GeezerVison

Sounds like upstream of prek is also reasonable to work with which is a major plus.

Where does the speedup for run come from? Is it just that run is also installing or is something else going on?

This is without the cache, so it is also installing all venvs required for running the linters. For day-to-day, it's probably not so much faster, except when we start changing configs or something that requires rebuilding them. As Hugo mentioned, I think, this doesn't do anything for speeding up the linters themselves. That is, mypy is probably going to be the slowest thing if it has to go through all files. While prek does have a fast path for some of the regular hooks, I don't think we have enough code for those to dominate times at all.

If I edit one file, git add it, and run without --all-files, it's about 1.21× faster (but still pretty quick):

$ hyperfine 'prek run' 'pre-commit run'
Benchmark 1: prek run
  Time (mean ± σ):     881.3 ms ± 210.7 ms    [User: 412.3 ms, System: 186.2 ms]
  Range (min … max):   584.5 ms … 1266.8 ms    10 runs
 
Benchmark 2: pre-commit run
  Time (mean ± σ):      1.063 s ±  0.114 s    [User: 0.671 s, System: 0.244 s]
  Range (min … max):    0.908 s …  1.292 s    10 runs
 
Summary
  prek run ran
    1.21 ± 0.32 times faster than pre-commit run

Even just the install speedup is nice to have. Yesterday I used my clone on my work machine for the first time in a while, and was kind of annoyed about having to wait for those hooks to install!

The improved disk usage offered by prek over pre-commit is a BIG and most welcome win AFAIK 💪

The number of times I've had to purge a scary amount of GBs from my pre-commit cache ... lost count 😬