[Perf] use/nvm_die_on_prefix: replicate npm config algorithm and remove npm config call

[ljharb]

Fixes #2251. Fixes #1774. Fixes #1694. Fixes #1277. Fixes #1261. Fixes #1242. Fixes #966. Fixes #926. Fixes #860. Fixes #781. Fixes #709. Closes #1826. Closes #730.

This PR removes the call to npm config get prefix on nvm use, instead partially replicating the algorithm npm itself uses to build up its config. Since we don't have to actually produce a config, merely detect if any option is set that would influence npm config get prefix (namely, prefix or globalconfig), the checks are simpler and can be done relatively quickly. Previously, #1679, 1458de7, 8ee6f30, and a1def71 implemented partial forms of this checking. This PR specifically adds the checks for npmrc files, which allows the call to npm to be removed.

[jleedev]

Here's the speedup I'm seeing with this change:

$ /usr/bin/time bash -c '. ~/pkg/ljharb_nvm/nvm.sh' 
0.29user 0.05system 0:00.34elapsed 98%CPU (0avgtext+0avgdata 4408maxresident)k
0inputs+8outputs (0major+3300minor)pagefaults 0swaps
$ /usr/bin/time bash -c '. ~/.nvm/nvm.sh' 
3.85user 1.27system 0:04.98elapsed 102%CPU (0avgtext+0avgdata 39664maxresident)k
0inputs+16outputs (0major+51708minor)pagefaults 0swaps

[ljharb]

Wow, that seems huge - thank you for the benchmarks!

[Schniz]

Wow!

[reasonablytall]

The current version of nvm doesn't run as long on my machine, but still quite an improvement! Running against nvm --version => 0.36.0

bash

$ hyperfine "bash -c '. ~/.nvm/nvm.sh'" "bash -c '. ~/proj/nvm/nvm.sh'"  
Benchmark #1: bash -c '. ~/.nvm/nvm.sh'
  Time (mean ± σ):     628.6 ms ±   7.8 ms    [User: 614.2 ms, System: 142.0 ms]
  Range (min … max):   619.3 ms … 641.8 ms    10 runs
 
Benchmark #2: bash -c '. ~/proj/nvm/nvm.sh'
  Time (mean ± σ):     346.4 ms ±   4.4 ms    [User: 319.1 ms, System: 112.4 ms]
  Range (min … max):   340.1 ms … 355.3 ms    10 runs

zsh

$ hyperfine "zsh -c '. ~/.nvm/nvm.sh'" "zsh -c '. ~/proj/nvm/nvm.sh'" 
Benchmark #1: zsh -c '. ~/.nvm/nvm.sh'
  Time (mean ± σ):     764.0 ms ±  11.7 ms    [User: 756.5 ms, System: 158.1 ms]
  Range (min … max):   745.8 ms … 780.0 ms    10 runs
 
Benchmark #2: zsh -c '. ~/proj/nvm/nvm.sh'
  Time (mean ± σ):     482.3 ms ±   6.8 ms    [User: 467.3 ms, System: 122.4 ms]
  Range (min … max):   475.5 ms … 496.4 ms    10 runs

I'm unsure where that >100ms between zsh and bash comes from though...

[MatthiasPortzel]

Seems to be roughly twice as fast, from around 400ms to around 200ms. (But I'm getting quite a bit of variance.)

I would love one day for nvm to be optimized down to the 30ms that it takes to start with --no-use, since right now it is the only slow part of my shell startup. But I can confirm that the remaining 200ms are not a result of prefix checking.

[ljharb]

@matthiassaihttam thanks! if you have any more information (performance metrics, which parts of nvm contribute how much to that extra 170ms, etc), filing a new issue with it would be greatly appreciated.

[ljharb]

v0.37.0 has been released with this change.