Note: After publishing, you may have to bypass your browser's cache to see the changes.
- Firefox / Safari: Hold Shift while clicking Reload, or press either Ctrl-F5 or Ctrl-R (⌘-R on a Mac)
- Google Chrome: Press Ctrl-Shift-R (⌘-Shift-R on a Mac)
- Edge: Hold Ctrl while clicking Refresh, or press Ctrl-F5.
/* Uploaded from branch master, commit 5d2cd81 */ // vim: ts=4 sw=4 et //<nowiki> function loadReplyLink( $, mw ) { var TIMESTAMP_REGEX = /\(UTC(?:(?:−|\+)\d+?(?:\.\d+)?)?\)\S*?\s*$/m; var EDIT_REQ_REGEX = /^((Semi|Template|Extended-confirmed)-p|P)rotected edit request on \d\d? \w+ \d{4}/; var EDIT_REQ_TPL_REGEX = /\{\{edit (template|fully|extended|semi)-protected\s*(\|.+?)*\}\}/; var LITERAL_SIGNATURE = "~~" + "~~"; // split up because it might get processed var i18n = { "en": { "rl-advert": " (using [[w:en:User:Enterprisey/reply-link|reply-link]])", "rl-error-status": "There was an error while replying! Please leave a note at " + "<a href='https://en.wikipedia.org/wiki/User_talk:Enterprisey/reply-link'>the script's talk page</a>" + " with any errors in <a href='https://en.wikipedia.org/wiki/WP:JSERROR'>the browser console</a>, if possible.", "rl-replying-to": "Replying to ", "rl-reloading": "automatically reloading", "rl-reload": "Reload", "rl-saved": "Reply saved!", "rl-cancel": "cancel ", "rl-placeholder": "Reply here!", "rl-reply": "Reply", "rl-preview": "Preview", "rl-cancel-button": "Cancel", "rl-started-reply": "You've started a reply but haven't posted it", "rl-loading": "Loading...", "rl-reply-label": "reply", "rl-to-label": " to ", "rl-auto-indent": "Automatically indent?" }, "pt": { "rl-advert": "(usando [[w:en:User:Enterprisey/reply-link|reply-link]])", "rl-error-status": "Ocorreu um erro ao responder! Por favor deixe um comentário na " + "<a href='https://en.wikipedia.org/wiki/User_talk:Enterprisey/reply-link'>página de discussão do script</a>" + " informando os erros que apareçam <a href='https://en.wikipedia.org/wiki/WP:JSERROR'>no console do navegador</a>, se possível.", "rl-replying-to": "Respondendo a ", "rl-reloading": "recarregando automaticamente", "rl-reload": "Recarregar", "rl-saved": "Resposta publicada!", "rl-cancel": "cancelar ", "rl-placeholder": "Responda aqui!", "rl-reply": "Responder", "rl-preview": "Prever", "rl-cancel-button": "Cancelar", "rl-started-reply": "Você começou a responder, mas não publicou sua resposta", "rl-loading": "Carregando...", "rl-reply-label": "responder", "rl-to-label": " a ", "rl-auto-indent": "Indentar automaticamente?" } }; var PARSOID_ENDPOINT = "https:" + mw.config.get( "wgServer" ) + "/api/rest_v1/page/html/"; var HEADER_SELECTOR = "h1,h2,h3,h4,h5,h6"; var MAX_UNICODE_DECIMAL = 1114111; var HEADER_REGEX = /^\s*=(=*)\s*(.+?)\s*\1=\s*$/gm; // T:TDYK, used at the end of loadReplyLink var TTDYK = "Template:Did_you_know_nominations"; var RFA_PG = "Wikipedia:Requests_for_adminship/"; // Threshold for indentation when we offer to outdent var OUTDENT_THRESH = 8; // All of the interface message keys that we explicitly load var INT_MSG_KEYS = [ "mycontris" ]; // Date format regexes in signatures (i.e. the "default date format") var DATE_FMT_RGX = { "//en.wikipedia.org": /\d\d:\d\d,\s\d{1,2}\s\w+?\s\d{4}/.source, "//simple.wikipedia.org": /\d\d:\d\d,\s\d{1,2}\s\w+?\s\d{4}/.source, "//en.wikisource.org": /\d\d:\d\d,\s\d{1,2}\s\w+?\s\d{4}/.source, "//pt.wikipedia.org": /\d\dh\d\dmin\sde \d{1,2} de \w+? de \d{4}/.source } // Shared API object var api; /* * Regex *sources* for a "userspace" link. Basically the * localized equivalent of User( talk)?|Special:Contributions/ * Initialized in buildUserspcLinkRgx, which is called near the top * of the closure in handleWrapperClick. * * Three subproperties: und for underscores instead of spaces (e.g. * "User_talk"), spc for spaces (e.g. "User talk"), and both for * a regex combining the two (used for matching on wikitext). */ var userspcLinkRgx = null; /** * This dictionary is some global state that holds three pieces of * information for each "(reply)" link (keyed by their unique IDs): * * - the indentation string for the comment (e.g. ":*::") * - the header tuple for the parent section, in the form of * [level, text, number], where: * - level is 1 for a h1, 2 for a h2, etc * - text is the text between the equal signs * - number is the zero-based index of the heading from the top * - sigIdx, or the zero-based index of the signature from the top * of the section * * This dictionary is populated in attachLinks, and unpacked in the * click handler for the links (defined in attachLinkAfterNode); the * values are then passed to doReply. */ var metadata = {}; /** * This global string flag is: * * - "AfD" if the current page is an AfD page * - "MfD" if the current page is an MfD page * - "TfD" if the current page is a TfD log page * - "CfD" if the current page is a CfD log page * - "FfD" if the current page is a FfD log page * - "" otherwise * * This flag is initialized in onReady and used in attachLinkAfterNode */ var xfdType; /** * The current page name, including namespace, because we may be reading it * a lot (especially in findUsernameInElem if we're on someone's user * talk page) */ var currentPageName; /** * A map for signatures that contain redirects, so that they can still * pass the sanity check. This will be updated manually, because I * don't want the overhead of a whole 'nother API call in the middle * of the reply process. If this map grows too much, though, I'll * consider switching to either a toolforge-hosted API or the * Wikipedia API. Used in doReply, for the username sanity check. */ var sigRedirectMapping = { "Salvidrim": "Salvidrim!" }; /** * When the reply is saved via API, this flag is set to true to * disable the onbeforeunload handler. */ var replyWasSaved = false; /** * Cache for getWikitext. Only useful in test mode. */ var getWikitextCache = {}; // Polyfill from https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/String/includes if( !String.prototype.includes ) { String.prototype.includes = function( search, start ) { if( search instanceof RegExp ) { throw TypeError('first argument must not be a RegExp'); } if( start === undefined ) { start = 0; } return this.indexOf( search, start ) !== -1; }; } /** * Get the formatted namespace name for a namespace ID. * Quick ref: user = 2, proj = 4 */ function fmtNs( nsId ) { return mw.config.get( "wgFormattedNamespaces" )[ nsId ]; } /** * Escapes a string for inclusion in a regex. */ function escapeForRegex( s ) { return s.replace( /[-\/\\^$*+?.()|[\]{}]/g, '\\$&' ); } /* * MediaWiki turns spaces before certain punctuation marks * into non-breaking spaces, so fix those. This is done by * the armorFrenchSpaces function in Mediawiki, in the file * /includes/parser/Sanitizer.php */ function deArmorFrenchSpaces( text ) { return text.replace( /\xA0([?:;!%»›])/g, " $1" ) .replace( /([«‹])\xA0/g, "$1 " ); } /** * Capitalize the first letter of a string. */ function capFirstLetter( someString ) { return someString.charAt( 0 ).toUpperCase() + someString.slice( 1 ); } /** * Namespace name to ID. * For example, nsNameToId( "Template" ) === 10. */ function nsNameToId( nsName ) { return mw.config.get( "wgNamespaceIds" )[ nsName.toLowerCase().replace( / /g, "_" ) ]; } /** * Canonical-ize a namespace. */ function canonicalizeNs( ns ) { return fmtNs( nsNameToId( ns ) ); } /** * This function converts any (index-able) iterable into a list. */ function iterableToList( nl ) { var len = nl.length; var arr = new Array( len ); for( var i = 0; i < len; i++ ) arr[i] = nl[i]; return arr; } /** * Process HTML character entities. * From https://stackoverflow.com/a/46851765 */ function processCharEntities( text ) { var el = document.createElement('div'); return text.replace( /\&[#0-9a-z]+;/gi, function ( enc ) { el.innerHTML = enc; return el.innerText } ); } /** * Process HTML character entities, MediaWiki style * From https://stackoverflow.com/a/46851765 */ function processCharEntitiesWikitext( text ) { var el = document.createElement('div'); return text.replace( /\&[#0-9a-z]+;/gi, function ( enc ) { if( /#\d+/.test( enc ) ) { if( parseInt( enc.slice( 1 ) ) > MAX_UNICODE_DECIMAL ) { return enc; } } el.innerHTML = enc; return el.innerText } ); } /** * When there's a panel being shown, this function sets the status * in the panel to the first argument. The callback function is * optional. */ function setStatus ( status, callback ) { var statusElement = $( "#reply-dialog-status" ); statusElement.fadeOut( function () { statusElement.html( status ).fadeIn( callback ); } ); } /** * Sets the panel status when an error happened. Good for use in * catch blocks. */ function setStatusError( e ) { console.error(e); setStatus( mw.msg( "rl-error-status" ) ); if( e.message ) { console.log( "Content request error: " + JSON.stringify( e.message ) ); } console.log( "DEBUG INFORMATION: '"+currentPageName+"' @ " + mw.config.get( "wgCurRevisionId" ),"parsoid",PARSOID_ENDPOINT+ encodeURIComponent(currentPageName).replace(/'/g,"%27")+"/"+mw.config.get("wgCurRevisionId") ); throw e; } /** * Given some wikitext, processes it to get just the text content. * This function should be identical to the MediaWiki function * that gets the wikitext between the equal signs and comes up * with the id's that anchor the headers. */ function wikitextToTextContent( wikitext ) { return decodeURIComponent( processCharEntities( wikitext ) ) .replace( /\[\[:?(?:[^\|\]]+?\|)?([^\]\|]+?)\]\]/g, "$1" ) .replace( /\{\{\s*tl\s*\|\s*(.+?)\s*\}\}/g, "{{$1}}" ) .replace( /\{\{\s*[Uu]\s*\|\s*(.+?)\s*\}\}/g, "$1" ) .replace( /('''?)(.+?)\1/g, "$2" ) .replace( /<s>(.+?)<\/s>/g, "$1" ) .replace( /<big>(.+?)<\/big>/g, "$1" ) .replace( /<span.*?>(.*?)<\/span>/g, "$1" ); } function wikitextHeaderEqualsDomHeader( wikitextHeader, domHeader ) { return wikitextToTextContent( wikitextHeader ) === deArmorFrenchSpaces( domHeader ); } /** * Finds and returns the div that is the immediate parent of the * first talk page header on the page, so that we can read all the * sections by iterating through its child nodes. */ function findMainContentEl() { // Which header are we looking for? var targetHeader = "h2"; if( xfdType || currentPageName.startsWith( RFA_PG ) ) targetHeader = "h3"; if( currentPageName.startsWith( TTDYK ) ) targetHeader = "h4"; // The element itself will be the text span in the h2; its // parent will be the h2; and the parent of the h2 is the // content container that we want var candidates = document.querySelectorAll( targetHeader + " > span.mw-headline" ); if( !candidates.length ) return null; var candidate = candidates[candidates.length-1].parentElement.parentElement; // Compatibility with User:Enterprisey/hover-edit-section // That script puts each section in its own div, so we need to // go out another level if it's running if( candidate.className === "hover-edit-section" ) { return candidate.parentElement; } else { return candidate; } } /** * Gets the wikitext of a page with the given title (namespace required). * Returns an object with keys "content" and "timestamp". */ function getWikitext( title, useCaching ) { if( useCaching === undefined ) useCaching = false; if( useCaching && getWikitextCache[ title ] ) { return $.when( getWikitextCache[ title ] ); } return $.getJSON( mw.util.wikiScript( "api" ), { format: "json", action: "query", prop: "revisions", rvprop: "content", rvslots: "main", rvlimit: 1, titles: title } ).then( function ( data ) { var pageId = Object.keys( data.query.pages )[0]; if( data.query.pages[pageId].revisions ) { var revObj = data.query.pages[pageId].revisions[0]; var result = { timestamp: revObj.timestamp, content: revObj.slots.main["*"] }; getWikitextCache[ title ] = result; return result; } return {}; } ); } /** * Creates userspcLinkRgx. Called in handleWrapperClick and the test * runner at the bottom. */ function buildUserspcLinkRgx() { var nsIdMap = mw.config.get( "wgNamespaceIds" ); var nsRgxFragments = []; var contribsSecondFrag = ":" + escapeForRegex( mw.messages.get( "mycontris" ) ) + "\\/"; for( var nsName in nsIdMap ) { if( !nsIdMap.hasOwnProperty( nsName ) ) continue; switch( nsIdMap[nsName] ) { case 2: case 3: nsRgxFragments.push( escapeForRegex( capFirstLetter( nsName ) ) + "\\s*:" ); break; case -1: nsRgxFragments.push( escapeForRegex( capFirstLetter( nsName ) ) + contribsSecondFrag ); break; } } userspcLinkRgx = {}; userspcLinkRgx.spc = "(?:" + nsRgxFragments.join( "|" ).replace( /_/g, " " ) + ")"; userspcLinkRgx.und = userspcLinkRgx.spc.replace( / /g, "_" ); userspcLinkRgx.both = userspcLinkRgx.spc.replace( / /g, "(?: |_)" ); } /** * Is there a signature (four tildes) present in the given text, * outside of a nowiki element? */ function hasSig( text ) { // no literal signature? if( !text.includes( LITERAL_SIGNATURE ) ) return false; // if there's a literal signature and no nowiki elements, // there must be a real signature if( !text.includes( "<nowiki>" ) ) return true; // Save all nowiki spans var nowikiSpanStarts = []; // list of ignored span beginnings var nowikiSpanLengths = []; // list of ignored span lengths var NOWIKI_RE = /<nowiki>.*?<\/nowiki>/g; var spanMatch; do { spanMatch = NOWIKI_RE.exec( text ); if( spanMatch ) { nowikiSpanStarts.push( spanMatch.index ); nowikiSpanLengths.push( spanMatch[0].length ); } } while( spanMatch ); // So that we don't check every ignore span every time var nowikiSpanStartIdx = 0; var LIT_SIG_RE = new RegExp( LITERAL_SIGNATURE, "g" ); var sigMatch; matchLoop: do { sigMatch = LIT_SIG_RE.exec( text ); if( sigMatch ) { // Check that we're not inside a nowiki for( var nwIdx = nowikiSpanStartIdx; nwIdx < nowikiSpanStarts.length; nwIdx++ ) { if( sigMatch.index > nowikiSpanStarts[nwIdx] ) { if ( sigMatch.index + sigMatch[0].length <= nowikiSpanStarts[nwIdx] + nowikiSpanLengths[nwIdx] ) { // Invalid sig continue matchLoop; } else { // We'll never encounter this span again, since // headers only get later and later in the wikitext nowikiSpanStartIdx = nwIdx; } } } // We aren't inside a nowiki return true; } } while( sigMatch ); return false; } /** * Given an Element object, attempt to recover a username from it. * Also will check up to two elements prior to the passed element. * Returns null if no username was found. Otherwise, returns an * object with these properties: * * - username: The username that we found. * - link: The DOM object for the link from which we got the * username. */ function findUsernameInElem( el ) { if( !el ) return null; var links; for( let i = 0; i < 3; i++ ) { if( el === null ) break; links = el.tagName.toLowerCase() === "a" ? [ el ] : el.querySelectorAll( "a" ); //console.log(i,"top of outer for in findUsernameInElem ",el, " links -> ",links); // Compatibility with "Comments in Local Time" if( el.className.includes( "localcomments" ) ) i--; // If we couldn't get any links, try again with prev elem if( !links ) continue; var link; // his name isn't zelda for( var j = 0; j < links.length; j++ ) { link = links[j]; //console.log(link,decodeURIComponent(link.getAttribute("href"))); if( link.className.includes( "mw-selflink" ) ) { return { username: currentPageName.replace( /.+:/, "" ) .replace( /_/g, " " ), link: link }; } // Also matches redlinks. Why people have redlinks in their sigs on // purpose, I may never know. //console.log( "^\\/(?:wiki\\/" + userspcLinkRgx.und + /(.+?)(?:\/.+?)?(?:#.+)?|w\/index\.php\?title=User(?:_talk)?:(.+?)&action=edit&redlink=1/.source + ")$" ) var sigLinkRe = new RegExp( "\\/(?:wiki\\/" + userspcLinkRgx.und + /(.+?)(?:\/.+?)?(?:#.+)?|w\/index\.php\?title=/.source + userspcLinkRgx.und + /(.+?)&action=edit&redlink=1/.source + ")$" ); var liveDecodedHref = decodeURIComponent( link.getAttribute( "href" ) ); if( liveDecodedHref.startsWith( "/" ) ) { liveDecodedHref = "https:" + mw.config.get( "wgServer" ) + liveDecodedHref; } var usernameMatch = sigLinkRe.exec( liveDecodedHref ); if( usernameMatch ) { //console.log("usernameMatch",usernameMatch) var rawUsername = usernameMatch[1] ? usernameMatch[1] : usernameMatch[2]; return { username: decodeURIComponent( rawUsername ).replace( /_/g, " " ), link: link }; } } // Go backwards one element and try again el = el.previousElementSibling; } return null; } /** * Given a reply-link-wrapper span, attempts to find who wrote * the comment that precedes it. For information about the return * value, see the documentation for findUsernameInElem. */ function getCommentAuthor( wrapper ) { var sigNode = wrapper.previousSibling; //console.log(sigNode,sigNode.style,sigNode.style ? sigNode.style.getPropertyValue("size"):""); var smallOrFake = sigNode.nodeType === 1 && ( sigNode.tagName.toLowerCase() === "small" || ( sigNode.tagName.toLowerCase() === "span" && sigNode.style && ( sigNode.style.getPropertyValue( "font-size" ) === "85%" || sigNode.style.getPropertyValue( "font-size" ).indexOf( "small" ) === 0 ) ) ); var possUserLinkElem = ( smallOrFake && sigNode.children.length > 1 ) ? sigNode.children[sigNode.children.length-1] : sigNode.previousElementSibling; return findUsernameInElem( possUserLinkElem ); } /** * Given the wikitext of a section, attempt to find the first edit * request template in it, and then mark that template as answered. * Returns the modified section wikitext. */ function markEditReqAnswered( sectionWikitext ) { var editReqMatch = EDIT_REQ_TPL_REGEX.exec( sectionWikitext ); if( !editReqMatch ) { console.error( "Couldn't find an edit request!" ); return sectionWikitext; } var ansParamMatch = /ans(wered)?=.*?(\||\}\})/.exec( editReqMatch[0] ); if( !ansParamMatch ) { sectionWikitext = sectionWikitext.replace( editReqMatch[0], editReqMatch[0].replace( "}}", "answered=yes}}" ) ); } else { var newEditReqTpl = editReqMatch[0].replace( ansParamMatch[0], "answered=yes" + ansParamMatch[2] ); sectionWikitext = sectionWikitext.replace( editReqMatch[0], newEditReqTpl ); } return sectionWikitext; } /** * Ascend until dd or li, or a p directly under div.mw-parser-output. * live is true if we're on the live DOM (and thus we have our own UI * elements to deal with) and false if we're on the psd DOM. */ function ascendToCommentContainer( startNode, live, recordPath ) { var currNode = startNode; if( recordPath === undefined ) recordPath = false; var path = []; var lcTag; var headerRegex = /h\d+/i; function hasHeaderAsAnyPreviousSibling( node ) { do { if( headerRegex.test( node.tagName ) ) { return true; } node = node.previousElementSibling; } while( node ); } function isActualContainer( node, nodeLcTag ) { if( nodeLcTag === undefined ) nodeLcTag = node.tagName.toLowerCase(); return /dd|li/.test( nodeLcTag ) || ( ( nodeLcTag === "p" || nodeLcTag === "div" ) && ( node.parentNode.className === "mw-parser-output" || hasHeaderAsAnyPreviousSibling( node ) || node.parentNode.className === "hover-edit-section" || ( node.parentNode.tagName.toLowerCase() === "section" && node.parentNode.dataset.mwSectionId ) ) ); } var smallContainerNodeLimit = live ? 3 : 1; do { currNode = currNode.parentNode; lcTag = currNode.tagName.toLowerCase(); if( lcTag === "html" ) { console.error( "ascendToCommentContainer reached root" ); break; } if( recordPath ) path.unshift( currNode ); //console.log( "checking isActualContainer for ", currNode, isActualContainer( currNode, lcTag ), // lcTag === "small", isActualContainer( currNode.parentNode ), // currNode.parentNode.childNodes, // currNode.parentNode.childNodes.length ); } while( !isActualContainer( currNode, lcTag ) && !( lcTag === "small" && isActualContainer( currNode.parentNode ) && currNode.parentNode.childNodes.length <= smallContainerNodeLimit ) ); //console.log("ascendToCommentContainer from ",startNode," terminating, r.v. ",recordPath?path:currNode); return recordPath ? path : currNode; } /** * Given a Parsoid DOM and a link in the live DOM that is the link at the * end of a signature, return the corresponding element in the Parsoid DOM * that represents the same comment, or null if none was found. * * psd = Parsoid, live = in the current, live page DOM. */ function getCorrCmt( psdDom, sigLinkElem ) { // First, define some helper functions // Does this node have a timestamp in it? function hasTimestamp( node ) { //console.log ("hasTimestamp ",node, node.nodeType === 3,node.textContent.trim(), // TIMESTAMP_REGEX.test( node.textContent.trim() ), // node.childNodes.length === 1, // node.childNodes.length && TIMESTAMP_REGEX.test( node.childNodes[0].textContent.trim()), // " => ",( node.nodeType === 3 && // TIMESTAMP_REGEX.test( node.textContent.trim() ) ) || // ( node.childNodes.length === 1 && // TIMESTAMP_REGEX.test( node.childNodes[0].textContent.trim() ) ) ); //console.log(node,node.textContent.trim(),TIMESTAMP_REGEX.test(node.textContent.trim())); var validTag = node.nodeType === 3 || ( node.nodeType === 1 && ( node.tagName.toLowerCase() === "small" || node.tagName.toLowerCase() === "span" ) ); return ( validTag && TIMESTAMP_REGEX.test( node.textContent.trim() ) || ( node.childNodes.length === 1 && TIMESTAMP_REGEX.test( node.childNodes[0].textContent.trim() ) ) ); } // Get prefix that's the actual comment function getPrefixComment( theNodes ) { var prefix = []; for( var j = 0; j < theNodes.length; j++ ) { prefix.push( theNodes[j] ); if( hasTimestamp( theNodes[j] ) ) break; } return prefix; } /** * From a "container elem" (like the whole dd, li, or p that has a * comment), get the prefix that ends in a timestamp (because other * comments might be after the timestamp), and return the text content. */ function surrTextContentFromElem( elem ) { var surrListElemNodes = elem.childNodes; // nodeType 8 is for comments return getPrefixComment( surrListElemNodes ) .map( function ( c ) { return ( c.nodeType !== 8 ) ? c.textContent : ""; } ) .join( "" ).trim(); } /** From a "container elem" (dd, li, or p), remove all but the first comment. */ function onlyFirstComment( container ) { //console.log("onlyFirstComment top container and container.childNodes",container,container.childNodes); if( container.childNodes.length === 1 && container.children[0].tagName.toLowerCase() === "small" ) { console.log( "[onlyFirstComment] container only had a small in it" ); container = container.children[0]; } var i, autosignedIdx, autosigned = container.querySelector( "small.autosigned" ); if( autosigned && ( autosignedIdx = iterableToList( container.childNodes ).includes( autosigned ) ) ) { i = autosignedIdx; } else { var childNodes = container.childNodes; for( i = 0; i < childNodes.length; i++ ) { if( hasTimestamp( childNodes[i] ) ) { //console.log( "[oFC] found a timestamp in ",childNodes[i]); break; } } if( i === childNodes.length ) { throw new Error( "[onlyFirstComment] No timestamp found" ); } } //console.log("[onlyFirstComment] killing all after ",i,container.childNodes[i]); i++; var elemToRemove; while( elemToRemove = container.childNodes[i] ) { container.removeChild( elemToRemove ); } } // End helper functions, begin actual code // We dump this object for debugging in the event of an error var corrCmtDebug = {}; // Convert live href to psd href (aka newHref) var newHref, liveHref = decodeURIComponent( sigLinkElem.getAttribute( "href" ) ); corrCmtDebug.liveHref = liveHref; if( sigLinkElem.className.includes( "mw-selflink" ) ) { newHref = "./" + currentPageName; } else { if( /^\/wiki/.test( liveHref ) ) { var hrefTokens = liveHref.split( ":" ); if( hrefTokens.length !== 2 ) throw new Error( "Malformed href" ); newHref = "./" + canonicalizeNs( hrefTokens[0].replace( /^\/wiki\//, "" ) ).replace( / /g, "_" ) + ":" + hrefTokens[1] .replace( /^Contributions%2F/, "Contributions/" ) .replace( /%2F/g, "/" ) .replace( /%23/g, "#" ) .replace( /%26/g, "&" ) .replace( /%3D/g, "=" ) .replace( /%2C/g, "," ); } else { var REDLINK_HREF_RGX = /^\/w\/index\.php\?title=(.+?)&action=edit&redlink=1$/; var redlinkMatch = REDLINK_HREF_RGX.exec( liveHref ); if( redlinkMatch ) { newHref = "./" + redlinkMatch[1]; } else { newHref = liveHref.replace( /_/g, '%20' ); } } } newHref = newHref.replace( /\\/g, "\\\\" ) .replace( /'/g, "\\'" ) .replace( /\?/g, "%3F" ); var livePath = ascendToCommentContainer( sigLinkElem, /* live */ true, /* recordPath */ true ); corrCmtDebug.newHref = newHref; corrCmtDebug.livePath = livePath; // Deal with the case where the comment has multiple links to // sigLinkElem's href; we will store the index of the link we want. // null means there aren't multiple links. if( liveHref ) { liveHref = liveHref.replace( /'/g, "\\'" ); } var liveDupeLinks = livePath[0].querySelectorAll( "a" + ( liveHref ? ( "[href='" + liveHref + "']" ) : ".mw-selflink" ) ); if( !liveDupeLinks ) throw new Error( "Couldn't select live dupe link" ); var liveDupeLinkIdx = ( liveDupeLinks.length > 1 ) ? iterableToList( liveDupeLinks ).indexOf( sigLinkElem ) : null; //console.log("liveDupeLinkIdx",liveDupeLinkIdx); //console.log("livePath[0]",livePath[0],livePath[0].childNodes); var liveClone = livePath[0].cloneNode( /* deep */ true ); // Remove our own UI elements var ourUiSelector = ".reply-link-wrapper,#reply-link-panel"; iterableToList( liveClone.querySelectorAll( ourUiSelector ) ).forEach( function ( n ) { n.parentNode.removeChild( n ); } ); //console.log("(BEFORE) liveClone",liveClone,liveClone.childNodes); onlyFirstComment( liveClone ); //console.log("(AFTER) liveClone",liveClone,liveClone.childNodes); // Process it a bit to make it look a bit more like the Parsoid output var liveAutoNumberedLinks = liveClone.querySelectorAll( "a.external.autonumber" ); for( var i = 0; i < liveAutoNumberedLinks.length; i++ ) { liveAutoNumberedLinks[i].textContent = ""; } var liveSelflinks = liveClone.querySelectorAll( "a.mw-selflink.selflink" ); for( var i = 0; i < liveSelflinks.length; i++ ) { liveSelflinks[i].href = "/wiki/" + currentPageName; } // "Comments in Local Time" compatibility: the text content is // gonna contain the modified time stamp, but the original time // stamp is still there var localCommentsSpan = liveClone.querySelector( "span.localcomments" ); if( localCommentsSpan ) { var dateNode = document.createTextNode( localCommentsSpan.getAttribute( "title" ) ); localCommentsSpan.parentNode.replaceChild( dateNode, localCommentsSpan ); } // User:Writ Keeper/Scripts/teahouseTalkbackLink.js compatibility: // get rid of the |C|TB that it adds var teahouseTalkbackLink = liveClone.querySelector( "a[id^=TBsubmit]" ); if( teahouseTalkbackLink ) { teahouseTalkbackLink.parentNode.removeChild( teahouseTalkbackLink.nextSibling ); for( var ttlIdx = 0; ttlIdx < 3; ttlIdx++ ) { teahouseTalkbackLink.parentNode.removeChild( teahouseTalkbackLink.previousSibling ); } teahouseTalkbackLink.parentNode.removeChild( teahouseTalkbackLink ); } var adminMarksClass = liveClone.querySelectorAll( "b.adminMark" ); if ( adminMarksClass.length > 0 ) { adminMarksClass.forEach( function ( currentValue, currentIndex, listObj ) { currentValue.parentNode.removeChild( currentValue ); } ); } // TODO: Optimization - surrTextContentFromElem does the prefixing // operation a second time, even though we already called onlyFirstComment // on it. var liveTextContent = surrTextContentFromElem( liveClone ); console.log("liveTextContent >>>>>"+liveTextContent + "<<<<<"); function normalizeTextContent( tc ) { return deArmorFrenchSpaces( tc ); } liveTextContent = normalizeTextContent( liveTextContent ); var selector = livePath.map( function ( node ) { return node.tagName.toLowerCase(); } ).join( " " ) + " a[href^='" + newHref + "']"; // TODO: Optimization opportunity - run querySelectorAll only on the // section that we know contains the comment var psdLinks = iterableToList( psdDom.querySelectorAll( selector ) ); console.log("(",liveDupeLinkIdx, ")",selector, " --> ", psdLinks); var oldPsdLinks = psdLinks, newHrefLen = newHref.length, hrefSubstr; psdLinks = []; for( var i = 0; i < oldPsdLinks.length; i++ ) { hrefSubstr = oldPsdLinks[i].getAttribute( "href" ).substring( newHrefLen ); if( !hrefSubstr || hrefSubstr.indexOf( "#" ) === 0 ) { psdLinks.push( oldPsdLinks[i] ); } } // Narrow down by entire textContent of list element var psdCorrLinks = []; // the corresponding link elem(s) if( liveDupeLinkIdx === null ) { for( var i = 0; i < psdLinks.length; i++ ) { var psdContainer = ascendToCommentContainer( psdLinks[i], /* live */ false, true ); //console.log("psdContainer",psdContainer); var psdTextContent = normalizeTextContent( surrTextContentFromElem( psdContainer[0] ) ); //console.log(i,">>>"+psdTextContent+"<<<"); if( psdTextContent === liveTextContent ) { psdCorrLinks.push( psdLinks[i] ); } /* else { //console.log(i,"len: psd live",psdTextContent.length,liveTextContent.length); for(var j = 0; j < Math.min(psdTextContent.length, liveTextContent.length); j++) { if(psdTextContent.charAt(j)!==liveTextContent.charAt(j)) { //console.log(i,j,"psd live", psdTextContent.codePointAt(j), liveTextContent.codePointAt( j ) ); break; } } } */ } } else { for( var i = 0; i < psdLinks.length; i++ ) { var psdContainer = ascendToCommentContainer( psdLinks[i], /* live */ false ); if( psdContainer.dataset.replyLinkGeCorrCo ) continue; var psdTextContent = normalizeTextContent( surrTextContentFromElem( psdContainer ) ); console.log(i,">>>"+psdTextContent+"<<<"); if( psdTextContent === liveTextContent ) { var psdDupeLinks = psdContainer.querySelectorAll( "a[href='" + newHref + "']" ); psdCorrLinks.push( psdDupeLinks[ liveDupeLinkIdx ] ); } // Flag to ensure we don't take a link from this container again psdContainer.dataset.replyLinkGeCorrCo = true; } } if( psdCorrLinks.length === 0 ) { console.error( "Failed to find a matching comment in the Parsoid DOM." ); return null; } else if( psdCorrLinks.length > 1 ) { console.error( "Found multiple matching comments in the Parsoid DOM." ); return null; } return psdCorrLinks[0]; } /** * Given a page title, the Parsoid output (GET /page/html endpoint) * of that page, page and a DOM object in the current page * corresponding to a link in a signature, locate the section * containing that comment. That section may not be in the provided * page! Returns an object with these properties: * * - page: The full title of the page directly containing the * comment (in its wikitext, not through transclusion). * - sectionName: The anticipated wikitext section name. Should * appear inside the equal signs at the above index. * - sectionDupeIdx: If there are multiple sections with the same * name, the 0-based index of the section with the comment among * those sections. Otherwise, 0. * - sectionLevel: The anticipated wikitext section level (e.g. * 2 for an h2) * - nearbyMwId: The Parsoid ID of some element near the * comment (in practice, a userspace link) for jumping purposes. * * Parsoid is abbreviated here as "psd" in variables and comments. */ function findSection( psdDomPageTitle, psdDomString, sigLinkElem ) { console.log("findSection(",psdDomPageTitle,", ...)"); //console.log(psdDomString); var domParser = new DOMParser(), psdDom = domParser.parseFromString( psdDomString, "text/html" ); var corrLink = getCorrCmt( psdDom, sigLinkElem ); if( corrLink === null ) { return $.when(); } //console.log("STEP 1 SUCCESS",corrLink); var corrCmt = ascendToCommentContainer( corrLink, /* live */ false ); // Ascend until we hit something in a transclusion var currNode = corrLink; var tsclnId = null; do { if( currNode.getAttribute( "about" ) && currNode.getAttribute( "about" ).indexOf( "#mwt" ) === 0 ) { tsclnId = currNode.getAttribute( "about" ); break; } currNode = currNode.parentNode; } while( currNode.tagName.toLowerCase() !== "html" ); //console.log( "tsclnId", tsclnId ); // Helper function: are we in a pseudo-section? (Unused, at the moment.) function inPseudo( headerElement ) { var currNodeIP = headerElement; // This requires Parsoid HTML v 2.0.0 do { if( currNodeIP.nodeType === 1 && currNodeIP.matches( "section" ) ) { return currNodeIP.dataset.mwSectionId < 0; break; } currNodeIP = currNodeIP.parentNode; } while( currNodeIP ); return false; } // Now, get the nearest header above us var currNode = corrCmt; var nearestHeader = null; var HTML_HEADER_RGX = /^h\d$/; do { if( HTML_HEADER_RGX.exec( currNode.tagName.toLowerCase() ) ) { // Commented because I don't think the !inPseudo requirement is necessary 2019-nov-01 //if( !inPseudo( currNode ) ) { nearestHeader = currNode; break; //} } var containedHeaders = currNode.querySelectorAll( HEADER_SELECTOR ); if( containedHeaders.length ) { var nearestHdrIdx = containedHeaders.length - 1; // Commented because I don't think the !inPseudo requirement is necessary 2019-nov-01 // TODO this is an extraordinarily silly while loop; it has been temporarily commented 2020-apr-25 //while( nearestHdrIdx >= 0 ){//&& inPseudo( containedHeaders[ nearestHdrIdx ] ) ) // nearestHdrIdx--; //} if( nearestHdrIdx >= 0 ) { nearestHeader = containedHeaders[ nearestHdrIdx ]; break; } } if( currNode.previousElementSibling ) { currNode = currNode.previousElementSibling; continue; } currNode = currNode.parentNode; } while( currNode.tagName.toLowerCase() !== "body" ); // Get the target page (page actually containing the comment) var targetPage; if( tsclnId === null ) { console.warn( "tsclnId === null" ); targetPage = psdDomPageTitle; } else { var tsclnInfoSel = "*[about='" + tsclnId + "'][typeof='mw:Transclusion']", infoJson = JSON.parse( psdDom.querySelector( tsclnInfoSel ) .dataset.mw ); // First, check the first and last wikitext segments to see if they have the header var firstWktxtSegIdx = 0; while( infoJson.parts[firstWktxtSegIdx].template && infoJson.parts[firstWktxtSegIdx].template.target.href.startsWith( "Template:" ) && firstWktxtSegIdx < infoJson.parts.length ) { firstWktxtSegIdx++; } if( firstWktxtSegIdx < infoJson.parts.length && typeof infoJson.parts[firstWktxtSegIdx] === typeof '' ) { var firstWktxtSeg = infoJson.parts[firstWktxtSegIdx]; var headerMatch = null; do { headerMatch = HEADER_REGEX.exec( firstWktxtSeg ); if( headerMatch ) { if( wikitextHeaderEqualsDomHeader( headerMatch[2], nearestHeader.textContent ) ) { targetPage = psdDomPageTitle; break; } } } while( headerMatch ); } } if( !targetPage ) { var lastWktxtSegIdx = infoJson.parts.length - 1; while( infoJson.parts[lastWktxtSegIdx].template && infoJson.parts[lastWktxtSegIdx].template.target.href.startsWith( "Template:" ) && lastWktxtSegIdx >= 0 ) { lastWktxtSegIdx--; } if( lastWktxtSegIdx >= 0 && typeof infoJson.parts[lastWktxtSegIdx] === typeof '' ) { var lastWktxtSeg = infoJson.parts[lastWktxtSegIdx]; var headerMatch = null; do { headerMatch = HEADER_REGEX.exec( lastWktxtSeg ); if( headerMatch ) { if( wikitextHeaderEqualsDomHeader( headerMatch[2], nearestHeader.textContent ) ) { targetPage = psdDomPageTitle; break; } } } while( headerMatch ); } } var recursiveCalls = $.when(); if( !targetPage ) { // Recurse on all non-top-level Templates! var pages = infoJson.parts.filter( function ( part ) { return part.template && part.template.target && part.template.target.href && ( !part.template.target.href.startsWith("./Template") || ( part.template.target.href.match( new RegExp( '/', 'g' ) ) || [] ).length >= 2 ); } ); if( pages.length ) { var pageNames = pages.map( function ( part ) { return part.template.target.href.substring( 2 ); // remove the ./ } ); var deferreds = pageNames.map( function ( pageName ) { return $.get( PARSOID_ENDPOINT + encodeURIComponent( pageName ) ) .then( function ( data ) { return data; } ); // truncate to first argument, which is the data } ); recursiveCalls = $.when.apply( $, deferreds ).then( function () { var results = arguments; // use keyword "arguments" to access deferred results var deferreds2 = []; if( pageNames.length !== results.length ) { console.error(pageNames,results); throw new Error( "pageNames.length !== results.length: " + pageNames.length + " " + results.length ); } for( var i = 0; i < pageNames.length; i++ ) { deferreds2.push( findSection( pageNames[i], results[i], sigLinkElem ) ); } return $.when.apply( $, deferreds2 ).then( function () { var results2 = Array.prototype.slice.call( arguments ); var namesAndResults2 = []; if( pageNames.length !== results2.length ) { throw new Error( "pageNames.length !== results2.length: " + pageNames.length + " " + results2.length ); } for( var i = 0; i < pageNames.length; i++ ) { if( results2[i] ) { namesAndResults2.push( [ pageNames[i], results2[i] ] ); } } if( namesAndResults2.length === 0 ) { return null; } else if( namesAndResults2.length === 1 ) { return namesAndResults2[0][1]; } else { var allSameName = namesAndResults2.every( function ( nameAndResult ) { return nameAndResult[0] === namesAndResults2[0][0]; } ); if( allSameName ) { return namesAndResults2[0][1]; } else { console.error( "WTF", namesAndResults2 ); } } } ); } ); } } return recursiveCalls.then( function ( data ) { if( data ) { return data; } else if( nearestHeader === null ) { return { page: targetPage, sectionName: "", sectionDupeIdx: 0, sectionLevel: 0, nearbyMwId: corrCmt.id }; } else { // We tried recursing, and it didn't work, so the // section must be on the current page targetPage = psdDomPageTitle; // Finally, get the index of our nearest header var allHeaders = iterableToList( psdDom.querySelectorAll( HEADER_SELECTOR ) ); var sectionDupeIdx = 0; for( var i = 0; i < allHeaders.length; i++ ) { if( allHeaders[i].textContent === nearestHeader.textContent ) { if( allHeaders[i] === nearestHeader ) { break; } else { sectionDupeIdx++; } } } var result = { page: targetPage, sectionName: nearestHeader.textContent, sectionDupeIdx: sectionDupeIdx, sectionLevel: nearestHeader.tagName.substring( 1 ), // that is, cut off the "h" at the beginning nearbyMwId: corrCmt.id }; //console.log("findSection return val: ",result); return result; } } ); } /** * Given some wikitext that's split into sections, return the full * wikitext (including header and newlines until the next header) of * the section with the given name. To get the content before the * first header, sectionName should be "". * * Performs a sanity check with the given section name. */ function getSectionWikitext( wikitext, sectionName, sectionDupeIdx ) { console.log("In getSectionWikitext, sectionName = >" + sectionName + "< (wikitext.length = " + wikitext.length + ")"); //console.log("wikitext (first 1000 chars) is " + dirtyWikitext.substring(0, 1000)); // There are certain locations where a header may appear in the // wikitext, but will not be present in the HTML; such as code // blocks or comments. So we keep track of those ranges // and ignore headings inside those. var ignoreSpanStarts = []; // list of ignored span beginnings var ignoreSpanLengths = []; // list of ignored span lengths var IGNORE_RE = /(<pre>[\s\S]+?<\/pre>)|(<source.+?>[\s\S]+?<\/source>)|(<!--[\s\S]+?-->)/g; var ignoreSpanMatch; do { ignoreSpanMatch = IGNORE_RE.exec( wikitext ); if( ignoreSpanMatch ) { //console.log("ignoreSpan ",ignoreSpanStarts.length," = ",ignoreSpanMatch); ignoreSpanStarts.push( ignoreSpanMatch.index ); ignoreSpanLengths.push( ignoreSpanMatch[0].length ); } } while( ignoreSpanMatch ); var startIdx = -1; // wikitext index of section start var endIdx = -1; // wikitext index of section end var headerCounter = 0; var headerMatch; // So that we don't check every ignore span every time var ignoreSpanStartIdx = 0; var dupeCount = 0; var lookingForEnd = false; if( sectionName === "" ) { // Getting first section startIdx = 0; lookingForEnd = true; } // Reset regex state, if for some reason we're not running this for the first time HEADER_REGEX.lastIndex = 0; headerMatchLoop: do { headerMatch = HEADER_REGEX.exec( wikitext ); if( headerMatch ) { // Check that we're not inside one of the "ignore" spans for( var igIdx = ignoreSpanStartIdx; igIdx < ignoreSpanStarts.length; igIdx++ ) { if( headerMatch.index > ignoreSpanStarts[igIdx] ) { if ( headerMatch.index + headerMatch[0].length <= ignoreSpanStarts[igIdx] + ignoreSpanLengths[igIdx] ) { console.log("(IGNORED, igIdx="+igIdx+") Header " + headerCounter + " (idx " + headerMatch.index + "): >" + headerMatch[0].trim() + "<"); // Invalid header continue headerMatchLoop; } else { // We'll never encounter this span again, since // headers only get later and later in the wikitext ignoreSpanStartIdx = igIdx; } } } //console.log("Header " + headerCounter + " (idx " + headerMatch.index + "): >" + headerMatch[0].trim() + "<"); // Note that if the lookingForEnd block were second, // then two consecutive matching section headers might // result in the wrong section being matched! if( lookingForEnd ) { endIdx = headerMatch.index; break; } else if( wikitextHeaderEqualsDomHeader( /* wikitext */ headerMatch[2], /* dom */ sectionName ) ) { if( dupeCount === sectionDupeIdx ) { startIdx = headerMatch.index; lookingForEnd = true; } else { dupeCount++; } } } headerCounter++; } while( headerMatch ); if( startIdx < 0 ) { throw( "Could not find section named \"" + sectionName + "\" (dupe idx " + sectionDupeIdx + ")" ); } // If we encountered no section after the target section, // then the target was the last one and the slice will go // until the end of wikitext if( endIdx < 0 ) { //console.log("[getSectionWikitext] endIdx negative, setting to " + wikitext.length); endIdx = wikitext.length; } //console.log("[getSectionWikitext] Slicing from " + startIdx + " to " + endIdx); return wikitext.slice( startIdx, endIdx ); } /** * Converts a signature index to a string index into the given * section wikitext. For example, if sigIdx is 1, then this function * will return the index in sectionWikitext pointing to right * after the second signature appearing in sectionWikitext. * * Returns -1 if we couldn't find anything. */ function sigIdxToStrIdx( sectionWikitext, sigIdx ) { console.log( "In sigIdxToStrIdx, sigIdx = " + sigIdx ); // There are certain regions that we skip while attaching links: // // - Spans with the class delsort-notice // - Divs with the class xfd-relist (and other divs) // // So, we grab the corresponding wikitext regions with regexes, // and store each region's start index in spanStartIndices, and // each region's length in spanLengths. Then, whenever we find a // signature with the right index, if it's included in one of // these regions, we skip it and move on. var spanStartIndices = []; var spanLengths = []; var DELSORT_SPAN_RE_TXT = /<small class="delsort-notice">(?:<small>.+?<\/small>|.)+?<\/small>/.source; var XFD_RELIST_RE_TXT = /<div class="xfd_relist"[\s\S]+?<\/div>(\s*|<!--.+?-->)*/.source; var STRUCK_RE_TXT = /<s>.+?<\/s>/.source; var SKIP_REGION_RE = new RegExp("(" + DELSORT_SPAN_RE_TXT + ")|(" + XFD_RELIST_RE_TXT + ")|(" + STRUCK_RE_TXT + ")", "ig"); var skipRegionMatch; do { skipRegionMatch = SKIP_REGION_RE.exec( sectionWikitext ); if( skipRegionMatch ) { spanStartIndices.push( skipRegionMatch.index ); spanLengths.push( skipRegionMatch[0].length ); } } while( skipRegionMatch ); //console.log(spanStartIndices,spanLengths); var dateFmtRgx = DATE_FMT_RGX[mw.config.get( "wgServer" )]; if( !dateFmtRgx ) { throw new Error( "Error! I don't know the native date format used by the server '" + mw.config.get( "wgServer" ) + "'!" ); } /* * I apologize for making you have to read this regex. * I made a summary, though: * * - a wikilink, without a ]] inside it * - some text, without a link to userspace or user talk space * - a timestamp * - as an alternative to all of the above, an autosigned script * and a timestamp * - some comments/whitespace or some non-whitespace * - finally, the end of the line * * It's also localized. */ var sigRgxSrc = "(?:" + /\[\[\s*(?:m:)?:?\s*/.source + "(" + userspcLinkRgx.both + /([^\]]|\](?!\]))*?/.source + ")" + /\]\]\)?/.source + "(" + /[^\[]|\[(?!\[)|\[\[/.source + "(?!" + userspcLinkRgx.both + "))*?" + DATE_FMT_RGX[mw.config.get( "wgServer" )] + /\s+\(UTC\)|class\s*=\s*"autosigned".+?\(UTC\)<\/small>/.source + ")" + /(\S*([ \t\f]|<!--.*?-->)*(?:\{\{.+?\}\})?(?!\S)|\s?\S+([ \t\f]|<!--.*?-->)*)$/.source; var sigRgx = new RegExp( sigRgxSrc, "igm" ); var matchIdx = 0; var match; var matchIdxEnd; var dstSpnIdx; sigMatchLoop: for( ; true ; matchIdx++ ) { match = sigRgx.exec( sectionWikitext ); if( !match ) { console.error("[sigIdxToStrIdx] out of matches"); return -1; } //console.log( "sig match (matchIdx = " + matchIdx + ") is >" + match[0] + "< (index = " + match.index + ")" ); matchIdxEnd = match.index + match[0].length; // Validate that we're not inside a delsort span for( dstSpnIdx = 0; dstSpnIdx < spanStartIndices.length; dstSpnIdx++ ) { //console.log(spanStartIndices[dstSpnIdx],match.index, // matchIdxEnd, spanStartIndices[dstSpnIdx] + // spanLengths[dstSpnIdx] ); if( match.index > spanStartIndices[dstSpnIdx] && ( matchIdxEnd <= spanStartIndices[dstSpnIdx] + spanLengths[dstSpnIdx] ) ) { // That wasn't really a match (as in, this match does not // correspond to any sig idx in the DOM), so we can't // increment matchIdx matchIdx--; continue sigMatchLoop; } } if( matchIdx === sigIdx ) { return match.index + match[0].length; } } } /** * Inserts fullReply on the next sensible line after strIdx in * sectionWikitext. indentLvl is the indentation level of the * comment we're replying to. * * This function essentially takes the indentation level and * position of the current comment, and looks for the first comment * that's indented strictly less than the current one. Then, it * puts the reply on the line right before that comment, and returns * the modified section wikitext. */ function insertTextAfterIdx( sectionWikitext, strIdx, indentLvl, fullReply ) { //console.log( "[insertTextAfterIdx] indentLvl = " + indentLvl ); // strIdx should point to the end of a line var counter = 0; while( ( sectionWikitext[ strIdx ] !== "\n" ) && ( counter++ <= 50 ) ) strIdx++; var slicedSecWikitext = sectionWikitext.slice( strIdx ); //console.log("slicedSecWikitext = >>" + slicedSecWikitext.slice(0,50) + "<<"); slicedSecWikitext = slicedSecWikitext.replace( /^\n/, "" ); var candidateLines = slicedSecWikitext.split( "\n" ); //console.log( "candidateLines =", candidateLines ); // number of the line in sectionWikitext that'll be right after reply var replyLine = 0; var INDENT_RE = /^[:*#]+/; if( slicedSecWikitext.trim().length > 0 ) { var currIndentation, currIndentationLvl, i; // Now, loop through all the comments replying to that // one and place our reply after the last one for( i = 0; i < candidateLines.length; i++ ) { if( candidateLines[i].trim() === "" ) { continue; } // Detect indentation level of current line currIndentation = INDENT_RE.exec( candidateLines[i] ); currIndentationLvl = currIndentation ? currIndentation[0].length : 0; //console.log(i + ">" + candidateLines[i] + "< => " + currIndentationLvl); if( currIndentationLvl <= indentLvl ) { // If it's an XfD, we might have found a relist // comment instead, so check for that if( xfdType && /<div class="xfd_relist"/.test( candidateLines[i] ) ) { // Our reply might go on the line above the xfd_relist line var potentialReplyLine = i; // Walk through the relist notice, line by line // After this loop, i will point to the line on which // the notice ends var NEW_COMMENTS_RE = /Please add new comments below this line/; while( !NEW_COMMENTS_RE.test( candidateLines[i] ) ) { i++; } // Relists are treated as if they're indented at level 1 if( 1 <= indentLvl ) { replyLine = potentialReplyLine; break; } } else { //console.log( "cIL <= iL, breaking" ); break; } } else { replyLine = i + 1; } } if( i === candidateLines.length ) { replyLine = i; } } else { // In this case, we may be replying to the last comment in a section replyLine = candidateLines.length; } // Walk backwards until non-empty line while( replyLine >= 1 && candidateLines[replyLine - 1].trim() === "" ) replyLine--; //console.log( "replyLine = " + replyLine ); // Splice into slicedSecWikitext slicedSecWikitext = candidateLines .slice( 0, replyLine ) .concat( [ fullReply ], candidateLines.slice( replyLine ) ) .join( "\n" ); // Remove extra newlines if( /\n\n\n+$/.test( slicedSecWikitext ) ) { slicedSecWikitext = slicedSecWikitext.trim() + "\n\n"; } // We may need an additional newline if the two slices don't have any var optionalNewline = ( !sectionWikitext.slice( 0, strIdx ).endsWith( "\n" ) && !slicedSecWikitext.startsWith( "\n" ) ) ? "\n" : ""; // Splice into sectionWikitext sectionWikitext = sectionWikitext.slice( 0, strIdx ) + optionalNewline + slicedSecWikitext; return sectionWikitext; } /** * Using the text in #reply-dialog-field, add a reply to the * current page. rplyToXfdNom is true if we're replying to an XfD nom, * in which case we should use an asterisk instead of a colon. * cmtAuthorDom is the username of the person who wrote the comment * we're replying to, parsed from the DOM. revObj is the object returned * by getWikitext for the page with the comment; findSectionResult is the * object returned by findSection for the comment. * * Returns a Deferred that resolves/rejects when the reply succeeds/fails. */ function doReply( indentation, header, sigIdx, cmtAuthorDom, rplyToXfdNom, revObj, findSectionResult ) { console.log("TOP OF doReply",header,findSectionResult); header = [ "" + findSectionResult.sectionLevel, findSectionResult.sectionName, findSectionResult.sectionDupeIdx ]; var deferred = $.Deferred(); var wikitext = revObj.content; try { // Generate reply in wikitext form var reply = document.getElementById( "reply-dialog-field" ).value.trim(); // Add a signature if one isn't already there if( !hasSig( reply ) ) { reply += " " + ( window.replyLinkSigPrefix ? window.replyLinkSigPrefix : "" ) + LITERAL_SIGNATURE; } var isUsingAutoIndentation = window.replyLinkAutoIndentation === "checkbox" ? ( !document.getElementById( "reply-link-option-auto-indent" ) || document.getElementById( "reply-link-option-auto-indent" ).checked ) : window.replyLinkAutoIndentation === "always"; if( isUsingAutoIndentation ) { var replyLines = reply.split( "\n" ); // If we're outdenting, reset indentation and add the // outdent template. This requires that there be at least // one character of indentation. var outdentCheckbox = document.getElementById( "reply-link-option-outdent" ); if( outdentCheckbox && outdentCheckbox.checked ) { replyLines[0] = "{" + "{od|" + indentation.slice( 0, -1 ) + "}}" + replyLines[0]; indentation = ""; } // Compose reply by adding indentation at the beginning of // each line (if not replying to an XfD nom) or {{pb}}'s // between lines (if replying to an XfD nom) var fullReply; if( rplyToXfdNom ) { // If there's a list in this reply, it's a bad idea to // use pb's, even though the markup'll probably be broken if( replyLines.some( function ( l ) { return l.substr( 0, 1 ) === "*"; } ) ) { fullReply = replyLines.map( function ( line ) { return indentation + "*" + line; } ).join( "\n" ); } else { fullReply = indentation + "* " + replyLines.join( "{{pb}}" ); } } else { fullReply = replyLines.map( function ( line ) { return indentation + ":" + line; } ).join( "\n" ); } } else { fullReply = reply; } // Prepare section metadata for getSectionWikitext call console.log( "in doReply, header =", header ); var sectionHeader, sectionIdx; if( header === null ) { sectionHeader = null, sectionIdx = -1; } else { sectionHeader = header[1], sectionDupeIdx = header[2]; } // Compatibility with User:Bility/copySectionLink if( document.querySelector( "span.mw-headline a#sectiontitlecopy0" ) ) { // If copySectionLink is active, the paragraph symbol at // the end is a fake sectionHeader = sectionHeader.replace( /\s*¶$/, "" ); } // Compatibility with the "auto-number headings" preference if( document.querySelector( "span.mw-headline-number" ) ) { sectionHeader = sectionHeader.replace( /^\d+ /, "" ); } var sectionWikitext = getSectionWikitext( wikitext, sectionHeader, sectionDupeIdx ); var oldSectionWikitext = sectionWikitext; // We'll String.replace old w/ new // Now, obtain the index of the end of the comment var strIdx = sigIdxToStrIdx( sectionWikitext, sigIdx ); // Check for a non-negative strIdx if( strIdx < 0 ) { throw( "Negative strIdx (signature not found in wikitext)" ); } // Determine the user who wrote the comment, for // edit-summary and sanity-check purposes var userRgx = new RegExp( /\[\[\s*(?:m:)?:?\s*/.source + userspcLinkRgx.both + /\s*(.+?)(?:\/.+?)?(?:#.+?)?\s*(?:\|.+?)?\]\]/.source, "ig" ); var userMatches = processCharEntitiesWikitext( sectionWikitext.slice( 0, strIdx ) ).match( userRgx ); var cmtAuthorWktxt = userRgx.exec( userMatches[userMatches.length - 1] )[1]; if( cmtAuthorWktxt === "DoNotArchiveUntil" ) { userRgx.lastIndex = 0; cmtAuthorWktxt = userRgx.exec( userMatches[userMatches.length - 2] )[1]; } // Normalize case, because that's what happens during // wikitext-to-HTML processing; also underscores to spaces function sanitizeUsername( u ) { u = u.charAt( 0 ).toUpperCase() + u.substr( 1 ); return u.replace( /_/g, " " ); } cmtAuthorWktxt = sanitizeUsername( cmtAuthorWktxt ); cmtAuthorDom = sanitizeUsername( cmtAuthorDom ); // Do a sanity check: is the sig username the same as the // DOM one? We attempt to check sigRedirectMapping in case // the naive check fails if( cmtAuthorWktxt !== cmtAuthorDom && processCharEntitiesWikitext( cmtAuthorWktxt ) !== cmtAuthorDom && sigRedirectMapping[ cmtAuthorWktxt ] !== cmtAuthorDom ) { throw new Error( "Sanity check on sig username failed! Found " + cmtAuthorWktxt + " but expected " + cmtAuthorDom + " (wikitext vs DOM)" ); } // Actually insert our reply into the section wikitext sectionWikitext = insertTextAfterIdx( sectionWikitext, strIdx, indentation.length, fullReply ); // Also, if the user wanted the edit request to be answered, // do that var editReqCheckbox = document.getElementById( "reply-link-option-edit-req" ); var markedEditReq = false; if( editReqCheckbox && editReqCheckbox.checked ) { sectionWikitext = markEditReqAnswered( sectionWikitext ); markedEditReq = true; } // If the user preferences indicate a dry run, print what the // wikitext would have been post-edit and bail out var dryRunCheckbox = document.getElementById( "reply-link-option-dry-run" ); if( window.replyLinkDryRun === "always" || ( dryRunCheckbox && dryRunCheckbox.checked ) ) { console.log( "~~~~~~ DRY RUN CONCLUDED ~~~~~~" ); console.log( sectionWikitext ); setStatus( "Check the console for the dry-run results." ); document.querySelector( "#reply-link-buttons button" ).disabled = false; deferred.resolve(); return deferred; } var newWikitext = wikitext.replace( oldSectionWikitext, sectionWikitext ); // Build summary var defaultSummmary = mw.msg( "rl-replying-to" ) + ( rplyToXfdNom ? xfdType + " nomination by " : "" ) + cmtAuthorWktxt + ( markedEditReq ? " and marking edit request as answered" : "" ); var customSummaryField = document.getElementById( "reply-link-summary" ); var summaryCore = defaultSummmary; if( window.replyLinkCustomSummary && customSummaryField.value ) { summaryCore = customSummaryField.value.trim(); } var summary = "/* " + sectionHeader + " */ " + summaryCore + mw.msg( "rl-advert" ); // Send another request, this time to actually edit the // page api.postWithToken( "csrf", { action: "edit", title: findSectionResult.page, summary: summary, text: newWikitext, basetimestamp: revObj.timestamp } ).done ( function ( data ) { // We put this function on the window object because we // give the user a "reload" link, and it'll trigger the function window.replyLinkReload = function () { window.location.hash = sectionHeader.replace( / /g, "_" ); if( findSectionResult.nearbyMwId ) { document.cookie = "parsoid_jump=" + findSectionResult.nearbyMwId; } window.location.reload( true ); }; if ( data && data.edit && data.edit.result && data.edit.result == "Success" ) { var needPurge = findSectionResult.page !== currentPageName; function finishReply( _ ) { var reloadHtml = window.replyLinkAutoReload ? mw.msg( "rl-reloading" ) : "<a href='javascript:window.replyLinkReload()' class='reply-link-reload'>" + mw.msg( "rl-reload" ) + "</a>"; setStatus( mw.msg( "rl-saved" ) + " (" + reloadHtml + ")" ); // Required to permit reload to happen, checked in onbeforeunload replyWasSaved = true; if( window.replyLinkAutoReload ) { window.replyLinkReload(); } deferred.resolve(); } if( needPurge ) { setStatus( "Reply saved! Purging..." ); api.post( { action: "purge", titles: currentPageName } ).done( finishReply ); } else { finishReply(); } } else { if( data && data.edit && data.edit.spamblacklist ) { setStatus( "Error! Your post contained a link on the <a href=" + "\"https://en.wikipedia.org/wiki/Wikipedia:Spam_blacklist\"" + ">spam blacklist</a>. Remove the link(s) to: " + data.edit.spamblacklist.split( "|" ).join( ", " ) + " to allow saving." ); document.querySelector( "#reply-link-buttons button" ).disabled = false; } else { setStatus( "While saving, the edit query returned an error." + " Check the browser console for more information." ); } deferred.reject(); } //console.log(data); document.getElementById( "reply-dialog-field" ).style["background-image"] = ""; } ).fail ( function( code, result ) { setStatus( "While replying, the edit failed." ); console.log(code); console.log(result); deferred.reject(); } ); } catch ( e ) { setStatusError( e ); deferred.reject(); } return deferred; } function handleWrapperClick ( linkLabel, parent, rplyToXfdNom ) { return function ( evt ) { $.when( mw.messages.exists( INT_MSG_KEYS[0] ) ? 1 : api.loadMessages( INT_MSG_KEYS ) ).then( function () { var newLink = this; var newLinkWrapper = this.parentNode; if( !userspcLinkRgx ) { buildUserspcLinkRgx(); } // Remove previous panel var prevPanel = document.getElementById( "reply-link-panel" ); if( prevPanel ) { prevPanel.remove(); } // Reset previous cancel links var cancelLinks = iterableToList( document.querySelectorAll( ".reply-link-wrapper a" ) ); cancelLinks.forEach( function ( el ) { if( el != newLink ) el.textContent = el.dataset.originalLabel; } ); // Handle disable action if( newLink.textContent === linkLabel ) { // Disable this link newLink.textContent = mw.msg( "rl-cancel" ) + linkLabel; } else { // We've already cancelled the reply newLink.textContent = linkLabel; evt.preventDefault(); return false; } // Figure out the username of the author // of the comment we're replying to var cmtAuthorAndLink = getCommentAuthor( newLinkWrapper ); try { var cmtAuthor = cmtAuthorAndLink.username, cmtLink = cmtAuthorAndLink.link; } catch ( e ) { setStatusError( e ); } // Create panel var panelEl = document.createElement( "div" ); panelEl.id = "reply-link-panel"; panelEl.innerHTML = "<textarea id='reply-dialog-field' class='mw-ui-input'" + " placeholder='" + mw.msg( "rl-placeholder" ) + "'></textarea>" + ( window.replyLinkCustomSummary ? "<label for='reply-link-summary'>Summary: </label>" + "<input id='reply-link-summary' class='mw-ui-input' placeholder='Edit summary' " + "value='Replying to " + cmtAuthor.replace( /'/g, "'" ) + "'/><br />" : "" ) + "<table style='border-collapse:collapse'><tr><td id='reply-link-buttons' style='width: " + ( window.replyLinkPreloadPing === "button" ? "325" : "255" ) + "px'>" + "<button id='reply-dialog-button' class='mw-ui-button mw-ui-progressive'>" + mw.msg( "rl-reply" ) + "</button> " + "<button id='reply-link-preview-button' class='mw-ui-button'>" + mw.msg( "rl-preview" ) + "</button>" + ( window.replyLinkPreloadPing === "button" ? " <button id='reply-link-ping-button' class='mw-ui-button'>Ping</button>" : "" ) + "<button id='reply-link-cancel-button' class='mw-ui-button mw-ui-quiet mw-ui-destructive'>" + mw.msg( "rl-cancel-button" ) + "</button></td>" + "<td id='reply-dialog-status'></span><div style='clear:left'></td></tr></table>" + "<div id='reply-link-options' class='gone-on-empty' style='margin-top: 0.5em'></div>" + "<div id='reply-link-preview' class='gone-on-empty' style='border: thin dashed gray; padding: 0.5em; margin-top: 0.5em'></div>"; parent.insertBefore( panelEl, newLinkWrapper.nextSibling ); var replyDialogField = document.getElementById( "reply-dialog-field" ); replyDialogField.style = "padding: 0.625em; min-height: 10em; margin-bottom: 0.75em; line-height: 1.3"; if( window.replyLinkPreloadPing === "always" && cmtAuthor && cmtAuthor !== mw.config.get( "wgUserName" ) && !/(\d+.){3}\d+/.test( cmtAuthor ) ) { replyDialogField.value = window.replyLinkPreloadPingTpl.replace( "##", cmtAuthor ); } // Fill up #reply-link-options function newOption( id, text, defaultOn ) { var newCheckbox = document.createElement( "input" ); newCheckbox.type = "checkbox"; newCheckbox.id = id; if( defaultOn ) { newCheckbox.checked = true; } var newLabel = document.createElement( "label" ); newLabel.htmlFor = id; newLabel.appendChild( document.createTextNode( text ) ); document.getElementById( "reply-link-options" ).appendChild( newCheckbox ); document.getElementById( "reply-link-options" ).appendChild( newLabel ); } // Fetch metadata about this specific comment var ourMetadata = metadata[this.id]; // If the dry-run option is "checkbox", add an option to make it // a dry run if( window.replyLinkDryRun === "checkbox" ) { newOption( "reply-link-option-dry-run", "Don't actually edit?", true ); } // If the current section header text indicates an edit request, // offer to mark it as answered if( ourMetadata[1] && EDIT_REQ_REGEX.test( ourMetadata[1][1] ) ) { newOption( "reply-link-option-edit-req", "Mark edit request as answered?", false ); } // If the previous comment was indented by OUTDENT_THRESH, // offer to outdent if( ourMetadata[0].length >= OUTDENT_THRESH ) { newOption( "reply-link-option-outdent", "Outdent?", false ); } if( window.replyLinkAutoIndentation === "checkbox" ) { newOption( "reply-link-option-auto-indent", mw.msg( "rl-auto-indent" ), true ); } /* Commented out because I could never get it to work // Autofill with a recommendation if we're replying to a nom if( rplyToXfdNom ) { replyDialogField.value = "'''Comment'''"; // Highlight the "Comment" part so the user can change it var range = document.createRange(); range.selectNodeContents( replyDialogField ); //range.setStart( replyDialogField, 3 ); // start of "Comment" //range.setEnd( replyDialogField, 10 ); // end of "Comment" var sel = window.getSelection(); sel.removeAllRanges(); sel.addRange( range ); }*/ // Close handler window.onbeforeunload = function ( e ) { if( !replyWasSaved && document.getElementById( "reply-dialog-field" ) && document.getElementById( "reply-dialog-field" ).value ) { var txt = mw.msg( "rl-started-reply" ); e.returnValue = txt; return txt; } }; // Called by the "Reply" button, Ctrl-Enter in the text area, and // Enter/Ctrl-Enter in the summary field function startReply() { // Change UI to make it clear we're performing an operation document.getElementById( "reply-dialog-field" ).style["background-image"] = "url(" + window.replyLinkPendingImageUrl + ")"; document.querySelector( "#reply-link-buttons button" ).disabled = true; setStatus( mw.msg( "rl-loading" ) ); var parsoidUrl = PARSOID_ENDPOINT + encodeURIComponent( currentPageName ) + "/" + mw.config.get( "wgCurRevisionId" ), findSectionResultPromise = $.get( parsoidUrl ) .then( function ( parsoidDomString ) { return findSection( currentPageName, parsoidDomString, cmtLink ); },console.error ); var revObjPromise = findSectionResultPromise.then( function ( findSectionResult ) { console.log( "findSectionResult ", findSectionResult ); return getWikitext( findSectionResult.page ); },console.error ); $.when( findSectionResultPromise, revObjPromise ).then( function ( findSectionResult, revObj ) { // ourMetadata contains data in the format: // [indentation, header, sigIdx] doReply( ourMetadata[0], ourMetadata[1], ourMetadata[2], cmtAuthor, rplyToXfdNom, revObj, findSectionResult ); }, function (e) { setStatusError(new Error(e))} ); } // Event listener for the "Reply" button document.getElementById( "reply-dialog-button" ) .addEventListener( "click", startReply ); // Event listener for the text area document.getElementById( "reply-dialog-field" ) .addEventListener( "keydown", function ( e ) { if( e.ctrlKey && ( e.keyCode == 10 || e.keyCode == 13 ) ) { startReply(); } } ); // Event listener for the "Preview" button document.getElementById( "reply-link-preview-button" ) .addEventListener( "click", function () { var reply = document.getElementById( "reply-dialog-field" ).value.trim(); // Add a signature if one isn't already there if( !hasSig( reply ) ) { reply += " " + ( window.replyLinkSigPrefix ? window.replyLinkSigPrefix : "" ) + LITERAL_SIGNATURE; } var sanitizedCode = encodeURIComponent( reply ); $.post( "https:" + mw.config.get( "wgServer" ) + "/w/api.php?action=parse&format=json&title=" + currentPageName + "&text=" + sanitizedCode + "&pst=1", function ( res ) { if ( !res || !res.parse || !res.parse.text ) return console.log( "Preview failed" ); document.getElementById( "reply-link-preview" ).innerHTML = res.parse.text['*']; // Add target="_blank" to links to make them open in a new tab by default var links = document.querySelectorAll( "#reply-link-preview a" ); for( var i = 0, n = links.length; i < n; i++ ) { links[i].setAttribute( "target", "_blank" ); } } ); } ); if( window.replyLinkPreloadPing === "button" ) { document.getElementById( "reply-link-ping-button" ) .addEventListener( "click", function () { replyDialogField.value = window.replyLinkPreloadPingTpl .replace( "##", cmtAuthor ) + replyDialogField.value; } ); } // Event listener for the "Cancel" button document.getElementById( "reply-link-cancel-button" ) .addEventListener( "click", function () { newLink.textContent = linkLabel; panelEl.remove(); } ); // Event listeners for the custom edit summary field if( window.replyLinkCustomSummary ) { document.getElementById( "reply-link-summary" ) .addEventListener( "keydown", function ( e ) { if( e.keyCode == 10 || e.keyCode == 13 ) { startReply(); } } ); } if( window.replyLinkTestInstantReply ) { startReply(); } }.bind( this ) ); // Cancel default event handler evt.preventDefault(); return false; } } /** * Adds a "(reply)" link after the provided text node, giving it * the provided element id. anyIndentation is true if there's any * indentation (i.e. indentation string is not the empty string) */ function attachLinkAfterNode( node, preferredId, anyIndentation ) { // Choose a parent node - walk up tree until we're under a dd, li, // p, or div. This walk is a bit unsafe, but this function should // only get called in a place where the walk will succeed. var parent = node; do { parent = parent.parentNode; } while( !( /^(p|dd|li|div|td)$/.test( parent.tagName.toLowerCase() ) ) ); // Determine whether we're replying to an XfD nom var rplyToXfdNom = false; if( xfdType === "AfD" || xfdType === "MfD" ) { // If the comment is non-indented, we are replying to a nom rplyToXfdNom = !anyIndentation; } else if( xfdType === "TfD" || xfdType === "FfD" ) { // If the sibling before the previous sibling of this node // is a h4, then this is a nom rplyToXfdNom = parent.previousElementSibling && parent.previousElementSibling.previousElementSibling && parent.previousElementSibling.previousElementSibling.nodeType === 1 && parent.previousElementSibling.previousElementSibling.tagName.toLowerCase() === "h4"; } else if( xfdType === "CfD" ) { // If our grandparent is a dl and our grandparent's previous // sibling is a h4, then this is a nom rplyToXfdNom = parent.parentNode.tagName.toLowerCase() === "dl" && parent.parentNode.previousElementSibling.nodeType === 1 && parent.parentNode.previousElementSibling.tagName.toLowerCase() === "h4"; } // Choose link label: if we're replying to an XfD, customize it var linkLabel = mw.msg( "rl-reply-label" ) + ( rplyToXfdNom ? mw.msg( "rl-to-label" ) + xfdType : "" ); // Construct new link var newLinkWrapper = document.createElement( "span" ); newLinkWrapper.className = "reply-link-wrapper"; var newLink = document.createElement( "a" ); newLink.href = "#"; newLink.id = preferredId; newLink.dataset.originalLabel = linkLabel; newLink.appendChild( document.createTextNode( linkLabel ) ); newLink.addEventListener( "click", handleWrapperClick( linkLabel, parent, rplyToXfdNom ) ); newLinkWrapper.appendChild( document.createTextNode( " (" ) ); newLinkWrapper.appendChild( newLink ); newLinkWrapper.appendChild( document.createTextNode( ")" ) ); // Insert new link into DOM parent.insertBefore( newLinkWrapper, node.nextSibling ); } /** * Uses attachLinkAfterTextNode to add a reply link after every * timestamp on the page. */ function attachLinks () { var mainContent = findMainContentEl(); if( !mainContent ) { console.error( "No main content element found; exiting." ); return; } var contentEls = mainContent.children; // Find the index of the first header in contentEls var headerIndex = 0; for( headerIndex = 0; headerIndex < contentEls.length; headerIndex++ ) { if( contentEls[ headerIndex ].tagName.toLowerCase().startsWith( "h" ) ) break; } // If we didn't find any headers at all, that's a problem and we // should bail if( mainContent.querySelector( "div.hover-edit-section" ) ) { headerIndex = 0; } else if( headerIndex === contentEls.length ) { console.error( "Didn't find any headers - hit end of loop!" ); return; } // We also should include the first header if( headerIndex > 0 ) { headerIndex--; } // Each element is a 2-element list of [level, node] var parseStack = iterableToList( contentEls ).slice( headerIndex ); parseStack.reverse(); parseStack = parseStack.map( function ( el ) { return [ "", el ]; } ); // Main parse loop var node; var currIndentation; // A string of symbols, like ":*::" var newIndentSymbol; var stackEl; // current element from the parse stack var idNum = 0; // used to make id's for the links var linkId = ""; // will be the element id for this link while( parseStack.length ) { stackEl = parseStack.pop(); node = stackEl[1]; currIndentation = stackEl[0]; // Compatibility with "Comments in Local Time" var isLocalCommentsSpan = node.nodeType === 1 && "span" === node.tagName.toLowerCase() && node.className.includes( "localcomments" ); var isSmall = node.nodeType === 1 && ( node.tagName.toLowerCase() === "small" || ( node.tagName.toLowerCase() === "span" && node.style && node.style.getPropertyValue( "font-size" ) === "85%" ) ); // Small nodes are okay, unless they're delsort notices var isOkSmallNode = isSmall && !node.className.includes( "delsort-notice" ); if( ( node.nodeType === 3 ) || isOkSmallNode || isLocalCommentsSpan ) { // If the current node has a timestamp, attach a link to it // Also, no links after timestamps, because it's just like // having normal text afterwards, which is rejected (because // that means someone put a timestamp in the middle of a // paragraph) var hasLinkAfterwardsNotInBlockEl = node.nextElementSibling && ( node.nextElementSibling.tagName.toLowerCase() === "a" || ( node.nextElementSibling.tagName.match( /^(span|small)$/i ) && node.nextElementSibling.querySelector( "a" ) ) ); if( TIMESTAMP_REGEX.test( node.textContent ) && ( node.previousSibling || isSmall ) && !hasLinkAfterwardsNotInBlockEl ) { linkId = "reply-link-" + idNum; attachLinkAfterNode( node, linkId, !!currIndentation ); idNum++; // Update global metadata dictionary metadata[linkId] = currIndentation; } } else if( node.nodeType === 1 && /^(div|p|dl|dd|ul|li|span|ol|table|tbody|tr|td)$/.test( node.tagName.toLowerCase() ) ) { switch( node.tagName.toLowerCase() ) { case "dl": newIndentSymbol = ":"; break; case "ul": newIndentSymbol = "*"; break; case "ol": newIndentSymbol = "#"; break; case "div": if( node.className.includes( "xfd_relist" ) ) { continue; } break; default: newIndentSymbol = ""; break; } var childNodes = node.childNodes; for( let i = 0, numNodes = childNodes.length; i < numNodes; i++ ) { parseStack.push( [ currIndentation + newIndentSymbol, childNodes[i] ] ); } } } // This loop adds two entries in the metadata dictionary: // the header data, and the sigIdx values var sigIdxEls = iterableToList( mainContent.querySelectorAll( HEADER_SELECTOR + ",span.reply-link-wrapper a" ) ); var currSigIdx = 0, j, numSigIdxEls, currHeaderEl, currHeaderData; var headerIdx = 0; // index of the current header var headerLvl = 0; // level of the current header for( j = 0, numSigIdxEls = sigIdxEls.length; j < numSigIdxEls; j++ ) { var headerTagNameMatch = /^h(\d+)$/.exec( sigIdxEls[j].tagName.toLowerCase() ); if( headerTagNameMatch ) { currHeaderEl = sigIdxEls[j]; // Test to make sure we're not in the table of contents if( currHeaderEl.parentNode.className === "toctitle" ) { continue; } // Reset signature counter currSigIdx = 0; // Dig down one level for the header text because // MW buries the text in a span inside the header var headlineEl = null; if( currHeaderEl.childNodes[0].className && currHeaderEl.childNodes[0].className.includes( "mw-headline" ) ) { headlineEl = currHeaderEl.childNodes[0]; } else { for( var i = 0; i < currHeaderEl.childNodes.length; i++ ) { if( currHeaderEl.childNodes[i].className && currHeaderEl.childNodes[i].className.includes( "mw-headline" ) ) { headlineEl = currHeaderEl.childNodes[i]; break; } } } var headerName = null; if( headlineEl ) { headerName = headlineEl.textContent; } if( headerName === null ) { console.error( currHeaderEl ); throw "Couldn't parse a header element!"; } headerLvl = headerTagNameMatch[1]; currHeaderData = [ headerLvl, headerName, headerIdx ]; headerIdx++; } else { // Save all the metadata for this link currIndentation = metadata[ sigIdxEls[j].id ]; metadata[ sigIdxEls[j].id ] = [ currIndentation, currHeaderData ? currHeaderData.slice(0) : null, currSigIdx ]; currSigIdx++; } } //console.log(metadata); // Disable links inside hatnotes, archived discussions var badRegionsSelector = [ "div.archived", "div.resolved", "table" ].map( function ( s ) { return s + " .reply-link-wrapper" } ).join( "," ); var insideArchived = mainContent.querySelectorAll( badRegionsSelector ); for( var i = 0; i < insideArchived.length; i++ ) { insideArchived[i].parentNode.removeChild( insideArchived[i] ); } } function runTestMode() { // We never want to make actual edits window.replyLinkDryRun = "always"; // Simulate having a panel open $( "#mw-content-text" ) .append( $( "<div>" ) .append( $( "<textarea>" ).attr( "id", "reply-dialog-field" ).val( "hi" ) ) .append( $( "<div>" ).attr( "id", "reply-link-buttons" ) .append( $( "<button> " ) ) ) ); mw.util.addCSS( ".reply-link-wrapper { background-color: orange; }" ); // Fetch content, Parsoid DOM, etc var parsoidUrl = PARSOID_ENDPOINT + encodeURIComponent( currentPageName ); $.when( $.get( parsoidUrl ), api.loadMessages( INT_MSG_KEYS ) ).then( function ( parsoidDomString, _ ) { buildUserspcLinkRgx(); // Statistics variables var successes = 0, failures = 0; // Run one test on a wrapper link function runOneTestOn( wrapper ) { try { var cmtAuthorAndLink = getCommentAuthor( wrapper ), cmtAuthor = cmtAuthorAndLink.username, cmtLink = cmtAuthorAndLink.link; var ourMetadata = metadata[ wrapper.children[0].id ]; findSection( currentPageName, parsoidDomString, cmtLink ).then( function ( findSectionResult ) { var revObjPromise = getWikitext( findSectionResult.page, /* useCaching */ true ); $.when( findSectionResult, revObjPromise ).then( function ( findSectionResult, revObj ) { doReply( ourMetadata[0], ourMetadata[1], ourMetadata[2], cmtAuthor, false, revObj, findSectionResult ).done( function () { wrapper.style.background = "green"; successes++; } ).fail( function () { wrapper.style.background = "red"; failures++; } ); }, function ( e ) { wrapper.style.background = "red"; failures++; } ); } ); } catch ( e ) { console.error( e ); wrapper.style.background = "red"; failures++; } } var wrappers = Array.from( document.querySelectorAll( ".reply-link-wrapper" ) ); function runOneTest() { var wrapper = wrappers.shift(); if( wrapper ) { runOneTestOn( wrapper ); setTimeout( runOneTest, 750 ); } else { var results = successes + " successes, " + failures + " failures"; $( "#mw-content-text" ).prepend( results ).append( results ); } } //console.log = function() {}; setTimeout( runOneTest, 0 ); } ); } function onReady() { var lang_code = mw.config.get( "wgUserLanguage" ) // Replace default English interface by translation if available var interface_messages = $.extend( {}, i18n.en, i18n[ lang_code.split('-')[0] ], i18n[ lang_code ] ); // Define interface messages mw.messages.set( interface_messages ); // Exit if history page or edit page if( mw.config.get( "wgAction" ) === "history" ) return; if( document.getElementById( "editform" ) ) return; api = new mw.Api(); mw.util.addCSS( "#reply-link-panel { padding: 1em; margin-left: 1.6em; "+ "max-width: 1200px; width: 66%; margin-top: 0.5em; }"+ ".gone-on-empty:empty { display: none; }" ); // Pre-load interface messages; we will check again when a (reply) // link is clicked api.loadMessages( INT_MSG_KEYS ); // Initialize the xfdType global variable, which must happen // before the call to attachLinks currentPageName = mw.config.get( "wgPageName" ); xfdType = ""; if( mw.config.get( "wgNamespaceNumber" ) === 4) { if( currentPageName.startsWith( "Wikipedia:Articles_for_deletion/" ) ) { xfdType = "AfD"; } else if( currentPageName.startsWith( "Wikipedia:Miscellany_for_deletion/" ) ) { xfdType = "MfD"; } else if( currentPageName.startsWith( "Wikipedia:Templates_for_discussion/Log/" ) ) { xfdType = "TfD"; } else if( currentPageName.startsWith( "Wikipedia:Categories_for_discussion/Log/" ) ) { xfdType = "CfD"; } else if( currentPageName.startsWith( "Wikipedia:Files_for_discussion/" ) ) { xfdType = "FfD"; } } // Default values for some preferences if( window.replyLinkAutoReload === undefined ) window.replyLinkAutoReload = true; if( window.replyLinkDryRun === undefined ) window.replyLinkDryRun = "never"; if( window.replyLinkPreloadPing === undefined ) window.replyLinkPreloadPing = "always"; if( window.replyLinkPreloadPingTpl === undefined ) window.replyLinkPreloadPingTpl = "{{u|##}}, "; if( window.replyLinkCustomSummary === undefined ) window.replyLinkCustomSummary = false; if( window.replyLinkTestMode === undefined ) window.replyLinkTestMode = false; if( window.replyLinkTestInstantReply === undefined) window.replyLinkTestInstantReply = false; if( window.replyLinkAutoIndentation === undefined ) window.replyLinkAutoIndentation = "checkbox"; // Insert "reply" links into DOM attachLinks(); // If test mode is enabled, create a link for that if( window.replyLinkTestMode ) { mw.util.addPortletLink( "p-cactions", "#", "reply-link test mode", "pt-reply-link-test" ) .addEventListener( "click", runTestMode ); } // This large string creats the "pending" texture window.replyLinkPendingImageUrl = "data:image/gif;base64,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"; } mw.loader.load( "mediawiki.ui.input", "text/css" ); mw.loader.using( [ "mediawiki.util", "mediawiki.api" ] ).then( function () { mw.hook( "wikipage.content" ).add( onReady ); } ); $.getScript('https://en.wikipedia.org/w/index.php?title=User:Enterprisey/parsoid-jump.js&action=raw&ctype=text%2Fjavascript'); // Return functions for testing return { "iterableToList": iterableToList, "sigIdxToStrIdx": sigIdxToStrIdx, "insertTextAfterIdx": insertTextAfterIdx, "wikitextToTextContent": wikitextToTextContent }; } // Export functions for testing if( typeof module === typeof {} ) { module.exports = { "loadReplyLink": loadReplyLink }; } // If we're in the right environment, load the script if( jQuery !== undefined && mediaWiki !== undefined ) { var currNamespace = mw.config.get( "wgNamespaceNumber" ); // Also enable on T:TDYK and its subpages var ttdykPage = mw.config.get( "wgPageName" ).indexOf( "Template:Did_you_know_nominations" ) === 0; // Normal "read" view and not a diff view var normalView = mw.config.get( "wgIsArticle" ) && !mw.config.get( "wgDiffOldId" ); if ( normalView && ( currNamespace % 2 === 1 || currNamespace === 4 || ttdykPage ) ) { loadReplyLink( jQuery, mediaWiki ); } } //</nowiki>