mirror of
https://gerrit.wikimedia.org/r/mediawiki/extensions/WikiEditor
synced 2024-11-30 19:15:43 +00:00
358 lines
12 KiB
JavaScript
358 lines
12 KiB
JavaScript
/* Highlight module for wikiEditor */
|
|
|
|
( function( $ ) { $.wikiEditor.modules.highlight = {
|
|
|
|
/**
|
|
* Core Requirements
|
|
*/
|
|
'req': [ 'iframe' ],
|
|
/**
|
|
* Configuration
|
|
*/
|
|
'cfg': {
|
|
'styleVersion': 3
|
|
},
|
|
/**
|
|
* Internally used event handlers
|
|
*/
|
|
'evt': {
|
|
'delayedChange': function( context, event ) {
|
|
if ( event.data.scope == 'realchange' ) {
|
|
$.wikiEditor.modules.highlight.fn.scan( context );
|
|
$.wikiEditor.modules.highlight.fn.mark( context, event.data.scope );
|
|
}
|
|
},
|
|
'ready': function( context, event ) {
|
|
$.wikiEditor.modules.highlight.fn.scan( context );
|
|
$.wikiEditor.modules.highlight.fn.mark( context, 'ready' );
|
|
}
|
|
},
|
|
/**
|
|
* Internally used functions
|
|
*/
|
|
'fn': {
|
|
/**
|
|
* Creates a highlight module within a wikiEditor
|
|
*
|
|
* @param config Configuration object to create module from
|
|
*/
|
|
'create': function( context, config ) {
|
|
context.modules.highlight.markersStr = '';
|
|
},
|
|
/**
|
|
* Scans text division for tokens
|
|
*
|
|
* @param division
|
|
*/
|
|
'scan': function( context, division ) {
|
|
// Remove all existing tokens
|
|
var tokenArray = context.modules.highlight.tokenArray = [];
|
|
// Scan text for new tokens
|
|
var text = context.fn.getContents();
|
|
// Perform a scan for each module which provides any expressions to scan for
|
|
// FIXME: This traverses the entire string once for every regex. Investigate
|
|
// whether |-concatenating regexes then traversing once is faster.
|
|
for ( var module in context.modules ) {
|
|
if ( module in $.wikiEditor.modules && 'exp' in $.wikiEditor.modules[module] ) {
|
|
for ( var exp in $.wikiEditor.modules[module].exp ) {
|
|
// Prepare configuration
|
|
var regex = $.wikiEditor.modules[module].exp[exp].regex;
|
|
var label = $.wikiEditor.modules[module].exp[exp].label;
|
|
var markAfter = $.wikiEditor.modules[module].exp[exp].markAfter || false;
|
|
// Search for tokens
|
|
var offset = 0, left, right, match;
|
|
while ( ( match = text.substr( offset ).match( regex ) ) != null ) {
|
|
right = ( left = offset + match.index ) + match[0].length;
|
|
tokenArray[tokenArray.length] = {
|
|
'offset': markAfter ? right : left,
|
|
'label': label,
|
|
'tokenStart': left,
|
|
'match': match
|
|
};
|
|
// Move to the right of this match
|
|
offset = right;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
// Sort by start
|
|
tokenArray.sort( function( a, b ) { return a.tokenStart - b.tokenStart; } );
|
|
// Let the world know, a scan just happened!
|
|
context.fn.trigger( 'scan' );
|
|
},
|
|
/**
|
|
* Marks up text with HTML
|
|
*
|
|
* @param division
|
|
* @param tokens
|
|
*/
|
|
// FIXME: What do division and tokens do?
|
|
// TODO: Document the scan() and mark() APIs somewhere
|
|
'mark': function( context, division, tokens ) {
|
|
// Reset markers
|
|
var markers = [];
|
|
|
|
// Recycle markers that will be skipped in this run
|
|
if ( context.modules.highlight.markers && division != '' ) {
|
|
for ( var i = 0; i < context.modules.highlight.markers.length; i++ ) {
|
|
if ( context.modules.highlight.markers[i].skipDivision == division ) {
|
|
markers.push( context.modules.highlight.markers[i] );
|
|
}
|
|
}
|
|
}
|
|
context.modules.highlight.markers = markers;
|
|
|
|
// Get all markers
|
|
context.fn.trigger( 'mark' );
|
|
markers.sort( function( a, b ) { return a.start - b.start || a.end - b.end; } );
|
|
|
|
// Serialize the markers array to a string and compare it with the one stored in the previous run - if they're
|
|
// equal, there's no markers to change
|
|
var markersStr = '';
|
|
for ( var i = 0; i < markers.length; i++ ) {
|
|
markersStr += markers[i].start + ',' + markers[i].end + ',' + markers[i].type + ',';
|
|
}
|
|
if ( context.modules.highlight.markersStr == markersStr ) {
|
|
// No change, bail out
|
|
return;
|
|
}
|
|
context.modules.highlight.markersStr = markersStr;
|
|
|
|
// Traverse the iframe DOM, inserting markers where they're needed - store visited markers here so we know which
|
|
// markers should be removed
|
|
var visited = [], v = 0;
|
|
for ( var i = 0; i < markers.length; i++ ) {
|
|
if ( typeof markers[i].skipDivision !== 'undefined' && ( division == markers[i].skipDivision ) ) {
|
|
continue;
|
|
}
|
|
|
|
// We want to isolate each marker, so we may need to split textNodes if a marker starts or ends halfway one.
|
|
var start = markers[i].start;
|
|
var s = context.fn.getOffset( start );
|
|
if ( !s ) {
|
|
// This shouldn't happen
|
|
continue;
|
|
}
|
|
var startNode = s.node;
|
|
|
|
// Don't wrap leading BRs, produces undesirable results
|
|
// FIXME: It's also possible that the offset is a bit high because getOffset() has incremented .length to
|
|
// fake the newline caused by startNode being in a P. In this case, prevent the textnode splitting below
|
|
// from making startNode an empty textnode, IE barfs on that
|
|
while ( startNode.nodeName == 'BR' || s.offset == startNode.nodeValue.length ) {
|
|
start++;
|
|
s = context.fn.getOffset( start );
|
|
startNode = s.node;
|
|
}
|
|
|
|
// The next marker starts somewhere in this textNode or at this BR
|
|
if ( s.offset > 0 && s.node.nodeName == '#text' ) {
|
|
// Split off the prefix - this leaves the prefix in the current node and puts the rest in a new node
|
|
// which is our start node
|
|
var newStartNode = startNode.splitText( s.offset < s.node.nodeValue.length ?
|
|
s.offset : s.node.nodeValue.length - 1
|
|
);
|
|
var oldStartNode = startNode;
|
|
startNode = newStartNode;
|
|
// Update offset objects. We don't need purgeOffsets(), simply manipulating the existing offset objects
|
|
// will suffice
|
|
// FIXME: This manipulates context.offsets directly, which is ugly, but the performance improvement vs.
|
|
// purgeOffsets() is worth it - this code doesn't set lastTextNode to newStartNode for offset objects
|
|
// with lastTextNode == oldStartNode, but that doesn't really matter
|
|
var subtracted = s.offset;
|
|
var oldLength = s.length;
|
|
|
|
var j, o;
|
|
// Update offset objects referring to oldStartNode
|
|
for ( j = start - subtracted; j < start; j++ ) {
|
|
if ( j in context.offsets ) {
|
|
o = context.offsets[j];
|
|
o.node = oldStartNode;
|
|
o.length = subtracted;
|
|
}
|
|
}
|
|
// Update offset objects referring to newStartNode
|
|
for ( j = start; j < start - subtracted + oldLength; j++ ) {
|
|
if ( j in context.offsets ) {
|
|
o = context.offsets[j];
|
|
o.node = newStartNode;
|
|
o.offset -= subtracted;
|
|
o.length -= subtracted;
|
|
o.lastTextNode = oldStartNode;
|
|
}
|
|
}
|
|
}
|
|
var end = markers[i].end;
|
|
// To avoid ending up at the first char of the next node, we grab the offset for end - 1 and add one to the
|
|
// offset
|
|
var e = context.fn.getOffset( end - 1 );
|
|
if ( !e ) {
|
|
// This shouldn't happen
|
|
continue;
|
|
}
|
|
var endNode = e.node;
|
|
if ( e.offset + 1 < e.length - 1 && endNode.nodeName == '#text' ) {
|
|
// Split off the suffix. This puts the suffix in a new node and leaves the rest in endNode
|
|
var oldEndNode = endNode;
|
|
var newEndNode = endNode.splitText( e.offset + 1 );
|
|
// Update offset objects
|
|
var subtracted = e.offset + 1;
|
|
var oldLength = e.length;
|
|
var j, o;
|
|
// Update offset objects referring to oldEndNode
|
|
for ( j = end - subtracted; j < end; j++ ) {
|
|
if ( j in context.offsets ) {
|
|
o = context.offsets[j];
|
|
o.node = oldEndNode;
|
|
o.length = subtracted;
|
|
}
|
|
}
|
|
// We have to insert this one, as it might not exist: we didn't call getOffset( end )
|
|
context.offsets[end] = {
|
|
'node': newEndNode,
|
|
'offset': 0,
|
|
'length': oldLength - subtracted,
|
|
'lastTextNode': oldEndNode
|
|
};
|
|
// Update offset objects referring to newEndNode
|
|
for ( j = end + 1; j < end - subtracted + oldLength; j++ ) {
|
|
if ( j in context.offsets ) {
|
|
o = context.offsets[j];
|
|
o.node = newEndNode;
|
|
o.offset -= subtracted;
|
|
o.length -= subtracted;
|
|
o.lastTextNode = oldEndNode;
|
|
}
|
|
}
|
|
}
|
|
// Don't wrap trailing BRs, doing that causes weird issues
|
|
if ( endNode.nodeName == 'BR' ) {
|
|
endNode = e.lastTextNode;
|
|
}
|
|
// If startNode and endNode have different parents, we need to pull endNode and all textnodes in between
|
|
// into startNode's parent and replace </p><p> with <br>
|
|
if ( startNode.parentNode != endNode.parentNode ) {
|
|
var startP = $( startNode ).closest( 'p' ).get( 0 );
|
|
var t = new context.fn.rawTraverser( startNode, startP, context.$content.get( 0 ), false );
|
|
var afterStart = startNode.nextSibling;
|
|
var lastP = startP;
|
|
var nextT = t.next();
|
|
while ( nextT && t.node != endNode ) {
|
|
t = nextT;
|
|
nextT = t.next();
|
|
// If t.node has a different parent, merge t.node.parentNode with startNode.parentNode
|
|
if ( t.node.parentNode != startNode.parentNode ) {
|
|
var oldParent = t.node.parentNode;
|
|
if ( afterStart ) {
|
|
if ( lastP != t.inP ) {
|
|
// We're entering a new <p>, insert a <br>
|
|
startNode.parentNode.insertBefore(
|
|
startNode.ownerDocument.createElement( 'br' ),
|
|
afterStart
|
|
);
|
|
}
|
|
// A <p> with just a <br> in it is an empty line, so let's not bother with unwrapping it
|
|
if ( !( oldParent.childNodes.length == 1 && oldParent.firstChild.nodeName == 'BR' ) ) {
|
|
// Move all children of oldParent into startNode's parent
|
|
while ( oldParent.firstChild ) {
|
|
startNode.parentNode.insertBefore( oldParent.firstChild, afterStart );
|
|
}
|
|
}
|
|
} else {
|
|
if ( lastP != t.inP ) {
|
|
// We're entering a new <p>, insert a <br>
|
|
startNode.parentNode.appendChild(
|
|
startNode.ownerDocument.createElement( 'br' )
|
|
);
|
|
}
|
|
// A <p> with just a <br> in it is an empty line, so let's not bother with unwrapping it
|
|
if ( !( oldParent.childNodes.length == 1 && oldParent.firstChild.nodeName == 'BR' ) ) {
|
|
// Move all children of oldParent into startNode's parent
|
|
while ( oldParent.firstChild ) {
|
|
startNode.parentNode.appendChild( oldParent.firstChild );
|
|
}
|
|
}
|
|
}
|
|
// Remove oldParent, which is now empty
|
|
oldParent.parentNode.removeChild( oldParent );
|
|
}
|
|
lastP = t.inP;
|
|
}
|
|
// Moving nodes around like this invalidates offset objects
|
|
// TODO: Update offset objects ourselves for performance. Requires rewriting this code block to be
|
|
// offset-based rather than traverser-based
|
|
}
|
|
// Now wrap everything between startNode and endNode (may be equal).
|
|
var ca1 = startNode, ca2 = endNode;
|
|
if ( ca1 && ca2 && ca1.parentNode ) {
|
|
var anchor = markers[i].getAnchor( ca1, ca2 );
|
|
if ( !anchor ) {
|
|
var commonAncestor = ca1.parentNode;
|
|
if ( markers[i].anchor == 'wrap') {
|
|
// We have to store things like .parentNode and .nextSibling because appendChild() changes these
|
|
var newNode = ca1.ownerDocument.createElement( 'span' );
|
|
var nextNode = ca2.nextSibling;
|
|
// Append all nodes between ca1 and ca2 (inclusive) to newNode
|
|
var n = ca1;
|
|
while ( n != nextNode ) {
|
|
var ns = n.nextSibling;
|
|
newNode.appendChild( n );
|
|
n = ns;
|
|
}
|
|
// Insert newNode in the right place
|
|
if ( nextNode ) {
|
|
commonAncestor.insertBefore( newNode, nextNode );
|
|
} else {
|
|
commonAncestor.appendChild( newNode );
|
|
}
|
|
anchor = newNode;
|
|
} else if ( markers[i].anchor == 'tag' ) {
|
|
anchor = commonAncestor;
|
|
}
|
|
$( anchor ).data( 'marker', markers[i] ).addClass( 'wikiEditor-highlight' );
|
|
// Allow the module adding this marker to manipulate it
|
|
markers[i].afterWrap( anchor, markers[i] );
|
|
|
|
} else {
|
|
// Update the marker object
|
|
$( anchor ).data( 'marker', markers[i] );
|
|
if ( typeof markers[i].onSkip == 'function' ) {
|
|
markers[i].onSkip( anchor );
|
|
}
|
|
}
|
|
visited[v++] = anchor;
|
|
}
|
|
}
|
|
// Remove markers that were previously inserted but weren't passed to this function - visited[] contains the
|
|
// visited elements in order and find() and each() preserve order
|
|
var j = 0;
|
|
context.$content.find( '.wikiEditor-highlight' ).each( function() {
|
|
if ( visited[j] == this ) {
|
|
// This marker is legit, leave it in
|
|
j++;
|
|
return true;
|
|
}
|
|
// Remove this marker
|
|
var marker = $(this).data( 'marker' );
|
|
if ( marker && typeof marker.skipDivision != 'undefined' && ( division == marker.skipDivision ) ) {
|
|
// Don't remove these either
|
|
return true;
|
|
}
|
|
if ( marker && typeof marker.beforeUnwrap == 'function' )
|
|
marker.beforeUnwrap( this );
|
|
if ( ( marker && marker.anchor == 'tag' ) || $(this).is( 'p' ) ) {
|
|
// Remove all classes
|
|
$(this).removeAttr( 'class' );
|
|
} else {
|
|
// Assume anchor == 'wrap'
|
|
$(this).replaceWith( this.childNodes );
|
|
}
|
|
context.fn.purgeOffsets();
|
|
});
|
|
|
|
}
|
|
}
|
|
|
|
}; })( jQuery );
|
|
|