fix: preserve chat stream moderation results - #3864
Conversation
Carry moderation reports into accumulated completions while preserving the last report across omitted or null values. Add sync and async coverage for reports, errors, and unmoderated streams.
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. |
Castiron custom code✅ No new custom-code files detected. 45 mixed files remain; 0 existing customizations changed. Compared 45 existing customizations unchanged
5 more in the full report. A changed generated baseline means this report cannot reliably identify which handwritten lines changed. Inspect the custom-code diffDownload the exact patch produced by this run (requires repository access): gh run download 35016545074 --repo openai/openai-python \
--name castiron-custom-code-35016545074-1 --dir /tmp/castiron-custom-code-35016545074-1
git apply --stat /tmp/castiron-custom-code-35016545074-1/custom-code.patch
cat /tmp/castiron-custom-code-35016545074-1/custom-code.patchOr reproduce it from an SDK checkout containing the vendored reporter: git fetch --no-tags origin 9640f2903d65c43f0e5aee58d60f46880ff0bd15 f9de450750b86f0b21fb032508e27410e77af789
python3 scripts/castiron/custom_code_report.py report \
--base 9640f2903d65c43f0e5aee58d60f46880ff0bd15 \
--head f9de450750b86f0b21fb032508e27410e77af789 --fetch --require-head-hash --public \
--out /tmp/castiron-custom-code-f9de450750b8
cat /tmp/castiron-custom-code-f9de450750b8/custom-code.patchThis is the current full custom patch for mixed files, not an attribution of only the handwritten lines changed by this PR. |
jbeckwith-oai
left a comment
jbeckwith-oai
left a comment
There was a problem hiding this comment.
Reviewed the accumulator and the sync/async regression matrix. Reconstructing the chunk-side moderation value as the completion-side generated model is the right type boundary; later non-null reports replace the snapshot, omitted/null moderation preserves the last report, and the initial-chunk conversion already carries an initial report. Coverage exercises success, both input/output errors, replacement, late delivery after finish, unmoderated streams, raw chunks, snapshots, final completions, and Pydantic v1. No blocking issues found.
markstuart-oai
left a comment
markstuart-oai
left a comment
There was a problem hiding this comment.
Reviewed 65ba4ab1825d29377f141d226edb36990c3f7bb1. No findings. The accumulator converts chunk-side moderation into the completion-side model, retains it through omitted/null chunks, and replaces it only when a new report arrives. Initial, late, successful and error reports remain consistent across sync/async snapshots and final completions. The conversion also isolates the snapshot from later mutation of the raw chunk while preserving extra fields. The change stays in the existing accumulation path without adding another abstraction.
Validation: 49 existing streaming/moderation tests passed, plus eight independent public-stream cases, two blocking-moderation controls and a snapshot-isolation probe. Exact-base controls demonstrate the late/replacement fix. Current CI passed Python 3.10/3.14, HTTPX2, lint and build checks. Local checks used Pydantic v2; a local v1 environment was unavailable. No live API calls were made.
Automated Release PR --- ## [3.15.0](openai/openai-python@v3.14.1...v3.15.0) (2026-09-18) ### Features * **api:** add agent session model settings ([openai#3882](openai#3882)) ([4b15817](openai@4b15817)) * **api:** add audio-mini model choices ([openai#3886](openai#3886)) ([a6eeb3f](openai@a6eeb3f)) * **api:** add compaction progress events ([openai#3866](openai#3866)) ([98e1d24](openai@98e1d24)) * **api:** add managed Responses WebSocket sessions ([openai#3887](openai#3887)) ([3b865af](openai@3b865af)) * **api:** add prompt-cache prewarming ([openai#3888](openai#3888)) ([8f7cdb8](openai@8f7cdb8)) ### Bug Fixes * preserve chat stream moderation results ([openai#3864](openai#3864)) ([b77076d](openai@b77076d)) ### Chores * **api:** clarify incoming SIP call ID usage ([openai#3885](openai#3885)) ([f0457ab](openai@f0457ab)) * **api:** update image request examples ([openai#3889](openai#3889)) ([bb79676](openai@bb79676)) ### Documentation * **realtime:** remove nonexistent type imports ([openai#3869](openai#3869)) ([2971aaf](openai@2971aaf)) --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please). Co-authored-by: openai-sdks[bot] <284451331+openai-sdks[bot]@users.noreply.github.com>
3 participants
Preserves moderation results in chat stream snapshots and final completions, including reports delivered after content. Later omitted or null fields retain the last report; unmoderated streams remain unchanged.
Validation: 324 tests passed (14 skipped), plus 19 focused tests with Pydantic v1. Ruff, mypy, changed-file formatting, and wheel/sdist builds pass.