nix-hashes workflow fails on transient infrastructure errors #30742

Description

The compute-hash matrix step in .github/workflows/nix-hashes.yml intermittently fails on transient infrastructure errors during nix build / bun install.

A sample of 8 recent failures shows three distinct transient patterns:

• 6/8 — APFS unlink race on macOS aarch64-darwin during nix's post-build cleanup:
  error: cannot unlink "/nix/store/.../node_modules/...": Directory not empty
• 1/8 — bun npm registry batch resolution failure (run 26936133588, x86_64-darwin):
  error: [email protected] failed to resolve (~25 packages)
• 1/8 — nix DNS failure: error: unable to download 'https://channels.nixos.org/...': Could not resolve hostname

Re-running the workflow normally succeeds in all cases.

Impact

When this fires, the update-hashes job is skipped and nix/hashes.json is not refreshed, breaking downstream Nix consumers until the workflow is manually re-run.

Proposed fix

Wrap the nix build + hash-extraction block in a 3-attempt retry loop with linear backoff (10s, 20s). Success criterion stays "hash extracted from fakeHash mismatch log"; structural failures still surface after the final attempt.

OpenCode version

dev @ 69cfc44db