* test(integration): red — Fix: SubprocessRunner.Build masks every build failure as a chmod ENOENT (refs be-4ecr4)
* feat(integration): green — Fix: SubprocessRunner.Build masks every build failure as a chmod ENOENT (refs be-4ecr4)
* fix(integration): stop orphaning the harness test's temp dir, pin the stdout half
Maintainer review fixups for PR #6715 (be-4ecr4), both minor:
- TestBuildErrorIsMaskedByChmod passes a synthetic *testing.T to Build, which
the framework never finalizes, so the Cleanup that its t.TempDir() registers
never ran and every execution orphaned a directory under $TMPDIR (measured:
one new /tmp entry per run). Remove the mkdtemp parent from the real t —
TempDir hands back <parent>/001, so deleting the binary's own directory would
leave the parent behind — guarded to a proper descendant of $TMPDIR so a
future layout change skips rather than deletes something we did not create.
- The PR makes two changes but pinned only one. Removing build.Stdout left all
three assertions satisfied, so the stdout capture could silently regress.
Assert on "[setup failed]", which `go test -c` demonstrably emits on stdout
in this scenario; the "stdout:" label would be a vacuous pin because it lives
in the format string and survives an unattached buffer. Verified by
counterfactual: dropping the wiring now reddens the test, and dropping the
early return still reddens it for the original masking reason.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: Julian Knutsen <julianknutsen@users.noreply.github.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>