Skip to content

Watch

In watch mode, Scrutin monitors your project and re-runs affected tests every time you save a file. It's on by default in the TUI and web: just run scrutin and start editing. Set watch.enabled = false in .scrutin/config.toml (or pass --set watch.enabled=false) to opt out. The plain, GitHub, JUnit, and list reporters are always one-shot regardless.

What happens on save

  1. The file watcher detects a change (with 50ms debounce to coalesce rapid saves)
  2. The dependency tracker determines which test files are affected
  3. If a source file changed, workers reload the package before running
  4. If only a test file changed, workers run it immediately without reloading
  5. Results stream into the frontend as they arrive

Dependency mapping

Scrutin doesn't re-run your entire test suite on every change. It tracks which test files depend on which source files.

R: Runtime instrumentation. On the first watch run (or when source / test files have changed since the last build), Scrutin instruments every function in the package namespace, runs each test file, and records which functions were called. The map is persisted in SQLite (.scrutin/state.db, tables dependencies and hashes) and invalidated when file contents change. Multi-suite aware: editing R/math.R correctly invalidates test files under both tests/testthat/ and inst/tinytest/.

Python: Static import analysis. Scrutin scans every .py file for import and from ... import statements and inverts the graph into a "source module → test files" index. Transitive: if test_x.py imports helpers.py which imports core.py, editing core.py triggers test_x.py. Dynamic imports (importlib.import_module) are missed; unresolved edits fall back to the filename heuristic or a full suite re-run. The Python index is rebuilt from scratch on every invocation (cheap enough that caching isn't worth the staleness risk).

Fallback: When no mapping exists for a changed file, all test files re-run.

Ignored files

The built-in walker also skips common build / VCS noise regardless of config (.git/, target/, node_modules/, __pycache__/, .pytest_cache/, .Rproj.user/, etc.). The [watch] ignore list adds user-supplied glob patterns on top, matched against paths relative to the project root. The default is [".git", "*.Rhistory"].

Add your own patterns:

[watch]
debounce_ms = 50
ignore = [".git", "*.Rhistory", "renv/"]

See Parallelism for how Scrutin schedules workers across suites, and the safe-spawn vs. fast-fork tradeoff.