Testing and Code Generation

Choose the narrowest test that exercises the changed behavior, then broaden only when shared infrastructure or integration risk justifies it.

Test Selection

Use -- --show-output or --nocapture only when the test's diagnostic output is needed.

Quick tests are scratch space for inspecting CST, formatter IR, or one analyzer query. Persistent behavior belongs in the subsystem's normal fixture directory before finishing.

Snapshot Workflow

Run the focused test to create pending snapshots. Treat unrelated test failures separately. Never update, add or delete snapshots manually. Let insta do that.

Agents must not run unfiltered cargo insta review; it needs a TTY. In a non-TTY agent shell, cargo-insta 1.44+ prints a filtered --snapshot diff without prompting:

cargo insta pending-snapshots --workspace
cargo insta review --workspace --snapshot '<path-from-list>'

After inspecting the full diff, accept or reject only that snapshot:

cargo insta accept --workspace --snapshot '<same-path>'
cargo insta reject --workspace --snapshot '<same-path>'

Repeat for each change, list pending snapshots again, then rerun the test. Review means checking behavior in the diff. A passing test proves only that output matches the snapshot.

Never use unfiltered cargo insta accept or cargo insta reject, cargo insta test --accept, cargo insta test --force-update-snapshots, or INSTA_UPDATE=always/force. They skip individual review and can modify unrelated work.

Orphaned Snapshots

Do not delete suspected orphan snapshots manually. Deletion is safe only after running the complete workspace snapshot suite without package, target, or test filters:

cargo insta test --workspace --unreferenced delete

For a scoped run, use --unreferenced warn or --unreferenced reject; incomplete test selection cannot prove that a snapshot is orphaned. Inspect every deletion from a complete run.

Biome CLI Behavior Tests

Every new or changed test that calls run_cli or a variant must snapshot the full session. Use assert_cli_snapshot, assert_cli_snapshot_with_redactor, or an insta-based wrapper. Scalar assertions do not replace the snapshot. Unit and helper tests that do not execute the CLI are exempt.

Snapshot after other assertions to capture the final filesystem, configuration, output, and result. Stabilize ordering or redact unstable values; do not omit the snapshot.

Keep related cases with the same arguments, configuration, source kind, and filesystem state in one fixture and usually one snapshot. Include valid, invalid, and boundary cases together. Keep related multi-file cases in one test. Split only for incompatible inputs or test setup.

For one migration rule, keep compatible cases in one JSONC fixture. Prefer overrides when they preserve the starting configuration.

Analyzer Fixtures

Place rule fixtures under the language analyzer's current tests/specs/<group>/<rule>/ hierarchy. The directory group must match the rule declaration.

Use focused source files for parser-dependent cases. Use .jsonc arrays when multiple independent script snippets share the same configuration and module semantics are not required. Use options.json in a subdirectory when tests need different rule configuration.

Diagnostic Expectation Comments

The test utilities recognize these marker texts in source comments:

should generate diagnostics
should not generate diagnostics

Current enforcement:

• a filename containing valid but not invalid must contain one of the expectation markers;
• when either marker is present, actual diagnostics must match it;
• an invalid or neutral filename without a marker is accepted, but adding the correct marker makes intent explicit;
• .snap, .json, .jsonc, and .md are exempt from the mandatory valid-file marker check;
• HTML-family workspace fixtures are checked from raw file content and should use a top-level HTML comment.

Put the marker at the top of a primary fixture. Sidecar files with neutral names do not need a marker unless they are independently analyzed as cases.

Parser Fixtures

Use the parser crate's established ok/ and error/ directories. A recovery regression should include valid syntax after the malformed construct to prove the parser resumes at the intended boundary.

Use just qt <parser-package> to inspect a CST during development; do not leave the quick test as the only regression coverage.

Formatter Fixtures

Use internal specs for behavior introduced or fixed by the change. External Prettier snapshots record comparison results but do not replace focused internal coverage.

The formatter harness performs its idempotency reformat during one test invocation for eligible files. Inspect both formatted output and any IR shown for a mismatch.

Required Code Generation

Root AGENTS.md is canonical for which artifacts must be committed and which full outputs CI Autofix may provide.

Do not run just ready in a dirty working tree: the recipe checks for a clean diff before and after its full verification sequence. Use the focused commands required by the current task, then just f and just l.

Completion Checklist

• A code change has focused persistent coverage.
• A bug fixture fails without the fix.
• Every changed snapshot was individually diffed before a filtered accept or reject.
• Every new or modified CLI behavior test creates a session snapshot.
• Related CLI cases are consolidated instead of fragmented across tiny snapshots.
• Expectation comments match fixture intent.
• Orphan snapshots were pruned through insta.
• Required generated artifacts are present.
• Narrow tests pass before broader checks run.

References

• Main test guide: CONTRIBUTING.md#testing
• Insta non-interactive review: https://github.com/mitsuhiko/insta/blob/1.48.0/CHANGELOG.md#1440
• Analyzer guide: crates/biome_analyze/CONTRIBUTING.md
• CLI snapshot harness: crates/biome_cli/tests/snap_test.rs
• Expectation enforcement: crates/biome_test_utils/src/lib.rs
• Formatter harness: crates/biome_formatter_test/src/spec.rs
• Generator recipes: justfile