Bazel docs disagree on whether plain bazel test reports individual test cases

Two Bazel docs pages state opposite things about what a plain bazel test reports, and the code doesn’t settle it.

• server/priv/docs/en/guides/features/test-insights/bazel.md:22 — “Ordinary bazel test still runs your tests, but does not report individual cases to Tuist and does not enforce quarantine policies.”
• server/priv/docs/en/guides/features/test-insights/flaky-tests/bazel.md:28 — “Ordinary bazel test still reports test results, but does not fetch or enforce Tuist quarantine policies.”

One of them is wrong. #13130 deliberately left both alone rather than pick by guesswork.

What the code establishes

Reporting is not gated on the CLI wrapper:

• Kura handles BEP TestResult events straight off the Build Event Service stream (kura/src/reapi/bep.rs:803-836).
• .bazelrc.tuist emits --bes_backend under the build command (BazelrcFile.swift:40-48), which Bazel applies to test too.
• BazelTestCommandService has no upload path — its collaborators are a config loader, a server-environment service, ListTestCasesService (a read), and a command runner. It fetches the quarantine policy and shells out.
• The server parses the JUnit report into individual cases via JunitReport.parse → Tests.create_test (test_report_ingestor.ex:29-49).

So tuist bazel test adds quarantine, not reporting.

What it doesn’t establish

Kura never receives the test.xml bytes. It resolves them by digest from its own REAPI store:

// kura/src/bazel_test_artifacts.rs:478-486
let key = blob_key(&format!("{}/{}", artifact.digest, artifact.size));
let manifest = self.store.manifest_for_key(ArtifactProducer::Reapi, project_handle, &key)

If the blob isn’t in Tuist’s cache it records bazel_test_artifact/missing, and the ingestor’s parse_report/1 falls through to %{test_suites: [], test_cases: []} — you get the invocation and the per-target summary, but nothing case-level.

That makes case-level reporting conditional on Bazel having uploaded the test outputs to Tuist’s CAS. The load-bearing question is whether a failing test’s outputs get there. Bazel doesn’t cache failed test results; if it also doesn’t upload their outputs, a plain bazel test would surface passing cases and silently drop failures — exactly the case test insights exist for.

Nothing in the repo answers it:

• kura/spec/e2e/ has no BES or test-result spec.
• server/test/tuist/bazel/test_report_ingestor_test.exs has no missing-artifact case.
• .bazelrc.tuist sets no upload flags; Kura’s own e2e harness passes --remote_upload_local_results=true explicitly (kura/spec/e2e/support.sh:445) rather than relying on the default.

How to settle it

Run a Bazel target with one passing and one deliberately failing case against a project with .bazelrc.tuist, using plain bazel test, then check whether each case appears in the dashboard. Worth covering both a cold run and a re-run, since a cached result behaves differently from a locally executed one.

Then:

1. Correct whichever sentence is wrong, and state the condition if reporting turns out to depend on outputs reaching the cache.
2. If failures genuinely don’t reach the CAS, that’s a product gap rather than a docs bug, and this issue should be re-scoped.
3. Consider an ingestor test for the missing-artifact path — today it silently produces zero cases.

🤖 Generated with Claude Code

No GitHub comments yet.