mirror of
https://github.com/Jermolene/TiddlyWiki5
synced 2024-12-26 10:00:34 +00:00
49a3cb8ede
Involving a bit of a refactoring of the parameters to the $tw.Tree.Macro constructor
186 lines
5.2 KiB
JavaScript
186 lines
5.2 KiB
JavaScript
/*\
|
|
title: $:/core/modules/parsers/newwikitextparser/newwikitextparser.js
|
|
type: application/javascript
|
|
module-type: parser
|
|
|
|
A new-school wikitext parser
|
|
|
|
\*/
|
|
(function(){
|
|
|
|
/*jslint node: true, browser: true */
|
|
/*global $tw: false */
|
|
"use strict";
|
|
|
|
/*
|
|
Define the wikitext renderer constructor
|
|
*/
|
|
var WikiTextRenderer = function(text,options) {
|
|
this.source = text || "";
|
|
this.sourceLength = this.source.length;
|
|
this.pos = 0;
|
|
this.wiki = options.wiki;
|
|
this.parser = options.parser;
|
|
this.tree = [];
|
|
this.dependencies = new $tw.Dependencies();
|
|
// Parse the text into blocks
|
|
while(this.pos < this.sourceLength) {
|
|
this.tree.push.apply(this.tree,this.parseBlock());
|
|
}
|
|
};
|
|
|
|
/*
|
|
Now make WikiTextRenderer inherit from the default Renderer class
|
|
*/
|
|
var Renderer = require("$:/core/modules/renderer.js").Renderer;
|
|
WikiTextRenderer.prototype = new Renderer();
|
|
WikiTextRenderer.constructor = WikiTextRenderer;
|
|
|
|
/*
|
|
Parse a block of text at the current position
|
|
*/
|
|
WikiTextRenderer.prototype.parseBlock = function() {
|
|
this.skipWhitespace();
|
|
if(this.pos >= this.sourceLength) {
|
|
return [];
|
|
}
|
|
// Look for a block rule
|
|
this.parser.blockRegExp.lastIndex = this.pos;
|
|
var match = this.parser.blockRegExp.exec(this.source);
|
|
if(this.parser.blockRules.length && match && match.index === this.pos) {
|
|
var rule;
|
|
for(var t=0; t<this.parser.blockRules.length; t++) {
|
|
if(match[t+1]) {
|
|
rule = this.parser.blockRules[t];
|
|
}
|
|
}
|
|
return rule ? rule.parse.call(this,match,true) : [];
|
|
} else {
|
|
// Treat it as a paragraph if we didn't find a block rule
|
|
return [$tw.Tree.Element("p",{},this.parseRun())];
|
|
}
|
|
};
|
|
|
|
WikiTextRenderer.prototype.skipWhitespace = function() {
|
|
var whitespaceRegExp = /(\s+)/mg;
|
|
whitespaceRegExp.lastIndex = this.pos;
|
|
var whitespaceMatch = whitespaceRegExp.exec(this.source);
|
|
if(whitespaceMatch && whitespaceMatch.index === this.pos) {
|
|
this.pos = whitespaceRegExp.lastIndex;
|
|
}
|
|
};
|
|
|
|
/*
|
|
Parse a run of text at the current position
|
|
terminatorRegExp: a regexp at which to stop the run
|
|
Returns an array of tree nodes
|
|
*/
|
|
WikiTextRenderer.prototype.parseRun = function(terminatorRegExp) {
|
|
var tree = [];
|
|
// Find the next occurrence of the terminator
|
|
terminatorRegExp = terminatorRegExp || /(\r?\n\r?\n)/mg;
|
|
terminatorRegExp.lastIndex = this.pos;
|
|
var terminatorMatch = terminatorRegExp.exec(this.source);
|
|
// Find the next occurrence of a runrule
|
|
this.parser.runRegExp.lastIndex = this.pos;
|
|
var runRuleMatch = this.parser.runRegExp.exec(this.source);
|
|
// Loop around until we've reached the end of the text
|
|
while(this.pos < this.sourceLength && (terminatorMatch || runRuleMatch)) {
|
|
// Return if we've found the terminator, and it precedes any run rule match
|
|
if(terminatorMatch) {
|
|
if(!runRuleMatch || runRuleMatch.index > terminatorMatch.index) {
|
|
if(terminatorMatch.index > this.pos) {
|
|
tree.push($tw.Tree.Text(this.source.substring(this.pos,terminatorMatch.index)));
|
|
}
|
|
this.pos = terminatorMatch.index;
|
|
return tree;
|
|
}
|
|
}
|
|
// Process any run rule, along with the text preceding it
|
|
if(runRuleMatch) {
|
|
// Preceding text
|
|
if(runRuleMatch.index > this.pos) {
|
|
tree.push($tw.Tree.Text(this.source.substring(this.pos,runRuleMatch.index)));
|
|
this.pos = runRuleMatch.index;
|
|
}
|
|
// Process the run rule
|
|
var rule;
|
|
for(var t=0; t<this.parser.runRules.length; t++) {
|
|
if(runRuleMatch[t+1]) {
|
|
rule = this.parser.runRules[t];
|
|
}
|
|
}
|
|
if(rule) {
|
|
tree.push.apply(tree,rule.parse.call(this,runRuleMatch,false));
|
|
}
|
|
// Look for the next run rule
|
|
this.parser.runRegExp.lastIndex = this.pos;
|
|
runRuleMatch = this.parser.runRegExp.exec(this.source);
|
|
}
|
|
}
|
|
// Process the remaining text
|
|
if(this.pos < this.sourceLength) {
|
|
tree.push($tw.tree.Text(this.source.substr(this.pos)));
|
|
}
|
|
this.pos = this.sourceLength;
|
|
return tree;
|
|
};
|
|
|
|
/*
|
|
Parse a run of text preceded by an optional class specifier `{{class}}`
|
|
*/
|
|
WikiTextRenderer.prototype.parseClassedRun = function(terminatorRegExp) {
|
|
var classRegExp = /\{\{([^\}]*)\}\}/mg,
|
|
className;
|
|
classRegExp.lastIndex = this.pos;
|
|
var match = classRegExp.exec(this.source);
|
|
if(match && match.index === this.pos) {
|
|
className = match[1];
|
|
this.pos = match.index + match[0].length;
|
|
}
|
|
var tree = this.parseRun(terminatorRegExp);
|
|
return {
|
|
"class": className,
|
|
tree: tree
|
|
};
|
|
};
|
|
|
|
/*
|
|
The wikitext parser assembles the rules and uses the wikitext renderer to do the parsing
|
|
*/
|
|
var WikiTextParser = function(options) {
|
|
this.wiki = options.wiki;
|
|
// Assemble the rule regexps
|
|
this.blockRules = [];
|
|
this.runRules = [];
|
|
var blockRegExpStrings = [],
|
|
runRegExpStrings = [],
|
|
rules = ($tw.plugins.moduleTypes.wikitextrule || []).slice(0);
|
|
for(var t=0; t<rules.length; t++) {
|
|
if(rules[t].blockParser) {
|
|
this.blockRules.push(rules[t]);
|
|
blockRegExpStrings.push("(" + rules[t].regExpString + ")");
|
|
}
|
|
if(rules[t].runParser) {
|
|
this.runRules.push(rules[t]);
|
|
runRegExpStrings.push("(" + rules[t].regExpString + ")");
|
|
}
|
|
}
|
|
this.blockRegExp = new RegExp(blockRegExpStrings.join("|"),"mg");
|
|
this.runRegExp = new RegExp(runRegExpStrings.join("|"),"mg");
|
|
};
|
|
|
|
/*
|
|
The wikitext parser constructs a wikitext renderer to do the work
|
|
*/
|
|
WikiTextParser.prototype.parse = function(type,text) {
|
|
return new WikiTextRenderer(text,{
|
|
wiki: this.wiki,
|
|
parser: this
|
|
});
|
|
};
|
|
|
|
exports["text/x-tiddlywiki-new"] = WikiTextParser;
|
|
|
|
})();
|