Performance

[cdthompson]

Running wist on my 2013 Macbook, it takes ~14s for wist to lint a 200-line BRS file. This is a lot longer than I expected but I don't have much experience with antlr.

Does this seem in line with performance you guys are seeing? If not, any hints on where I should start to isolate bottlenecks?

[nishtahir]

Thanks for sharing your feedback. We have been experiencing similar performance issues as well although not quite as bad as what you are seeing.

A roughly 500 line *.brs takes 12.06s on a similar Macbook pro.

The root cause seems to be mostly associated with the antlr JS runtime just being much slower than other platforms or possibly some optimizations we can make in the grammar

antlr/antlr4#1540

One option I plan on investigating is to compile the grammar for a C++ target and cross compile it to a wasm module as mentioned here

antlr/antlr4#1375

Any thoughts? @vannuysm

[cdthompson]

I did the minimum to get emscripten solution parsing, but didn't incorporate the listeners for linting rules. It brought the time for my above test case down to 2s but seems there's a lot of work to expose the listeners across the C++/JS boundary. For now I'm going to focus on using wist in a CI environment exclusively, then perhaps later come back to helping improve this. I'll leave this issue open and let you both decide whether to track it or close.

[nishtahir]

I did mostly the same thing here. My results were similar to yours with ~1.5s to parse a 1000 line*.brs file. I'm currently in the process of writing a wrapper around the generated ANTLR AST so we can plugin the rules engine.

[nishtahir]

So far we have some parts of wist ported over to use emscripten. i've created a branch experimental/web-assembly-parser for this.

[cdthompson]

Thanks for this effort