Release suite: sibling paths from the exported variables, a refusal to install over a linked node_modules, and the lanes a macOS host cannot run named instead of failed

[puskuruk]

Four gaps the release suite showed on a macOS host whose checkouts are not laid out as siblings, found running npm run release:check for #7-#10. None changes what a lane tests; each removes a failure that was about the harness, not the release.

• test/gea-commonjs.mjs, test/gea-host-namespace-members.mjs and test/native-rtc.mjs resolved core as ../core beside the compiler checkout, although the compiler lane exports GEA_CORE_ROOT. They now read the variable first and fall back to the sibling. Verified from a worktree with no core beside it: all three pass with the variable set.
• scripts/release-setup.mjs ran npm ci in a worktree whose node_modules was a link to another checkout's install, and npm emptied that checkout. Setup now refuses such a directory, naming the link's target.
• The perf lane required the benchmarks checkout on every host, then named itself as not run off Linux x86-64. The checkout is required only where the lane runs.
• The parallel lane shipped the native fixtures to dashwin-geastack from macOS and failed when that machine was unreachable. When neither GEA_PARALLEL_HOST nor the default host answers an ssh -o BatchMode=yes probe, the fixtures build and run on this host, which has the sanitizers they need; the choice is printed. Verified: --only=parallel passes on this Mac with the host unreachable (656 s for the native step alone; beside a gate run and a container build the stress fixture under ThreadSanitizer exceeded the runner's 300 s cap, so the lane wants the machine to itself).
• The three run-gea-companion-*.sh tests compile the generated C++ of a gea-companion macOS build under apple's targets/macos/generated/gea-companion, which no lane produces. They are named in the summary as not run, as the suite already does for host-bound tests, until a lane produces that build.

Related

• Honour the release suite's compiler variables in the four tests that still resolved the compiler as a sibling checkout core#73 makes core's own scripts honour GEA_COMPILER_CLI and GEA_COMPILER_RUNTIME the same way.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}