Fix duplicate messages in ACP client connection #466
Merged
Conversation
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
The issue was a race condition between streaming chunks and the final message. When agent.send() completed, it would immediately send a final sessionUpdate with the complete response. However, streaming chunk tasks created via asyncio.create_task() were not awaited, so they could still be in flight, causing the final message to arrive before or interleaved with the last streaming chunk. Solution: 1. Track all streaming tasks in a list as they're created 2. After agent.send() completes, await all streaming tasks with asyncio.gather() 3. Only send final message if streaming wasn't active (avoiding duplicates) This ensures the final message is never sent before streaming completes, and eliminates duplicate messages since the last streaming chunk already contains the complete response. Fixes duplicate message issue in ACP client connections.
When streaming is enabled, the final chunk already contains the complete response text, so sending an additional final message causes duplicate messages to be received by the client. This fix introduces a `streaming_enabled` flag that tracks whether streaming was set up for a prompt. When streaming is enabled, we skip sending the redundant final message, preventing duplicates. This approach is simpler than tracking and awaiting streaming tasks, and eliminates the race condition between the final streaming chunk and the final message. Fixes issue where ACP clients received messages twice.
The issue was that streaming chunk tasks were created but not awaited before returning PromptResponse(stopReason=END_TURN). This caused chunks to potentially arrive AFTER the turn was marked complete, leading to duplicate/out-of-order messages. Solution: 1. Track all streaming tasks in a list as they're created 2. After agent.send() completes, await all tasks with asyncio.gather() 3. Skip final sessionUpdate if streaming was enabled (last chunk has complete response) 4. Then return PromptResponse with END_TURN This ensures proper message ordering: All streaming chunks → (optional final message) → END_TURN Fixes duplicate ACP messages issue.
The previous fix incorrectly checked streaming_enabled (whether a listener was registered) instead of checking if chunks were actually sent. This caused no notifications to be sent when streaming was enabled but no chunks arrived (e.g., with passthrough model). Now we check if streaming_tasks is empty: - If chunks were streamed: skip final update (last chunk has everything) - If no chunks were streamed: send final update (even if listener registered) This fixes test failures where client received no sessionUpdate notifications.
evalstate
deleted the
claude/fix-duplicate-acp-messages-011CUy6TpaSgEh8qeVVJJ2SK
branch
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
No description provided.