Repository navigation
fix(review): separate future follow-ups from intake blockers - #3982
Conversation
The flat required-action list made future-stage conditions block intake. Add structured non-blocking follow-ups, trusted-schema cache invalidation, and fail-closed queue validation.
Haidian Submission ValidationSubmission validationResult: PASS Warnings:
|
CocoSgt
left a comment
There was a problem hiding this comment.
Exact head d2b0a8b4fcba412e21ca3b0d3e4901fa28422248 reproduces and fixes the fail-closed/fail-close ambiguity without weakening current gates: current participant-controlled repairs remain blocking, while explicit future-triggered items require blocking_now=false, a nonempty trigger, and a bounded owner enum. Schema-invalid or ambiguous items still fail closed; the worker invalidates legacy cached reviews by schema version and uses the trusted current-main schema. Independent Python 3.9.6 verification passed 56/56 related review/encoding tests, three-script py_compile, schema JSON parsing, and git diff --check; required exact-head CI is SUCCESS. This aligns the existing documented repository-intake boundary and does not alter the 60-point floor, mandatory rejection, four local gates, publication semantics, participant path scope, or exact-head checks.
… 0.2.0 (open-city-ai#3982 separates untriggered follow-ups from intake blockers); no substantive content change
Summary
0.2.0and add structuredconditional_followupswith requiredblocking_now=false,trigger, andownerrequired_next_actions_zh/ per-dimension repairs for current participant-controlled blockers, while rendering future-stage conditions in a separate non-blocking sectionRoot cause and safety boundary
The flat
required_next_actions_zhlist had to carry both current repairs and future-stage conditions. The normalizer and queue therefore converted any condition intorequest-changes, even when the review itself called it non-blocking. This patch adds an explicit machine-readable distinction instead of inferring ownership or timing from keywords.A current unsupported claim is still a blocker: the participant must correct the claim now. Only work whose stated trigger has not occurred belongs in
conditional_followups.The authoritative PR comment now renders current blockers and conditional follow-ups separately. This does not duplicate #3743, which remains the separate accept-path delivery fix for posting the full authoritative comment.
Verification
python -m unittest discover -s tests -p 'test_*review*.py' -v— 89/89 passedpy_compilefor the three changed review scripts — passedgit diff --checkand schema JSON parse — passed452b367ff3c19fc5079a38fc60999c86fd084faausing the trusted0.2.0schema — succeeded; reviewed package hash remainedc8eca57bd34497af50bc53db1d625d1a6299d1cef334906024d3311816fe54ebA broader sparse-worktree run passed 287/289 non-symlink/non-gallery tests; the two remaining failures were unrelated existing/sparse-fixture checks (cover separator text and generated gallery state without all submissions materialized).
Rollout note
This code intentionally does not mutate existing PR reviews or labels. Existing
changes-requestedPRs need a maintainer requeue or manual intake decision after this lands; the schema-version check then prevents reuse of their old cached advisory result.Addresses #3980.