An idea to speed up stubs

[felipecrs]

Hi, I was doing some experiments, and I was able to speed up stubs by around 10%.

However, to make it work for real it will require a new flag or environment variable in pkgx, like --swallow-first-arg or PKGX_SWALLOW_FIRST_ARG=1.

The idea is simple: instead of relying on sh to run pkgx, use just env instead:

❯ cat ./test-sh
#!/usr/bin/env sh

exec pkgx echo "$0" "$@"
❯ cat ./test-env
#!/usr/bin/env -S pkgx echo

Benchmarks:

❯ pkgx hyperfine --warmup 2 --shell none --runs 1000 './test-sh' './test-env'
Benchmark 1: ./test-sh
  Time (mean ± σ):       7.1 ms ±   0.5 ms    [User: 3.9 ms, System: 4.1 ms]
  Range (min … max):     6.5 ms …  15.1 ms    1000 runs

Benchmark 2: ./test-env
  Time (mean ± σ):       6.5 ms ±   0.2 ms    [User: 3.9 ms, System: 3.5 ms]
  Range (min … max):     5.9 ms …   8.0 ms    1000 runs

Summary
  ./test-env ran
    1.09 ± 0.09 times faster than ./test-sh

But once the flag is implemented, then the stubs can look like:

#!/usr/bin/env -S pkgx --swallow-first-arg echo

Or something shorter like pkgx -# echo.

[felipecrs]

Another option would be to implement some kind of pkgx interpreter, like:

#!/usr/bin/env pkgx

git@latest

pkgx would need to interpret this file as a list of args to pkgx itself and concatenate user-provided args with it.

Maybe this would be less hacky, but maybe also slower, given pkgx would need to in fact open the file and read it.

[jhheider]

that's cute. whether or not stubs have a lasting lifetime, using pkgx as a shebang is good. since this currently works:

#!/usr/bin/env -S pkgx +foo +bar

something like:

#!/usr/bin/env -S pkgx +foo +bar --exec foo

could work.

[mxcl]

I like it, we could add this.

[mxcl]

Why -#? Is there precedence? Note I am not dismissing it, just inquiring for research reasons.

[felipecrs]

No, it's just a symbol for a flag that people are unlikely to confuse.

[mxcl]

Having talked to ChatGPT I think -# is a good idea. Will try an implementation.

[felipecrs]

Having talked to ChatGPT I think -# is a good idea.

That's surprising. lol

[mxcl]

Ah ok ofc, we actually need to drop the last parameter, --pop seems more appropriate

[felipecrs]

Related: we could improve an additional ~10% by dropping the need of env:

> cd /tmp

> echo "#!/usr/local/bin/pkgx echo" | tee no-env-shim

> echo "#!/usr/bin/env -S pkgx echo" | tee env-shim

> chmod +x no-env-shim env-shim 

> pkgx hyperfine --warmup 10 --shell none --runs 1000 ./env-shim ./no-env-shim
Benchmark 1: ./env-shim
  Time (mean ± σ):       6.7 ms ±   0.4 ms    [User: 3.8 ms, System: 3.6 ms]
  Range (min … max):     5.9 ms …  12.4 ms    1000 runs
 
Benchmark 2: ./no-env-shim
  Time (mean ± σ):       6.2 ms ±   0.4 ms    [User: 3.8 ms, System: 3.0 ms]
  Range (min … max):     5.6 ms …   8.6 ms    1000 runs
 
Summary
  ./no-env-shim ran
    1.07 ± 0.09 times faster than ./env-shim

At the cost of shim's portability. It will probably work ok so long pkgx is always installed to /usr/local/bin/pkgx, but if pkgx is moved, the shims will stop working too.

Probably not worth it in general, but under certain cases, it may be.