Release v0.0.290

Summary

TestCompareDynamic_ZeroAllocHotPath (pkg/registry/file/dynamicpathdetector/tests) is flaky. It was the cause of the intermittent pr-created / test failures (e.g. on #339, which is unrelated — it only touches summary-storage aggregation).

Root cause

The test measured allocations by diffing the process-global runtime.MemStats.Mallocs counter around a 10,000-iteration loop:

runtime.ReadMemStats(&before)
for i := 0; i < iters; i++ { _ = CompareDynamic(...) }
runtime.ReadMemStats(&after)
require.Equal(t, uint64(0), after.Mallocs-before.Mallocs)

Mallocs is process-wide, so a single allocation from a background runtime goroutine (GC assist, sysmon, the testing timer) landing inside the measurement window flips the delta to 1 and fails the assertion. The microsecond-scale window makes it rare but nonzero — the failure was observed flipping between different subtests across runs (literal_mismatch in CI, literal_exact_match locally), the signature of a flake rather than a regression.

Fix

Switch to testing.AllocsPerRun, which pins GOMAXPROCS(1) during measurement and integer-divides total mallocs by the run count. A stray sub-runs allocation now floors to 0 instead of flaking, while a genuine per-call allocation still yields ~1.0 and fails the zero-alloc contract. The contract is preserved; only the noise is removed.

Verification

• 30/30 isolated runs green (previously ~PASS/PASS/FAIL/FAIL/PASS).
• Full go test ./... suite green.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Tests
  • Updated a performance-oriented test to use a more reliable allocation check.
  • The zero-allocation assertion now uses per-run allocation measurement with clearer failure messaging.