Optimizations for the generated parser

[austin-schick]

Changes made

• Use a while loop and break to simulate the C parser's goto behavior, instead of creating and calling a done() function
• Stop running PyBytes_FromStringAndSize when tokenizing. I verified that .bytes was never read during execution. But it's possible that this wasn't a valid change.
• Add the nodejs profiling script to source control
• Small improvements to scripts and tests

Performance improvements

Here's are the times I see on my laptop for the datetime tests:

First PEG parser:                      193ms
Generated PEG parser (before this PR): 130ms
Generated PEG parser (after this PR):   88ms
Legacy parser:                          55ms

And here's an additional datapoint from my pyperformance tests:

First PEG parser:                      2.8s
Generated PEG parser (before this PR): 2.6s
Generated PEG parser (after this PR):  2.1s
Legacy parser:                         1.2s

So I think we're sitting around 1.7x slower than the legacy parser after these improvements.

[austin-schick]

This was necessary to get the nodejs profiler running, but didn't seem necessary in the browser. I didn't have the time to dig into exactly why that discrepancy exists.

[PierreQuentel]

Once again impressive work Austin !

I am currently working on a separate branch generated_peg_parser and making good progress to pass all the tests in the built-in test suite. I am not a GIT expert, is it possible to apply this PR to the branch ?

[austin-schick]

Thank you, and yes! I updated this PR to merge into the generated_peg_parser branch. It looks like there are a few merge conflicts -- I should be able to look into those this week.

[austin-schick]

Merge conflicts fixed! I verified that datetime still parses, but I didn't run the full test suite on this branch. What's your current process for running the built-in test suite on the generated parser?

[PierreQuentel]

I ran the test suite with the page /tests/parse_tests/index_generated_parser.html.

After fixing a bug in pegen.js, I have merged your PR. The whole test suite passes, slightly slower than with the legacy parser, but the difference is negligible compared to the benefits of the new parser.

The next step is to test all the scripts in the gallery, the documentation etc. before I can merge the branch into master.

Thanks once again Austin, this is a major step forward for Brython !

[austin-schick]

Very exciting! I'm happy to have been able to help out some. Thanks for all your work on this Pierre!