Repository navigation
fix: backport 0.12.1 steer batch (#32396 with its ordering fix #32826) - #32886
Conversation
) A turn that ended on nothing_to_do kept every later text chunk once a person's steer was pending, because it might answer them. When the turn then ended without the backend's consumption echo, the steer was requeued and answered by its own turn, and the kept text (in the field, the backend's alternation placeholder) was saved as an assistant row above the real answer. Such text is now held instead of streamed. The consumption echo releases it through the normal text path; a turn that ends without the echo leaves it dropped and counted in the turn-end contract warning. Closes #27655 (cherry picked from commit 2e0c482)
After a quiet end, text that arrives while a person's steer is pending is held until the backend echoes the steer's consumption (codex does that only at the terminal). Tool rows after that text rendered live, so the released text was replayed after all of them and the saved transcript was out of order. The text held so far now goes out just before the next tool call, so the transcript keeps the model's order and every tool event still runs live. Text written before a tool call is the model's work, so it is kept even if the echo never comes; only text with no tool call after it stays held and is dropped then. Refs #32396 (cherry picked from commit 9777445)
UX Review (Opus 5.5) — ⏭️ skippedRevision |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: this is a clean backport of two fixes already merged on Notes
[DESIGN-REVIEWED] 1d34b7a Description read[DESCRIPTION-READ] e453bc63641de188ec873b54b73f8d42c2e33689af492181bc5e86b5d66ccdf0 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS This backports two fixes already merged on What this change shipsInventory (7 items) — 6 justifiedIntent: give 0.12.x users a chat transcript after a quiet end where a pending steer leaves no placeholder reply and keeps the model's order. FIX.
[FIRST-PRINCIPLES-REVIEWED] 1d34b7a Description read[DESCRIPTION-READ] d55c3e3e550281cdc3da5a583aec58dcfc086ae357e19c31b841f667a2ace17b That sha256 names the PR description this verdict read. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Problem / Motivation
Goal: Bring #32396 and its ordering fix #32826, both merged on
main, ontorelease/0.12.0together, one commit each, in a PR of their own.Why it matters
#32396 was dropped from batch 45 (#32653) because of the ordering bug #32826 fixes. They ship together so the release never carries one without the other.
Not a goal
main; nothing is changed only here.CHANGELOG.mdchange.What changed (motivation → approach → change)
Each main squash commit was cherry-picked with
-xontorelease/0.12.0@2e1fff086c, in merge order onmain. Both applied cleanly, with no prerequisite and no hand-resolved hunk; each commit's added and removed lines are identical to its main commit.2e0c482d4e2f840a594b9777445d5b1d34b7ae0fbuilder-security checked: each change is the one already merged on
main; no bind, listener, new route, auth mechanism or storage grant is added.Backwards compatibility
Compatible: each change is the one already merged on
main, and its compatibility note applies unchanged. #32826's known trade-off carries over: in the rare turn whose steer is never echoed, text written before a tool call is kept.Removes nothing: no setting, key or caller is removed.
Tests
Proof: not provable -- other: a backport of main commits whose proofs ran on their own PRs
release/*, so the related set was run here:test_nothing_to_do_directive.py(both PRs' tests),test_chat_runner_turn_plumbing.py,test_chat_runner_composition*.pyand thetest_*steer*.pyfiles,-n 4 --dist loadgroup. 1366 passed, 2 failed. The 2, intest_folder_steering_provider_dedup.py, fail the same way on a cleanrelease/0.12.0@2e1fff086con this host.git merge-treeagainst the open batch fix: backport 0.12.1 wide batch 3 (15 user-facing fixes) #32836, which also toucheschat_runner.pyandsession.md, reports no conflict.Manual verification
N/A - each fix was verified on its own main PR. The 0.12.1 RC soak is the end-to-end check.
Related Issues
Backport to the 0.12 line for 0.12.1 of #32396 and #32826. #32396 was held back from #32653 for this pairing.
Sibling sweep
Siblings: none -- each source PR carried its own sibling sweep; this PR only carries those fixes.
Pattern harvest
Not generalizable: this PR only carries fixes already on
main. The source PRs have their own harvest.Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.