Testing

Test-driven by default. Prove the problem before solving it.

Core Workflow: Bug First

1. Reproduce — Run the failing scenario and observe the actual behavior
2. Write a failing test — Minimal reproduction as a test case
3. Commit the failing test — Use Skill(commit) with message like test: reproduce #issue-description
4. Fix the code — Only now attempt the fix
5. Verify — Run pnpm test and confirm the test passes
6. Commit the fix — Separate commit from the test

This applies to all bug fixes. The failing test proves the bug exists and prevents regressions.

Fixture Tests

Noodle uses directory-based fixtures in testdata/. Read references/fixtures.md for full format, helper API, and code patterns.

Creating a New Fixture

./scripts/scaffold-fixture.sh <package-dir> <fixture-name> [state-count]

Fill in the generated input.json/input.ndjson and expected.md, then:

pnpm fixtures:hash:sync

Bug Fixtures

See references/fixtures.md for full bug fixture format (frontmatter fields, expected error sections, and the fix workflow).

Key Commands

Running Specific Tests

# Single package
go test ./loop

# Single test
go test ./loop -run TestLoopDirectoryFixtures

# Bypass cache
go test -count=1 ./parse

# With race detector
go test -race ./...

# Verbose
go test -v ./tui

Integration Tests

Long-running or external-dependency tests use testing.Short():

if testing.Short() {
    t.Skip("skipping integration test in short mode")
}

pnpm test:short skips these. pnpm test runs everything.