perf(test): make suite sharding configurable

[benvinegar]

Problem

Hunk's test runner can shard only its default Linux suite. Other suites and CI platforms cannot reuse the independent-process supervision without duplicating it, while naïvely sharding filtered or coverage runs produces incorrect failures or output.

Approach

• add named test groups to the existing independent-process shard runner
• route test:integration through the same supervised runner
• allow CI to explicitly enable measured shard counts on non-Linux platforms
• keep resource-intensive PTY tests and non-Linux developer runs serial by default
• force filtered and coverage-producing invocations to remain serial so zero-match shards cannot fail focused runs and shards cannot clobber shared coverage output

This keeps Bun's isolated --parallel mode disabled because it is incompatible with OpenTUI's native FFI initialization.

Validation

• ./node_modules/.bin/bun test scripts/test/run-test-suite.test.ts
• ./node_modules/.bin/bun run test
• ./node_modules/.bin/bun run test:integration
• focused integration invocation with --test-name-pattern
• bun run format:check
• bun run lint
• bun run typecheck
• bun run deps:check

An explicit two-shard PTY benchmark completed locally, but a hosted-runner responsiveness test exposed resource contention. PTY therefore remains serial by default; explicit sharding is available only for measured environments. Windows command construction is unit-tested here and exercised by the next PR.

Stack

1. perf(test): make suite sharding configurable #1057 — configurable test sharding (this PR)
2. perf(ci): parallelize validation lanes #1058 — parallel CI validation lanes
3. perf(nix): skip unchanged lock generation #1059 — faster Nix validation

Merge in this order.

This PR description was generated by Pi using OpenAI GPT-5.6 Sol

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
hunk-web	Ignored	Preview	Sep 8, 2026 5:56pm UTC

[greptile-apps]

Greptile Summary

This PR extends the test-suite runner to support named test groups and uses it to shard the PTY integration suite.

• Defines default and integration test-pattern groups.
• Keeps filtered and coverage-producing invocations serial.
• Allows explicit cross-platform shard counts while retaining Linux-only automatic sharding.
• Adds unit coverage for group resolution, serial-execution detection, shard overrides, and command construction.
• Replaces the direct integration-test command with the grouped runner.

Confidence Score: 5/5

The PR appears safe to merge with no actionable correctness, security, or repository-rule violations identified.

The grouped runner preserves existing argument and runtime contracts, current CI reaches the intended Linux sharding path, and platform-specific overrides remain explicit opt-ins.

Important Files Changed

Filename	Overview
scripts/test/run-test-suite.ts	Adds named test groups, serial-mode detection for filtering and coverage, and explicit cross-platform sharding.
scripts/test/run-test-suite.test.ts	Covers group selection, invalid selectors, shard-count behavior, serial-mode flags, and integration command construction.
package.json	Routes the PTY integration test command through the grouped sharding runner.
.changeset/fast-checks-rest.md	Adds an empty Changeset appropriate for internal CI and test-runner maintenance.

_{Reviews (1): Last reviewed commit: "perf(test): shard PTY integration suite" | Re-trigger Greptile}

[benvinegar]

Force-pushed the stack after CI showed that concurrent PTY processes can starve a responsiveness test on hosted runners. PTY tests now stay serial by default; the runner still supports explicit measured shard counts and keeps filtered/coverage runs safe.

This comment was generated by Pi using OpenAI GPT-5.6 Sol

[benvinegar]

Fixed the file-backed reporter finding: --reporter-outfile now forces serial execution just like coverage and filtered runs, covering both --reporter-outfile=path and split argument forms in the unit test. The stack was cascade-rebased and pushed.

Validation: Bun 1.4.2 runner tests (9 passed, 33 assertions), typecheck, and lint.

This comment was generated by Pi using OpenAI GPT-5.6 Sol