The three plan.py call sites (tree, explain, correct) that originally used
container.resolve(DecisionService) were already fixed in master. This commit
completes the fix by aligning the Robot Framework test mocks to match the
corrected production code:
- robot/helper_m3_decision_validation_smoke.py: mock_container.resolve →
mock_container.decision_service (plan correct dry-run path)
- robot/helper_m4_correction_subplan_smoke.py: container.resolve →
container.decision_service (correction subplan path)
Both helpers previously passed silently because MagicMock auto-creates any
attribute, masking the mismatch between mock setup and actual code. Added
spec=Container to both mock containers so that accessing non-existent
attributes raises AttributeError, matching real container behavior.
Updated comment wording in features/container_resolve_crash.feature to
clarify that the bug is fixed and the scenarios serve as permanent
regression guards. The @tdd_expected_fail tag was already removed by the
TDD branch before this fix branch was created; the permanent @tdd_issue
and @tdd_issue_647 tags remain per TDD tag lifecycle rules.
Added CHANGELOG.md entry describing the fix.
ISSUES CLOSED: #647
hurui200320