ci(patch): Recover the benchmark check's exit status and find the coverage executable

[hassila]

Two workflows fail on every pull request in this repository, independent of the change under review; #8 shows both.

Benchmark comparison always reports an unknown failure. The compare step in swift-benchmark-delta.yml runs under GitHub's default bash -e. swift package benchmark baseline check exits non-zero for any threshold deviation, so the step aborted at that line, before the stderr mapping that turns the result into same, improvement, regression, crash or missing baseline. exitStatus kept its initial value of 1 and the job failed with "unknown failure" whatever the benchmark actually said; the workflow has never passed in its eight-run history. A set +e / set -e guard around the check lets the mapping run. With the mapping reachable, a crash or a missing baseline now also gets a comment producer: the failure text step covers every status other than same, regression and improvement, with distinct wording for those two.

Coverage export finds no executable. swift-code-coverage.yml looked for .build/debug/<repository name>PackageTests.xctest, but the package is named FixedPointDecimal, so that path never existed and the job relied on a recursive find for any .xctest file. That worked until the swift:latest container moved to Swift 6.4, whose default Swift Build backend emits per-target -test-runner and .so artifacts on Linux and no .xctest at all. The find came back empty, and because the file test was unquoted, [ -f ] evaluated true and llvm-cov export ran without a file ("No filenames specified!"). The coverage run now uses --build-system native (supported, deprecated in 6.4), which keeps the single <Package>PackageTests.xctest executable, and the export step finds it by package name, keeps the recursive fallback, quotes the path, and lists the build directory before failing if nothing is found.

Verification. Shell control flow of both steps checked by hand against the follow-up steps; SwiftPM's behaviour for --build-system native and the profile path taken from the 6.4.0 release sources. No Linux Swift 6.4 build was run locally (no container runtime on the machine), so this PR's own CI run is the first real test of the coverage path. Codex (gpt-6-astra) blocked the first version of the coverage fix, which assumed a .swift-testing executable that Swift Build never produces; the rework above answers that and its two should-fix points, and its final pass found nothing blocking.

🤖 Generated with Claude Code

[gemini-code-assist]

Note

Gemini is unable to generate a review for this pull request due to the file types involved not being currently supported.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.83%. Comparing base (ab45157) to head (8c2ea99).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main       #9      +/-   ##
==========================================
+ Coverage   96.79%   96.83%   +0.04%     
==========================================
  Files          35       35              
  Lines        4795     4795              
==========================================
+ Hits         4641     4643       +2     
+ Misses        154      152       -2     

see 2 files with indirect coverage changes
see 2 files with indirect coverage changes

---

Continue to review full report in Codecov by Harness.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ab45157...8c2ea99. Read the comment docs.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[github-actions]

Pull request had a regression
Pull request benchmark comparison [ubuntu-latest] with 'main' run at 2026-09-21T09:42:36+00:00