mediawiki-extensions-Visual.../modules/parser/ext.core.QuoteTransformer.js

262 lines
7.8 KiB
JavaScript
Raw Normal View History

/*
* MediaWiki-compatible italic/bold handling as a token stream transformation.
*
* @author Gabriel Wicke <gwicke@wikimedia.org>
*/
function QuoteTransformer ( dispatcher ) {
this.reset();
this.register( dispatcher );
}
// constants
QuoteTransformer.prototype.quoteAndNewlineRank = 2.1;
QuoteTransformer.prototype.anyRank = 2.101; // Just after regular quote and newline
QuoteTransformer.prototype.reset = function ( ) {
// A chunk starts with a token context around a quote token and is
// (optionally) followed by non-quote tokens. The quote token and its
// context is later replaced with the actual tag token for italic or bold.
this.currentChunk = [];
// List of chunks, each starting with a (potentially) bold or italic token
// and followed by plain tokens.
this.chunks = [];
// References to chunks in which the first token context / quote token
// should be converted to italic or bold tokens.
this.italics = [];
this.bolds = [];
this.isActive = false;
};
2011-12-13 10:25:18 +00:00
// Register this transformer with the TokenTransformer
QuoteTransformer.prototype.register = function ( dispatcher ) {
this.dispatcher = dispatcher;
// Register for QUOTE tag tokens
dispatcher.addTransform( this.onQuote.bind(this),
this.quoteAndNewlineRank, 'tag', 'mw-quote' );
};
// Make a copy of the token context
QuoteTransformer.prototype._startNewChunk = function ( ) {
this.chunks.push( this.currentChunk );
this.currentChunk = [];
this.currentChunk.pos = this.chunks.length - 1;
};
2011-12-13 10:25:18 +00:00
// Handle QUOTE tags. These are collected in italic/bold lists depending on
// the length of quote string. Actual analysis and conversion to the
// appropriate tag tokens is deferred until the next NEWLINE token triggers
// onNewLine.
2012-02-07 11:53:29 +00:00
QuoteTransformer.prototype.onQuote = function ( token, frame, prevToken ) {
var qlen = token.value.length,
tokens = [], // output tokens
ctx = {
token: token,
frame: frame,
prevToken: prevToken
},
ctx2 = {
frame: frame,
prevToken: prevToken
};
if ( ! this.isActive ) {
this.dispatcher.addTransform( this.onNewLine.bind(this),
this.quoteAndNewlineRank, 'newline' );
// Treat end-of-input just the same as a newline
this.dispatcher.addTransform( this.onNewLine.bind(this),
this.quoteAndNewlineRank, 'end' );
// register for any token if not yet active
this.dispatcher.addTransform( this.onAny.bind(this), this.anyRank, 'any' );
this.isActive = true;
}
this._startNewChunk();
2011-12-13 10:25:18 +00:00
switch (qlen) {
case 2:
this.currentChunk.push(ctx);
this.italics.push(this.currentChunk);
break;
case 3:
this.currentChunk.push(ctx);
this.bolds.push(this.currentChunk);
break;
case 4:
this.currentChunk.push( "'" );
this._startNewChunk();
this.currentChunk.push(ctx);
this.bolds.push(this.currentChunk);
break;
case 5:
2011-12-13 10:25:18 +00:00
// The order of italic vs. bold does not matter. Those are
// processed in a fixed order, and any nesting issues are fixed up
// by the HTML 5 tree builder. This does not always result in the
// prettiest result, but at least it is always correct and very
// convenient.
this.currentChunk.push(ctx);
this.italics.push(this.currentChunk);
this._startNewChunk();
ctx2.token = { attribs: token.attribs };
this.currentChunk.push(ctx2);
this.bolds.push(this.currentChunk);
break;
default: // longer than 5, only use the last 5 ticks
var newvalue = token.value.substr(0, qlen - 5 );
this.currentChunk.push ( newvalue );
this._startNewChunk();
this.currentChunk.push(ctx);
this.italics.push(this.currentChunk);
this._startNewChunk();
ctx2.token = { attribs: ctx.token.attribs };
this.currentChunk.push(ctx2);
this.bolds.push(this.currentChunk);
break;
}
2011-12-13 10:25:18 +00:00
return {};
};
2012-02-07 11:53:29 +00:00
QuoteTransformer.prototype.onAny = function ( token, frame, prevToken ) {
//console.warn('qt onAny: ' + JSON.stringify(token, null, 2));
this.currentChunk.push( token );
return {};
};
2011-12-13 10:25:18 +00:00
// Handle NEWLINE tokens, which trigger the actual quote analysis on the
// collected quote tokens so far.
2012-02-07 11:53:29 +00:00
QuoteTransformer.prototype.onNewLine = function ( token, frame, prevToken ) {
var res;
if( ! this.isActive ) {
// Nothing to do, quick abort.
return { token: token };
}
//token.rank = this.quoteAndNewlineRank;
//console.warn('chunks: ' + JSON.stringify( this.chunks, null, 2 ) );
//console.warn("onNewLine: " + this.italics.length + 'i/b' + this.bolds.length);
// balance out tokens, convert placeholders into tags
if (this.italics.length % 2 && this.bolds.length % 2) {
var firstsingleletterword = -1,
firstmultiletterword = -1,
firstspace = -1;
for (var j = 0; j < this.bolds.length; j++) {
var ctx = this.bolds[j][0];
//console.warn("balancing!" + JSON.stringify(ctx.prevToken, null, 2));
if (ctx.prevToken) {
if (ctx.prevToken.constructor === String) {
var lastchar = prevToken[ctx.prevToken.length - 1],
secondtolastchar = ctx.prevToken[ctx.prevToken.length - 2];
if (lastchar === ' ' && firstspace === -1) {
firstspace = j;
} else if (lastchar !== ' ') {
if ( secondtolastchar === ' ' &&
firstsingleletterword === -1)
{
firstsingleletterword = j;
} else if ( firstmultiletterword == -1) {
firstmultiletterword = j;
}
}
} else if ( ( ctx.prevToken.constructor === NlTk ||
2012-02-01 18:37:48 +00:00
ctx.prevToken.constructor === TagTk ) &&
2011-12-13 10:25:18 +00:00
firstmultiletterword == -1 ) {
// This is an approximation, as the original doQuotes
// operates on the source and just looks at space vs.
// non-space. At least some tags are thus recognized as
// words in the original implementation.
firstmultiletterword = j;
}
}
}
// now see if we can convert a bold to an italic and
// an apostrophe
if (firstsingleletterword > -1) {
this.convertBold(firstsingleletterword);
} else if (firstmultiletterword > -1) {
this.convertBold(firstmultiletterword);
} else if (firstspace > -1) {
this.convertBold(firstspace);
}
}
this.quotesToTags( this.italics, 'i' );
this.quotesToTags( this.bolds, 'b' );
this.currentChunk.push( token );
this._startNewChunk();
//console.warn('chunks: ' + JSON.stringify( this.chunks, null, 2 ) );
// return all collected tokens including the newline
res = { tokens: [].concat.apply([], this.chunks) };
// prepare for next line
this.reset();
// remove 'end', 'newline' and 'any' registrations
this.dispatcher.removeTransform( this.quoteAndNewlineRank, 'end' );
this.dispatcher.removeTransform( this.quoteAndNewlineRank, 'newline' );
this.dispatcher.removeTransform( this.anyRank, 'any' );
return res;
};
2011-12-13 10:25:18 +00:00
// Convert a bold token to italic to balance an uneven number of both bold and
// italic tags. In the process, one quote needs to be converted back to text.
QuoteTransformer.prototype.convertBold = function ( i ) {
var chunk = this.bolds[i],
textToken = "'";
//console.warn('convertbold!');
if ( chunk.pos ) {
this.chunks[chunk.pos - 1].push( textToken );
} else {
// prepend another chunk
this.chunks.unshift( [ textToken ] );
}
// delete from bolds
this.bolds.splice(i, 1);
this.italics.push(chunk);
this.italics.sort(function(a,b) { return a.pos - b.pos; } );
};
2011-12-13 10:25:18 +00:00
// Convert italics/bolds into tags
QuoteTransformer.prototype.quotesToTags = function ( chunks, name ) {
var toggle = true,
t,
j,
out = [];
for (j = 0; j < chunks.length; j++) {
//console.warn( 'quotesToTags ' + name + ': ' + JSON.stringify( chunks, null, 2 ) );
t = chunks[j][0].token;
//console.warn( 'quotesToTags t: ' + JSON.stringify( t, null, 2));
if(toggle) {
2012-02-01 18:37:48 +00:00
chunks[j][0] = new TagTk( name, t.attribs );
} else {
2012-02-01 18:37:48 +00:00
chunks[j][0] = new EndTagTk( name, t.attribs );
}
toggle = !toggle;
}
if (!toggle) {
// Add end tag
2012-02-01 18:37:48 +00:00
this.currentChunk.push( new EndTagTk( name ) );
}
};
if (typeof module == "object") {
module.exports.QuoteTransformer = QuoteTransformer;
}