1
0
mirror of https://github.com/Jermolene/TiddlyWiki5 synced 2024-12-01 22:09:57 +00:00
TiddlyWiki5/core/modules/parsers/wikiparser/wikiparser.js
Robin Munn 215bd4e015
Avoid skipping extra whitespace in wikiparser.js (#7835)
When wikiparser parses text looking for a pragma block, it skips
whitespace before looking for the next pragma. If no pragma is found,
we should return the parse position to the original location so that the
skipped whitespace can be parsed as a text node. This allows the
attribute `join=" and "` to parse as " and " rather than "and ".
2023-11-14 22:10:58 +00:00

464 lines
15 KiB
JavaScript

/*\
title: $:/core/modules/parsers/wikiparser/wikiparser.js
type: application/javascript
module-type: parser
The wiki text parser processes blocks of source text into a parse tree.
The parse tree is made up of nested arrays of these JavaScript objects:
{type: "element", tag: <string>, attributes: {}, children: []} - an HTML element
{type: "text", text: <string>} - a text node
{type: "entity", value: <string>} - an entity
{type: "raw", html: <string>} - raw HTML
Attributes are stored as hashmaps of the following objects:
{type: "string", value: <string>} - literal string
{type: "indirect", textReference: <textReference>} - indirect through a text reference
{type: "macro", macro: <TBD>} - indirect through a macro invocation
\*/
(function(){
/*jslint node: true, browser: true */
/*global $tw: false */
"use strict";
/*
type: content type of text
text: text to be parsed
options: see below:
parseAsInline: true to parse text as inline instead of block
wiki: reference to wiki to use
_canonical_uri: optional URI of content if text is missing or empty
configTrimWhiteSpace: true to trim whitespace
*/
var WikiParser = function(type,text,options) {
this.wiki = options.wiki;
var self = this;
// Check for an externally linked tiddler
if($tw.browser && (text || "") === "" && options._canonical_uri) {
this.loadRemoteTiddler(options._canonical_uri);
text = $tw.language.getRawString("LazyLoadingWarning");
}
// Save the parse text
this.type = type || "text/vnd.tiddlywiki";
this.source = text || "";
this.sourceLength = this.source.length;
// Flag for ignoring whitespace
this.configTrimWhiteSpace = options.configTrimWhiteSpace !== undefined ? options.configTrimWhiteSpace : false;
// Parser mode
this.parseAsInline = options.parseAsInline;
// Set current parse position
this.pos = 0;
// Start with empty output
this.tree = [];
// Assemble the rule classes we're going to use
var pragmaRuleClasses, blockRuleClasses, inlineRuleClasses;
if(options.rules) {
pragmaRuleClasses = options.rules.pragma;
blockRuleClasses = options.rules.block;
inlineRuleClasses = options.rules.inline;
} else {
// Setup the rule classes if we don't have them already
if(!this.pragmaRuleClasses) {
WikiParser.prototype.pragmaRuleClasses = $tw.modules.createClassesFromModules("wikirule","pragma",$tw.WikiRuleBase);
this.setupRules(WikiParser.prototype.pragmaRuleClasses,"$:/config/WikiParserRules/Pragmas/");
}
pragmaRuleClasses = this.pragmaRuleClasses;
if(!this.blockRuleClasses) {
WikiParser.prototype.blockRuleClasses = $tw.modules.createClassesFromModules("wikirule","block",$tw.WikiRuleBase);
this.setupRules(WikiParser.prototype.blockRuleClasses,"$:/config/WikiParserRules/Block/");
}
blockRuleClasses = this.blockRuleClasses;
if(!this.inlineRuleClasses) {
WikiParser.prototype.inlineRuleClasses = $tw.modules.createClassesFromModules("wikirule","inline",$tw.WikiRuleBase);
this.setupRules(WikiParser.prototype.inlineRuleClasses,"$:/config/WikiParserRules/Inline/");
}
inlineRuleClasses = this.inlineRuleClasses;
}
// Instantiate the pragma parse rules
this.pragmaRules = this.instantiateRules(pragmaRuleClasses,"pragma",0);
// Instantiate the parser block and inline rules
this.blockRules = this.instantiateRules(blockRuleClasses,"block",0);
this.inlineRules = this.instantiateRules(inlineRuleClasses,"inline",0);
// Parse any pragmas
var topBranch = this.parsePragmas();
// Parse the text into inline runs or blocks
if(this.parseAsInline) {
topBranch.push.apply(topBranch,this.parseInlineRun());
} else {
topBranch.push.apply(topBranch,this.parseBlocks());
}
// Return the parse tree
};
/*
*/
WikiParser.prototype.loadRemoteTiddler = function(url) {
var self = this;
$tw.utils.httpRequest({
url: url,
type: "GET",
callback: function(err,data) {
if(!err) {
var tiddlers = self.wiki.deserializeTiddlers(".tid",data,self.wiki.getCreationFields());
$tw.utils.each(tiddlers,function(tiddler) {
tiddler["_canonical_uri"] = url;
});
if(tiddlers) {
self.wiki.addTiddlers(tiddlers);
}
}
}
});
};
/*
*/
WikiParser.prototype.setupRules = function(proto,configPrefix) {
var self = this;
if(!$tw.safeMode) {
$tw.utils.each(proto,function(object,name) {
if(self.wiki.getTiddlerText(configPrefix + name,"enable") !== "enable") {
delete proto[name];
}
});
}
};
/*
Instantiate an array of parse rules
*/
WikiParser.prototype.instantiateRules = function(classes,type,startPos) {
var rulesInfo = [],
self = this;
$tw.utils.each(classes,function(RuleClass) {
// Instantiate the rule
var rule = new RuleClass(self);
rule.is = {};
rule.is[type] = true;
rule.init(self);
var matchIndex = rule.findNextMatch(startPos);
if(matchIndex !== undefined) {
rulesInfo.push({
rule: rule,
matchIndex: matchIndex
});
}
});
return rulesInfo;
};
/*
Skip any whitespace at the current position. Options are:
treatNewlinesAsNonWhitespace: true if newlines are NOT to be treated as whitespace
*/
WikiParser.prototype.skipWhitespace = function(options) {
options = options || {};
var whitespaceRegExp = options.treatNewlinesAsNonWhitespace ? /([^\S\n]+)/mg : /(\s+)/mg;
whitespaceRegExp.lastIndex = this.pos;
var whitespaceMatch = whitespaceRegExp.exec(this.source);
if(whitespaceMatch && whitespaceMatch.index === this.pos) {
this.pos = whitespaceRegExp.lastIndex;
}
};
/*
Get the next match out of an array of parse rule instances
*/
WikiParser.prototype.findNextMatch = function(rules,startPos) {
// Find the best matching rule by finding the closest match position
var matchingRule,
matchingRulePos = this.sourceLength;
// Step through each rule
for(var t=0; t<rules.length; t++) {
var ruleInfo = rules[t];
// Ask the rule to get the next match if we've moved past the current one
if(ruleInfo.matchIndex !== undefined && ruleInfo.matchIndex < startPos) {
ruleInfo.matchIndex = ruleInfo.rule.findNextMatch(startPos);
}
// Adopt this match if it's closer than the current best match
if(ruleInfo.matchIndex !== undefined && ruleInfo.matchIndex <= matchingRulePos) {
matchingRule = ruleInfo;
matchingRulePos = ruleInfo.matchIndex;
}
}
return matchingRule;
};
/*
Parse any pragmas at the beginning of a block of parse text
*/
WikiParser.prototype.parsePragmas = function() {
var currentTreeBranch = this.tree;
while(true) {
var savedPos = this.pos;
// Skip whitespace
this.skipWhitespace();
// Check for the end of the text
if(this.pos >= this.sourceLength) {
break;
}
// Check if we've arrived at a pragma rule match
var nextMatch = this.findNextMatch(this.pragmaRules,this.pos);
// If not, just exit
if(!nextMatch || nextMatch.matchIndex !== this.pos) {
this.pos = savedPos;
break;
}
// Process the pragma rule
var subTree = nextMatch.rule.parse();
if(subTree.length > 0) {
// Quick hack; we only cope with a single parse tree node being returned, which is true at the moment
currentTreeBranch.push.apply(currentTreeBranch,subTree);
subTree[0].children = [];
currentTreeBranch = subTree[0].children;
}
}
return currentTreeBranch;
};
/*
Parse a block from the current position
terminatorRegExpString: optional regular expression string that identifies the end of plain paragraphs. Must not include capturing parenthesis
*/
WikiParser.prototype.parseBlock = function(terminatorRegExpString) {
var terminatorRegExp = terminatorRegExpString ? new RegExp(terminatorRegExpString + "|\\r?\\n\\r?\\n","mg") : /(\r?\n\r?\n)/mg;
this.skipWhitespace();
if(this.pos >= this.sourceLength) {
return [];
}
// Look for a block rule that applies at the current position
var nextMatch = this.findNextMatch(this.blockRules,this.pos);
if(nextMatch && nextMatch.matchIndex === this.pos) {
return nextMatch.rule.parse();
}
// Treat it as a paragraph if we didn't find a block rule
var start = this.pos;
var children = this.parseInlineRun(terminatorRegExp);
var end = this.pos;
return [{type: "element", tag: "p", children: children, start: start, end: end }];
};
/*
Parse a series of blocks of text until a terminating regexp is encountered or the end of the text
terminatorRegExpString: terminating regular expression
*/
WikiParser.prototype.parseBlocks = function(terminatorRegExpString) {
if(terminatorRegExpString) {
return this.parseBlocksTerminated(terminatorRegExpString);
} else {
return this.parseBlocksUnterminated();
}
};
/*
Parse a block from the current position to the end of the text
*/
WikiParser.prototype.parseBlocksUnterminated = function() {
var tree = [];
while(this.pos < this.sourceLength) {
tree.push.apply(tree,this.parseBlock());
}
return tree;
};
/*
Parse blocks of text until a terminating regexp is encountered. Wrapper for parseBlocksTerminatedExtended that just returns the parse tree
*/
WikiParser.prototype.parseBlocksTerminated = function(terminatorRegExpString) {
var ex = this.parseBlocksTerminatedExtended(terminatorRegExpString);
return ex.tree;
};
/*
Parse blocks of text until a terminating regexp is encountered
*/
WikiParser.prototype.parseBlocksTerminatedExtended = function(terminatorRegExpString) {
var terminatorRegExp = new RegExp(terminatorRegExpString,"mg"),
result = {
tree: []
};
// Skip any whitespace
this.skipWhitespace();
// Check if we've got the end marker
terminatorRegExp.lastIndex = this.pos;
var match = terminatorRegExp.exec(this.source);
// Parse the text into blocks
while(this.pos < this.sourceLength && !(match && match.index === this.pos)) {
var blocks = this.parseBlock(terminatorRegExpString);
result.tree.push.apply(result.tree,blocks);
// Skip any whitespace
this.skipWhitespace();
// Check if we've got the end marker
terminatorRegExp.lastIndex = this.pos;
match = terminatorRegExp.exec(this.source);
}
if(match && match.index === this.pos) {
this.pos = match.index + match[0].length;
result.match = match;
}
return result;
};
/*
Parse a run of text at the current position
terminatorRegExp: a regexp at which to stop the run
options: see below
Options available:
eatTerminator: move the parse position past any encountered terminator (default false)
*/
WikiParser.prototype.parseInlineRun = function(terminatorRegExp,options) {
if(terminatorRegExp) {
return this.parseInlineRunTerminated(terminatorRegExp,options);
} else {
return this.parseInlineRunUnterminated(options);
}
};
WikiParser.prototype.parseInlineRunUnterminated = function(options) {
var tree = [];
// Find the next occurrence of an inline rule
var nextMatch = this.findNextMatch(this.inlineRules,this.pos);
// Loop around the matches until we've reached the end of the text
while(this.pos < this.sourceLength && nextMatch) {
// Process the text preceding the run rule
if(nextMatch.matchIndex > this.pos) {
this.pushTextWidget(tree,this.source.substring(this.pos,nextMatch.matchIndex),this.pos,nextMatch.matchIndex);
this.pos = nextMatch.matchIndex;
}
// Process the run rule
tree.push.apply(tree,nextMatch.rule.parse());
// Look for the next run rule
nextMatch = this.findNextMatch(this.inlineRules,this.pos);
}
// Process the remaining text
if(this.pos < this.sourceLength) {
this.pushTextWidget(tree,this.source.substr(this.pos),this.pos,this.sourceLength);
}
this.pos = this.sourceLength;
return tree;
};
WikiParser.prototype.parseInlineRunTerminated = function(terminatorRegExp,options) {
var ex = this.parseInlineRunTerminatedExtended(terminatorRegExp,options);
return ex.tree;
};
WikiParser.prototype.parseInlineRunTerminatedExtended = function(terminatorRegExp,options) {
options = options || {};
var tree = [];
// Find the next occurrence of the terminator
terminatorRegExp.lastIndex = this.pos;
var terminatorMatch = terminatorRegExp.exec(this.source);
// Find the next occurrence of a inlinerule
var inlineRuleMatch = this.findNextMatch(this.inlineRules,this.pos);
// Loop around until we've reached the end of the text
while(this.pos < this.sourceLength && (terminatorMatch || inlineRuleMatch)) {
// Return if we've found the terminator, and it precedes any inline rule match
if(terminatorMatch) {
if(!inlineRuleMatch || inlineRuleMatch.matchIndex >= terminatorMatch.index) {
if(terminatorMatch.index > this.pos) {
this.pushTextWidget(tree,this.source.substring(this.pos,terminatorMatch.index),this.pos,terminatorMatch.index);
}
this.pos = terminatorMatch.index;
if(options.eatTerminator) {
this.pos += terminatorMatch[0].length;
}
return {
match: terminatorMatch,
tree: tree
};
}
}
// Process any inline rule, along with the text preceding it
if(inlineRuleMatch) {
// Preceding text
if(inlineRuleMatch.matchIndex > this.pos) {
this.pushTextWidget(tree,this.source.substring(this.pos,inlineRuleMatch.matchIndex),this.pos,inlineRuleMatch.matchIndex);
this.pos = inlineRuleMatch.matchIndex;
}
// Process the inline rule
tree.push.apply(tree,inlineRuleMatch.rule.parse());
// Look for the next inline rule
inlineRuleMatch = this.findNextMatch(this.inlineRules,this.pos);
// Look for the next terminator match
terminatorRegExp.lastIndex = this.pos;
terminatorMatch = terminatorRegExp.exec(this.source);
}
}
// Process the remaining text
if(this.pos < this.sourceLength) {
this.pushTextWidget(tree,this.source.substr(this.pos),this.pos,this.sourceLength);
}
this.pos = this.sourceLength;
return {
tree: tree
};
};
/*
Push a text widget onto an array, respecting the configTrimWhiteSpace setting
*/
WikiParser.prototype.pushTextWidget = function(array,text,start,end) {
if(this.configTrimWhiteSpace) {
text = $tw.utils.trim(text);
}
if(text) {
array.push({type: "text", text: text, start: start, end: end});
}
};
/*
Parse zero or more class specifiers `.classname`
*/
WikiParser.prototype.parseClasses = function() {
var classRegExp = /\.([^\s\.]+)/mg,
classNames = [];
classRegExp.lastIndex = this.pos;
var match = classRegExp.exec(this.source);
while(match && match.index === this.pos) {
this.pos = match.index + match[0].length;
classNames.push(match[1]);
match = classRegExp.exec(this.source);
}
return classNames;
};
/*
Amend the rules used by this instance of the parser
type: `only` keeps just the named rules, `except` keeps all but the named rules
names: array of rule names
*/
WikiParser.prototype.amendRules = function(type,names) {
names = names || [];
// Define the filter function
var target;
if(type === "only") {
target = true;
} else if(type === "except") {
target = false;
} else {
return;
}
// Define a function to process each of our rule arrays
var processRuleArray = function(ruleArray) {
for(var t=ruleArray.length-1; t>=0; t--) {
if((names.indexOf(ruleArray[t].rule.name) === -1) === target) {
ruleArray.splice(t,1);
}
}
};
// Process each rule array
processRuleArray(this.pragmaRules);
processRuleArray(this.blockRules);
processRuleArray(this.inlineRules);
};
exports["text/vnd.tiddlywiki"] = WikiParser;
})();