Skip to content

PLAN-add-tests.md

Latest commit

 

History

History
69 lines (54 loc) · 3.76 KB

PLAN-add-tests.md

File metadata and controls

69 lines (54 loc) · 3.76 KB

Add Tests to Verify Usage Tracking

Context

No tests exist yet (listed as TODO in README). The goal is to verify that Claude Code usage is correctly tracked: JSONL parsing, token aggregation, cost calculations, and DB queries all produce correct numbers. We'll use in-memory SQLite (better-sqlite3 supports :memory:) for fast, isolated tests.

Approach

  • Framework: Vitest (natural fit since Vite is already in the project; handles ESM + TypeScript natively)
  • DB strategy: Add a single setDb() function to schema.ts to allow injecting an in-memory database. All query functions already call getDb() which returns the singleton, so no other production code changes needed.
  • Test types: Unit tests for pure functions + integration tests for the parse-to-DB-to-query pipeline

Files to Create

File Purpose
vitest.config.ts Separate from client vite.config.ts (which has root: 'src/client' and React plugin)
src/server/test/setup.ts setupTestDb() / teardownTestDb() helpers using :memory: DB
src/server/test/fixtures.ts JSONL test data constants (various models, streaming dedup, subagents, etc.)
src/server/parser/jsonl.test.ts Parser unit + integration tests (~18 tests)
src/server/db/queries.test.ts DB query and cost calculation tests (~20 tests)

Files to Modify

File Change
src/server/db/schema.ts Add setDb(database) function (4 lines)
package.json Add vitest devDependency + "test" / "test:watch" scripts

Test Coverage

Parser tests (jsonl.test.ts)

  • Pure functions: isSubagentFile, extractProjectFromPath, extractSessionExternalIdFromPath, extractParentSessionExternalId
  • Integration: Basic session parse -> DB verify, streaming message deduplication (same ID appears multiple times, last wins), skipping file-history-snapshot lines, handling invalid JSON gracefully, custom title extraction, sync state tracking, incremental sync, upsert idempotency, subagent parsing + parent linking

Query tests (queries.test.ts)

  • CRUD: upsertSession insert + update semantics (keeps earliest start, latest end), insertMessages bulk + upsert on conflict
  • Cost calculations (all 5 pricing tiers with hand-calculated expected values):
    • Sonnet 4: input=$3, cache_write=$3.75, cache_read=$0.30, output=$15
    • Sonnet 4.5 (<=200K context): $3/$3.75/$0.30/$15
    • Sonnet 4.5 (>200K context): $6/$7.50/$0.60/$22.50
    • Opus 4.5: $5/$6.25/$0.50/$25
    • Haiku 4.5: $1/$1.25/$0.10/$5
    • Unknown model falls through to default pricing
  • Aggregation: getSessionStats with date/project/customTitle filters, agent-* exclusion, subagentCount; getDailyStats date grouping; getSummary totals + cache savings; getProjects/getCustomTitles distinct lists; getSubagentsBySessionId; cleanupOrphanedSubagentSessions

Implementation Sequence

  1. Install vitest: npm install --save-dev vitest
  2. Add npm scripts: "test": "vitest run", "test:watch": "vitest"
  3. Create vitest.config.ts at project root
  4. Modify schema.ts - add setDb() function
  5. Create src/server/test/setup.ts
  6. Create src/server/test/fixtures.ts
  7. Create src/server/db/queries.test.ts
  8. Create src/server/parser/jsonl.test.ts
  9. Run npm test to verify

Verification

npm test            # run all tests once
npm run test:watch  # watch mode during development

Issues

No issues encountered during implementation. All 50 tests passed on the first run (170ms total):

  • queries.test.ts: 28 tests passed
  • jsonl.test.ts: 22 tests passed

The in-memory SQLite approach worked seamlessly -- the single setDb() addition to schema.ts was sufficient to redirect all DB operations to the test database without any other production code changes.
⚡