Skip to content

BUGS.md

Latest commit

 

History

History
36 lines (31 loc) · 2.07 KB

BUGS.md

File metadata and controls

36 lines (31 loc) · 2.07 KB
Raw

Bugs & Gotchas

A running log of non-obvious bugs/gotchas, their fixes, and why — so we don't rediscover them.

Newest first. Each entry: symptom → cause → fix → why.

npm install "$FRONTEND_SYSTEM_TGZ" keeps re-dirtying package.json

  • Found: 2026-06-24.
  • Status: known gotcha (intentional policy; cleaner automation deferred).
  • Symptom: after the design-system setup/reinstall step (npm install "$FRONTEND_SYSTEM_TGZ", e.g. when bumping the pin), git status shows package.json modified with a line like: 
    "@wl/frontend-system": "file:../../../../../nix/store/<hash>-frontend-system-<ver>/frontend-system.tgz"
    It looks like a real change to commit. It is not — discard it (git checkout package.json).
  • Cause: installing a tarball by path makes npm record it as a file: dependency with the resolved path. That path points into the local Nix store, and both the hash and version are baked in — so it changes on every design-system release, won't resolve on any other machine, and disappears after nix-collect-garbage. Committing it would pin the repo to a store path that only exists here, only right now.
  • Fix: don't track it. The reproducible pin lives in devenv.lock (the frontend-system flake input → a git rev); that's the file to commit when bumping. The package.json store-path line is local install output and stays uncommitted. This was settled earlier in 81964da "chore: don't commit the arch-specific design-system store path".
  • Why (the principle): source of truth for the design-system version is the flake lock, not package.json. package.json only carries the peer/runtime deps; the design-system tarball is delivered and pinned out-of-band by devenv.
  • Deferred improvement: the per-install churn is friction (easy to commit by accident). A just recipe that reinstalls and then resets the package.json line, or an npm flag / .npmrc tweak that avoids writing the resolved path, would remove the trap. Not worth it under the current manual bump-the-pin flow.