[WRONG BRANCH] release: 2.71.0 - #6228
Merged
Merged
Conversation
Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
* docs(devlog): plan 2.70.0 release * docs(devlog): fold release plan audit * docs(devlog): record Sonnet 5.5 rollout and 2.70.0 release outcome
…very Claude family (#6217) * docs(devlog): plan Claude contract hardening and PR carries * docs(devlog): fold hardening plan audit * fix(anthropic): apply the live sampling and tool-choice contract to every Claude family A live probe of api.anthropic.com on 2026-09-29 showed the Sonnet 5.5 rules from #6210 were narrower than the upstream contract: - Opus 4.7+, Sonnet 5+ and every Fable return 400 on any non-default temperature/top_p, with or without thinking. The adapter only dropped them when it sent thinking, so a no-reasoning request with temperature failed. - Haiku 4.5, Sonnet 4.5/4.6 and Opus 4.5/4.6 take either field alone but 400 when both are sent; top_p is now dropped when both are present. - Fable 5.1 rejects forced tool_choice like Opus 5.5; it now degrades to auto. - Opus 5.5 and Fable reject both thinking disabled and between_tools, so the web-search and vision sidecars send a low effort and no thinking field for them (verified to end_turn with text inside a 1,024-token budget). Plan, audit and evidence: devlog/_plan/260929_dev_next_hardening_carry/.
* fix(link): reject forged ss owner tuples before trusting a PID parseListenEntriesFromSs attributed a listener with /pid=(\d+)/ over the whole row. ss prints comm unescaped between quotes, so a crafted 15-byte task name containing an embedded quote plus pid= text lands a forged pid field ahead of the real tuple — a local process could make reclaim/join logic blame (or kill) an innocent PID. Replace the row-wide regex with a strict users:(...) tuple parser: quoted name taken verbatim, then pid= plus only the field keys ss is known to emit (fd, ino, sk, v6only) inside the tuple boundary. Any grammar deviation — a second quoted segment, an unknown key, trailing garbage — drops the row's attribution entirely rather than trusting a partial parse. Shared sockets still report every owner tuple. * fix(link): require fd= on each accepted ss owner tuple --------- Co-authored-by: luvs01 <27862058+luvs01@users.noreply.github.com>
…6195) * fix(codex): preserve drift-heal ownership veto before config writes healCodexConfigDrift re-injected missing routing roots whenever the journal said they had drifted, without checking service-home ownership — the same admission unattended sync requires. A heal tick inside a no-longer-owned home could write config keys over a client that now belongs to someone else. Gate the heal on admitCodexWrite: a service-home refusal returns not-healed before the injector is even called, and beforeClientWrite re-runs the admission so ownership lost mid-write still throws. Other refusal authorities (config/generation/external-provider) do not block — they are not this heal's verdict. * fix(codex): veto drift heal on unowned homes at both write boundaries --------- Co-authored-by: luvs01 <27862058+luvs01@users.noreply.github.com>
An untrusted transcript can hold tens of thousands of unmatched <conversation> openings. The greedy /<conversation>[\s\S]*<\/conversation>/i scan re-walks the tail for every opening position, giving quadratic regex backtracking and event-loop blocking inside request handling. Replace it with two fixed-tag scans: an anchored case-insensitive /<conversation>/i exec, then a /<\/conversation>/gi search resumed from after that opening. Both run in place over the transcript — no lowercased copy, whose second body-sized string would roughly double peak memory for requests that can reach hundreds of MiB. Regression tests pin the behavior with deterministic probes instead of a wall-clock bound: every transcript scan must be one of the two fixed tag patterns (a greedy [\s\S]* pass fails that assertion even though it only scans once), and no transcript-sized normalized copy may be allocated. Co-authored-by: luvs01 <27862058+luvs01@users.noreply.github.com>
…uota design Review follow-up for #6206: two citations named files without their directory, and the admission contract did not say what happens on a 3xx response (CodeRabbit thread PRRT_kwDOS-0Gi86mz4CC).
Review follow-up for #6209: the reference said a saturated host no longer delays the probe past its ceilings, but the PR's own measurement shows p90 3.0s at full saturation. State what the boost does and does not guarantee.
Review follow-up for #6198: the comment carried two literal question marks where a dash was meant.
… budget The case registers 32 profiles through the real transaction path; each registration performs several fsync'd atomic writes (and ACL hardening on Windows), and those writes are what the test asserts. It normally takes 0.45s on windows-latest, drifted to 2.7-6.9s on dev dispatch runs, and hit 35.0s against its 30s budget in Cross-platform CI run 36499924172 (windows 8/9), turning the dev tip red with no code change. BULK_DURABLE_IO_BUDGET_MS (180s on Windows, 90s elsewhere) is the budget tests/helpers/test-budget.ts defines for this kind of work; no assertion depends on it, so a regression still fails.
chore(release): integrate six reviewed PRs for 2.71.0
Contributor
|
✅ Deterministic PR hygiene checks passed. |
Contributor
|
Navigate logical layers of code changes, visualize relationships, and explore their blast radius. Note Currently processing new changes in this PR. This may take a few minutes, please wait... ⚙️ Run configurationConfiguration used: Repository: lidge-jun/opencodex/.coderabbit.yaml Review profile: ASSERTIVE Plan: Advanced Run ID: ⛔ Files ignored due to path filters (1)
📒 Files selected for processing (100)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
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 |
Contributor
⏳ DRAFT
What to do
Its title has been prefixed with |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Promote
devatd161c0e83etomainas 2.71.0. Candidatedevatd161c0e83eis #6224, which landed #6206, #6201, #6209, #6094, #5905 and #6198, plus the eight commits merged since 2.70.0 (#6217 Claude sampling/tool-choice contract, #6193, #6194, #6195, #6218 and housekeeping). The tree equals the candidate; its version sources already read 2.71.0. The-s oursmerge keepsmainhistory linear with the promotion.Plan:
devlog/_plan/260929_release_2_71_0/041_wp4_execution.md.GUI changes in this promotion (from #6094 and #5905):
Verification
git diff --quiet d161c0e83e HEAD: tree equal.bun scripts/release-version-sources.ts check 2.71.0: 4 sources carry it.dcfbb2f708: Cross-platform CI pull_request 36525839698 and workflow_dispatch lane=all 36526719414 (windows 1-9) success; Service lifecycle 36525839767 success. The release dispatch waits for push-event Cross-platform CI and Service lifecycle on this merge SHA.Checklist