Make it possible to run jsdom in a worker

[lawnsea]

For my research project Treehouse, I forked jsdom and got it to run in a web worker. This was two years ago. I'm interested in doing this again, but in a way that is maintainable, rather than just a proof-of-concept.

At a high level, this is what was necessary to get jsdom 0.2.4 running in a worker:

• remove node dependencies that cannot be met in the browser (fs, e.g.)*
• provide pure-JS implementations of node dependencies where possible (url, e.g.)*
• use util.inherits instead of assigning to __proto__
• use Object.defineProperty instead of __defineSetter__/Getter__
• replace uses of variables named window and document with win and doc**

I also refactored the code to use AMD and build with RequireJS. I don't think this is necessary now (and may not have been necessary then): it should be possible to use the cjsTranslate method to build for the browser.

I would like to keep the diff from my fork to the original as minimal as possible. So, I'd like to upstream those changes that support browser compatibility in a non-breaking way. Would y'all be open to that?

* browserify may help with these
** I'm embarrassed to admit I don't remember why this was necessary

EDIT: I'm aware of this thread and #245

[domenic]

Wow, this is great.

I think the path these days is to use browserify. So that should eliminate the need to provide pure-JS implementations of the node dependencies, since browserify comes with those; it may also solve the node dependencies that cannot be met. And should obviate the need to use AMD at all; just produce a browserify UMD bundle, which last I heard are worker compatible (or if they're not, that should be an easy PR to get accepted).

The other thing that gives you is the ability to use the browser field in package.json to selectively overwrite modules inside jsdom. That'll be quite useful I imagine, although it could require breaking apart jsdom into smaller modules than exist currently.

use util.inherits instead of assigning to __proto__

I'd prefer not to add the super_ property that util.inherits does, but I'd be totally open to converting to the correct

Subclass.prototype = Object.create(Superclass.prototype, {
  constructor: {
    value: Subclass,
    writable: true,
    configurable: true
  }
});

use Object.defineProperty instead of __defineSetter__/Getter__

Yes please!

replace uses of variables named window and document with win and doc

Hopefully not necessary, but if we can figure it out, I don't see a real problem with it.

[lawnsea]

Ok, great! Thanks for the feedback, @domenic.

I'm going to do the switch to Object.defineProperty and the new inheritance scheme first. Then I'll tackle browserify. Shall I leave this issue open as a tracking bug?

[domenic]

Yeah sounds good! I'll merge as things come in.

[lawnsea]

Ok, I've surveyed all external dependencies and tried requiring them in a worker via browserify.

The following don't work and will likely never work:

• contexify: V8 only; there is no browser alternative
• canvas: depends on cairo; a non-trivial fix would be to implement a JS-only canvas shim backed by a typed array

The following don't work but can likely be fixed:

• http/http: uses window instead of self. fixed in Use 'global' instead of window. Fixes #36 browserify/http-browserify#37
• cssstyle: uses fs.readdirSync to read in property files. we'll want to do that at build time instead.
• request: browserify fails with Error: ENOENT, open 'tls' while resolving "tls" from file /Users/lon/shared/dev/src/treehousejs/jsdom/node_modules/request/node_modules/forever-agent/index.js

[lawnsea]

Ok, it looks like we may be able to use iriscouch/browser-request instead of mikeal/request.

[lawnsea]

@domenic, do you have any preference for how I should accomplish the browserify build? I could just add an entry to scripts in package.json. Or, we could get crazy with grunt, gulp, or whatever.

[lawnsea]

🎆 🍻

I can build a browserify bundle that works with this worker (in Chrome Stable latest):

var jsdom = require('../lib/jsdom');

jsdom.env({
  url: 'http://example.com',
  html: '<p><a class="the-link" href="https://github.com/tmpvar/jsdom">jsdom\'s Homepage</a></p>',
  done: function (errors, window) {
    var link = window.document.querySelector('a.the-link');
    postMessage(link.innerHTML);
  }
});

That was fun. Next step is to get the tests running in a worker and find out how many we fail.

[domenic]

Woah, nice!

do you have any preference for how I should accomplish the browserify build? I could just add an entry to scripts in package.json. Or, we could get crazy with grunt, gulp, or whatever.

What kind of build is required? I was kind of hoping we'd just use the browser field in package.json to override a bunch of dependencies.

[lawnsea]

Yep! That's how it works. The only real wrinkle is that OS X needs the maximum number of files bumped up.

It's possible we will need something more advanced in the future, but adding a browserify entry to scripts should do the job for now.

[lawnsea]

Ok, I hacked on a copy of test/runner and got it to run the tests. Here's the current status:

• N/M: N of M tests passed
• ok: all tests passed
• fail: the suite fails to run to completion
• N/A: this suite is commented out in test/runner

"level1/core": ok
"level1/html": ok
"level1/svg": ok
"level2/core": ok
"level2/html": 6/708
"level2/style": 0/15
"level2/extra": ok
"level2/events": ok
"level3/core": N/A
"level3/ls": N/A
"level3/xpath": 0/93
"window/index": ok
"window/history": 0/5
"window/script": 0/10
"window/console": 0/2
"window/frame": fail
"sizzle/index": fail
"jsdom/index": fail
"jsdom/parsing": 0/11
"jsdom/env": 9/25
"jsdom/utils": ok
"jsonp/jsonp": 0/1
"browser/css": 0/1
"browser/index": 30/34