Benchmarking support to find low-hanging optimizations?

[jschaf]

Happy Thanksgiving! In the spirit, thank you for this wonderful linting tool.

How might I go about benchmarking protolint to improve the speed? I have 130 proto files, totaling 31 kLOC, and it takes about 3.4 seconds for protolint to check them. That means:

• protolint parses about 9 kLOC/second.

By comparison, Go compiles at about 44 kLOC/second, so I suspect there are opportunities to improve the parsing speed.

I'm interested because protolint is the slowest lint I've added to our repo by about 2x. I want to improve interactive performance for a better dev experience.

Here's the trace showing the performance of various lints:

I did a bit of profiling by adding pprof locally to protolint but didn't find anything conclusive. The majority of time is spent parsing. The rule execution only accounts for 2% of the time.

[yoheimuta]

@jschaf Thank you for the profiling insights!
I agree that there's room for improvement in protolint's processing speed.

The current implementation prioritizes ease of development over efficiency, so this result is understandable.
On reflection, this line could be a potential culprit, as it results in protolint parsing the same file repeatedly for each rule application.