Release terminal trajectory capture owners - #1052
Merged
Merged
Conversation
bradhilton
marked this pull request as ready for review
September 30, 2026 01:27
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Completed HTTP responses can keep an entire captured trajectory alive. HTTPX's response/stream cycle retains ART's capture state, which still owns the trajectory, request and copied response body after capture finishes.
Release those private owners on finish or discard, including failure paths. The recorded exchange keeps its request and response; repeated finish/add calls stay inert. This applies to all adapters using
CaptureStateand changes no GC settings.Nine focused lifecycle tests pass, including the actual OpenAI/HTTPX path, streamed delivery, cancellation and error identity. All nine fail on the parent. Existing capture coverage: 78 passed; seven optional Brotli cases skipped locally. This proves the retention mechanism, not a measured reduction in live GC pauses.