Runner Pipeline

CLI Flow

When you run repterm, the following pipeline executes:

1. Parse arguments — Load configuration (record, workers, timeout, verbose, prompt-lines, slow-threshold, recording-dir)
2. Discover tests — discoverTests(paths) finds .ts/.js test files. setup.ts/setup.js files in a directory are auto-loaded before that directory's tests
3. Load test files — Registers suites and tests into the global registry
4. Filter — filterSuites(allSuites, recordEnabled) applies the --record filter
5. Execute — workers === 1: sequential / workers > 1: parallel scheduler
6. Report — Stream results and print summary; exit 1 on any failure

Single-Worker Execution

Tests execute using an onion model:

Suite
  → beforeAll
  → Tests in order
    → beforeEach (+ lazy fixtures)
    → test function (with timeout via Promise.race)
    → afterEach (cleanup fixtures)
  → Child suites (recursively)
  → afterAll

Each test follows this lifecycle:

1. Notify Reporter (onTestStart)
2. Build terminal (recording / PTY-only / spawn based on config)
3. Lazy-load fixtures based on test function parameter names
4. Run test function with timeout guard
5. Produce RunResult (with optional recording path)
6. Cleanup: clear steps, run afterEach, close terminal

Multi-Worker Execution

With --workers <n> (n > 1), the scheduler distributes work:

1. Main process creates worker processes and waits for ready signals
2. Suites are dispatched to idle workers via IPC
3. Each worker runs its assigned suite and streams results back
4. Main process aggregates results and recycles workers

Reporter

The reporter handles test output:

• onTestStart — Prints suite hierarchy
• onTestResult — Prints pass/fail with duration (highlighted if > slowThreshold)
• onRunComplete — Prints summary and failure details (stack traces shown with --verbose)

Options: verbose (boolean), colors (default true), slowThreshold (default 50ms).

Artifacts

Recording artifacts are managed automatically:

• Base directory: /tmp/repterm (configurable via --recording-dir)
• Each run gets a unique ID (timestamp + random suffix)
• .cast files are produced for recording tests
• Recording paths are passed to the Reporter for display

Default Configuration