Modul:Text
--[=[ 2015-05-22 Text utilities ]=]
local Text = { } local PatternCJK = false local PatternCombined = false local PatternLatin = false local PatternTerminated = false local QuoteLang = false local QuoteType = false local RangesLatin = false local SeekQuote = false
local function factoryQuote()
-- Create quote definitions QuoteLang = { af = "bd", ar = "la", be = "labd", bg = "bd", ca = "la", cs = "bd", da = "bd", de = "bd", dsb = "bd", et = "bd", el = "lald", en = "ld", es = "la", eu = "la", fi = "rd", fr = "la", ga = "ld", he = "ldla", hr = "bd", hsb = "bd", hu = "bd", hy = "labd", id = "rd", is = "bd", it = "ld", ja = "x300C", ka = "bd", ko = "ld", lt = "bd", lv = "bd", nl = "ld", nn = "la", no = "la", pl = "bdla", pt = "lald", ro = "bdla", ru = "labd", sk = "bd", sl = "bd", sq = "la", sr = "bd", sv = "rd", th = "ld", tr = "ld", uk = "la", zh = "ld", ["de-ch"] = "la", ["en-gb"] = "lsld", ["en-us"] = "ld", ["fr-ch"] = "la", ["it-ch"] = "la", ["pt-br"] = "ldla", ["zh-tw"] = "x300C", ["zh-zh"] = "ld" } QuoteType = { bd = { { 8222, 8220 }, { 8218, 8217 } }, bdla = { { 8222, 8220 }, { 171, 187 } }, la = { { 171, 187 }, { 8249, 8250 } }, labd = { { 171, 187 }, { 8222, 8220 } }, lald = { { 171, 187 }, { 8220, 8221 } }, ld = { { 8220, 8221 }, { 8216, 8217 } }, ldla = { { 8220, 8221 }, { 171, 187 } }, lsld = { { 8216, 8217 }, { 8220, 8221 } }, rd = { { 8221, 8221 }, { 8217, 8217 } }, x300C = { { 0x300C, 0x300D }, { 0x300E, 0x300F } } } return r
end -- factoryQuote()
local function fiatQuote( apply, alien, advance )
-- Quote text -- Parameter: -- apply -- string, with text -- alien -- string, with language code -- advance -- number, with level 1 or 2 local r = apply local suite if not QuoteLang then factoryQuote() end suite = QuoteLang[ alien ] if not suite then local slang = alien:match( "^(%l+)-" ) if slang then suite = QuoteLang[ slang ] end if not suite then suite = QuoteLang[ "en" ] end end if suite then local quotes = QuoteType[ suite ] if quotes then quotes = quotes[ advance ] if quotes then r = mw.ustring.format( "%s%s%s", mw.ustring.char( quotes[ 1 ] ), apply, mw.ustring.char( quotes[ 2 ] ) ) end else mw.log( "fiatQuote() " .. suite ) end end return r
end -- fiatQuote()
Text.concatParams = function ( args, apply, adapt )
-- Concat list items into one string -- Parameter: -- args -- table (sequence) with numKey=string -- apply -- string (optional); separator (default: "|") -- adapt -- string (optional); format including "%s" -- Returns: string local collect = { } for k, v in pairs( args ) do if type( k ) == "number" then v = mw.text.trim( v ) if v ~= "" then if adapt then v = mw.ustring.format( adapt, v ) end table.insert( collect, v ) end end end return table.concat( collect, apply or "|" )
end -- Text.concatParams()
Text.containsCJK = function ( analyse )
-- Is any CJK code within? -- Parameter: -- analyse -- string -- Returns: true, if CJK detected local r if not patternCJK then patternCJK = mw.ustring.char( 91, 13312, 45, 40959, 131072, 45, 178207, 93 ) end if mw.ustring.find( analyse, patternCJK ) then r = true else r = false end return r
end -- Text.containsCJK()
Text.isLatinRange = function ( adjust )
-- Are characters expected to be latin or symbols within latin texts? -- Precondition: -- adjust -- string, or nil for initialization -- Returns: true, if valid for latin only local r if not RangesLatin then RangesLatin = { { 7, 687 }, { 7531, 7578 }, { 7680, 7935 }, { 8194, 8250 } } end if not PatternLatin then local range PatternLatin = "^[" for i = 1, #RangesLatin do range = RangesLatin[ i ] PatternLatin = PatternLatin .. mw.ustring.char( range[ 1 ], 45, range[ 2 ] ) end -- for i PatternLatin = PatternLatin .. "]*$" end if adjust then if mw.ustring.match( adjust, PatternLatin ) then r = true else r = false end end return r
end -- Text.isLatinRange()
Text.isQuote = function ( ask )
-- Is this character any quotation mark? -- Parameter: -- ask -- string, with single character -- Returns: true, if ask is quotation mark local r if not SeekQuote then SeekQuote = mw.ustring.char( 34, -- " 39, -- ' 171, -- laquo 187, -- raquo 8216, -- lsquo 8217, -- rsquo 8218, -- sbquo 8220, -- ldquo 8221, -- rdquo 8222, -- bdquo 8249, -- lsaquo 8250, -- rsaquo 0x300C, -- CJK 0x300D, -- CJK 0x300E, -- CJK 0x300F ) -- CJK end if ask == "" then r = false elseif mw.ustring.find( SeekQuote, ask, 1, true ) then r = true else r = false end return r
end -- Text.isQuote()
Text.listToText = function ( args, adapt )
-- Format list items similar to mw.text.listToText() -- Parameter: -- args -- table (sequence) with numKey=string -- adapt -- string (optional); format including "%s" -- Returns: string local collect = { } for k, v in pairs( args ) do if type( k ) == "number" then v = mw.text.trim( v ) if v ~= "" then if adapt then v = mw.ustring.format( adapt, v ) end table.insert( collect, v ) end end end return mw.text.listToText( collect )
end -- Text.listToText()
Text.quoteUnquoted = function ( apply, alien, advance )
-- Quote text, if not yet quoted -- Parameter: -- apply -- string, with text -- alien -- string, with language code, or nil -- advance -- number, with level 1 or 2, or nil -- Returns: string; possibly quoted local r = mw.text.trim( apply ) local s = mw.ustring.sub( r, 1, 1 ) if s ~= "" and not Text.isQuote( s ) then s = mw.ustring.sub( r, -1, 1 ) if not Text.isQuote( s ) then local mode, slang if type( alien ) == "string" then slang = alien:lower() else slang = mw.language.getContentLanguage():getCode() end if advance == 2 then mode = 2 else mode = 1 end r = fiatQuote( r, slang, mode ) end end return r
end -- Text.quoteUnquoted()
Text.removeDiacritics = function ( adjust )
-- Remove all diacritics -- Parameter: -- adjust -- string -- Returns: string; all latin letters should be ASCII -- or basic greek or cyrillic or symbols etc. local cleanup, decomposed if not PatternCombined then PatternCombined = mw.ustring.char( 91, 0x0300, 45, 0x036F, 0x1AB0, 45, 0x1AFF, 0x1DC0, 45, 0x1DFF, 0xFE20, 45, 0xFE2F, 93 ) end decomposed = mw.ustring.toNFD( adjust ) cleanup = mw.ustring.gsub( decomposed, PatternCombined, "" ) return mw.ustring.toNFC( cleanup )
end -- Text.removeDiacritics()
Text.sentenceTerminated = function ( analyse )
-- Is string terminated by dot, question or exclamation mark? -- Quotation, link termination and so on granted -- Parameter: -- analyse -- string -- Returns: true, if sentence terminated local r if not PatternTerminated then PatternTerminated = mw.ustring.char( 91, 12290, 65281, 65294, 65311 ) .. "!%.%?…][\"'%]‹›«»‘’“”]*$" end if mw.ustring.find( analyse, PatternTerminated ) then r = true else r = false end return r
end -- Text.sentenceTerminated()
Text.ucfirstAll = function ( adjust )
-- Capitalize all words -- Precondition: -- adjust -- string -- Returns: string with all first letters in upper case local r = " " .. adjust local i = 1 local c, j, m if adjust:find( "&" ) then r = r:gsub( "&", "&" ) :gsub( "<", "<" ) :gsub( ">", ">" ) :gsub( " ", " " ) :gsub( " ", " " ) :gsub( "", "" ) :gsub( "", "" ) :gsub( "", "" ) :gsub( "", "" ) m = true end while i do i = mw.ustring.find( r, "%W%l", i ) if i then j = i + 1 c = mw.ustring.upper( mw.ustring.sub( r, j, j ) ) r = string.format( "%s%s%s", mw.ustring.sub( r, 1, i ), c, mw.ustring.sub( r, i + 2 ) ) i = j end end -- while i r = r:sub( 2 ) if m then r = r:gsub( "&", "&" ) :gsub( "<", "<" ) :gsub( ">", ">" ) :gsub( " ", " " ) :gsub( " ", " " ) :gsub( "", "" ) :gsub( "", "" ) :gsub( "", "" ) :gsub( "", "" ) :gsub( "&#X(%x+);", "&#x%1;" ) end return r
end -- Text.ucfirstAll()
Text.uprightNonlatin = function ( adjust )
-- Ensure non-italics for non-latin text parts
-- One single greek letter might be granted
-- Precondition:
-- adjust -- string
-- Returns: string with non-latin parts enclosed in
local r
Text.isLatinRange()
if mw.ustring.match( adjust, PatternLatin ) then
-- latin only, horizontal dashes, quotes
r = adjust
else
local c
local j = false
local k = 1
local m = false
local n = mw.ustring.len( adjust )
local span = "%s%s%s"
local flat = function ( a )
-- isLatin
local range
for i = 1, #RangesLatin do
range = RangesLatin[ i ]
if a >= range[ 1 ] and a <= range[ 2 ] then
return true
end
end -- for i
end -- flat()
local form = function ( a )
return string.format( span,
r,
mw.ustring.sub( adjust, k, j - 1 ),
mw.ustring.sub( adjust, j, a ) )
end -- form()
r = ""
for i = 1, n do
c = mw.ustring.codepoint( adjust, i, i )
if c > 64 or c == 38 or c == 60 then -- '&' '<'
if flat( c ) then
if j then
if m then
if i == m then
-- single greek letter.
j = false
end
m = false
end
if j then
local nx = i - 1
local s = ""
for ix = nx, 1, -1 do
c = mw.ustring.sub( adjust, ix, ix )
if c == " " or c == "(" then
nx = nx - 1
s = c .. s
else
break -- for ix
end
end -- for ix
r = form( nx ) .. s
j = false
k = i
end
end
elseif not j then
j = i
if c >= 880 and c <= 1023 then
-- single greek letter?
m = i + 1
else
m = false
end
end
elseif m then
m = m + 1
end
end -- for i
if j and ( not m or m < n ) then
r = form( n )
else
r = r .. mw.ustring.sub( adjust, k )
end
end
return r
end -- Text.uprightNonlatin()
Text.test = function ( about )
local r if about == "quote" then factoryQuote() r = { } r.QuoteLang = QuoteLang r.QuoteType = QuoteType end return r
end -- Text.test()
-- Export local p = { }
function p.concatParams( frame )
local args local template = frame.args.template if type( template ) == "string" then template = mw.text.trim( template ) template = ( template == "1" ) end if template then args = frame:getParent().args else args = frame.args end return Text.concatParams( args, frame.args.separator, frame.args.format )
end
function p.containsCJK( frame )
return Text.containsCJK( frame.args[ 1 ] or "" ) and "1" or ""
end
function p.isLatinRange( frame )
return Text.isLatinRange( frame.args[1] or "" ) and "1" or ""
end
function p.isQuote( frame )
return Text.isQuote( frame.args[1] or "" ) and "1" or ""
end
function p.listToText( frame )
local args local template = frame.args.template if type( template ) == "string" then template = mw.text.trim( template ) template = ( template == "1" ) end if template then args = frame:getParent().args else args = frame.args end return Text.listToText( args, frame.args.format )
end
function p.quoteUnquoted( frame )
local slang = frame.args[2] if type( slang ) == "string" then slang = mw.text.trim( slang ) end return Text.quoteUnquoted( frame.args[1] or "", slang, tonumber( frame.args[3] ) )
end
function p.removeDiacritics( frame )
return Text.removeDiacritics( frame.args[1] or "" )
end
function p.sentenceTerminated( frame )
return Text.sentenceTerminated( frame.args[ 1 ] or "" ) and "1" or ""
end
function p.ucfirstAll( frame )
return Text.ucfirstAll( frame.args[ 1 ] or "" )
end
function p.uprightNonlatin( frame )
return Text.uprightNonlatin( frame.args[ 1 ] or "" )
end
function p.zip(frame) local lists = {} local seps = {} local defaultsep = frame.args["sep"] or "" local innersep = frame.args["isep"] or "" local outersep = frame.args["osep"] or ""
-- Parameter parsen for k, v in pairs(frame.args) do local knum = tonumber(k) if knum then lists[knum] = v else if string.sub(k, 1, 3) == "sep" then local sepnum = tonumber(string.sub(k, 4)) if sepnum then seps[sepnum] = v end end end end -- sofern keine expliziten Separatoren angegeben sind, den Standardseparator verwenden for i = 1, math.max(#seps, #lists) do if not seps[i] then seps[i] = defaultsep end end
-- Listen splitten local maxListLen = 0 for i = 1, #lists do lists[i] = mw.text.split(lists[i], seps[i]) if #lists[i] > maxListLen then maxListLen = #lists[i] end end
local result = "" for i = 1, maxListLen do if i ~= 1 then result = result .. outersep end for j = 1, #lists do if j ~= 1 then result = result .. innersep end result = result .. (lists[j][i] or "") end end return result end
p.Text = function ()
return Text
end -- p.Text
return p