Files

Fusion c990c115a3 test(dashboard): consolidate low-signal TaskDetailModal matrix tests

Collapse CSS-class presence checks and column-gate permutations into
table-driven variants. The rest of this file covers real user flows
(retry uniqueness FN-bug, paste/drag upload, plan approval, inline
editing, agent assignment, etc.) and was left intact.

- mobile responsive structure 6→1 (class-presence aggregate)
- Commits tab visibility 3→1 (it.each over done-task variants)
- Refinement button column gate 5→2 (it.each over columns)
- Workflow tab visibility 4→1 (it.each over data states)

5,676 → 5,433 LOC (-243). 190 tests still passing.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-04-24 10:03:57 -07:00

3.6 KiB

Raw Blame History

Phase 0 Baseline — Test Suite Measurement

Captured 2026-04-24. Three consecutive pnpm test runs from a warm-ish state (cache existed from prior work). No code changes between runs.

Wall-clock per full run

Run	Duration	Exit	Notes
1	151s	❌ 1	2 failed tests (see Flakes)
2	137s	✅ 0	9,680 tests passed
3	95s	✅ 0	fully warm cache

Warm-cache floor ≈ 95s. Full-rebuild ceiling ≈ 151s. Variance driven mostly by vitest collect + core's SQLite setup.

Per-package (run 2, representative)

Package	Files	Wall (s)	Test CPU (s)	Notes
core	67	91.4	274.0	dominant wall cost; heavy SQLite, worker-pool capped
dashboard	329	33.9	220.2	massively parallel, highest test count
engine	64	5.9	15.9	lean
cli	41	3.7	12.5	lean
tui	4	2.1	2.5
desktop	14	0.9	0.9
pi-claude-cli	9	0.6	0.2
mobile	3	0.5	<0.1
plugin-sdk	1	0.17	<0.01
7 plugins	8	<0.4 ea	trivial
TOTAL	540	≈137s	≈526s CPU

Where the time goes — core (67%) + dashboard (25%) = 92% of wall-clock. Everything else combined is < 10s.

Flakes detected

Run 1 failed; runs 2 & 3 were green → 1/3 flake rate at the suite level. Two distinct flaky tests, both in packages/dashboard:

src/__tests__/routes-diff.test.ts > GET /api/tasks/:id/diff > returns 404 when task not found
- Mode: 5000ms testTimeout exceeded. Test just calls a handler with a mock store — no legitimate reason to take >5s. Almost certainly state leakage or fake-timer hang.
app/components/__tests__/QuickEntryBox.test.tsx > Rich creation features > resets all state after successful creation (disclosure resets to collapsed)
- Mode: assertion textarea.value === "" got "Task to reset". State leak from a prior test in the same file (textarea re-used across iterations before reset).

Both are in the __tests__/ variant of the dashboard dual-test-dir duplication (see test-cull-duplicates.md). Note: these files may be the newer/larger replacements — deleting the companion flat-dir copies is still safe, but these two cannot just be deleted since they cover real endpoints. Fix or skip candidates for Phase 4.

Confirmed Phase 1 duplicates

9 component/hook test files are duplicated between packages/dashboard/app/**/*.test.tsx (flat) and packages/dashboard/app/**/__tests__/*.test.tsx. Full table in .planning/test-cull-duplicates.md. Back-of-envelope LOC the flat copies alone: ~5,150 LOC; some __tests__ copies are smaller and are the redundant side. Cheap win — just need to diff each pair and delete the less-complete one.

Headline numbers to beat

540 test files → target <300 after Phases 1+2
137s warm-ish wall time → target < 90s
Flake rate 1/3 → target 0/5
9,680 passing tests → target unchanged in meaningful coverage, but LOC down ≥ 30%

Raw artifacts

.planning/test-baseline-run1.log (failed — includes flake stack traces)
.planning/test-baseline-run2.log (green)
.planning/test-baseline-run3.log (green)
.planning/test-baseline-times.txt

Ready for Phase 1

Next step: diff each of the 9 duplicate pairs and delete the redundant copy. QuickScriptsDropdown, useProjects, useUsageData, useActivityLog have the flat version larger — suggests the __tests__ copy is the stale one for those. For the other five, __tests__ is larger.

Awaiting go-ahead before deletions.

3.6 KiB Raw Blame History