/*! Title parser * This parser will remove "The", "A" and "An" from the beginning of a book * or movie title, so it sorts by the second word or number * Demo: http://jsfiddle.net/Mottie/abkNM/5/ */ /*global jQuery: false */ ;(function($){ "use strict"; // basic list from http://en.wikipedia.org/wiki/Article_%28grammar%29 $.tablesorter.ignoreArticles = { "en" : "the, a, an", "de" : "der, die, das, des, dem, den, ein, eine, einer, eines, einem, einen", "nl" : "de, het, de, een", "es" : "el, la, lo, los, las, un, una, unos, unas", "pt" : "o, a, os, as, um, uma, uns, umas", "fr" : "le, la, l'_, les, un, une, des", "it" : "il, lo, la, l'_, i, gli, le, un', uno, una, un", "hu" : "a, az, egy" }; // To add a custom parser, define: // $.tablesorter.ignoreArticles['xx'] = "A, B, C"; // and then set the language id 'xx' in the headers option // ignoreArticles : 'xx' $.tablesorter.addParser({ id: 'ignoreArticles', is: function() { return false; }, format: function(s, table, cell, cellIndex) { var c = table.config, art, lang; if ( !(c.headers && c.headers[cellIndex] && c.headers[cellIndex].ignoreArticlesRegex) ) { // initialize - save regex in c.headers[cellIndex].ignoreArticles if (!c.headers) { c.headers = {}; } if (!c.headers[cellIndex]) { c.headers[cellIndex] = {}; } lang = $.tablesorter.getData(c.$headers.eq(cellIndex), c.headers[cellIndex], 'ignoreArticles'); art = ($.tablesorter.ignoreArticles[lang] || "the, a, an" ) + ""; c.headers[cellIndex].ignoreArticlesRegex = new RegExp('^(' + $.trim( art.split(/\s*\,\s*/).join('\\s|') + "\\s" ).replace("_\\s","") + ')', 'i'); } return (s || '').replace(c.headers[cellIndex].ignoreArticlesRegex, ''); }, type: 'text' }); })(jQuery);