WIP: use ragel options to minimize states and use gotos

[jtarchie]

What problem is this PR intended to solve?

No problem.

Experimenting with the features of ragel for faster parsing tables for gumbo.
I'd like to see if the tests passed through CI.

Have you included adequate test coverage?

This shouldn't affect the functionality.
This relies on ragel and the C compiler for better optimizations in parsing.

Does this change affect the behavior of either the C or the Java implementations?

As far as I know, just the C implementation.

[stevecheckoway]

It's nice to see someone looking at this!

I played with the various options a few years ago and couldn't get better output. Looks like this choice of options ends up with 50k more lines of code to compile.

I'm curious how this choice impacts the size of the binary, compilation time, and runtime parsing performance.

[flavorjones]

Thanks, @jtarchie! I kicked off CI and have an HTML5 benchmark on my dev machine that I'll run a comparison with.

[jtarchie]

There are file differences:

• char_ref.c goes from 692K -> 1.4M in size
• the object file goes from 298K -> 870K
• with -O3, the object file is 367K (taking 12s to compile over the 0.5s without any -O flag above)

The object file shows the size difference that would affect the binary.

This may not give the speed increase needed, based on the changelog ragel originally made this faster. This might have already hit the speed sweet spot. It might not be the bottleneck it once was.

[jtarchie]

=Is this something I can help explore still?

[flavorjones]

@jtarchie Thanks for asking. I definitely want to understand the tradeoffs between size, compilation speed (less of an issue these days), and runtime performance. It's on my list of things to explore next month when I have some free time to work on OSS.