[HC-013] Hard cutover tests into behavior-first suites and generated artifact boundaries

[doublemover]

Hard cutover policy: tests validate language behavior first and reports second. No generated report churn as primary confidence, no fixture layout that hides semantic ownership, no old behavior fixtures kept as passing cases.

Source facts already established:

• The repo has 5,266 tracked files.
• tests/ contains 4,366 tracked files.
• tests/tooling/fixtures/ contains 2,508 tracked files.
• test:fast runs 12 selected execution fixtures, runtime acceptance fast, and one focused replay.
• Many fixtures and reports validate command surfaces, report contracts, readiness packets, and generated evidence.

Required test tree:

• tests/native/parser/
  • positive/
  • negative/
  • snapshots/
• tests/native/sema/
  • types/
  • ownership/
  • objc/
  • control_flow/
  • errors/
  • concurrency/
  • negative/
• tests/native/lowering/
  • expressions/
  • statements/
  • objc_runtime/
  • ownership/
  • errors/
• tests/native/ir/
  • module/
  • function/
  • metadata/
  • runtime_calls/
• tests/native/runtime/
  • dispatch/
  • object_model/
  • storage/
  • arc/
  • blocks/
  • errors/
  • concurrency/
• tests/native/e2e/
  • smoke/
  • feature_matrix/
  • negative_execution/
• tests/tooling/
  • workflow/
  • schemas/
  • generators/
  • docs/
  • release/
• tests/fixtures/canonical/: hand-authored fixtures.
• tests/fixtures/generated/: generated fixtures with generator provenance.

Required work:

• Move native fixtures to the behavior-first tree above.
• Rename fast validation to smoke validation.
• Add a stronger default validation command that covers parser, sema, lowering, IR, runtime, and e2e samples.
• Delete passing fixtures for old modes, runtime shims, and fallback dispatch behavior.
• Convert generated report checks into schema and generator tests.
• Keep canonical hand-authored fixtures separate from generated fixtures.
• Add fixture metadata that records owning compiler phase and expected diagnostic code.
• Make every support claim link to an executable behavior test.

Maximum effort shared hoist pass:

• Hoist fixture discovery to one Python module.
• Hoist expected diagnostic parsing to one helper.
• Hoist execution result comparison to one helper.
• Hoist schema validation to one helper.
• Hoist generated fixture provenance checks to one helper.

Acceptance criteria:

• tests/native is organized by compiler phase and behavior family.
• Old mode and shim fixtures are negative or deleted.
• Default validation runs a representative behavior matrix.
• Report-only tests cannot create a support claim without executable coverage.
• Test docs describe smoke, default, full, and release gates with exact commands.

[doublemover]

Current checkout evaluation after the hard-cutover integration commits: PARTIAL, keep open.

Landed work:

• tests/native exists and the behavior fixture boundary test passed during integration.
• tests/fixtures/canonical and tests/fixtures/generated boundaries exist.

Remaining blockers from the current checkout:

• The old tests/tooling/fixtures/native corpus still exists and still contains old runtime/dispatch expectations, including objc3_msgsend_i32 outputs.
• Fixture audits still find fallback/shim/legacy/lane/proof wording across tooling fixtures and old generated outputs.
• The behavior-first tree is present, but the canonical fixture migration is not complete.

This should close only after old-mode/shim/fallback positive fixtures are either converted to canonical rejection/strict-error coverage or removed from the active acceptance surface.

[doublemover]

Checkpoint pushed: f4b6e455d.

Behavior-test progress in this checkpoint:

• Updated runtime dispatch fixtures away from retired msgsend/fallback wording and toward strict resolved-or-error expectations.
• Added runtime probe coverage for exact strict dispatch status/result diagnostic payloads.
• Added canonical runnable object probe assertions for live unknown-selector strict errors and cached strict-error behavior.
• Kept generated fixture/replay updates tied to the live runtime dispatch host-link contract.

Validation:

• $env:PYTHONPATH='scripts'; python -m pytest tests/tooling/test_objc3c_library_cli_parity.py::test_parity_source_mode_work_key_changes_with_backend_and_runtime_contract tests/tooling/test_objc3c_library_cli_parity.py::test_parity_source_mode_default_work_key_is_deterministic_for_same_inputs tests/native/test_behavior_fixture_boundaries.py tests/tooling/runtime -q PASS (7 passed)
• npm run objc3c -- test-runtime-acceptance-fast PASS (21/21 cases)

Not closing. The broader behavior-first test tree reorganization, canonical/generated fixture split, smoke/default/full command reshape, and report-only claim restrictions remain open.

[doublemover]

Checkpoint 47cc548e2 removes the runtime-dispatch blocker for behavior-first runtime coverage:

• added live strict dispatch coverage for malformed metadata, unsupported return type, and unsupported argument layout via tests/tooling/runtime/strict_dispatch_error_status_probe.cpp
• extended runtime-library acceptance to compile/run that probe
• kept runtime dispatch behavior covered in the fast acceptance bundle
• validation passed: python scripts/check_objc3c_runtime_acceptance.py --case runtime-library, the corrective runtime checks, and npm run objc3c -- test-runtime-acceptance-fast

Keeping this open. The broader #8144 acceptance still needs the behavior-first fixture tree migration, generated/canonical fixture boundary completion, default/smoke/full/release command documentation, and report-only support-claim restrictions.

[doublemover]

Checkpoint pushed: 3a430e0e5.

Behavior-first test progress in this checkpoint:

• Added shared behavior fixture discovery/metadata validation in scripts/objc3c_tooling/behavior_fixtures.py.
• Added executable behavior matrix gate in scripts/check_objc3c_behavior_matrix.py, wired as npm run objc3c -- test-behavior-matrix.
• Reshaped public validation so test-default delegates to test-smoke, and test-smoke now runs behavior matrix + runtime acceptance fast + focused replay.
• Removed the public test-fast action and public-test-fast suite surface; remaining test-fast text is confined to intentional negative source-hygiene fixtures.
• Updated canonical manifest support claims so every current parser/sema/lowering/IR/runtime/e2e claim links to executable behavior coverage.
• Converted the stale runtime dispatch fixture from a positive fallback-shaped expectation into strict runtime error coverage.
• Updated docs/runbooks for default, smoke, full, nightly/release, and behavior-matrix command names.

Validation run before commit:

• python scripts/check_objc3c_behavior_matrix.py PASS (12/12 fixtures)
• python -m pytest tests/native/test_behavior_fixture_boundaries.py -q PASS (7 passed)
• npm run objc3c -- test-smoke PASS
• python scripts/build_objc3c_native_docs.py --check PASS
• python scripts/render_objc3c_public_command_surface.py --check PASS
• python -m scripts.source_hygiene PASS
• git diff --check PASS
• npm run objc3c -- lint PASS
• python scripts/shared_compiler_runtime_acceptance_harness.py --check-catalog PASS

Keeping this open. Remaining blockers are the full retirement/migration of the legacy tests/tooling/fixtures/native corpus, broader canonical/generated fixture coverage for all old generated/report checks, and final enforcement that report-only surfaces cannot imply support without executable behavior coverage.

[doublemover]

Checkpoint pushed: ac31f1519.

Behavior/evidence boundary progress related to this issue:

• scripts/validate_capability_docs.py now requires every canonical manifest support_claims entry to be backed by an existing behavior fixture and npm run objc3c -- test-behavior-matrix.
• Implemented capability rows in docs/support/capability_matrix.json now link the current parser, sema, lowering, IR, runtime, and e2e behavior claims to metadata-backed fixtures under tests/native.
• Added tests/tooling/test_capability_docs_validator.py so report-only commands cannot satisfy support claims.

Validation:

• python scripts/validate_capability_docs.py --check PASS
• $env:PYTHONPATH='scripts'; python -m pytest tests/tooling/test_objc3c_shared_json_schema.py tests/tooling/test_capability_docs_validator.py -q PASS (13 passed)
• npm run objc3c -- lint PASS

Keeping this open. This closes the support-claim-to-executable-behavior gap for the current matrix, but the old tests/tooling/fixtures/native corpus migration/removal and broader generated/canonical fixture cleanup remain.

[doublemover]

Behavior/fixture checkpoint (May 6, 2026): landed and pushed 893a4af12.

Completed in this checkpoint:

• Reclassified runtime-dispatch retired-surface fixture metadata in tests/fixtures/canonical/manifest.json and tests/native/runtime/dispatch/message_send_runtime_dispatch.meta.json.
• Tightened tests/native/test_behavior_fixture_boundaries.py so retired tags in the canonical manifest must match fixture metadata and remain strict negative/rejection coverage.
• Cleaned stale positive wording from runtime-dispatch negative fixture docs/sources under tests/tooling/fixtures/native/execution.

Validation evidence:

• python -m pytest tests/native/test_behavior_fixture_boundaries.py PASS, 8 passed.
• python scripts/check_objc3c_behavior_matrix.py --phase runtime PASS, 2 runtime fixtures passed.
• git diff --cached --check PASS.

Current closure status: still open. The broader fixture corpus still needs more generated/native old-mode cleanup.

[doublemover]

Fixture cleanup checkpoint (May 6, 2026): landed and pushed 1455128eb.

Completed in this checkpoint:

• Replaced legacy canonical migration fixture residue with strict parser negative coverage:
  • added tests/native/parser/negative/legacy_boolean_and_null_aliases_rejected.objc3
  • added matching metadata
  • removed the old tooling legacy migration positive fixture
• Tightened canonical fixture manifest and behavior-boundary test coverage.

Validation evidence:

• python -m pytest tests/native/test_behavior_fixture_boundaries.py PASS, 9 passed.
• python scripts/check_objc3c_behavior_matrix.py --phase parser PASS, 3 parser fixtures passed.
• git diff --cached --check PASS.

Current closure status: still open. This removes another stale fixture cluster, but broader generated/native fixture cleanup remains.