`vite fmt` and `vite lint` unexpectedly run postinstall script

[wojtekmaj]

vite fmt and vite lint commands unexpectedly run postinstall script, consequently taking much longer than anticipated:

    "lint": "oxlint && oxfmt --check",
    "lint:oldschool": "vite lint && vite fmt --check",

% hyperfine "pnpm lint" "pnpm lint:oldschool"
Benchmark 1: pnpm lint
  Time (mean ± σ):      1.533 s ±  0.077 s    [User: 9.325 s, System: 0.664 s]
  Range (min … max):    1.442 s …  1.660 s    10 runs
 
Benchmark 2: pnpm lint:oldschool
  Time (mean ± σ):      4.954 s ±  2.620 s    [User: 5.006 s, System: 4.751 s]
  Range (min … max):    0.777 s …  9.800 s    10 runs

Summary
  pnpm lint ran
    3.23 ± 1.72 times faster than pnpm lint:oldschool

[serhalp]

This might be stating the obvious, but it's my understanding that vite+ runs pnpm i intentionally, which in turn runs the postinstall script even when the install step was a no-op, which is just how pnpm works.

pnpm tries to skip the postinstall step as well when it isn't "needed", with some heuristics. We're probably hitting a case that https://pnpm.io/settings#sideeffectscache doesn't handle. (See also perhaps pnpm/pnpm#9949 and pnpm/pnpm#5002.)

[serhalp]

... My bad, it appears sideEffectsCache only applies to dependencies' postinstall scripts, not your own workspace's packages'.

[fengmk2]

Currently, the built-in commands of vite+ will implicitly execute vite install once, which may be slow here. I will note this down first, and later I will conduct a performance analysis and post the results when I have them.