mirror of
https://gerrit.wikimedia.org/r/mediawiki/extensions/Scribunto
synced 2024-11-27 09:40:12 +00:00
a0a1b05b74
The following are now correctly escaped: * Blank lines (including those with only tabs) * ---- at the start of a line Bug: 53658 Change-Id: Ib000ff4f21f76c310741de89de0e0b66f6450344
283 lines
6.7 KiB
Lua
283 lines
6.7 KiB
Lua
local mwtext = {}
|
|
local php
|
|
local options
|
|
|
|
local util = require 'libraryUtil'
|
|
local checkType = util.checkType
|
|
local checkTypeForNamedArg = util.checkTypeForNamedArg
|
|
|
|
function mwtext.setupInterface( opts )
|
|
-- Boilerplate
|
|
mwtext.setupInterface = nil
|
|
php = mw_interface
|
|
mw_interface = nil
|
|
options = opts
|
|
|
|
-- Register this library in the "mw" global
|
|
mw = mw or {}
|
|
mw.text = mwtext
|
|
|
|
package.loaded['mw.text'] = mwtext
|
|
end
|
|
|
|
function mwtext.trim( s, charset )
|
|
charset = charset or '\t\r\n\f '
|
|
s = mw.ustring.gsub( s, '^[' .. charset .. ']*(.-)[' .. charset .. ']*$', '%1' )
|
|
return s
|
|
end
|
|
|
|
local htmlencode_map = {
|
|
['>'] = '>',
|
|
['<'] = '<',
|
|
['&'] = '&',
|
|
['"'] = '"',
|
|
["'"] = ''',
|
|
['\194\160'] = '&#nbsp;',
|
|
}
|
|
local htmldecode_map = {}
|
|
for k, v in pairs( htmlencode_map ) do
|
|
htmldecode_map[v] = k
|
|
end
|
|
local decode_named_entities = nil
|
|
|
|
function mwtext.encode( s, charset )
|
|
charset = charset or '<>&"\'\194\160'
|
|
s = mw.ustring.gsub( s, '[' .. charset .. ']', function ( m )
|
|
if not htmlencode_map[m] then
|
|
local e = string.format( '&#%d;', mw.ustring.codepoint( m ) )
|
|
htmlencode_map[m] = e
|
|
htmldecode_map[e] = m
|
|
end
|
|
return htmlencode_map[m]
|
|
end )
|
|
return s
|
|
end
|
|
|
|
function mwtext.decode( s, decodeNamedEntities )
|
|
local dec
|
|
if decodeNamedEntities then
|
|
if decode_named_entities == nil then
|
|
decode_named_entities = php.getEntityTable()
|
|
setmetatable( decode_named_entities, { __index = htmldecode_map } )
|
|
end
|
|
dec = decode_named_entities
|
|
else
|
|
dec = htmldecode_map
|
|
end
|
|
-- string.gsub is safe here, because only ASCII chars are in the pattern
|
|
s = string.gsub( s, '(&(#?x?)([a-zA-Z0-9]+);)', function ( m, flg, name )
|
|
if not dec[m] then
|
|
local n = nil
|
|
if flg == '#' then
|
|
n = tonumber( name, 10 )
|
|
elseif flg == '#x' then
|
|
n = tonumber( name, 16 )
|
|
end
|
|
if n and n <= 0x10ffff then
|
|
n = mw.ustring.char( n )
|
|
if n then
|
|
htmldecode_map[m] = n
|
|
htmlencode_map[n] = m
|
|
end
|
|
end
|
|
end
|
|
return dec[m]
|
|
end )
|
|
return s
|
|
end
|
|
|
|
local nowikiRepl1 = {
|
|
['"'] = '"',
|
|
['&'] = '&',
|
|
["'"] = ''',
|
|
['<'] = '<',
|
|
['='] = '=',
|
|
['>'] = '>',
|
|
['['] = '[',
|
|
[']'] = ']',
|
|
['{'] = '{',
|
|
['|'] = '|',
|
|
['}'] = '}',
|
|
}
|
|
|
|
local nowikiRepl2 = {
|
|
["\n#"] = "\n#", ["\r#"] = "\r#",
|
|
["\n*"] = "\n*", ["\r*"] = "\r*",
|
|
["\n:"] = "\n:", ["\r:"] = "\r:",
|
|
["\n;"] = "\n;", ["\r;"] = "\r;",
|
|
["\n "] = "\n ", ["\r "] = "\r ",
|
|
["\n\n"] = "\n ", ["\r\n"] = " \n",
|
|
["\n\r"] = "\n ", ["\r\r"] = "\r ",
|
|
["\n\t"] = "\n	", ["\r\t"] = "\r	",
|
|
}
|
|
|
|
local nowikiReplMagic = {}
|
|
for sp, esc in pairs( {
|
|
[' '] = ' ',
|
|
['\t'] = '	',
|
|
['\r'] = ' ',
|
|
['\n'] = ' ',
|
|
['\f'] = '',
|
|
} ) do
|
|
nowikiReplMagic['ISBN' .. sp] = 'ISBN' .. esc
|
|
nowikiReplMagic['RFC' .. sp] = 'RFC' .. esc
|
|
nowikiReplMagic['PMID' .. sp] = 'PMID' .. esc
|
|
end
|
|
|
|
function mwtext.nowiki( s )
|
|
-- string.gsub is safe here, because we're only caring about ASCII chars
|
|
s = string.gsub( s, '["&\'<=>%[%]{|}]', nowikiRepl1 )
|
|
s = '\n' .. s
|
|
s = string.gsub( s, '[\r\n][#*:; \n\r\t]', nowikiRepl2 )
|
|
s = string.gsub( s, '([\r\n])%-%-%-%-', '%1----' )
|
|
s = string.sub( s, 2 )
|
|
s = string.gsub( s, '__', '__' )
|
|
s = string.gsub( s, '://', '://' )
|
|
s = string.gsub( s, 'ISBN%s', nowikiReplMagic )
|
|
s = string.gsub( s, 'RFC%s', nowikiReplMagic )
|
|
s = string.gsub( s, 'PMID%s', nowikiReplMagic )
|
|
for k, v in pairs( options.nowiki_protocols ) do
|
|
s = string.gsub( s, k, v )
|
|
end
|
|
|
|
return s
|
|
end
|
|
|
|
function mwtext.tag( name, attrs, content )
|
|
local named = false
|
|
if type( name ) == 'table' then
|
|
named = true
|
|
name, attrs, content = name.name, name.attrs, name.content
|
|
checkTypeForNamedArg( 'tag', 'name', name, 'string' )
|
|
checkTypeForNamedArg( 'tag', 'attrs', attrs, 'table', true )
|
|
else
|
|
checkType( 'tag', 1, name, 'string' )
|
|
checkType( 'tag', 2, attrs, 'table', true )
|
|
end
|
|
|
|
local ret = { '<' .. name }
|
|
for k, v in pairs( attrs or {} ) do
|
|
if type( k ) ~= 'string' then
|
|
error( "bad named argument attrs to 'tag' (keys must be strings, found " .. type( k ) .. ")", 2 )
|
|
end
|
|
if string.match( k, '[\t\r\n\f /<>"\'=]' ) then
|
|
error( "bad named argument attrs to 'tag' (invalid key '" .. k .. "')", 2 )
|
|
end
|
|
local tp = type( v )
|
|
if tp == 'boolean' then
|
|
if v then
|
|
ret[#ret+1] = ' ' .. k
|
|
end
|
|
elseif tp == 'string' or tp == 'number' then
|
|
ret[#ret+1] = string.format( ' %s="%s"', k, mwtext.encode( tostring( v ) ) )
|
|
else
|
|
error( "bad named argument attrs to 'tag' (value for key '" .. k .. "' may not be " .. tp .. ")", 2 )
|
|
end
|
|
end
|
|
|
|
local tp = type( content )
|
|
if content == nil then
|
|
ret[#ret+1] = '>'
|
|
elseif content == false then
|
|
ret[#ret+1] = ' />'
|
|
elseif tp == 'string' or tp == 'number' then
|
|
ret[#ret+1] = '>'
|
|
ret[#ret+1] = content
|
|
ret[#ret+1] = '</' .. name .. '>'
|
|
else
|
|
if named then
|
|
checkTypeForNamedArg( 'tag', 'content', content, 'string, number, nil, or false' )
|
|
else
|
|
checkType( 'tag', 3, content, 'string, number, nil, or false' )
|
|
end
|
|
end
|
|
|
|
return table.concat( ret )
|
|
end
|
|
|
|
function mwtext.unstrip( s )
|
|
return php.unstrip( s )
|
|
end
|
|
|
|
function mwtext.split( text, pattern, plain )
|
|
local ret = {}
|
|
for m in mwtext.gsplit( text, pattern, plain ) do
|
|
ret[#ret+1] = m
|
|
end
|
|
return ret
|
|
end
|
|
|
|
function mwtext.gsplit( text, pattern, plain )
|
|
local s, l = 1, mw.ustring.len( text )
|
|
return function ()
|
|
if s then
|
|
local e, n = mw.ustring.find( text, pattern, s, plain )
|
|
local ret
|
|
if not e then
|
|
ret = mw.ustring.sub( text, s )
|
|
s = nil
|
|
elseif n < e then
|
|
-- Empty separator!
|
|
ret = mw.ustring.sub( text, s, e )
|
|
if e < l then
|
|
s = e + 1
|
|
else
|
|
s = nil
|
|
end
|
|
else
|
|
ret = e > s and mw.ustring.sub( text, s, e - 1 ) or ''
|
|
s = n + 1
|
|
end
|
|
return ret
|
|
end
|
|
end, nil, nil
|
|
end
|
|
|
|
function mwtext.listToText( list, separator, conjunction )
|
|
separator = separator or options.comma
|
|
conjunction = conjunction or options['and']
|
|
local n = #list
|
|
|
|
local ret
|
|
if n > 1 then
|
|
local tmp = list[n-1]
|
|
list[n-1] = list[n-1] .. conjunction .. list[n]
|
|
ret = table.concat( list, separator, 1, n - 1 )
|
|
list[n-1] = tmp
|
|
else
|
|
ret = tostring( list[1] or '' )
|
|
end
|
|
|
|
return ret
|
|
end
|
|
|
|
function mwtext.truncate( text, length, ellipsis, adjustLength )
|
|
local l = mw.ustring.len( text )
|
|
if l <= math.abs( length ) then
|
|
return text
|
|
end
|
|
|
|
ellipsis = ellipsis or options.ellipsis
|
|
local elen = 0
|
|
if adjustLength then
|
|
elen = mw.ustring.len( ellipsis )
|
|
end
|
|
|
|
local ret
|
|
if math.abs( length ) <= elen then
|
|
ret = ellipsis
|
|
elseif length > 0 then
|
|
ret = mw.ustring.sub( text, 1, length - elen ) .. ellipsis
|
|
else
|
|
ret = ellipsis .. mw.ustring.sub( text, length + elen )
|
|
end
|
|
|
|
if mw.ustring.len( ret ) < l then
|
|
return ret
|
|
else
|
|
return text
|
|
end
|
|
end
|
|
|
|
return mwtext
|