Make in-browser filesystem API asynchronous

[jvilk]

While we can make do with a synchronous filesystem API for now, it would be beneficial to switch over to an asynchronous API.

The benefits are as follows:

• No more blocking the UI thread for file downloads.
• The ability to strategically queue up file downloads to make certain things faster, e.g. asynchronous class loading.

• 41 Integrating ourselves with cloud services will be much simpler, as I believe those APIs are asynchronous. At least, filepicker.io seems to be, although that's a paid service.

Note that I am talking about the filesystem API, and not the Node API.

[jvilk]

This has become my next task. By doing this change, I plan to fix:

• 205

• 142

• Half of #206
• Enable #41 to be possible for GSoC students :smile:

The filesystem will become a straightforward Unix-style filesystem. The root will be completely in-memory, with file providers such as localStorage, indexedDB, cloud providers, temporary storage, and the webserver mounted somewhere in the in-memory hierarchy.

There will be no more complex 'composed' file sources that cause headaches and issues.

Eventually, once this transformation is complete and stable, I plan to break it away from Doppio into its own repository, as it will be useful in its own right. Perhaps I will also implement other asynchronous Node filesystem functions.

[jvilk]

I'm currently integrating BrowserFS in the doppio-bfs branch. Here's what's remaining:

• src/testing.coffee needs to switch over to asynchronous commands.
• Our tab completion code (which should be rewritten, imo) needs to switch over to async commands.
• Java native methods need to switch over to async commands, although they are not a functional issue so far since everything uses PreloadFile, which supports this use case.

Everything else, though, I believe I have converted. It works quite well. :smile:

[perimosocordiae]

Bullet points 1 and 2 taken care of.

[jvilk]

<3

[perimosocordiae]

Current blocker: BFS doesn't use the correct error codes, and the current hacks mean that we're getting internal errors instead of exceptions.

[jvilk]

Oof. Do you know which error codes are actually important? If it's just one or two, we can hack them in for now.

[perimosocordiae]

Looks like it's just ENOENT, for now at least. See the above commit.

The hackfix gets us to the point where we're trying to create files in the /sys readonly mount. We'll need to modify our tests to get them working again.

[perimosocordiae]

Woo! All tests pass in the browser! I think we're ready to merge.

[perimosocordiae]

Merged #238, closing.