test: add /v1/responses restart regression (issue #64 responses variant) - #122
Merged
Conversation
Add a /v1/responses restart test mirroring the existing openai and anthropic cases: a thinking-bearing session is compressed live, then a fresh host primes from the persisted view and the pre-LLM acp_status must show the block. This fails on acp-kernel 0.0.32 (blocks lost after restart) because the kernel keyed responses reasoning core ids on the host-generated item id, which the primeFold mirror cannot reproduce. Passes with acp-kernel 0.0.33 (ranxianglei/acp-kernel#107), which keys the id on the reasoning text like the anthropic/openai codecs. Bumps acp-kernel 0.0.32 -> 0.0.33.
This was referenced Aug 22, 2026
Open
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Problem
On
/v1/responses(e.g. SGLang, Qwen with thinking), compressed blocksdisappear after a restart:
acp_statusshowsBlocks: noneuntil thefirst live provider request refolds the session. The existing restart
tests only covered
openai-completionsandanthropic-messages— therewas no
openai-responsescase.Root cause
In
acp-kernel,responsesToCorekeyed the reasoning core id on thehost-generated
item.id(rs_<hash>), while the anthropic/openai codecskey it on the thinking text. The primeFold mirror
(
viewToResponsesCore) emits reasoning items without an id, so thekernel fell back to a content hash — a different core id than the
host's. On restart the span fingerprints mismatched, every in-stream
compress replay was rejected, and the blocks were lost.
Fix
The fix is in the kernel: ranxianglei/acp-kernel#107 (key the
responses reasoning core id on the reasoning text, consistent with
anthropic/openai). This PR adds the missing restart regression test and
bumps
acp-kernel0.0.32 -> 0.0.33.Tests
provider+responses: restart primeFold rebuilds blocks from thinking-bearing turns (issue #64 responses).acp-kernel0.0.32 (0 blocks after restart, expected 1).the post-restart wire payload).
Dependency
Requires ranxianglei/acp-kernel#107 to be merged and published as
0.0.33first — CI will fail to installacp-kernel@0.0.33until then.