description("Tests mailto: URL canonicalization"); cases = [ ["addr1", "addr1"], ["addr1@foo.com", "addr1@foo.com"], // Trailing whitespace is stripped. ["addr1 \t ", "addr1"], ["addr1?to=jon", "addr1?to=jon"], ["addr1,addr2", "addr1,addr2"], ["addr1, addr2", "addr1, addr2"], ["addr1%2caddr2", "addr1%2caddr2"], // U+10300 input as UTF-16 surrogate pair, expected as escaped UTF-8 ["\uD800\uDF00", "%F0%90%8C%80"], ["addr1?", "addr1?"], // Null character should be escaped to %00 // Not sure how to test null characters in JavaScript. They appear to get // stripped out. // ["addr1\\0addr2?foo", "addr1%00addr2?foo"], ]; for (var i = 0; i < cases.length; ++i) { shouldBe("canonicalize('mailto:" + cases[i][0] + "')", "'mailto:" + cases[i][1] + "'"); } // Invalid -- UTF-8 encoded surrogate value. shouldBeFalse("canonicalize('mailto:\xed\xa0\x80') == 'mailto:%EF%BF%BD'"); var successfullyParsed = true;