fix(ci): portable hash parsing in nix-hashes workflow#11533
fix(ci): portable hash parsing in nix-hashes workflow#11533rekram1-node merged 1 commit intoanomalyco:devfrom
Conversation
The sed -E regex with \s+ behaves differently on BSD sed (macOS) vs GNU sed (Linux), causing the hash extraction to fail on darwin runners and include 'got:' prefix in the output. Replace complex sed parsing with portable grep -oE to extract sha256 hashes directly. Use tail -n1 to get the 'got:' hash (second match) since nix outputs 'specified:' hash first. Ref: anomalyco#11530 (comment)
|
Thanks for your contribution! This PR doesn't have a linked issue. All PRs must reference an existing issue. Please:
See CONTRIBUTING.md for details. |
|
The following comment was made by an LLM, it may be inaccurate: The only other PR found is #11530 ("ci: fix nix hash issue"), which is actually the issue that PR #11533 is meant to close (as stated in the description: "Closes #11530"). This is not a duplicate - #11530 is the issue being addressed by the current PR. No duplicate PRs found |
|
wanna just add the token change onto this pr too: |
|
I think it'll fix that push failure |
@rekram1-node, @gigamonster256: that PR with #11530 and the nix packaging pipeline should work again for all platform. |
You want me to integrate your changes also in that PR? |
|
I'm a little bit away from knowing why the darwin vs linux build hashes differ, looks like the linux build of the darwin node_modules is pulling in extra deps Details...Only in /nix/var/nix/builds/nix-44933-3678005781/build/tmp.BqqdABf3oP/which-typed-array: 1.1.20@@@1 |
The PR explain it: |
|
I understand that, but... why? for instance the linux based darwin node_modules installs sst-darwin-arm64 but the darwin based one doesnt. That seems wrong. |
The dependencies tree has leaves that depend on the platform: fs notification, etc.. And most package managers in the JS ecosystem does not handle it strictly and do performance trade-off at computing the "merged" tree of the really needed deps. |
3 participants
Summary
Problem
PR #11495 introduced a regression where darwin hashes in
nix/hashes.jsonincludegot:prefix (e.g.,got:sha256-...instead ofsha256-...).The
sed -Eregex using\s+behaves differently on BSD sed (macOS) vs GNU sed (Linux), causing the substitution to fail on darwin runners.Ref: #11530 (comment)
Solution
Replace complex sed parsing with portable
grep -oEto extract sha256 hashes directly. Usetail -n1to get thegot:hash (second match) since nix outputsspecified:hash first.Ref: #11530