Introduce the bruteforce parser (interpreter?).

[eddyb]

Check out src/lyg.rs for an example involving pattern-matching parse nodes, and the two tests (ported over from the gll crate) for how to use the interpret and what the output is like.

I'm open to suggestions on the API and formatting, it's a bit arbitrary at the moment.

[CAD97]

Everything looks fine. Obviously src/slow_bruteforce_interpreter.rs is the tricky part.

[eddyb]

I think I'd like this to bake a while longer, at least to get to test it with wg-grammar and maybe try out some imperative disambiguation? Could result in a wildly different handle API.

@oli-obk also expressed interest in experimenting with the "web frontend" idea.

EDIT: some initial testing with wg-grammar results in a larger slowdown than expected, will investigate.
EDIT2: oops, that was a limitation in cyclotron, got 1000x speedup (or more) on some files.

Force push: need to reread

[CAD97]

Generally: Everything outside slow_bruteforce_interpreter.rs looks good to me so far. I think some of the FIXME in that file can be pretty easily addressed, though. (Generally, it looks fine, but I've only really scanned that file so far.)

[CAD97]

Everything looks good here

[eddyb]

Need to sync these with gll, additions were made there.

[eddyb]

This needs to be reverted/replaced with self.expand_repeats(cx).node_shape(cx, named_rules).

[eddyb]

This needs to be dealt with ASAP.

[eddyb]

So this is the nastiest part of parsing atm, an emulated stack to avoid blowing up the real one.

(Using a queue instead of a stack looks too much like GLL for my own comfort, otherwise I would've probably done it that way)

I'm wondering how I can make this more readable - maybe keep more things explicitly on the emulated stack? Use an enum with variants as an explicit state machine?

[eddyb]

Might be cheaper to use e.g. im::OrdSet for this.