Is it really faster?

[martenlienen]

I am coming from pyenv for python version management, which uses shims like asdf. And sure enough, the shims add roughly 100ms every time I invoke python or another shim command.

$ time python -c ""
python -c ""  0.07s user 0.04s system 101% cpu 0.111 total
$ time ~/.pyenv/versions/3.12.1/bin/python -c ""
~/.pyenv/versions/3.12.1/bin/python -c ""  0.02s user 0.00s system 97% cpu 0.017 total

rtx will save me this time by putting the correct python on PATH and avoiding the shims. However, as far as I can tell, it does so by executing eval "$(/home/ml/.cargo/bin/rtx hook-env -s zsh)"; as a pre-command and chpwd hook in zsh. Let's see how long that takes:

$ hyperfine '/home/ml/.cargo/bin/rtx hook-env -s zsh'
Benchmark 1: /home/ml/.cargo/bin/rtx hook-env -s zsh
  Time (mean ± σ):     118.6 ms ±   6.2 ms    [User: 112.9 ms, System: 11.7 ms]
  Range (min … max):   108.7 ms … 130.0 ms    23 runs

Do I understand it correctly that rtx saves me 100ms to run python by costing me 100ms before I run any command or change directory? Also, how is rtx hook-env so slow? Shouldn't printing a constant string in rust take less than 1ms?

[jdx]

Can you run it with RTX_DEBUG=1 set?

[jdx]

I suppose I could try to reproduce your vanilla setup with no config files and no plugins but I really don't see why that would make it load slower.

[martenlienen]

It is still not as fast as on your machine but significantly faster now. -N tells hyperfine to run the command without shell involvement to measure short runtimes accurately (though the resulting time seems to be the same either way for me).

# Compiled from commit immediately before your PR
$ hyperfine -N '/home/ml/.cargo/bin/rtx hook-env -s zsh'
Benchmark 1: /home/ml/.cargo/bin/rtx hook-env -s zsh
  Time (mean ± σ):      97.4 ms ±   6.1 ms    [User: 92.0 ms, System: 11.6 ms]
  Range (min … max):    89.8 ms … 116.2 ms    25 runs

# With your PR
$ hyperfine -N '/home/ml/.cargo/bin/rtx hook-env -s zsh'
Benchmark 1: /home/ml/.cargo/bin/rtx hook-env -s zsh
  Time (mean ± σ):       4.8 ms ±   0.7 ms    [User: 3.9 ms, System: 6.5 ms]
  Range (min … max):     3.8 ms …   7.7 ms    389 runs

Do you have any ideas how I could get it to sub-1ms as well?

[jdx]

you're about where I am now (to be clear it's time in zsh that said <.01s):

rtx  git-diff [!] ❯ hyperfine -N "rtx hook-env -s zsh" --warmup 20
Benchmark 1: rtx hook-env -s zsh
  Time (mean ± σ):       4.0 ms ±   0.2 ms    [User: 3.7 ms, System: 7.2 ms]
  Range (min … max):     3.7 ms …   4.8 ms    703 runs

What I think I should do is setup the tracing crate so you could install rtx with tracing and we could figure out exactly where it is spending time since I suspect this could be improved further.

[martenlienen]

Sounds good! When it is ready, I can recompile and run rtx again with tracing.

[jdx]

the good news is that a complex setup with lots of plugins and config files doesn't seem to impact performance at all compared to yours that has nothing