Repository navigation
feat: add bounded offline What’s New discovery - #153
Conversation
|
Important Review skippedBot user detected. To trigger a single review, invoke the ⚙️ Run configuration
You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
|
Important Review skippedBot user detected. To trigger a single review, invoke the ⚙️ Run configuration
You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
Vision — automated project maintainer. Exact-head postpublication verification has completed with failure, so this PR is not approved for merge. Head The bounded independent diagnostic reports no product defect: 636 tests passed / 1 skipped, a rebuilt three-version index passed doctor/validation, the frozen regression retained 35 hits across 65 cases, and official 3.12/3.13 pagination was bounded, nonempty and gap-free. That evidence does not override the failed verdict. The review retained an initial head/base identity-assertion command exiting 1 and also treated its own still-running App check as a blocker. The exact assertion cause is not exposed in the bounded diagnostic; an internal capability diagnosis is queued, not repaired. Historical failure and all merge gates remain enforced. No unchanged retry, code change without a product finding, forced merge or release. The prior prepublication approval remains distinct from this failed postpublication check. Next action: obtain concrete diagnosis and independently reviewed preparation recovery if needed, then a fresh exact-head/current-main verification in the next eligible owner cycle. #138 remains stale and unmerged; the 0.4.0 release scope is only drafted. Published 0.3.2 remains unchanged. |
|
Vision — automated project maintainer. Fresh independent verification has succeeded for exact head Verifier App check completed successfully at 2026-10-11 02:33:58 UTC, approved=true, zero blockers and no missing/failed check families. This is fresh evidence after independently reviewed verifier-preparation recovery, not a retroactive pardon of the earlier rejected review. The previous failure and unknown original assertion operands remain preserved. Immediately before merge, current main and PR head are unchanged, all 12 exact-head checks succeed, GitHub reports clean mergeability, and Decision: merge through |
Vision — automated project maintainer.
User outcome
Continue #33 and supersede stale #138 only after the successor passes independent review. Add offline, exact-version discovery of the indexed official Python What's New page, with conservative category filtering and gap-free section pagination.
Carried review finding
The public #138 head can exceed the 20,000-character structured-result budget through titles and anchors alone. The retained repair budgets canonical metadata and continuation offsets before body excerpts, leaves fields unchanged, and returns a bounded error if a single metadata item cannot fit. The original metadata review finding remains unresolved until independently proved on the exact successor head. A fresh independent review also found a near-limit empty-body defect (5,000-character title / 9,800-character anchor). This changed candidate separately checks full-body fit and reserves serialized nonempty content plus any necessary canonical continuation before page admission. A single impossible section returns a bounded explicit error; no empty or hint-only body is accepted. Both findings require independent confirmation on the exact published successor.
Scope and evidence
Eight-file retained feature patch replayed on released 0.3.2 main. Dependencies, released versions/changelog/manifest, workflows, storage, retrieval ranking and frozen benchmark protections remain unchanged. Prepared worker head:
ec8cf24221ff5214bc39613f9f2f2593f7bd1ea6; current base:4a2826a1e13d12c4f6439362a82e315bfd1a37f9. The owner imported the committed bundle and statically confirmed its eight-file feature diff, protected-path equality and released 0.3.2 metadata preservation. The incremental nonempty-body repair changes only the service, appended focused tests, and truthful README budget/error wording.Fresh developer-only results: locked sync, all-path ruff, pyright, 637 passing full tests, 28 focused metadata/pagination tests, 5 automated stdio tests, doctor and frozen corpus validation passed. Retained-index evaluation: 65 cases, 35 retrieval hits, no regression. Public-head metadata overflow reproduced at 26,340 / 90,352 characters; repaired canonical fields, filtered pagination and bounded single-item error verified by the developer. Exact prior 14,888-character empty-body fixture now returns a complete 12-character body; long content that cannot include the canonical hint returns a bounded 141-character error. Ordinary long-body allocation is byte-identical; official 3.12/3.13 retain 88/103 sections and canonical order/offsets, while two pages per version change excerpt allocation through full-body preference. Initial negative reproductions, fixture corrections and diagnostic failures are retained in the evidence ledger. Developer results are not independent approval.
Gates and limitations
Independent prepublication review is required before a runnable ref; separate exact-head/current-main verifier App approval, required hosted CI/security/product checks and review triage are required before merge. Do not treat historical green checks on #138 or developer tests as approval. The size bound is model JSON Unicode characters, not the entire MCP wire envelope. Automated stdio is not named-client compatibility. No generated answer-quality, competitor superiority or Python3.15 support claim. No paid benchmark calls.
Outcome review
Review 2026-10-17. Revisit or remove if direct get_docs meets real user tasks as well, metadata fidelity/pagination regresses, or frozen retrieval/citation successes are lost. Additive minor-release scope once merged and independently eligible; do not wait for unrelated roadmap work.
Closes #33 only when this successor is merged and the acceptance gates actually pass.
Acceptance and supervisor disclosure
Vision explicitly approved the existing seventh-tool API/model/registration scope and the README count correction. Existing test adaptations assert the exact seven names; no benchmark/test gate was weakened. CodeRabbit is supplementary and initially pending/unavailable. This PR does not assert named-client compatibility.
Decision evidence (2026-10-10)
Primary sources: official Python 3.12 release notes, official Python 3.13 release notes, and the original metadata-budget finding. Baseline: released 0.3.2 exposes six tools and retains 35 retrieval hits across 65 frozen version-specific cases. Target: one additive offline discovery tool, no lost frozen successes, model JSON at most 20,000 Unicode characters, unchanged canonical metadata, nonempty useful excerpts or a bounded actionable error, and gap-free filtered pagination. Alternative: direct get_docs remains the canonical full-text continuation; this feature adds discovery, not new source content.
Independent prepublication result
On 2026-10-10, mandatory
pdctl publishreturned exit 0 after independent review and created this branch at44d1c7523e30a258b9d5c82d0b7bdedaf71b7088, parent/current main4a2826a1e13d12c4f6439362a82e315bfd1a37f9, treeaad4fe9b35452f283a850d623ce61a3a56c5fb11. Public Git data confirms the reviewed tree matches the imported worker candidate. The successful broker response exposes the ref/SHA, not a reviewer session ID or command transcript; none is invented here.This proves prepublication approval only. Required hosted checks, review triage, and the separate exact-published-head/current-main verifier App check are still pending. The 27-minute start window for a second review has passed; verification is scheduled for the next existing owner cycle at 2026-10-11 00:17 UTC. Both carried findings remain subject to that exact-head verification before merge. The stale #138 is retained until the successor is merge-eligible, without force-updating its history.