mirror of
https://gerrit.wikimedia.org/r/mediawiki/extensions/VisualEditor
synced 2024-11-15 18:39:52 +00:00
877d1e3359
De-duplicates the two implementations we had. Bug: T253584 Change-Id: If5d15dd4e7e0d3704ca8d75c0a25c529b06f17fb
270 lines
9.5 KiB
JavaScript
270 lines
9.5 KiB
JavaScript
/*!
|
|
* Parsoid utilities.
|
|
*
|
|
* @copyright 2011-2020 VisualEditor Team and others; see http://ve.mit-license.org
|
|
*/
|
|
|
|
mw.libs.ve = mw.libs.ve || {};
|
|
|
|
/**
|
|
* Resolve a URL relative to a given base.
|
|
*
|
|
* Copied from ve.resolveUrl
|
|
*
|
|
* @param {string} url URL to resolve
|
|
* @param {HTMLDocument} base Document whose base URL to use
|
|
* @return {string} Resolved URL
|
|
*/
|
|
mw.libs.ve.resolveUrl = function ( url, base ) {
|
|
var node = base.createElement( 'a' );
|
|
node.setAttribute( 'href', url );
|
|
// If doc.baseURI isn't set, node.href will be an empty string
|
|
// This is crazy, returning the original URL is better
|
|
return node.href || url;
|
|
};
|
|
|
|
/**
|
|
* Decode a URI component into a mediawiki article title
|
|
*
|
|
* N.B. Illegal article titles can result from fairly reasonable input (e.g. "100%25beef");
|
|
* see https://phabricator.wikimedia.org/T137847 .
|
|
*
|
|
* @param {string} s String to decode
|
|
* @param {boolean} [preserveUnderscores] Don't convert underscores to spaces
|
|
* @return {string} Decoded string, or original string if decodeURIComponent failed
|
|
*/
|
|
mw.libs.ve.decodeURIComponentIntoArticleTitle = function ( s, preserveUnderscores ) {
|
|
try {
|
|
s = decodeURIComponent( s );
|
|
} catch ( e ) {
|
|
return s;
|
|
}
|
|
if ( preserveUnderscores ) {
|
|
return s;
|
|
}
|
|
return s.replace( /_/g, ' ' );
|
|
};
|
|
|
|
/**
|
|
* Unwrap Parsoid sections
|
|
*
|
|
* @param {HTMLElement} element Parent element, e.g. document body
|
|
* @param {string} [keepSection] Section to keep
|
|
*/
|
|
mw.libs.ve.unwrapParsoidSections = function ( element, keepSection ) {
|
|
Array.prototype.forEach.call( element.querySelectorAll( 'section[data-mw-section-id]' ), function ( section ) {
|
|
var parent = section.parentNode,
|
|
sectionId = section.getAttribute( 'data-mw-section-id' );
|
|
// Copy section ID to first child (should be a heading)
|
|
// Pseudo-sections (with negative section IDs) may not have a heading
|
|
if ( sectionId !== null && +sectionId > 0 ) {
|
|
section.firstChild.setAttribute( 'data-mw-section-id', sectionId );
|
|
}
|
|
if ( keepSection !== undefined && sectionId === keepSection ) {
|
|
return;
|
|
}
|
|
while ( section.firstChild ) {
|
|
parent.insertBefore( section.firstChild, section );
|
|
}
|
|
parent.removeChild( section );
|
|
} );
|
|
};
|
|
|
|
/**
|
|
* Strip legacy (non-HTML5) IDs; typically found as section IDs inside
|
|
* headings.
|
|
*
|
|
* @param {HTMLElement} element Parent element, e.g. document body
|
|
*/
|
|
mw.libs.ve.stripParsoidFallbackIds = function ( element ) {
|
|
Array.prototype.forEach.call( element.querySelectorAll( 'span[typeof="mw:FallbackId"][id]:empty' ), function ( legacySpan ) {
|
|
legacySpan.parentNode.removeChild( legacySpan );
|
|
} );
|
|
};
|
|
|
|
mw.libs.ve.restbaseIdRegExp = /^mw[a-zA-Z0-9\-_]{2,6}$/;
|
|
|
|
mw.libs.ve.stripRestbaseIds = function ( doc ) {
|
|
var restbaseIdRegExp = mw.libs.ve.restbaseIdRegExp;
|
|
Array.prototype.forEach.call( doc.querySelectorAll( '[id^="mw"]' ), function ( element ) {
|
|
if ( element.id.match( restbaseIdRegExp ) ) {
|
|
element.removeAttribute( 'id' );
|
|
}
|
|
} );
|
|
};
|
|
|
|
/**
|
|
* Fix fragment links which should be relative to the current document
|
|
*
|
|
* This prevents these links from trying to navigate to another page,
|
|
* or open in a new window.
|
|
*
|
|
* Call this after ve.targetLinksToNewWindow, as it removes the target attribute.
|
|
* Call this after LinkCache.styleParsoidElements, as it breaks that method by including the query string.
|
|
*
|
|
* @param {HTMLElement} container Parent element, e.g. document body
|
|
* @param {mw.Title} docTitle Current title, only links to this title will be normalized
|
|
* @param {string} [prefix] Prefix to add to fragment and target ID to avoid collisions
|
|
*/
|
|
mw.libs.ve.fixFragmentLinks = function ( container, docTitle, prefix ) {
|
|
var docTitleText = docTitle.getPrefixedText();
|
|
prefix = prefix || '';
|
|
Array.prototype.forEach.call( container.querySelectorAll( 'a[href*="#"]' ), function ( el ) {
|
|
var target, title,
|
|
fragment = new mw.Uri( el.href ).fragment,
|
|
targetData = mw.libs.ve.getTargetDataFromHref( el.href, el.ownerDocument );
|
|
|
|
if ( targetData.isInternal ) {
|
|
title = mw.Title.newFromText( targetData.title );
|
|
if ( title && title.getPrefixedText() === docTitleText ) {
|
|
|
|
if ( !fragment ) {
|
|
// Special case for empty fragment, even if prefix set
|
|
el.setAttribute( 'href', '#' );
|
|
} else {
|
|
if ( prefix ) {
|
|
target = container.querySelector( '#' + $.escapeSelector( fragment ) );
|
|
// There may be multiple links to a specific target, so check the target
|
|
// hasn't already been fixed (in which case it would be null)
|
|
if ( target ) {
|
|
target.setAttribute( 'id', prefix + fragment );
|
|
}
|
|
}
|
|
el.setAttribute( 'href', '#' + prefix + fragment );
|
|
}
|
|
el.removeAttribute( 'target' );
|
|
|
|
}
|
|
}
|
|
} );
|
|
};
|
|
|
|
/**
|
|
* Parse URL to get title it points to.
|
|
*
|
|
* @param {string} href
|
|
* @param {HTMLDocument|string} doc Document whose base URL to use, or base URL as a string.
|
|
* @return {Object} Information about the given href
|
|
* @return {string} return.title
|
|
* The title of the internal link, else the original href if href is external
|
|
* @return {string} return.rawTitle
|
|
* The title without URL decoding and underscore normalization applied
|
|
* @return {boolean} return.isInternal
|
|
* True if the href pointed to the local wiki, false if href is external
|
|
*/
|
|
mw.libs.ve.getTargetDataFromHref = function ( href, doc ) {
|
|
var relativeBase, relativeBaseRegex, relativeHref, isInternal, matches, data, uri;
|
|
|
|
function regexEscape( str ) {
|
|
return str.replace( /([.?*+^$[\]\\(){}|-])/g, '\\$1' );
|
|
}
|
|
|
|
// Protocol relative href
|
|
relativeHref = href.replace( /^https?:/i, '' );
|
|
// Paths without a host portion are assumed to be internal
|
|
isInternal = !/^\/\//.test( relativeHref );
|
|
|
|
// Check if this matches the server's article path
|
|
// Protocol relative base
|
|
relativeBase = mw.libs.ve.resolveUrl( mw.config.get( 'wgArticlePath' ), doc ).replace( /^https?:/i, '' );
|
|
relativeBaseRegex = new RegExp( regexEscape( relativeBase ).replace( regexEscape( '$1' ), '(.*)' ) );
|
|
matches = relativeHref.match( relativeBaseRegex );
|
|
if ( matches && matches[ 1 ].split( '#' )[ 0 ].indexOf( '?' ) === -1 ) {
|
|
// Take the relative path
|
|
href = matches[ 1 ];
|
|
isInternal = true;
|
|
}
|
|
|
|
// Check if this matches the server's script path (as used by red links)
|
|
relativeBase = mw.libs.ve.resolveUrl( mw.config.get( 'wgScript' ), doc ).replace( /^https?:/i, '' );
|
|
if ( relativeHref.indexOf( relativeBase ) === 0 ) {
|
|
uri = new mw.Uri( relativeHref );
|
|
if ( Object.keys( uri.query ).length === 1 && uri.query.title ) {
|
|
href = uri.query.title;
|
|
isInternal = true;
|
|
} else if ( Object.keys( uri.query ).length === 3 && uri.query.title && uri.query.action === 'edit' && uri.query.redlink === '1' ) {
|
|
href = uri.query.title;
|
|
isInternal = true;
|
|
} else {
|
|
href = relativeHref;
|
|
isInternal = false;
|
|
}
|
|
}
|
|
|
|
// This href doesn't necessarily come from Parsoid (and it might not have the "./" prefix), but
|
|
// this method will work fine.
|
|
data = mw.libs.ve.parseParsoidResourceName( href );
|
|
data.isInternal = isInternal;
|
|
return data;
|
|
};
|
|
|
|
/**
|
|
* Expand a string of the form jquery.foo,bar|jquery.ui.baz,quux to
|
|
* an array of module names like [ 'jquery.foo', 'jquery.bar',
|
|
* 'jquery.ui.baz', 'jquery.ui.quux' ]
|
|
*
|
|
* Implementation of ResourceLoaderContext::expandModuleNames
|
|
* TODO: Consider upstreaming this to MW core.
|
|
*
|
|
* @param {string} moduleNames Packed module name list
|
|
* @return {string[]} Array of module names
|
|
*/
|
|
mw.libs.ve.expandModuleNames = function ( moduleNames ) {
|
|
var modules = [];
|
|
|
|
moduleNames.split( '|' ).forEach( function ( group ) {
|
|
var matches, prefix, suffixes;
|
|
if ( group.indexOf( ',' ) === -1 ) {
|
|
// This is not a set of modules in foo.bar,baz notation
|
|
// but a single module
|
|
modules.push( group );
|
|
} else {
|
|
// This is a set of modules in foo.bar,baz notation
|
|
matches = group.match( /(.*)\.([^.]*)/ );
|
|
if ( !matches ) {
|
|
// Prefixless modules, i.e. without dots
|
|
modules = modules.concat( group.split( ',' ) );
|
|
} else {
|
|
// We have a prefix and a bunch of suffixes
|
|
prefix = matches[ 1 ];
|
|
suffixes = matches[ 2 ].split( ',' ); // [ 'bar', 'baz' ]
|
|
suffixes.forEach( function ( suffix ) {
|
|
modules.push( prefix + '.' + suffix );
|
|
} );
|
|
}
|
|
}
|
|
} );
|
|
return modules;
|
|
};
|
|
|
|
/**
|
|
* Split Parsoid resource name into the href prefix and the page title.
|
|
*
|
|
* @param {string} resourceName Resource name, from a `href` or `resource` attribute
|
|
* @return {Object} Object with the following properties:
|
|
* @return {string} return.title Full page title in text form (with namespace, and spaces instead of underscores)
|
|
* @return {string} return.rawTitle The title without URL decoding and underscore normalization applied
|
|
*/
|
|
mw.libs.ve.parseParsoidResourceName = function ( resourceName ) {
|
|
// Resource names are always prefixed with './' to prevent the MediaWiki namespace from being
|
|
// interpreted as a URL protocol, consider e.g. 'href="./File:Foo.png"'.
|
|
// (We accept input without the prefix, so this can also take plain page titles.)
|
|
var matches = resourceName.match( /^(\.\/|)(.*)$/ );
|
|
return {
|
|
// '%' and '?' are valid in page titles, but normally URI-encoded. This also changes underscores
|
|
// to spaces.
|
|
title: mw.libs.ve.decodeURIComponentIntoArticleTitle( matches[ 2 ] ),
|
|
rawTitle: matches[ 2 ]
|
|
};
|
|
};
|
|
|
|
/**
|
|
* Extract the page title from a Parsoid resource name.
|
|
*
|
|
* @param {string} resourceName Resource name, from a `href` or `resource` attribute
|
|
* @return {string} Full page title in text form (with namespace, and spaces instead of underscores)
|
|
*/
|
|
mw.libs.ve.normalizeParsoidResourceName = function ( resourceName ) {
|
|
return mw.libs.ve.parseParsoidResourceName( resourceName ).title;
|
|
};
|