gecko-dev/netwerk/test/unit/test_mozTXTToHTMLConv.js

396 строки
10 KiB
JavaScript
Исходник Ответственный История

Этот файл содержит неоднозначные символы Юникода!

Этот файл содержит неоднозначные символы Юникода, которые могут быть перепутаны с другими в текущей локали. Если это намеренно, можете спокойно проигнорировать это предупреждение. Используйте кнопку Экранировать, чтобы подсветить эти символы.

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this file,
* You can obtain one at http://mozilla.org/MPL/2.0/. */
/**
* Test that mozITXTToHTMLConv works properly.
*/
"use strict";
function run_test() {
let converter = Cc["@mozilla.org/txttohtmlconv;1"].getService(
Ci.mozITXTToHTMLConv
);
const scanTXTtests = [
// -- RFC1738
{
input: "RFC1738: <URL:http://mozilla.org> then",
url: "http://mozilla.org",
},
{
input: "RFC1738: <URL:mailto:john.doe+test@mozilla.org> then",
url: "mailto:john.doe+test@mozilla.org",
},
// -- RFC2396E
{
input: "RFC2396E: <http://mozilla.org/> then",
url: "http://mozilla.org/",
},
{
input: "RFC2396E: <john.doe+test@mozilla.org> then",
url: "mailto:john.doe+test@mozilla.org",
},
// -- abbreviated
{
input: "see www.mozilla.org maybe",
url: "http://www.mozilla.org",
},
{
input: "mail john.doe+test@mozilla.org maybe",
url: "mailto:john.doe+test@mozilla.org",
},
// -- delimiters
{
input: "see http://www.mozilla.org/maybe today", // Spaces
url: "http://www.mozilla.org/maybe",
},
{
input: 'see "http://www.mozilla.org/maybe today"', // Double quotes
url: "http://www.mozilla.org/maybetoday", // spaces ignored
},
{
input: "see <http://www.mozilla.org/maybe today>", // Angle brackets
url: "http://www.mozilla.org/maybetoday", // spaces ignored
},
// -- freetext
{
input: "I mean http://www.mozilla.org/.",
url: "http://www.mozilla.org/",
},
{
input: "you mean http://mozilla.org:80, right?",
url: "http://mozilla.org:80",
},
{
input: "go to http://mozilla.org; then go home",
url: "http://mozilla.org",
},
{
input: "http://mozilla.org! yay!",
url: "http://mozilla.org",
},
{
input: "er, http://mozilla.com?",
url: "http://mozilla.com",
},
{
input: "http://example.org- where things happen",
url: "http://example.org",
},
{
input: "see http://mozilla.org: front page",
url: "http://mozilla.org",
},
{
input: "'http://mozilla.org/': that's the url",
url: "http://mozilla.org/",
},
{
input: "some special http://mozilla.org/?x=.,;!-:x",
url: "http://mozilla.org/?x=.,;!-:x",
},
{
// escape & when producing html
input: "'http://example.org/?test=true&success=true': ok",
url: "http://example.org/?test=true&amp;success=true",
},
{
input: "bracket: http://localhost/[1] etc.",
url: "http://localhost/",
},
{
input: "bracket: john.doe+test@mozilla.org[1] etc.",
url: "mailto:john.doe+test@mozilla.org",
},
{
input: "parenthesis: (http://localhost/) etc.",
url: "http://localhost/",
},
{
input: "parenthesis: (john.doe+test@mozilla.org) etc.",
url: "mailto:john.doe+test@mozilla.org",
},
{
input: "(thunderbird)http://mozilla.org/thunderbird",
url: "http://mozilla.org/thunderbird",
},
{
input: "(mail)john.doe+test@mozilla.org",
url: "mailto:john.doe+test@mozilla.org",
},
{
input: "()http://mozilla.org",
url: "http://mozilla.org",
},
{
input:
"parenthesis included: http://kb.mozillazine.org/Performance_(Thunderbird) etc.",
url: "http://kb.mozillazine.org/Performance_(Thunderbird)",
},
{
input: "parenthesis slash bracket: (http://localhost/)[1] etc.",
url: "http://localhost/",
},
{
input: "parenthesis bracket: (http://example.org[1]) etc.",
url: "http://example.org",
},
{
input: "ipv6 1: https://[1080::8:800:200C:417A]/foo?bar=x test",
url: "https://[1080::8:800:200C:417A]/foo?bar=x",
},
{
input: "ipv6 2: http://[::ffff:127.0.0.1]/#yay test",
url: "http://[::ffff:127.0.0.1]/#yay",
},
{
input: "ipv6 parenthesis port: (http://[2001:db8::1]:80/) test",
url: "http://[2001:db8::1]:80/",
},
{
input:
"test http://www.map.com/map.php?t=Nova_Scotia&markers=//Not_a_survey||description=plm2 test",
url:
"http://www.map.com/map.php?t=Nova_Scotia&amp;markers=//Not_a_survey||description=plm2",
},
{
input: "bug#1509493 (john@mozilla.org)john@mozilla.org test",
url: "mailto:john@mozilla.org",
text: "john@mozilla.org",
},
{
input: "bug#1509493 {john@mozilla.org}john@mozilla.org test",
url: "mailto:john@mozilla.org",
text: "john@mozilla.org",
},
];
const scanTXTglyph = [
// Some "glyph" testing (not exhaustive, the system supports 16 different
// smiley types).
{
input: "this is superscript: x^2",
results: ["<sup", "2", "</sup>"],
},
{
input: "this is plus-minus: +/-",
results: ["&plusmn;"],
},
{
input: "this is a smiley :)",
results: ["moz-smiley-s1"],
},
{
input: "this is a smiley :-)",
results: ["moz-smiley-s1"],
},
{
input: "this is a smiley :-(",
results: ["moz-smiley-s2"],
},
];
const scanTXTstrings = [
"underline", // ASCII
"äöüßáéíóúî", // Latin-1
"a\u0301c\u0327c\u030Ce\u0309n\u0303t\u0326e\u0308d\u0323",
// áçčẻñțëḍ Latin
"\u016B\u00F1\u0257\u0119\u0211\u0142\u00ED\u00F1\u0119",
// Pseudo-ese ūñɗęȑłíñę
"\u01DDu\u0131\u0283\u0279\u01DDpun", // Upside down ǝuıʃɹǝpun
"\u03C5\u03C0\u03BF\u03B3\u03C1\u03AC\u03BC\u03BC\u03B9\u03C3\u03B7",
// Greek υπογράμμιση
"\u0441\u0438\u043B\u044C\u043D\u0443\u044E", // Russian сильную
"\u0C2C\u0C32\u0C2E\u0C46\u0C56\u0C28", // Telugu బలమైన
"\u508D\u7DDA\u3059\u308B", // Japanese 傍線する
"\uD841\uDF0E\uD841\uDF31\uD841\uDF79\uD843\uDC53\uD843\uDC78",
// Chinese (supplementary plane)
"\uD801\uDC14\uD801\uDC2F\uD801\uDC45\uD801\uDC28\uD801\uDC49\uD801\uDC2F\uD801\uDC3B",
// Deseret 𐐔𐐯𐑅𐐨𐑉𐐯𐐻
];
const scanTXTstructs = [
{
delimiter: "/",
tag: "i",
class: "moz-txt-slash",
},
{
delimiter: "*",
tag: "b",
class: "moz-txt-star",
},
{
delimiter: "_",
tag: "span",
class: "moz-txt-underscore",
},
{
delimiter: "|",
tag: "code",
class: "moz-txt-verticalline",
},
];
const scanHTMLtests = [
{
input: "http://foo.example.com",
shouldChange: true,
},
{
input: " <a href='http://a.example.com/'>foo</a>",
shouldChange: false,
},
{
input: "<abbr>see http://abbr.example.com</abbr>",
shouldChange: true,
},
{
input: "<!-- see http://comment.example.com/ -->",
shouldChange: false,
},
{
input: "<!-- greater > -->",
shouldChange: false,
},
{
input: "<!-- lesser < -->",
shouldChange: false,
},
{
input:
"<style id='ex'>background-image: url(http://example.com/ex.png);</style>",
shouldChange: false,
},
{
input: "<style>body > p, body > div { color:blue }</style>",
shouldChange: false,
},
{
input: "<script>window.location='http://script.example.com/';</script>",
shouldChange: false,
},
{
input: "<head><title>http://head.example.com/</title></head>",
shouldChange: false,
},
{
input: "<header>see http://header.example.com</header>",
shouldChange: true,
},
{
input: "<iframe src='http://iframe.example.com/' />",
shouldChange: false,
},
{
input: "broken end <script",
shouldChange: false,
},
];
function hrefLink(url) {
return ' href="' + url + '"';
}
function linkText(plaintext) {
return ">" + plaintext + "</a>";
}
for (let i = 0; i < scanTXTtests.length; i++) {
let t = scanTXTtests[i];
let output = converter.scanTXT(t.input, Ci.mozITXTToHTMLConv.kURLs);
let link = hrefLink(t.url);
let text;
if (t.text) {
text = linkText(t.text);
}
if (!output.includes(link)) {
do_throw(
"Unexpected conversion by scanTXT: input=" +
t.input +
", output=" +
output +
", link=" +
link
);
}
if (text && !output.includes(text)) {
do_throw(
"Unexpected conversion by scanTXT: input=" +
t.input +
", output=" +
output +
", text=" +
text
);
}
}
for (let i = 0; i < scanTXTglyph.length; i++) {
let t = scanTXTglyph[i];
let output = converter.scanTXT(
t.input,
Ci.mozITXTToHTMLConv.kGlyphSubstitution
);
for (let j = 0; j < t.results.length; j++) {
if (!output.includes(t.results[j])) {
do_throw(
"Unexpected conversion by scanTXT: input=" +
t.input +
", output=" +
output +
", expected=" +
t.results[j]
);
}
}
}
for (let i = 0; i < scanTXTstrings.length; ++i) {
for (let j = 0; j < scanTXTstructs.length; ++j) {
let input =
scanTXTstructs[j].delimiter +
scanTXTstrings[i] +
scanTXTstructs[j].delimiter;
let expected =
"<" +
scanTXTstructs[j].tag +
' class="' +
scanTXTstructs[j].class +
'">' +
'<span class="moz-txt-tag">' +
scanTXTstructs[j].delimiter +
"</span>" +
scanTXTstrings[i] +
'<span class="moz-txt-tag">' +
scanTXTstructs[j].delimiter +
"</span>" +
"</" +
scanTXTstructs[j].tag +
">";
let actual = converter.scanTXT(input, Ci.mozITXTToHTMLConv.kStructPhrase);
Assert.equal(encodeURIComponent(actual), encodeURIComponent(expected));
}
}
for (let i = 0; i < scanHTMLtests.length; i++) {
let t = scanHTMLtests[i];
let output = converter.scanHTML(t.input, Ci.mozITXTToHTMLConv.kURLs);
let changed = t.input != output;
if (changed != t.shouldChange) {
do_throw(
"Unexpected change by scanHTML: changed=" +
changed +
", shouldChange=" +
t.shouldChange +
", \ninput=" +
t.input +
", \noutput=" +
output
);
}
}
}