Avoid running `git diff` repeatedly

[JelleZijlstra]

Summary

I was exploring prek on a huge work repo (tens of thousands of files) and noticed that it's about 3x slower than pre-commit. Looking at the trace log revealed that prek seems to be running git diff once per hook. Usually that would be fast but for us it often takes a few seconds, and that adds up when there's a few dozen hooks.

Could you instead run git diff once per invocation and reuse the result among hooks?

Platform

Darwin 25.2.0

Version

0.2.27

.pre-commit-config.yaml

No response

Log file

No response

[JelleZijlstra]

Dug some more and it looks like pre-commit also runs git diff after it ran a hook, but not after a skipped hook (because hooks that run may have changed files, but skipped hooks didn't).

On my repo I was able to improve git diff performance and now prek is comparable in speed to pre-commit, though still a little slower most of the time.

[j178]

Thanks! prek runs git diff after each hook to check whether any files were modified. There’s probably some room to optimize that, but I don’t really have a better idea right now.

About the skipped hooks: I think that’s a regression from #1232, and it should be fixed in #1335. Could you try that out? If prek is still slow after that, could you share a verbose log (-vvv or ~/.cache/prek/prek.log) so I can take a closer look? Thanks!

[yihong0618]

if we use code for git diff instead of subcommad is that help tiny performace imporve?

[JelleZijlstra]

Thanks! Skipping git diff for skipped checks should help a lot, I'll give it another try soon!