This repo benchmarks the change where we read packageManager field and determine version from that instead of running <pm> --version.

Note: This is with a locally published version from the PR (nrwl/nx#29249)

The getPackageManagerVersion function itself sees a 16x speedup if we parse package.json first and skip creating a process.

hyperfine -n before 'node run.mjs --before' -n after 'node run.mjs'
Benchmark 1: before
  Time (mean ± σ):      8.991 s ±  0.612 s    [User: 5.987 s, System: 1.682 s]
  Range (min … max):    8.339 s … 10.018 s    10 runs
 
Benchmark 2: after
  Time (mean ± σ):      97.1 ms ±  15.5 ms    [User: 95.4 ms, System: 20.5 ms]
  Range (min … max):    84.5 ms … 131.2 ms    34 runs
 
  Warning: Statistical outliers were detected. Consider re-running this benchmark on a quiet system without any interferences from other programs. It might help to use the '--warmup' or '--prepare' options.
 
Summary
  after ran
   92.57 ± 16.10 times faster than before

---

To measure impact on graph creation, I've added some plugins, which all call getPackageManagerCommand, which then calls getPackageManagerVersion. I'm just running nx report without daemon, so the plugins load always.

BEFORE (on "slow" branch, no packageManager field):

NX_DAEMON=false hyperfine "nx report"
Benchmark 1: nx report
  Time (mean ± σ):     594.0 ms ±  30.4 ms    [User: 301.6 ms, System: 96.7 ms]
  Range (min … max):   559.4 ms … 667.5 ms    10 runs

AFTER:

NX_DAEMON=false hyperfine "nx report"
Benchmark 1: nx report
  Time (mean ± σ):     520.4 ms ±  54.1 ms    [User: 244.4 ms, System: 82.1 ms]
  Range (min … max):   485.0 ms … 667.5 ms    10 runs

You can see, about 80ms difference. This difference grows larger the more plugins you use.