(function (Prism) { Prism.languages.xquery = Prism.languages.extend('markup', { 'xquery-comment': { pattern: /\(:[\s\S]*?:\)/, greedy: true, alias: "comment" }, 'string': { pattern: /(["'])(?:\1\1|(?!\1)[\s\S])*\1/, greedy: true }, 'extension': { pattern: /\(#.+?#\)/, alias: 'symbol' }, 'variable': /\$[\w-:]+/, 'axis': { pattern: /(^|[^-])(?:ancestor(?:-or-self)?|attribute|child|descendant(?:-or-self)?|following(?:-sibling)?|parent|preceding(?:-sibling)?|self)(?=::)/, lookbehind: true, alias: 'operator' }, 'keyword-operator': { pattern: /(^|[^:-])\b(?:and|castable as|div|eq|except|ge|gt|idiv|instance of|intersect|is|le|lt|mod|ne|or|union)\b(?=$|[^:-])/, lookbehind: true, alias: 'operator' }, 'keyword': { pattern: /(^|[^:-])\b(?:as|ascending|at|base-uri|boundary-space|case|cast as|collation|construction|copy-namespaces|declare|default|descending|else|empty (?:greatest|least)|encoding|every|external|for|function|if|import|in|inherit|lax|let|map|module|namespace|no-inherit|no-preserve|option|order(?: by|ed|ing)?|preserve|return|satisfies|schema|some|stable|strict|strip|then|to|treat as|typeswitch|unordered|validate|variable|version|where|xquery)\b(?=$|[^:-])/, lookbehind: true }, 'function': /[\w-]+(?::[\w-]+)*(?=\s*\()/, 'xquery-element': { pattern: /(element\s+)[\w-]+(?::[\w-]+)*/, lookbehind: true, alias: 'tag' }, 'xquery-attribute': { pattern: /(attribute\s+)[\w-]+(?::[\w-]+)*/, lookbehind: true, alias: 'attr-name' }, 'builtin': { pattern: /(^|[^:-])\b(?:attribute|comment|document|element|processing-instruction|text|xs:(?:anyAtomicType|anyType|anyURI|base64Binary|boolean|byte|date|dateTime|dayTimeDuration|decimal|double|duration|ENTITIES|ENTITY|float|gDay|gMonth|gMonthDay|gYear|gYearMonth|hexBinary|ID|IDREFS?|int|integer|language|long|Name|NCName|negativeInteger|NMTOKENS?|nonNegativeInteger|nonPositiveInteger|normalizedString|NOTATION|positiveInteger|QName|short|string|time|token|unsigned(?:Byte|Int|Long|Short)|untyped(?:Atomic)?|yearMonthDuration))\b(?=$|[^:-])/, lookbehind: true }, 'number': /\b\d+(?:\.\d+)?(?:E[+-]?\d+)?/, 'operator': [ /[+*=?|@]|\.\.?|:=|!=|<[=<]?|>[=>]?/, { pattern: /(\s)-(?=\s)/, lookbehind: true } ], 'punctuation': /[[\](){},;:/]/ }); Prism.languages.xquery.tag.pattern = /<\/?(?!\d)[^\s>\/=$<%]+(?:\s+[^\s>\/=]+(?:=(?:("|')(?:\\[\s\S]|{(?!{)(?:{(?:{[^}]*}|[^}])*}|[^}])+}|(?!\1)[^\\])*\1|[^\s'">=]+))?)*\s*\/?>/i; Prism.languages.xquery['tag'].inside['attr-value'].pattern = /=(?:("|')(?:\\[\s\S]|{(?!{)(?:{(?:{[^}]*}|[^}])*}|[^}])+}|(?!\1)[^\\])*\1|[^\s'">=]+)/i; Prism.languages.xquery['tag'].inside['attr-value'].inside['punctuation'] = /^="|"$/; Prism.languages.xquery['tag'].inside['attr-value'].inside['expression'] = { // Allow for two levels of nesting pattern: /{(?!{)(?:{(?:{[^}]*}|[^}])*}|[^}])+}/, inside: { rest: Prism.languages.xquery }, 'alias': 'language-xquery' }; // The following will handle plain text inside tags var stringifyToken = function (token) { if (typeof token === 'string') { return token; } if (typeof token.content === 'string') { return token.content; } return token.content.map(stringifyToken).join(''); }; var walkTokens = function (tokens) { var openedTags = []; for (var i = 0; i < tokens.length; i++) { var token = tokens[i]; var notTagNorBrace = false; if (typeof token !== 'string') { if (token.type === 'tag' && token.content[0] && token.content[0].type === 'tag') { // We found a tag, now find its kind if (token.content[0].content[0].content === ' 0 && openedTags[openedTags.length - 1].tagName === stringifyToken(token.content[0].content[1])) { // Pop matching opening tag openedTags.pop(); } } else { if (token.content[token.content.length - 1].content === '/>') { // Autoclosed tag, ignore } else { // Opening tag openedTags.push({ tagName: stringifyToken(token.content[0].content[1]), openedBraces: 0 }); } } } else if ( openedTags.length > 0 && token.type === 'punctuation' && token.content === '{' && // Ignore `{{` (!tokens[i + 1] || tokens[i + 1].type !== 'punctuation' || tokens[i + 1].content !== '{') && (!tokens[i - 1] || tokens[i - 1].type !== 'plain-text' || tokens[i - 1].content !== '{') ) { // Here we might have entered an XQuery expression inside a tag openedTags[openedTags.length - 1].openedBraces++; } else if (openedTags.length > 0 && openedTags[openedTags.length - 1].openedBraces > 0 && token.type === 'punctuation' && token.content === '}') { // Here we might have left an XQuery expression inside a tag openedTags[openedTags.length - 1].openedBraces--; } else if (token.type !== 'comment') { notTagNorBrace = true } } if (notTagNorBrace || typeof token === 'string') { if (openedTags.length > 0 && openedTags[openedTags.length - 1].openedBraces === 0) { // Here we are inside a tag, and not inside an XQuery expression. // That's plain text: drop any tokens matched. var plainText = stringifyToken(token); // And merge text with adjacent text if (i < tokens.length - 1 && (typeof tokens[i + 1] === 'string' || tokens[i + 1].type === 'plain-text')) { plainText += stringifyToken(tokens[i + 1]); tokens.splice(i + 1, 1); } if (i > 0 && (typeof tokens[i - 1] === 'string' || tokens[i - 1].type === 'plain-text')) { plainText = stringifyToken(tokens[i - 1]) + plainText; tokens.splice(i - 1, 1); i--; } if (/^\s+$/.test(plainText)) { tokens[i] = plainText; } else { tokens[i] = new Prism.Token('plain-text', plainText, null, plainText); } } } if (token.content && typeof token.content !== 'string') { walkTokens(token.content); } } }; Prism.hooks.add('after-tokenize', function (env) { if (env.language !== 'xquery') { return; } walkTokens(env.tokens); }); }(Prism));