        -- use this (or external version) when there is some place to link to
--        prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode})
 
    text="[[" .. handler.link .. "|" .. handler.label .. "]]" .. handler.separator .. id_copy;        -- because no place to link to yet

    if false == valid_ismn then
        text = text .. ' ' .. set_error( 'bad_ismn' )                            -- add an error message if the ismn is invalid
    end
   
    return text;
end


--[[--------------------------< I S S N >----------------------------------------------------------------------

Validate and format an issn.  This code fixes the case where an editor has included an ISSN in the citation but
has separated the two groups of four digits with a space.  When that condition occurred, the resulting link looked
like this:

    |issn=0819 4327 gives: [http://www.worldcat.org/issn/0819 4327 0819 4327]  -- can't have spaces in an external link
   
This code now prevents that by inserting a hyphen at the issn midpoint.  It also validates the issn for length
and makes sure that the checkdigit agrees with the calculated value.  Incorrect length (8 digits), characters
other than 0-9 and X, or checkdigit / calculated value mismatch will all cause a check issn error message.  The
issn is always displayed with a hyphen, even if the issn was given as a single group of 8 digits.

]]

local function issn(id, e)
    local issn_copy = id;        -- save a copy of unadulterated issn; use this version for display if issn does not validate
    local handler;
    local text;
    local valid_issn = true;
   
    if e then
         handler = cfg.id_handlers['EISSN'];
    else
         handler = cfg.id_handlers['ISSN'];
    end

    id=id:gsub( "[%s-–]", "" );                                    -- strip spaces, hyphens, and endashes from the issn

    if 8 ~= id:len() or nil == id:match( "^%d*X?$" ) then        -- validate the issn: 8 digits long, containing only 0-9 or X in the last position
        valid_issn=false;                                        -- wrong length or improper character
    else
        valid_issn=is_valid_isxn(id, 8);                        -- validate issn
    end

    if true == valid_issn then
        id = string.sub( id, 1, 4 ) .. "-" .. string.sub( id, 5 );    -- if valid, display correctly formatted version
    else
        id = issn_copy;                                            -- if not valid, use the show the invalid issn with error message
    end
   
    text = external_link_id({link = handler.link, label = handler.label,
        prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode})
 
    if false == valid_issn then
        text = text .. ' ' .. set_error( 'bad_issn', e and 'e' or '' )            -- add an error message if the issn is invalid
    end
   
    return text
end


--[[--------------------------< A M A Z O N >------------------------------------------------------------------

Formats a link to Amazon.  Do simple error checking: asin must be mix of 10 numeric or uppercase alpha
characters.  If a mix, first character must be uppercase alpha; if all numeric, asins must be 10-digit
isbn. If 10-digit isbn, add a maintenance category so a bot or awb script can replace |asin= with |isbn=.
Error message if not 10 characters, if not isbn10, if mixed and first character is a digit.

]]

local function amazon(id, domain)
    local err_cat = ""

    if not id:match("^[%d%u][%d%u][%d%u][%d%u][%d%u][%d%u][%d%u][%d%u][%d%u][%d%u]$") then
        err_cat =  ' ' .. set_error ('bad_asin');                                -- asin is not a mix of 10 uppercase alpha and numeric characters
    else
        if id:match("^%d%d%d%d%d%d%d%d%d[%dX]$") then                                -- if 10-digit numeric (or 9 digits with terminal X)
            if check_isbn( id ) then                                                -- see if asin value is isbn10
                add_maint_cat ('ASIN');
            elseif not is_set (err_cat) then
                err_cat =  ' ' .. set_error ('bad_asin');                        -- asin is not isbn10
            end
        elseif not id:match("^%u[%d%u]+$") then
            err_cat =  ' ' .. set_error ('bad_asin');                            -- asin doesn't begin with uppercase alpha
        end
    end
    if not is_set(domain) then
        domain = "com";
    elseif in_array (domain, {'jp', 'uk'}) then            -- Japan, United Kingdom
        domain = "co." .. domain;
    elseif in_array (domain, {'au', 'br', 'mx'}) then    -- Australia, Brazil, Mexico
        domain = "com." .. domain;
    end
    local handler = cfg.id_handlers['ASIN'];
    return external_link_id({link=handler.link,
        label=handler.label, prefix=handler.prefix .. domain .. "/dp/",
        id=id, encode=handler.encode, separator = handler.separator}) .. err_cat;
end


--[[--------------------------< A R X I V >--------------------------------------------------------------------

See: http://arxiv.org/help/arxiv_identifier

format and error check arXiv identifier.  There are three valid forms of the identifier:
the first form, valid only between date codes 9108 and 0703 is:
    arXiv:<archive>.<class>/<date code><number><version>
where:
    <archive> is a string of alpha characters - may be hyphenated; no other punctuation
    <class> is a string of alpha characters - may be hyphenated; no other punctuation
    <date code> is four digits in the form YYMM where YY is the last two digits of the four-digit year and MM is the month number January = 01
        first digit of YY for this form can only 9 and 0
    <number> is a three-digit number
    <version> is a 1 or more digit number preceded with a lowercase v; no spaces (undocumented)
   
the second form, valid from April 2007 through December 2014 is:
    arXiv:<date code>.<number><version>
where:
    <date code> is four digits in the form YYMM where YY is the last two digits of the four-digit year and MM is the month number January = 01
    <number> is a four-digit number
    <version> is a 1 or more digit number preceded with a lowercase v; no spaces

the third form, valid from January 2015 is:
    arXiv:<date code>.<number><version>
where:
    <date code> and <version> are as defined for 0704-1412
    <number> is a five-digit number
]]

local function arxiv (id, class)
    local handler = cfg.id_handlers['ARXIV'];
    local year, month, version;
    local err_cat = '';
    local text;
   
    if id:match("^%a[%a%.%-]+/[90]%d[01]%d%d%d%d$") or id:match("^%a[%a%.%-]+/[90]%d[01]%d%d%d%dv%d+$") then    -- test for the 9108-0703 format w/ & w/o version
        year, month = id:match("^%a[%a%.%-]+/([90]%d)([01]%d)%d%d%d[v%d]*$");
        year = tonumber(year);
        month = tonumber(month);
        if ((not (90 < year or 8 > year)) or (1 > month or 12 < month)) or        -- if invalid year or invalid month
            ((91 == year and 7 > month) or (7 == year and 3 < month)) then        -- if years ok, are starting and ending months ok?
                err_cat = ' ' .. set_error( 'bad_arxiv' );                        -- set error message
        end
    elseif id:match("^%d%d[01]%d%.%d%d%d%d$") or id:match("^%d%d[01]%d%.%d%d%d%dv%d+$") then    -- test for the 0704-1412 w/ & w/o version
        year, month = id:match("^(%d%d)([01]%d)%.%d%d%d%d[v%d]*$");
        year = tonumber(year);
        month = tonumber(month);
        if ((7 > year) or (14 < year) or (1 > month or 12 < month)) or            -- is year invalid or is month invalid? (doesn't test for future years)
            ((7 == year) and (4 > month)) then --or                                    -- when year is 07, is month invalid (before April)?
                err_cat = ' ' .. set_error( 'bad_arxiv' );                        -- set error message
        end
    elseif id:match("^%d%d[01]%d%.%d%d%d%d%d$") or id:match("^%d%d[01]%d%.%d%d%d%d%dv%d+$") then    -- test for the 1501- format w/ & w/o version
        year, month = id:match("^(%d%d)([01]%d)%.%d%d%d%d%d[v%d]*$");
        year = tonumber(year);
        month = tonumber(month);
        if ((15 > year) or (1 > month or 12 < month)) then                        -- is year invalid or is month invalid? (doesn't test for future years)
            err_cat = ' ' .. set_error( 'bad_arxiv' );                            -- set error message
        end
    else
        err_cat = ' ' .. set_error( 'bad_arxiv' );                                -- arXiv id doesn't match any format
    end

    text = external_link_id({link = handler.link, label = handler.label,
            prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode, access=handler.access}) .. err_cat;

    if is_set (class) then
        class = ' [[' .. '//arxiv.org/archive/' .. class .. ' ' .. class .. ']]';    -- external link within square brackets, not wikilink
    else
        class = '';                                                                -- empty string for concatenation
    end
   
    return text .. class;
end

--[[--------------------------< B I O R X I V >-----------------------------------------------------------------

Format bioRxiv id and do simple error checking.  BiorXiv ids are exactly 6 digits.
The bioRxiv id is the number following the last slash in the bioRxiv-issued DOI:
http://dx.doi.org/10.1101/078733 -> 078733

]]

local function biorxiv(id)
    local handler = cfg.id_handlers['BIORXIV'];
    local err_cat =  '';                                -- presume that bioRxiv id is valid
   
    if nil == id:match("^%d%d%d%d%d%d$") then                -- if bioRxiv id has anything but six digits
        err_cat = ' ' .. set_error( 'bad_biorxiv');    -- set an error message
    end
   
    return external_link_id({link = handler.link, label = handler.label,
            prefix=handler.prefix,id=id,separator=handler.separator,
            encode=handler.encode, access=handler.access}) .. err_cat;
end

--[[--------------------------< N O R M A L I Z E _ L C C N >--------------------------------------------------

lccn normalization (http://www.loc.gov/marc/lccn-namespace.html#normalization)
1. Remove all blanks.
2. If there is a forward slash (/) in the string, remove it, and remove all characters to the right of the forward slash.
3. If there is a hyphen in the string:
    a. Remove it.
    b. Inspect the substring following (to the right of) the (removed) hyphen. Then (and assuming that steps 1 and 2 have been carried out):
        1. All these characters should be digits, and there should be six or less. (not done in this function)
        2. If the length of the substring is less than 6, left-fill the substring with zeroes until the length is six.

Returns a normalized lccn for lccn() to validate.  There is no error checking (step 3.b.1) performed in this function.
]]

local function normalize_lccn (lccn)
    lccn = lccn:gsub ("%s", "");                                    -- 1. strip whitespace

    if nil ~= string.find (lccn,'/') then
        lccn = lccn:match ("(.-)/");                                -- 2. remove forward slash and all character to the right of it
    end

    local prefix
    local suffix
    prefix, suffix = lccn:match ("(.+)%-(.+)");                        -- 3.a remove hyphen by splitting the string into prefix and suffix

    if nil ~= suffix then                                            -- if there was a hyphen
        suffix=string.rep("0", 6-string.len (suffix)) .. suffix;    -- 3.b.2 left fill the suffix with 0s if suffix length less than 6
        lccn=prefix..suffix;                                        -- reassemble the lccn
    end
   
    return lccn;
    end


--[[--------------------------< L C C N >----------------------------------------------------------------------

Format LCCN link and do simple error checking.  LCCN is a character string 8-12 characters long. The length of
the LCCN dictates the character type of the first 1-3 characters; the rightmost eight are always digits.
http://info-uri.info/registry/OAIHandler?verb=GetRecord&metadataPrefix=reg&identifier=info:lccn/

length = 8 then all digits
length = 9 then lccn[1] is lower case alpha
length = 10 then lccn[1] and lccn[2] are both lower case alpha or both digits
length = 11 then lccn[1] is lower case alpha, lccn[2] and lccn[3] are both lower case alpha or both digits
length = 12 then lccn[1] and lccn[2] are both lower case alpha

]]

local function lccn(lccn)
    local handler = cfg.id_handlers['LCCN'];
    local err_cat =  '';                                -- presume that LCCN is valid
    local id = lccn;                                    -- local copy of the lccn

    id = normalize_lccn (id);                            -- get canonical form (no whitespace, hyphens, forward slashes)
    local len = id:len();                                -- get the length of the lccn

    if 8 == len then
        if id:match("[^%d]") then                        -- if LCCN has anything but digits (nil if only digits)
            err_cat = ' ' .. set_error( 'bad_lccn' );    -- set an error message
        end
    elseif 9 == len then                                -- LCCN should be adddddddd
        if nil == id:match("%l%d%d%d%d%d%d%d%d") then            -- does it match our pattern?
            err_cat = ' ' .. set_error( 'bad_lccn' );    -- set an error message
        end
    elseif 10 == len then                                -- LCCN should be aadddddddd or dddddddddd
        if id:match("[^%d]") then                            -- if LCCN has anything but digits (nil if only digits) ...
            if nil == id:match("^%l%l%d%d%d%d%d%d%d%d") then    -- ... see if it matches our pattern
                err_cat = ' ' .. set_error( 'bad_lccn' );    -- no match, set an error message
            end
        end
    elseif 11 == len then                                -- LCCN should be aaadddddddd or adddddddddd
        if not (id:match("^%l%l%l%d%d%d%d%d%d%d%d") or id:match("^%l%d%d%d%d%d%d%d%d%d%d")) then    -- see if it matches one of our patterns
            err_cat = ' ' .. set_error( 'bad_lccn' );    -- no match, set an error message
        end
    elseif 12 == len then                                -- LCCN should be aadddddddddd
        if not id:match("^%l%l%d%d%d%d%d%d%d%d%d%d") then    -- see if it matches our pattern
            err_cat = ' ' .. set_error( 'bad_lccn' );    -- no match, set an error message
        end
    else
        err_cat = ' ' .. set_error( 'bad_lccn' );        -- wrong length, set an error message
    end

    if not is_set (err_cat) and nil ~= lccn:find ('%s') then
        err_cat = ' ' .. set_error( 'bad_lccn' );        -- lccn contains a space, set an error message
    end

    return external_link_id({link = handler.link, label = handler.label,
            prefix=handler.prefix,id=lccn,separator=handler.separator, encode=handler.encode}) .. err_cat;
end


--[[--------------------------< P M I D >----------------------------------------------------------------------

Format PMID and do simple error checking.  PMIDs are sequential numbers beginning at 1 and counting up.  This
code checks the PMID to see that it contains only digits and is less than test_limit; the value in local variable
test_limit will need to be updated periodically as more PMIDs are issued.

]]

local function pmid(id)
    local test_limit = 33000000;                        -- update this value as PMIDs approach
    local handler = cfg.id_handlers['PMID'];
    local err_cat =  '';                                -- presume that PMID is valid
   
    if id:match("[^%d]") then                            -- if PMID has anything but digits
        err_cat = ' ' .. set_error( 'bad_pmid' );        -- set an error message
    else                                                -- PMID is only digits
        local id_num = tonumber(id);                    -- convert id to a number for range testing
        if 1 > id_num or test_limit < id_num then        -- if PMID is outside test limit boundaries
            err_cat = ' ' .. set_error( 'bad_pmid' );    -- set an error message
        end
    end
   
    return external_link_id({link = handler.link, label = handler.label,
            prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode}) .. err_cat;
end


--[[--------------------------< I S _ E M B A R G O E D >------------------------------------------------------

Determines if a PMC identifier's online version is embargoed. Compares the date in |embargo= against today's date.  If embargo date is
in the future, returns the content of |embargo=; otherwise, returns and empty string because the embargo has expired or because
|embargo= was not set in this cite.

]]

local function is_embargoed (embargo)
    if is_set (embargo) then
        local lang = mw.getContentLanguage();
        local good1, embargo_date, good2, todays_date;
        good1, embargo_date = pcall( lang.formatDate, lang, 'U', embargo );
        good2, todays_date = pcall( lang.formatDate, lang, 'U' );
   
        if good1 and good2 then                                                    -- if embargo date and today's date are good dates
            if tonumber( embargo_date ) >= tonumber( todays_date ) then            -- is embargo date is in the future?
                return embargo;                                                    -- still embargoed
            else
                add_maint_cat ('embargo')
                return '';                                                        -- unset because embargo has expired
            end
        end
    end
    return '';                                                                    -- |embargo= not set return empty string
end


--[[--------------------------< P M C >------------------------------------------------------------------------

Format a PMC, do simple error checking, and check for embargoed articles.

The embargo parameter takes a date for a value. If the embargo date is in the future the PMC identifier will not
be linked to the article.  If the embargo date is today or in the past, or if it is empty or omitted, then the
PMC identifier is linked to the article through the link at cfg.id_handlers['PMC'].prefix.

PMC embargo date testing is done in function is_embargoed () which is called earlier because when the citation
has |pmc=<value> but does not have a |url= then |title= is linked with the PMC link.  Function is_embargoed ()
returns the embargo date if the PMC article is still embargoed, otherwise it returns an empty string.

PMCs are sequential numbers beginning at 1 and counting up.  This code checks the PMC to see that it contains only digits and is less
than test_limit; the value in local variable test_limit will need to be updated periodically as more PMCs are issued.

]]

local function pmc(id, embargo)
    local test_limit = 8000000;                            -- update this value as PMCs approach
    local handler = cfg.id_handlers['PMC'];
    local err_cat =  '';                                -- presume that PMC is valid
   
    local text;

    if id:match("[^%d]") then                            -- if PMC has anything but digits
        err_cat = ' ' .. set_error( 'bad_pmc' );            -- set an error message
    else                                                -- PMC is only digits
        local id_num = tonumber(id);                    -- convert id to a number for range testing
        if 1 > id_num or test_limit < id_num then        -- if PMC is outside test limit boundaries
            err_cat = ' ' .. set_error( 'bad_pmc' );        -- set an error message
        end
    end
   
    if is_set (embargo) then                                                    -- is PMC is still embargoed?
        text="[[" .. handler.link .. "|" .. handler.label .. "]]" .. handler.separator .. id .. err_cat;    -- still embargoed so no external link
    else
        text = external_link_id({link = handler.link, label = handler.label,            -- no embargo date or embargo has expired, ok to link to article
            prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode, access=handler.access}) .. err_cat;
    end
    return text;
end


--[[--------------------------< D O I >------------------------------------------------------------------------

Formats a DOI and checks for DOI errors.

DOI names contain two parts: prefix and suffix separated by a forward slash.
    Prefix: directory indicator '10.' followed by a registrant code
    Suffix: character string of any length chosen by the registrant

This function checks a DOI name for: prefix/suffix.  If the doi name contains spaces or endashes, or, if it ends
with a period or a comma, this function will emit a bad_doi error message.

DOI names are case-insensitive and can incorporate any printable Unicode characters so the test for spaces, endash,
and terminal punctuation may not be technically correct but it appears, that in practice these characters are rarely
if ever used in doi names.

]]

local function doi(id, inactive, access)
    local cat = ""
    local handler = cfg.id_handlers['DOI'];
   
    local text;
    if is_set(inactive) then
        local inactive_year = inactive:match("%d%d%d%d") or '';        -- try to get the year portion from the inactive date
        text = "[[" .. handler.link .. "|" .. handler.label .. "]]:" .. id;
        if is_set(inactive_year) then
            table.insert( z.error_categories, "!Páginas com DOI inativos desde " .. inactive_year );
        else
            table.insert( z.error_categories, "!Páginas com DOI inativos" );    -- when inactive doesn't contain a recognizable year
        end
        inactive = " (" .. cfg.messages['inactive'] .. " " .. inactive .. ")"
    else
        text = external_link_id({link = handler.link, label = handler.label,
            prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode, access=access})
        inactive = ""
    end

    if nil == id:match("^10%.[^%s–]-/[^%s–]-[^%.,]$") then    -- doi must begin with '10.', must contain a fwd slash, must not contain spaces or endashes, and must not end with period or comma
        cat = ' ' .. set_error( 'bad_doi' );
    end
    return text .. inactive .. cat
end


--[[--------------------------< H D L >------------------------------------------------------------------------

Formats an HDL with minor error checking.

HDL names contain two parts: prefix and suffix separated by a forward slash.
    Prefix: character string using any character in the UCS-2 character set except '/'
    Suffix: character string of any length using any character in the UCS-2 character set chosen by the registrant

This function checks a HDL name for: prefix/suffix.  If the HDL name contains spaces, endashes, or, if it ends
with a period or a comma, this function will emit a bad_hdl error message.

HDL names are case-insensitive and can incorporate any printable Unicode characters so the test for endashes and
terminal punctuation may not be technically correct but it appears, that in practice these characters are rarely
if ever used in HDLs.

]]

local function hdl(id, access)
    local handler = cfg.id_handlers['HDL'];
   
    local text = external_link_id({link = handler.link, label = handler.label,
            prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode, access=access})

    if nil == id:match("^[^%s–]-/[^%s–]-[^%.,]$") then                            -- hdl must contain a fwd slash, must not contain spaces, endashes, and must not end with period or comma
        text = text .. ' ' .. set_error( 'bad_hdl' );
    end
    return text;
end


--[[--------------------------< O P E N L I B R A R Y >--------------------------------------------------------

Formats an OpenLibrary link, and checks for associated errors.

]]

local function openlibrary(id, access)
    local code = id:match("^%d+([AMW])$");                                        -- only digits followed by 'A', 'M', or 'W'
    local handler = cfg.id_handlers['OL'];

    if ( code == "A" ) then
        return external_link_id({link=handler.link, label=handler.label,
            prefix=handler.prefix .. 'authors/OL',
            id=id, separator=handler.separator,    encode = handler.encode,
            access = access})
    elseif ( code == "M" ) then
        return external_link_id({link=handler.link, label=handler.label,
            prefix=handler.prefix .. 'books/OL',
            id=id, separator=handler.separator,    encode = handler.encode,
            access = access})
    elseif ( code == "W" ) then
        return external_link_id({link=handler.link, label=handler.label,
            prefix=handler.prefix .. 'works/OL',
            id=id, separator=handler.separator,    encode = handler.encode,
            access = access})
    else
        return external_link_id({link=handler.link, label=handler.label,
            prefix=handler.prefix .. 'OL',
            id=id, separator=handler.separator,    encode = handler.encode,
            access = access}) .. ' ' .. set_error( 'bad_ol' );
    end
end


--[[--------------------------< M E S S A G E _ I D >----------------------------------------------------------

Validate and format a usenet message id.  Simple error checking, looks for 'id-left@id-right' not enclosed in
'<' and/or '>' angle brackets.

]]

local function message_id (id)
    local handler = cfg.id_handlers['USENETID'];

    local text = external_link_id({link = handler.link, label = handler.label,
        prefix=handler.prefix,id=id,separator=handler.separator, encode=handler.encode})
 
    if not id:match('^.+@.+$') or not id:match('^[^<].*[^>]$')then                -- doesn't have '@' or has one or first or last character is '< or '>'
        text = text .. ' ' .. set_error( 'bad_message_id' )                        -- add an error message if the message id is invalid
    end
   
    return text
end


--[[--------------------------< O C L C >----------------------------------------------------------------------

Validate and format an oclc id.  https://www.oclc.org/batchload/controlnumber.en.html

]]

local function oclc (id)
    local handler = cfg.id_handlers['OCLC'];
    local number;
    local err_msg = '';                                                            -- empty string for concatenation
   
    if id:match('^ocm%d%d%d%d%d%d%d%d$') then                                    -- ocm prefix and 8 digits; 001 field (12 characters)
        number = id:match('ocm(%d+)');                                            -- get the number
    elseif id:match('^ocn%d%d%d%d%d%d%d%d%d$') then                                -- ocn prefix and 9 digits; 001 field (12 characters)
        number = id:match('ocn(%d+)');                                            -- get the number
    elseif id:match('^on%d%d%d%d%d%d%d%d%d%d+$') then                            -- on prefix and 10 or more digits; 001 field (12 characters)
        number = id:match('^on(%d%d%d%d%d%d%d%d%d%d+)$');                        -- get the number
    elseif id:match('^%(OCoLC%)[1-9]%d*$') then                                    -- (OCoLC) prefix and variable number digits; no leading zeros; 035 field
        number = id:match('%(OCoLC%)([1-9]%d*)');                                -- get the number
        if 9 < number:len() then
            number = nil;                                                        -- contrain to 1 to 9 digits; change this when oclc issues 10-digit numbers
        end
    elseif id:match('^%d+$') then                                                -- no prefix
        number = id;                                                            -- get the number
        if 10 < number:len() then
            number = nil;                                                        -- contrain to 1 to 10 digits; change this when oclc issues 11-digit numbers
        end
    end

    if number then                                                                -- proper format
        id = number;                                                            -- exclude prefix, if any, from external link
    else
        err_msg = ' ' .. set_error( 'bad_oclc' )                                -- add an error message if the id is malformed
    end
   
    local text = external_link_id({link=handler.link, label=handler.label,
        prefix=handler.prefix, id=id, separator=handler.separator, encode=handler.encode}) .. err_msg;

    return text;
end


--[[--------------------------< B I B C O D E >--------------------------------------------------------------------

Validates (sort of) and formats a bibcode id.

Format for bibcodes is specified here: http://adsabs.harvard.edu/abs_doc/help_pages/data.html#bibcodes

But, this: 2015arXiv151206696F is apparently valid so apparently, the only things that really matter are length, 19 characters
and first four digits must be a year.  This function makes these tests:
    length must be 19 characters
    characters in position
        1–4 must be digits and must represent a year in the range of 1000 – next year
        5 must be a letter
        6 must be letter, ampersand, or dot (ampersand cannot directly precede a dot; &. )
        7–8 must be letter, digit, ampersand, or dot (ampersand cannot directly precede a dot; &. )
        9–18 must be letter, digit, or dot
        19 must be a letter or dot

]]

local function bibcode (id, access)
    local handler = cfg.id_handlers['BIBCODE'];
    local err_type;
    local year;

    local text = external_link_id({link=handler.link, label=handler.label,
        prefix=handler.prefix, id=id, separator=handler.separator, encode=handler.encode,
        access=access});
   
    if 19 ~= id:len() then
        err_type = 'length';
    else
        year = id:match ("^(%d%d%d%d)[%a][%a&%.][%a&%.%d][%a&%.%d][%a%d%.]+[%a%.]$")    --
        if not year then                                                        -- if nil then no pattern match
            err_type = 'value';                                                    -- so value error
        else
            local next_year = tonumber(os.date ('%Y'))+1;                        -- get the current year as a number and add one for next year
            year = tonumber (year);                                                -- convert year portion of bibcode to a number
            if (1000 > year) or (year > next_year) then
                err_type = 'year';                                                -- year out of bounds
            end
            if id:find('&%.') then
                err_type = 'journal';                                            -- journal abbreviation must not have '&.' (if it does its missing a letter)
            end
        end
    end

    if is_set (err_type) then                                                    -- if there was an error detected
        text = text .. ' ' .. set_error( 'bad_bibcode', {err_type});
    end
    return text;
end

--[[--------------------------< C I T E S E E R X >------------------------------------------------------------

CiteSeerX use their own notion of "doi" (not to be confused with the identifiers resolved via doi.org).

The description of the structure of this identifier can be found at Help_talk:Citation_Style_1#CiteSeerX_id_structure
]]

local function citeseerx (id)
    local handler = cfg.id_handlers['CITESEERX'];
    local matched;
   
    local text = external_link_id({link=handler.link, label=handler.label, q=handler.q, redirect=handler.redirect,
        prefix=handler.prefix, id=id, separator=handler.separator, encode=handler.encode,
        access=handler.access});
   
    matched = id:match ("^10%.1%.1%.[1-9]%d?%d?%d?%.[1-9]%d?%d?%d?$");
    if not matched then
        text = text .. ' ' .. set_error( 'bad_citeseerx' );
    end
    return text;
end

--[[--------------------------< B U I L D _ I D _ L I S T >--------------------------------------------------------

Takes a table of IDs created by extract_ids() and turns it into a table of formatted ID outputs.

inputs:
    id_list – table of identifiers built by extract_ids()
    options – table of various template parameter values used to modify some manually handled identifiers

]]

local function build_id_list( id_list, options )
    local new_list, handler = {};

    local function fallback(k) return { __index = function(t,i) return cfg.id_handlers[k][i] end } end;
   
    for k, v in pairs( id_list ) do                                                -- k is uc identifier name as index to cfg.id_handlers; e.g. cfg.id_handlers['ISBN'], v is a table
       
        -- fallback to read-only cfg
        handler = setmetatable( { ['id'] = v, ['access'] = options.IdAccessLevels[k] }, fallback(k) );

        if handler.mode == 'external' then
            table.insert( new_list, {handler.label, external_link_id( handler ) } );
        elseif handler.mode == 'internal' then
            table.insert( new_list, {handler.label, internal_link_id( handler ) } );
        elseif handler.mode ~= 'manual' then
            error( cfg.messages['unknown_ID_mode'] );
        elseif k == 'BIBCODE' then
            table.insert( new_list, {handler.label, bibcode( v, handler.access ) } );
        elseif k == 'BIORXIV' then
            table.insert( new_list, {handler.label, biorxiv( v ) } );
        elseif k == 'CITESEERX' then
            table.insert( new_list, {handler.label, citeseerx( v ) } );
        elseif k == 'DOI' then
            table.insert( new_list, {handler.label, doi( v, options.DoiBroken, handler.access ) } );
        elseif k == 'HDL' then
            table.insert( new_list, {handler.label, hdl( v, handler.access ) } );
        elseif k == 'ARXIV' then
            table.insert( new_list, {handler.label, arxiv( v, options.Class ) } );
        elseif k == 'ASIN' then
            table.insert( new_list, {handler.label, amazon( v, options.ASINTLD ) } );
        elseif k == 'LCCN' then
            table.insert( new_list, {handler.label, lccn( v ) } );
        elseif k == 'OL' or k == 'OLA' then
            table.insert( new_list, {handler.label, openlibrary( v, handler.access ) } );
        elseif k == 'PMC' then
            table.insert( new_list, {handler.label, pmc( v, options.Embargo ) } );
        elseif k == 'PMID' then
            table.insert( new_list, {handler.label, pmid( v ) } );
        elseif k == 'OCLC' then
            table.insert( new_list, {handler.label, oclc( v ) } );
        elseif k == 'ISMN' then
            table.insert( new_list, {handler.label, ismn( v ) } );
        elseif k == 'ISSN' then
            table.insert( new_list, {handler.label, issn( v ) } );
        elseif k == 'EISSN' then
            table.insert( new_list, {handler.label, issn( v, true ) } );        -- true distinguishes eissn from issn
        elseif k == 'ISBN' then
            local ISBN = internal_link_id( handler );
            local check;
            local err_type = '';
--            if not check_isbn( v ) and not is_set(options.IgnoreISBN) then
--                ISBN = ISBN .. set_error( 'bad_isbn', {}, false, " ", "" );
--            end
            check, err_type = check_isbn( v );
            if not check then
                if is_set(options.IgnoreISBN) then                                -- ISBN is invalid; if |ignore-isbn-error= set
                    add_maint_cat ('ignore_isbn_err');                            -- ad a maint category
                else
                    ISBN = ISBN .. set_error( 'bad_isbn', {err_type}, false, " ", "" );    -- else display an error message
                end
            elseif is_set(options.IgnoreISBN) then                                -- ISBN is OK; if |ignore-isbn-error= set
                    add_maint_cat ('ignore_isbn_err');                            -- because |ignore-isbn-error= unnecessary
            end
            table.insert( new_list, {handler.label, ISBN } );               
        elseif k == 'USENETID' then
            table.insert( new_list, {handler.label, message_id( v ) } );
        else
            error( cfg.messages['unknown_manual_ID'] );
        end
    end
   
    local function comp( a, b )    -- used in following table.sort()
        return a[1] < b[1];
    end
   
    table.sort( new_list, comp );
    for k, v in ipairs( new_list ) do
        new_list[k] = v[2];
    end
   
    return new_list;
end


--[[--------------------------< E X T R A C T _ I D S >------------------------------------------------------------

Populates ID table from arguments using configuration settings. Loops through cfg.id_handlers and searches args for
any of the parameters listed in each cfg.id_handlers['...'].parameters.  If found, adds the parameter and value to
the identifier list.  Emits redundant error message is more than one alias exists in args

]]

local function extract_ids( args )
    local id_list = {};                                                            -- list of identifiers found in args
    for k, v in pairs( cfg.id_handlers ) do                                        -- k is uc identifier name as index to cfg.id_handlers; e.g. cfg.id_handlers['ISBN'], v is a table
        v = select_one( args, v.parameters, 'redundant_parameters' );            -- v.parameters is a table of aliases for k; here we pick one from args if present
        if is_set(v) then id_list[k] = v; end                                    -- if found in args, add identifier to our list
    end
    return id_list;
end

--[[--------------------------< E X T R A C T _ I D _ A C C E S S _ L E V E L S >--------------------------------------

Fetches custom id access levels from arguments using configuration settings.
Parameters which have a predefined access level (e.g. arxiv) do not use this
function as they are directly rendered as free without using an additional parameter.

]]

local function extract_id_access_levels( args, id_list )
    local id_accesses_list = {};
    for k, v in pairs( cfg.id_handlers ) do
        local access_param = v.custom_access;
        local k_lower = string.lower(k);
        if is_set(access_param) then
            local access_level = args[access_param];
            if is_set(access_level) then
                if not in_array (access_level:lower(), cfg.keywords['id-access']) then
                    table.insert( z.message_tail, { set_error( 'invalid_param_val', {access_param, access_level}, true ) } );   
                    access_level = nil;
                end
                if not is_set(id_list[k]) then
                    table.insert( z.message_tail, { set_error( 'param_access_requires_param', {k_lower}, true ) } );
                end
                if is_set(access_level) then
                    access_level = access_level:lower();
                end
                id_accesses_list[k] = access_level;
            end
        end
    end
    return id_accesses_list;
end


--[[--------------------------< S E T _ S E L E C T E D _ M O D U L E S >--------------------------------------

Sets local cfg table and imported functions table to same (live or sandbox) as that used by the other modules.

]]

local function set_selected_modules (cfg_table_ptr, utilities_page_ptr)
    cfg = cfg_table_ptr;

    is_set = utilities_page_ptr.is_set;                                            -- import functions from select Module:Citation/CS1/Utilities module
    in_array = utilities_page_ptr.in_array;
    set_error = utilities_page_ptr.set_error;
    select_one = utilities_page_ptr.select_one;
    add_maint_cat = utilities_page_ptr.add_maint_cat;
    substitute = utilities_page_ptr.substitute;

    z = utilities_page_ptr.z;                                                    -- table of tables in Module:Citation/CS1/Utilities
end



return {
    build_id_list = build_id_list,
    extract_ids = extract_ids,
    extract_id_access_levels = extract_id_access_levels,
    is_embargoed = is_embargoed;
    set_selected_modules = set_selected_modules;
    }