aboutsummaryrefslogtreecommitdiff
path: root/lib/dokuwiki/inc/parser/parser.php
diff options
context:
space:
mode:
Diffstat (limited to 'lib/dokuwiki/inc/parser/parser.php')
-rw-r--r--lib/dokuwiki/inc/parser/parser.php956
1 files changed, 0 insertions, 956 deletions
diff --git a/lib/dokuwiki/inc/parser/parser.php b/lib/dokuwiki/inc/parser/parser.php
deleted file mode 100644
index a78b08a29..000000000
--- a/lib/dokuwiki/inc/parser/parser.php
+++ /dev/null
@@ -1,956 +0,0 @@
-<?php
-if(!defined('DOKU_INC')) die('meh.');
-require_once DOKU_INC . 'inc/parser/lexer.php';
-require_once DOKU_INC . 'inc/parser/handler.php';
-
-
-/**
- * Define various types of modes used by the parser - they are used to
- * populate the list of modes another mode accepts
- */
-global $PARSER_MODES;
-$PARSER_MODES = array(
- // containers are complex modes that can contain many other modes
- // hr breaks the principle but they shouldn't be used in tables / lists
- // so they are put here
- 'container' => array('listblock','table','quote','hr'),
-
- // some mode are allowed inside the base mode only
- 'baseonly' => array('header'),
-
- // modes for styling text -- footnote behaves similar to styling
- 'formatting' => array('strong', 'emphasis', 'underline', 'monospace',
- 'subscript', 'superscript', 'deleted', 'footnote'),
-
- // modes where the token is simply replaced - they can not contain any
- // other modes
- 'substition' => array('acronym','smiley','wordblock','entity',
- 'camelcaselink', 'internallink','media',
- 'externallink','linebreak','emaillink',
- 'windowssharelink','filelink','notoc',
- 'nocache','multiplyentity','quotes','rss'),
-
- // modes which have a start and end token but inside which
- // no other modes should be applied
- 'protected' => array('preformatted','code','file','php','html','htmlblock','phpblock'),
-
- // inside this mode no wiki markup should be applied but lineendings
- // and whitespace isn't preserved
- 'disabled' => array('unformatted'),
-
- // used to mark paragraph boundaries
- 'paragraphs' => array('eol')
-);
-
-//-------------------------------------------------------------------
-
-/**
-* Sets up the Lexer with modes and points it to the Handler
-* For an intro to the Lexer see: wiki:parser
-*/
-class Doku_Parser {
-
- var $Handler;
-
- var $Lexer;
-
- var $modes = array();
-
- var $connected = false;
-
- function addBaseMode(& $BaseMode) {
- $this->modes['base'] = & $BaseMode;
- if ( !$this->Lexer ) {
- $this->Lexer = & new Doku_Lexer($this->Handler,'base', true);
- }
- $this->modes['base']->Lexer = & $this->Lexer;
- }
-
- /**
- * PHP preserves order of associative elements
- * Mode sequence is important
- */
- function addMode($name, & $Mode) {
- if ( !isset($this->modes['base']) ) {
- $this->addBaseMode(new Doku_Parser_Mode_base());
- }
- $Mode->Lexer = & $this->Lexer;
- $this->modes[$name] = & $Mode;
- }
-
- function connectModes() {
-
- if ( $this->connected ) {
- return;
- }
-
- foreach ( array_keys($this->modes) as $mode ) {
-
- // Base isn't connected to anything
- if ( $mode == 'base' ) {
- continue;
- }
-
- $this->modes[$mode]->preConnect();
-
- foreach ( array_keys($this->modes) as $cm ) {
-
- if ( $this->modes[$cm]->accepts($mode) ) {
- $this->modes[$mode]->connectTo($cm);
- }
-
- }
-
- $this->modes[$mode]->postConnect();
- }
-
- $this->connected = true;
- }
-
- function parse($doc) {
- if ( $this->Lexer ) {
- $this->connectModes();
- // Normalize CRs and pad doc
- $doc = "\n".str_replace("\r\n","\n",$doc)."\n";
- $this->Lexer->parse($doc);
- $this->Handler->_finalize();
- return $this->Handler->calls;
- } else {
- return false;
- }
- }
-
-}
-
-//-------------------------------------------------------------------
-/**
- * This class and all the subclasses below are
- * used to reduce the effort required to register
- * modes with the Lexer. For performance these
- * could all be eliminated later perhaps, or
- * the Parser could be serialized to a file once
- * all modes are registered
- *
- * @author Harry Fuecks <hfuecks@gmail.com>
-*/
-class Doku_Parser_Mode {
-
- var $Lexer;
-
- var $allowedModes = array();
-
- // returns a number used to determine in which order modes are added
- function getSort() {
- trigger_error('getSort() not implemented in '.get_class($this), E_USER_WARNING);
- }
-
- // Called before any calls to connectTo
- function preConnect() {}
-
- // Connects the mode
- function connectTo($mode) {}
-
- // Called after all calls to connectTo
- function postConnect() {}
-
- function accepts($mode) {
- return in_array($mode, (array) $this->allowedModes );
- }
-
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_base extends Doku_Parser_Mode {
-
- function Doku_Parser_Mode_base() {
- global $PARSER_MODES;
-
- $this->allowedModes = array_merge (
- $PARSER_MODES['container'],
- $PARSER_MODES['baseonly'],
- $PARSER_MODES['paragraphs'],
- $PARSER_MODES['formatting'],
- $PARSER_MODES['substition'],
- $PARSER_MODES['protected'],
- $PARSER_MODES['disabled']
- );
- }
-
- function getSort() {
- return 0;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_footnote extends Doku_Parser_Mode {
-
- function Doku_Parser_Mode_footnote() {
- global $PARSER_MODES;
-
- $this->allowedModes = array_merge (
- $PARSER_MODES['container'],
- $PARSER_MODES['formatting'],
- $PARSER_MODES['substition'],
- $PARSER_MODES['protected'],
- $PARSER_MODES['disabled']
- );
-
- unset($this->allowedModes[array_search('footnote', $this->allowedModes)]);
- }
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern(
- '\x28\x28(?=.*\x29\x29)',$mode,'footnote'
- );
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern(
- '\x29\x29','footnote'
- );
- }
-
- function getSort() {
- return 150;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_header extends Doku_Parser_Mode {
-
- function preConnect() {
- //we're not picky about the closing ones, two are enough
- $this->Lexer->addSpecialPattern(
- '[ \t]*={2,}[^\n]+={2,}[ \t]*(?=\n)',
- 'base',
- 'header'
- );
- }
-
- function getSort() {
- return 50;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_notoc extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern('~~NOTOC~~',$mode,'notoc');
- }
-
- function getSort() {
- return 30;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_nocache extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern('~~NOCACHE~~',$mode,'nocache');
- }
-
- function getSort() {
- return 40;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_linebreak extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern('\x5C{2}(?:[ \t]|(?=\n))',$mode,'linebreak');
- }
-
- function getSort() {
- return 140;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_eol extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $badModes = array('listblock','table');
- if ( in_array($mode, $badModes) ) {
- return;
- }
- // see FS#1652, pattern extended to swallow preceding whitespace to avoid issues with lines that only contain whitespace
- $this->Lexer->addSpecialPattern('(?:^[ \t]*)?\n',$mode,'eol');
- }
-
- function getSort() {
- return 370;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_hr extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern('\n[ \t]*-{4,}[ \t]*(?=\n)',$mode,'hr');
- }
-
- function getSort() {
- return 160;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_formatting extends Doku_Parser_Mode {
- var $type;
-
- var $formatting = array (
- 'strong' => array (
- 'entry'=>'\*\*(?=.*\*\*)',
- 'exit'=>'\*\*',
- 'sort'=>70
- ),
-
- 'emphasis'=> array (
- 'entry'=>'//(?=[^\x00]*[^:])', //hack for bugs #384 #763 #1468
- 'exit'=>'//',
- 'sort'=>80
- ),
-
- 'underline'=> array (
- 'entry'=>'__(?=.*__)',
- 'exit'=>'__',
- 'sort'=>90
- ),
-
- 'monospace'=> array (
- 'entry'=>'\x27\x27(?=.*\x27\x27)',
- 'exit'=>'\x27\x27',
- 'sort'=>100
- ),
-
- 'subscript'=> array (
- 'entry'=>'<sub>(?=.*</sub>)',
- 'exit'=>'</sub>',
- 'sort'=>110
- ),
-
- 'superscript'=> array (
- 'entry'=>'<sup>(?=.*</sup>)',
- 'exit'=>'</sup>',
- 'sort'=>120
- ),
-
- 'deleted'=> array (
- 'entry'=>'<del>(?=.*</del>)',
- 'exit'=>'</del>',
- 'sort'=>130
- ),
- );
-
- function Doku_Parser_Mode_formatting($type) {
- global $PARSER_MODES;
-
- if ( !array_key_exists($type, $this->formatting) ) {
- trigger_error('Invalid formatting type '.$type, E_USER_WARNING);
- }
-
- $this->type = $type;
-
- // formatting may contain other formatting but not it self
- $modes = $PARSER_MODES['formatting'];
- $key = array_search($type, $modes);
- if ( is_int($key) ) {
- unset($modes[$key]);
- }
-
- $this->allowedModes = array_merge (
- $modes,
- $PARSER_MODES['substition'],
- $PARSER_MODES['disabled']
- );
- }
-
- function connectTo($mode) {
-
- // Can't nest formatting in itself
- if ( $mode == $this->type ) {
- return;
- }
-
- $this->Lexer->addEntryPattern(
- $this->formatting[$this->type]['entry'],
- $mode,
- $this->type
- );
- }
-
- function postConnect() {
-
- $this->Lexer->addExitPattern(
- $this->formatting[$this->type]['exit'],
- $this->type
- );
-
- }
-
- function getSort() {
- return $this->formatting[$this->type]['sort'];
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_listblock extends Doku_Parser_Mode {
-
- function Doku_Parser_Mode_listblock() {
- global $PARSER_MODES;
-
- $this->allowedModes = array_merge (
- $PARSER_MODES['formatting'],
- $PARSER_MODES['substition'],
- $PARSER_MODES['disabled'],
- $PARSER_MODES['protected'] #XXX new
- );
-
- // $this->allowedModes[] = 'footnote';
- }
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('\n {2,}[\-\*]',$mode,'listblock');
- $this->Lexer->addEntryPattern('\n\t{1,}[\-\*]',$mode,'listblock');
-
- $this->Lexer->addPattern('\n {2,}[\-\*]','listblock');
- $this->Lexer->addPattern('\n\t{1,}[\-\*]','listblock');
-
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern('\n','listblock');
- }
-
- function getSort() {
- return 10;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_table extends Doku_Parser_Mode {
-
- function Doku_Parser_Mode_table() {
- global $PARSER_MODES;
-
- $this->allowedModes = array_merge (
- $PARSER_MODES['formatting'],
- $PARSER_MODES['substition'],
- $PARSER_MODES['disabled'],
- $PARSER_MODES['protected']
- );
- }
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('\n\^',$mode,'table');
- $this->Lexer->addEntryPattern('\n\|',$mode,'table');
- }
-
- function postConnect() {
- $this->Lexer->addPattern('\n\^','table');
- $this->Lexer->addPattern('\n\|','table');
- $this->Lexer->addPattern('[\t ]*:::[\t ]*(?=[\|\^])','table');
- $this->Lexer->addPattern('[\t ]+','table');
- $this->Lexer->addPattern('\^','table');
- $this->Lexer->addPattern('\|','table');
- $this->Lexer->addExitPattern('\n','table');
- }
-
- function getSort() {
- return 60;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_unformatted extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('<nowiki>(?=.*</nowiki>)',$mode,'unformatted');
- $this->Lexer->addEntryPattern('%%(?=.*%%)',$mode,'unformattedalt');
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern('</nowiki>','unformatted');
- $this->Lexer->addExitPattern('%%','unformattedalt');
- $this->Lexer->mapHandler('unformattedalt','unformatted');
- }
-
- function getSort() {
- return 170;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_php extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('<php>(?=.*</php>)',$mode,'php');
- $this->Lexer->addEntryPattern('<PHP>(?=.*</PHP>)',$mode,'phpblock');
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern('</php>','php');
- $this->Lexer->addExitPattern('</PHP>','phpblock');
- }
-
- function getSort() {
- return 180;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_html extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('<html>(?=.*</html>)',$mode,'html');
- $this->Lexer->addEntryPattern('<HTML>(?=.*</HTML>)',$mode,'htmlblock');
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern('</html>','html');
- $this->Lexer->addExitPattern('</HTML>','htmlblock');
- }
-
- function getSort() {
- return 190;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_preformatted extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- // Has hard coded awareness of lists...
- $this->Lexer->addEntryPattern('\n (?![\*\-])',$mode,'preformatted');
- $this->Lexer->addEntryPattern('\n\t(?![\*\-])',$mode,'preformatted');
-
- // How to effect a sub pattern with the Lexer!
- $this->Lexer->addPattern('\n ','preformatted');
- $this->Lexer->addPattern('\n\t','preformatted');
-
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern('\n','preformatted');
- }
-
- function getSort() {
- return 20;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_code extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('<code(?=.*</code>)',$mode,'code');
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern('</code>','code');
- }
-
- function getSort() {
- return 200;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_file extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('<file(?=.*</file>)',$mode,'file');
- }
-
- function postConnect() {
- $this->Lexer->addExitPattern('</file>','file');
- }
-
- function getSort() {
- return 210;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_quote extends Doku_Parser_Mode {
-
- function Doku_Parser_Mode_quote() {
- global $PARSER_MODES;
-
- $this->allowedModes = array_merge (
- $PARSER_MODES['formatting'],
- $PARSER_MODES['substition'],
- $PARSER_MODES['disabled'],
- $PARSER_MODES['protected'] #XXX new
- );
- #$this->allowedModes[] = 'footnote';
- #$this->allowedModes[] = 'preformatted';
- #$this->allowedModes[] = 'unformatted';
- }
-
- function connectTo($mode) {
- $this->Lexer->addEntryPattern('\n>{1,}',$mode,'quote');
- }
-
- function postConnect() {
- $this->Lexer->addPattern('\n>{1,}','quote');
- $this->Lexer->addExitPattern('\n','quote');
- }
-
- function getSort() {
- return 220;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_acronym extends Doku_Parser_Mode {
- // A list
- var $acronyms = array();
- var $pattern = '';
-
- function Doku_Parser_Mode_acronym($acronyms) {
- usort($acronyms,array($this,'_compare'));
- $this->acronyms = $acronyms;
- }
-
- function preConnect() {
- if(!count($this->acronyms)) return;
-
- $bound = '[\x00-\x2f\x3a-\x40\x5b-\x60\x7b-\x7f]';
- $acronyms = array_map('Doku_Lexer_Escape',$this->acronyms);
- $this->pattern = '(?<=^|'.$bound.')(?:'.join('|',$acronyms).')(?='.$bound.')';
- }
-
- function connectTo($mode) {
- if(!count($this->acronyms)) return;
-
- if ( strlen($this->pattern) > 0 ) {
- $this->Lexer->addSpecialPattern($this->pattern,$mode,'acronym');
- }
- }
-
- function getSort() {
- return 240;
- }
-
- /**
- * sort callback to order by string length descending
- */
- function _compare($a,$b) {
- $a_len = strlen($a);
- $b_len = strlen($b);
- if ($a_len > $b_len) {
- return -1;
- } else if ($a_len < $b_len) {
- return 1;
- }
-
- return 0;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_smiley extends Doku_Parser_Mode {
- // A list
- var $smileys = array();
- var $pattern = '';
-
- function Doku_Parser_Mode_smiley($smileys) {
- $this->smileys = $smileys;
- }
-
- function preConnect() {
- if(!count($this->smileys) || $this->pattern != '') return;
-
- $sep = '';
- foreach ( $this->smileys as $smiley ) {
- $this->pattern .= $sep.'(?<=\W|^)'.Doku_Lexer_Escape($smiley).'(?=\W|$)';
- $sep = '|';
- }
- }
-
- function connectTo($mode) {
- if(!count($this->smileys)) return;
-
- if ( strlen($this->pattern) > 0 ) {
- $this->Lexer->addSpecialPattern($this->pattern,$mode,'smiley');
- }
- }
-
- function getSort() {
- return 230;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_wordblock extends Doku_Parser_Mode {
- // A list
- var $badwords = array();
- var $pattern = '';
-
- function Doku_Parser_Mode_wordblock($badwords) {
- $this->badwords = $badwords;
- }
-
- function preConnect() {
-
- if ( count($this->badwords) == 0 || $this->pattern != '') {
- return;
- }
-
- $sep = '';
- foreach ( $this->badwords as $badword ) {
- $this->pattern .= $sep.'(?<=\b)(?i)'.Doku_Lexer_Escape($badword).'(?-i)(?=\b)';
- $sep = '|';
- }
-
- }
-
- function connectTo($mode) {
- if ( strlen($this->pattern) > 0 ) {
- $this->Lexer->addSpecialPattern($this->pattern,$mode,'wordblock');
- }
- }
-
- function getSort() {
- return 250;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_entity extends Doku_Parser_Mode {
- // A list
- var $entities = array();
- var $pattern = '';
-
- function Doku_Parser_Mode_entity($entities) {
- $this->entities = $entities;
- }
-
- function preConnect() {
- if(!count($this->entities) || $this->pattern != '') return;
-
- $sep = '';
- foreach ( $this->entities as $entity ) {
- $this->pattern .= $sep.Doku_Lexer_Escape($entity);
- $sep = '|';
- }
- }
-
- function connectTo($mode) {
- if(!count($this->entities)) return;
-
- if ( strlen($this->pattern) > 0 ) {
- $this->Lexer->addSpecialPattern($this->pattern,$mode,'entity');
- }
- }
-
- function getSort() {
- return 260;
- }
-}
-
-//-------------------------------------------------------------------
-// Implements the 640x480 replacement
-class Doku_Parser_Mode_multiplyentity extends Doku_Parser_Mode {
-
- function connectTo($mode) {
-
- $this->Lexer->addSpecialPattern(
- '(?<=\b)(?:[1-9]|\d{2,})[xX]\d+(?=\b)',$mode,'multiplyentity'
- );
-
- }
-
- function getSort() {
- return 270;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_quotes extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- global $conf;
-
- $ws = '\s/\#~:+=&%@\-\x28\x29\]\[{}><"\''; // whitespace
- $punc = ';,\.?!';
-
- if($conf['typography'] == 2){
- $this->Lexer->addSpecialPattern(
- "(?<=^|[$ws])'(?=[^$ws$punc])",$mode,'singlequoteopening'
- );
- $this->Lexer->addSpecialPattern(
- "(?<=^|[^$ws]|[$punc])'(?=$|[$ws$punc])",$mode,'singlequoteclosing'
- );
- $this->Lexer->addSpecialPattern(
- "(?<=^|[^$ws$punc])'(?=$|[^$ws$punc])",$mode,'apostrophe'
- );
- }
-
- $this->Lexer->addSpecialPattern(
- "(?<=^|[$ws])\"(?=[^$ws$punc])",$mode,'doublequoteopening'
- );
- $this->Lexer->addSpecialPattern(
- "\"",$mode,'doublequoteclosing'
- );
-
-
- }
-
- function getSort() {
- return 280;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_camelcaselink extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern(
- '\b[A-Z]+[a-z]+[A-Z][A-Za-z]*\b',$mode,'camelcaselink'
- );
- }
-
- function getSort() {
- return 290;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_internallink extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- // Word boundaries?
- $this->Lexer->addSpecialPattern("\[\[.+?\]\]",$mode,'internallink');
- }
-
- function getSort() {
- return 300;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_media extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- // Word boundaries?
- $this->Lexer->addSpecialPattern("\{\{[^\}]+\}\}",$mode,'media');
- }
-
- function getSort() {
- return 320;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_rss extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern("\{\{rss>[^\}]+\}\}",$mode,'rss');
- }
-
- function getSort() {
- return 310;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_externallink extends Doku_Parser_Mode {
- var $schemes = array();
- var $patterns = array();
-
- function preConnect() {
- if(count($this->patterns)) return;
-
- $ltrs = '\w';
- $gunk = '/\#~:.?+=&%@!\-';
- $punc = '.:?\-;,';
- $host = $ltrs.$punc;
- $any = $ltrs.$gunk.$punc;
-
- $this->schemes = getSchemes();
- foreach ( $this->schemes as $scheme ) {
- $this->patterns[] = '\b(?i)'.$scheme.'(?-i)://['.$any.']+?(?=['.$punc.']*[^'.$any.'])';
- }
-
- $this->patterns[] = '\b(?i)www?(?-i)\.['.$host.']+?\.['.$host.']+?['.$any.']+?(?=['.$punc.']*[^'.$any.'])';
- $this->patterns[] = '\b(?i)ftp?(?-i)\.['.$host.']+?\.['.$host.']+?['.$any.']+?(?=['.$punc.']*[^'.$any.'])';
- }
-
- function connectTo($mode) {
-
- foreach ( $this->patterns as $pattern ) {
- $this->Lexer->addSpecialPattern($pattern,$mode,'externallink');
- }
- }
-
- function getSort() {
- return 330;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_filelink extends Doku_Parser_Mode {
-
- var $pattern;
-
- function preConnect() {
-
- $ltrs = '\w';
- $gunk = '/\#~:.?+=&%@!\-';
- $punc = '.:?\-;,';
- $host = $ltrs.$punc;
- $any = $ltrs.$gunk.$punc;
-
- $this->pattern = '\b(?i)file(?-i)://['.$any.']+?['.
- $punc.']*[^'.$any.']';
- }
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern(
- $this->pattern,$mode,'filelink');
- }
-
- function getSort() {
- return 360;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_windowssharelink extends Doku_Parser_Mode {
-
- var $pattern;
-
- function preConnect() {
- $this->pattern = "\\\\\\\\\w+?(?:\\\\[\w$]+)+";
- }
-
- function connectTo($mode) {
- $this->Lexer->addSpecialPattern(
- $this->pattern,$mode,'windowssharelink');
- }
-
- function getSort() {
- return 350;
- }
-}
-
-//-------------------------------------------------------------------
-class Doku_Parser_Mode_emaillink extends Doku_Parser_Mode {
-
- function connectTo($mode) {
- // pattern below is defined in inc/mail.php
- $this->Lexer->addSpecialPattern('<'.PREG_PATTERN_VALID_EMAIL.'>',$mode,'emaillink');
- }
-
- function getSort() {
- return 340;
- }
-}
-
-
-//Setup VIM: ex: et ts=4 enc=utf-8 :