Summary
Keep every individual test in the isolated full-suite inventory below one second without weakening behavioral coverage.
Problem to solve
The full isolated Vitest inventory contains tests whose wall time is dominated by repeated module initialization, oversized fixtures, real timeout waits, or broad plugin discovery. These outliers make local and CI feedback slower, obscure genuine regressions, and can push a monolithic profiling run into OOM before it reports the complete inventory.
Proposed solution
Add a bounded full-suite duration-report path, then remove the measured outliers at their owning boundaries: reuse shared harness warmup, defer heavy imports, keep plugin artifacts lightweight, add core-registry fast paths, reduce fixtures while preserving size-limit contracts, and replace real deadline sleeps with deterministic timer coverage. The final isolated inventory should report zero individual tests above 1,000 ms.
Alternatives considered
- Raising the threshold hides the regressions and preserves the slow feedback loop.
- Running only a hand-picked offender list misses new outliers elsewhere in the 89-plan suite.
- Increasing worker counts makes timing noisier and reintroduces the memory-pressure failure mode.
Impact
Affected: OpenClaw contributors and CI runners.
Severity: Medium developer-productivity and CI-capacity cost.
Frequency: Every broad test or duration-profile run.
Consequence: Slower iteration, harder performance triage, and OOM-prone inventory runs.
Evidence/examples
The isolated inventory previously exposed repeated one-time initialization costs and real timer waits above one second. A bounded Testbox run avoids the monolithic OOM and produces per-plan RSS and individual-test timing evidence.
Additional information
This is maintainer-requested test/runtime performance work. It does not change release versions or published artifacts.
Summary
Keep every individual test in the isolated full-suite inventory below one second without weakening behavioral coverage.
Problem to solve
The full isolated Vitest inventory contains tests whose wall time is dominated by repeated module initialization, oversized fixtures, real timeout waits, or broad plugin discovery. These outliers make local and CI feedback slower, obscure genuine regressions, and can push a monolithic profiling run into OOM before it reports the complete inventory.
Proposed solution
Add a bounded full-suite duration-report path, then remove the measured outliers at their owning boundaries: reuse shared harness warmup, defer heavy imports, keep plugin artifacts lightweight, add core-registry fast paths, reduce fixtures while preserving size-limit contracts, and replace real deadline sleeps with deterministic timer coverage. The final isolated inventory should report zero individual tests above 1,000 ms.
Alternatives considered
Impact
Affected: OpenClaw contributors and CI runners.
Severity: Medium developer-productivity and CI-capacity cost.
Frequency: Every broad test or duration-profile run.
Consequence: Slower iteration, harder performance triage, and OOM-prone inventory runs.
Evidence/examples
The isolated inventory previously exposed repeated one-time initialization costs and real timer waits above one second. A bounded Testbox run avoids the monolithic OOM and produces per-plan RSS and individual-test timing evidence.
Additional information
This is maintainer-requested test/runtime performance work. It does not change release versions or published artifacts.