mediawiki-extensions-Discus.../modules/modifier.js
Bartosz Dziewoński 8c6928aacf modifier: Handle empty nodes in appendSignature()
Bug: T292664
Change-Id: I0003528076e3981d639d337affcccbf394f59224
2021-10-11 22:17:46 +02:00

553 lines
17 KiB
JavaScript

'use strict';
/* global $:off */
/**
* @external CommentItem
*/
var utils = require( './utils.js' );
/**
* Add an attribute to a list item to remove pre-whitespace in Parsoid
*
* @param {HTMLElement} listItem List item element
*/
function whitespaceParsoidHack( listItem ) {
// HACK: Setting data-parsoid removes the whitespace after the list item,
// which makes nested lists work.
// This is undocumented behaviour and probably very fragile.
listItem.setAttribute( 'data-parsoid', '{}' );
}
/**
* Remove extra linebreaks from a wikitext string
*
* @param {string} wikitext Wikitext
* @return {string}
*/
function sanitizeWikitextLinebreaks( wikitext ) {
return utils.htmlTrim( wikitext )
.replace( /\r/g, '\n' )
.replace( /\n+/g, '\n' );
}
/**
* Given a comment and a reply link, add the reply link to its document's DOM tree, at the end of
* the comment.
*
* @param {CommentItem} comment Comment item
* @param {HTMLElement} linkNode Reply link
*/
function addReplyLink( comment, linkNode ) {
var target = comment.range.endContainer;
// Insert the link before trailing whitespace.
// In the MediaWiki parser output, <ul>/<dl> nodes are preceded by a newline. Normally it isn't
// visible on the page. But if we insert an inline element (the reply link) after it, it becomes
// meaningful and gets rendered, which results in additional spacing before some reply links.
// Split the text node, so that we can insert the link before the trailing whitespace.
if ( target.nodeType === Node.TEXT_NODE ) {
target.splitText( target.textContent.match( /\s*$/ ).index );
}
target.parentNode.insertBefore( linkNode, target.nextSibling );
}
/**
* Given a comment, add a list item to its document's DOM tree, inside of which a reply to said
* comment can be added.
*
* The DOM tree is suitably rearranged to ensure correct indentation level of the reply (wrapper
* nodes are added, and other nodes may be moved around).
*
* @param {CommentItem} comment Comment item
* @return {HTMLElement}
*/
function addListItem( comment ) {
var listTypeMap = {
li: 'ul',
dd: 'dl'
};
// 1. Start at given comment
// 2. Skip past all comments with level greater than the given
// (or in other words, all replies, and replies to replies, and so on)
// 3. Add comment with level of the given comment plus 1
var curComment = comment;
while ( curComment.replies.length ) {
curComment = curComment.replies[ curComment.replies.length - 1 ];
}
// Tag names for lists and items we're going to insert
// TODO Add an option to prefer bulleted lists (ul/li)
var itemType = 'dd';
var listType = listTypeMap[ itemType ];
var desiredLevel = comment.level + 1;
var target = curComment.range.endContainer;
// target is a text node or an inline element at the end of a "paragraph" (not necessarily paragraph node).
// First, we need to find a block-level parent that we can mess with.
// If we can't find a surrounding list item or paragraph (e.g. maybe we're inside a table cell
// or something), take the parent node and hope for the best.
var parent = utils.closestElement( target, [ 'li', 'dd', 'p' ] ) || target.parentNode;
while ( target.parentNode !== parent ) {
target = target.parentNode;
}
// parent is a list item or paragraph (hopefully)
// target is an inline node within it
// If the comment is fully covered by some wrapper element, insert replies outside that wrapper.
// This will often just be a paragraph node (<p>), but it can be a <div> or <table> that serves
// as some kind of a fancy frame, which are often used for barnstars and announcements.
var covered = utils.getFullyCoveredSiblings( curComment );
if ( curComment.level === 1 && covered ) {
target = covered[ covered.length - 1 ];
parent = target.parentNode;
}
// If we can't insert a list directly inside this element, insert after it.
// The covered wrapper check above handles most cases, but we still need this sometimes, such as:
// * If the comment starts in the middle of a list, then ends with an unindented p/pre, the
// wrapper check doesn't adjust the parent
// * If the comment consists of multiple list items (starting with a <dt>, so that the comment is
// considered to be unindented, that is level === 1), but not all of them, the wrapper check
// adjusts the parent to be the list, and the rest of the algorithm doesn't handle that well
if (
parent.tagName.toLowerCase() === 'p' ||
parent.tagName.toLowerCase() === 'pre' ||
parent.tagName.toLowerCase() === 'ul' ||
parent.tagName.toLowerCase() === 'dl'
) {
parent = parent.parentNode;
target = target.parentNode;
}
// HACK: Skip past our own reply buttons
if ( target.nextSibling && target.nextSibling.className && target.nextSibling.className.indexOf( 'ext-discussiontools-init-replylink-buttons' ) !== -1 ) {
target = target.nextSibling;
}
// Instead of just using curComment.level, consider indentation of lists within the
// comment (T252702)
var curLevel = utils.getIndentLevel( target, curComment.rootNode ) + 1;
var item, list;
if ( desiredLevel === 1 ) {
// Special handling for top-level comments
item = target.ownerDocument.createElement( 'div' );
item.discussionToolsModified = 'new';
parent.insertBefore( item, target.nextSibling );
// TODO: We should not insert a <div>, instead we need a function that returns parent and target,
// so that we can insert nodes in this place in other code
} else if ( curLevel < desiredLevel ) {
// Insert more lists after the target to increase nesting.
// Parsoid puts HTML comments (and other "rendering-transparent nodes", e.g. category links)
// which appear at the end of the line in wikitext outside the paragraph,
// but we usually shouldn't insert replies between the paragraph and such comments. (T257651)
// Skip over comments and whitespace, but only update target when skipping past comments.
var pointer = target;
while (
pointer.nextSibling && (
utils.isRenderingTransparentNode( pointer.nextSibling ) ||
(
pointer.nextSibling.nodeType === Node.TEXT_NODE &&
utils.htmlTrim( pointer.nextSibling.textContent ) === '' &&
// If more that two lines of whitespace are detected, the following HTML
// comments are not considered to be part of the reply (T264026)
!/(\r?\n){2,}/.test( pointer.nextSibling.textContent )
)
)
) {
pointer = pointer.nextSibling;
if ( utils.isRenderingTransparentNode( pointer ) ) {
target = pointer;
}
}
// Insert required number of wrappers
while ( curLevel < desiredLevel ) {
list = target.ownerDocument.createElement( listType );
list.discussionToolsModified = 'new';
item = target.ownerDocument.createElement( itemType );
item.discussionToolsModified = 'new';
whitespaceParsoidHack( item );
parent.insertBefore( list, target.nextSibling );
list.appendChild( item );
target = item;
parent = list;
curLevel++;
}
} else {
// Split the ancestor nodes after the target to decrease nesting.
var newNode;
do {
if ( !target || !parent ) {
throw new Error( 'Can not decrease nesting any more' );
}
// If target is the last child of its parent, no need to split it
if ( target.nextSibling ) {
// Create new identical node after the parent
newNode = parent.cloneNode( false );
parent.discussionToolsModified = 'split';
parent.parentNode.insertBefore( newNode, parent.nextSibling );
// Move nodes following target to the new node
while ( target.nextSibling ) {
newNode.appendChild( target.nextSibling );
}
}
target = parent;
parent = parent.parentNode;
// Decrease nesting level if we escaped outside of a list
if ( listTypeMap[ target.tagName.toLowerCase() ] ) {
curLevel--;
}
} while ( curLevel >= desiredLevel );
// parent is now a list, target is a list item
if ( itemType === target.tagName.toLowerCase() ) {
item = target.ownerDocument.createElement( itemType );
item.discussionToolsModified = 'new';
whitespaceParsoidHack( item );
parent.insertBefore( item, target.nextSibling );
} else {
// This is the wrong type of list, split it one more time
// If target is the last child of its parent, no need to split it
if ( target.nextSibling ) {
// Create new identical node after the parent
newNode = parent.cloneNode( false );
parent.discussionToolsModified = 'split';
parent.parentNode.insertBefore( newNode, parent.nextSibling );
// Move nodes following target to the new node
while ( target.nextSibling ) {
newNode.appendChild( target.nextSibling );
}
}
target = parent;
parent = parent.parentNode;
// Insert a list of the right type in the middle
list = target.ownerDocument.createElement( listType );
list.discussionToolsModified = 'new';
item = target.ownerDocument.createElement( itemType );
item.discussionToolsModified = 'new';
whitespaceParsoidHack( item );
parent.insertBefore( list, target.nextSibling );
list.appendChild( item );
}
}
return item;
}
/**
* Undo the effects of #addListItem, also removing or merging any affected parent nodes.
*
* @param {HTMLElement} node
*/
function removeAddedListItem( node ) {
while ( node && node.discussionToolsModified ) {
var nextNode;
if ( node.discussionToolsModified === 'new' ) {
nextNode = node.previousSibling || node.parentNode;
// Remove this node
delete node.discussionToolsModified;
node.parentNode.removeChild( node );
} else if ( node.discussionToolsModified === 'split' ) {
// Children might be split too, if so, descend into them afterwards
if ( node.lastChild && node.lastChild.discussionToolsModified === 'split' ) {
node.discussionToolsModified = 'done';
nextNode = node.lastChild;
} else {
delete node.discussionToolsModified;
nextNode = node.parentNode;
}
// Merge the following sibling node back into this one
while ( node.nextSibling.firstChild ) {
node.appendChild( node.nextSibling.firstChild );
}
node.parentNode.removeChild( node.nextSibling );
} else {
nextNode = node.parentNode;
}
node = nextNode;
}
}
/**
* Unwrap a top level list, converting list item text to paragraphs
*
* Assumes that the list has a parent node.
*
* @param {Node} list DOM node, will be wrapepd if it is a list element (dl/ol/ul)
* @param {DocumentFragment|null} fragment Containing document fragment if list has no parent
*/
function unwrapList( list, fragment ) {
var doc = list.ownerDocument,
container = fragment || list.parentNode,
referenceNode = list;
if ( !(
list.nodeType === Node.ELEMENT_NODE && (
list.tagName.toLowerCase() === 'dl' ||
list.tagName.toLowerCase() === 'ol' ||
list.tagName.toLowerCase() === 'ul'
)
) ) {
// Not a list, leave alone (e.g. auto-generated ref block)
return;
}
// If the whole list is a template return it unmodified (T253150)
if ( utils.getTranscludedFromElement( list ) ) {
return;
}
var insertBefore;
while ( list.firstChild ) {
if ( list.firstChild.nodeType === Node.ELEMENT_NODE ) {
// Move <dd> contents to <p>
var p = doc.createElement( 'p' );
while ( list.firstChild.firstChild ) {
// If contents is a block element, place outside the paragraph
// and start a new paragraph after
if ( utils.isBlockElement( list.firstChild.firstChild ) ) {
if ( p.firstChild ) {
insertBefore = referenceNode.nextSibling;
referenceNode = p;
container.insertBefore( p, insertBefore );
}
insertBefore = referenceNode.nextSibling;
referenceNode = list.firstChild.firstChild;
container.insertBefore( list.firstChild.firstChild, insertBefore );
p = doc.createElement( 'p' );
} else {
p.appendChild( list.firstChild.firstChild );
}
}
if ( p.firstChild ) {
insertBefore = referenceNode.nextSibling;
referenceNode = p;
container.insertBefore( p, insertBefore );
}
list.removeChild( list.firstChild );
} else {
// Text node / comment node, probably empty
insertBefore = referenceNode.nextSibling;
referenceNode = list.firstChild;
container.insertBefore( list.firstChild, insertBefore );
}
}
container.removeChild( list );
}
/**
* Add another list item after the given one.
*
* @param {HTMLElement} previousItem
* @return {HTMLElement}
*/
function addSiblingListItem( previousItem ) {
var listItem = previousItem.ownerDocument.createElement( previousItem.tagName );
whitespaceParsoidHack( listItem );
previousItem.parentNode.insertBefore( listItem, previousItem.nextSibling );
return listItem;
}
// TODO: No longer used in the client
function createWikitextNode( doc, wt ) {
var span = doc.createElement( 'span' );
span.setAttribute( 'typeof', 'mw:Transclusion' );
span.setAttribute( 'data-mw', JSON.stringify( { parts: [ wt ] } ) );
return span;
}
/**
* Check whether wikitext contains a user signature.
*
* @param {string} wikitext
* @return {boolean}
*/
function isWikitextSigned( wikitext ) {
wikitext = utils.htmlTrim( wikitext );
// Contains ~~~~ (four tildes), but not ~~~~~ (five tildes), at the end.
return /([^~]|^)~~~~$/.test( wikitext );
}
/**
* Check whether HTML node contains a user signature.
*
* @param {HTMLElement} container
* @return {boolean}
*/
function isHtmlSigned( container ) {
// Good enough?…
var matches = container.querySelectorAll( 'span[typeof="mw:Transclusion"][data-mw*="~~~~"]' );
if ( matches.length === 0 ) {
return false;
}
var lastSig = matches[ matches.length - 1 ];
// Signature must be at the end of the comment - there must be no sibling following this node, or its parents
var node = lastSig;
while ( node ) {
// Skip over whitespace nodes
while (
node.nextSibling &&
node.nextSibling.nodeType === Node.TEXT_NODE &&
utils.htmlTrim( node.nextSibling.textContent ) === ''
) {
node = node.nextSibling;
}
if ( node.nextSibling ) {
return false;
}
node = node.parentNode;
}
return true;
}
/**
* Append a user signature to the comment in the container.
*
* @param {HTMLElement} container
*/
function appendSignature( container ) {
var doc = container.ownerDocument,
signature = mw.msg( 'discussiontools-signature-prefix' ) + '~~~~';
// If the last node isn't a paragraph (e.g. it's a list created in visual mode), then
// add another paragraph to contain the signature.
if ( !container.lastChild || container.lastChild.nodeName.toLowerCase() !== 'p' ) {
container.appendChild( doc.createElement( 'p' ) );
}
// If the last node is empty, trim the signature to prevent leading whitespace triggering
// preformatted text (T269188, T276612)
if ( !container.lastChild.firstChild ) {
signature = signature.replace( /^ +/, '' );
}
// Sign the last line
container.lastChild.appendChild(
createWikitextNode( doc, signature )
);
}
/**
* Add a reply to a specific comment
*
* TODO: No longer used in the client
*
* @param {CommentItem} comment Comment being replied to
* @param {HTMLElement} container Container of comment DOM nodes
*/
function addReply( comment, container ) {
var newParsoidItem;
// Transfer comment DOM to Parsoid DOM
// Wrap every root node of the document in a new list item (dd/li).
// In wikitext mode every root node is a paragraph.
// In visual mode the editor takes care of preventing problematic nodes
// like <table> or <h2> from ever occurring in the comment.
while ( container.childNodes.length ) {
if ( !newParsoidItem ) {
newParsoidItem = addListItem( comment );
} else {
newParsoidItem = addSiblingListItem( newParsoidItem );
}
newParsoidItem.appendChild( container.firstChild );
}
}
/**
* Create a container of comment DOM nodes from wikitext
*
* TODO: No longer used in the client
*
* @param {CommentItem} comment Comment being replied to
* @param {string} wikitext Wikitext
*/
function addWikitextReply( comment, wikitext ) {
var doc = comment.range.endContainer.ownerDocument,
container = doc.createElement( 'div' );
wikitext = sanitizeWikitextLinebreaks( wikitext );
wikitext.split( '\n' ).forEach( function ( line ) {
var p = doc.createElement( 'p' );
p.appendChild( createWikitextNode( doc, line ) );
container.appendChild( p );
} );
if ( !isWikitextSigned( wikitext ) ) {
appendSignature( container );
}
addReply( comment, container );
}
/**
* Create a container of comment DOM nodes from HTML
*
* TODO: No longer used in the client
*
* @param {CommentItem} comment Comment being replied to
* @param {string} html HTML
*/
function addHtmlReply( comment, html ) {
var doc = comment.range.endContainer.ownerDocument,
container = doc.createElement( 'div' );
container.innerHTML = html;
// Remove empty lines
// This should really be anything that serializes to empty string in wikitext,
// (e.g. <h2></h2>) but this will catch most cases
// Create a non-live child node list, so we don't have to worry about it changing
// as nodes are removed.
var childNodeList = Array.prototype.slice.call( container.childNodes );
childNodeList.forEach( function ( node ) {
if ( node.nodeName.toLowerCase() === 'p' && !utils.htmlTrim( node.innerHTML ) ) {
container.removeChild( node );
}
} );
if ( !isHtmlSigned( container ) ) {
appendSignature( container );
}
addReply( comment, container );
}
module.exports = {
addReplyLink: addReplyLink,
addListItem: addListItem,
removeAddedListItem: removeAddedListItem,
addSiblingListItem: addSiblingListItem,
unwrapList: unwrapList,
createWikitextNode: createWikitextNode,
addWikitextReply: addWikitextReply,
addHtmlReply: addHtmlReply,
isWikitextSigned: isWikitextSigned,
isHtmlSigned: isHtmlSigned,
appendSignature: appendSignature,
sanitizeWikitextLinebreaks: sanitizeWikitextLinebreaks
};