Issue 7296 - Implement lightweight URL parsing

To test the performance:

  (function()
  {
    "use strict";

    let {parseURL} = require("./lib/url");

    let startTime = Date.now();
    for (let i = 0; i < 10000000; i++)
      //new URL("https://example.com/foo/?bar=1#lambda");
      parseURL("https://example.com/foo/?bar=1#lambda");
    console.log(Date.now() - startTime);
  })();

This at least 10x as fast and even 25x or more for longer URLs.

[Manish Jethani, 2019-02-21 11:24:37]

Patch Set 1

This is a lightweight version of the URL object. I still need to add more tests
from https://cs.chromium.org/chromium/src/url/gurl_unittest.cc

The idea is that the extension, which already has a valid, normalized, properly
encoded URL string, will use the new `parseURL()` function instead of the native
`URL` object. It needs only the protocol and the hostname, none of the other
stuff.

[Manish Jethani, 2019-02-21 15:29:17]

Patch Sets 2-4

The only question I have in my mind is whether the URL passed to the
`onBeforeRequest` handler is always normalized. I will have to check.

[hub, 2019-02-21 16:24:08]

https://codereview.adblockplus.org/30013574/diff/30013591/lib/url.js
File lib/url.js (right):

https://codereview.adblockplus.org/30013574/diff/30013591/lib/url.js#newcode115
lib/url.js:115: return match ? new URLInfo(url, match[1], match[2]) : new
URLInfo(url);
if the parsing fails, then we only have a valid `_href` in URLInfo. I'm
wondering if instead we shouldn't just return null instead.

[Manish Jethani, 2019-02-23 07:32:06]

Patch Set 5: Make protocol field required

https://codereview.adblockplus.org/30013574/diff/30013591/lib/url.js
File lib/url.js (right):

https://codereview.adblockplus.org/30013574/diff/30013591/lib/url.js#newcode115
lib/url.js:115: return match ? new URLInfo(url, match[1], match[2]) : new
URLInfo(url);
On 2019/02/21 16:24:08, hub wrote:
> if the parsing fails, then we only have a valid `_href` in URLInfo. I'm
> wondering if instead we shouldn't just return null instead.

Actually since the function assumes that the URL will always be valid and
normalized, `match` should never be null (and if it is then an error should be
thrown).

I have updated the patch so it works this way now.

[Manish Jethani, 2019-02-23 08:11:06]

On 2019/02/21 15:29:17, Manish Jethani wrote:

> The only question I have in my mind is whether the URL passed to the
> `onBeforeRequest` handler is always normalized. I will have to check.

Confirmed that on both Chrome and Firefox the URL passed to the
`onBeforeRequest` handler is normalized.

i.e. "//example.com:80?" becomes "http://example.com/?", any non-ASCII
characters in the URL are properly encoded using IDNA or percent-encoding. We
can use this implementation of `parseURL()` with such URLs.

[Manish Jethani, 2019-02-23 08:27:29]

Patch Set 6: Remove non-normalized URLs from tests

[hub, 2019-02-25 20:01:17]

LGTM