During streaming, I still see truncated sentences. The se...#35
Merged
lukemarsden merged 5 commits intomainfrom Apr 6, 2026
Merged
During streaming, I still see truncated sentences. The se...#35lukemarsden merged 5 commits intomainfrom
lukemarsden merged 5 commits intomainfrom
Conversation
…cated streaming When tool calls arrived during streaming, the preceding text appeared truncated because the NewEntry handler sent the tool_call entry without flushing the text entry's pending throttled content first. The stale-pending flush only existed in throttled_send_message_added, which NewEntry bypasses. Changes: - Extract flush_stale_pending_for_thread() helper from inline stale-pending flush - Call it in NewEntry handler before sending, ensuring text content is complete - Bypass 100ms throttle for tool_call entries (infrequent, need prompt delivery) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Spec-Ref: helix-specs@df2833ab1:001694_during-streaming-i-still
… stale pending The previous fix (flush_stale_pending_for_thread) was insufficient because the throttle buffer's pending_content is a snapshot from the last EntryUpdated event. When AcpThread::push_entry() calls flush_streaming_text() before emitting NewEntry, the final tokens get flushed into the Markdown entity but no EntryUpdated is emitted — so the pending snapshot is stale. Fix: in the NewEntry handler, re-read ALL preceding entries' current content directly from the thread model (which has the complete text after flush_streaming_text()) and send fresh message_added events for each. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Spec-Ref: helix-specs@393a7da43:001694_during-streaming-i-still
Upstream PR zed-industries#51499 added a StreamingTextBuffer that rate-limits how many bytes get revealed into the Markdown entity per tick, creating a smooth typewriter animation. However, content_only(cx) reads markdown.source() which only returns revealed bytes. This means EntryUpdated -> WebSocket sync sends incomplete content, causing the Go accumulator's baseline to drift behind reality. When patches are computed from this stale baseline, the frontend sees backwards edits that truncate text before tool calls. Fix: each 16ms timer tick now drains the entire pending buffer instead of limiting to bytes_to_reveal_per_tick. The timer still batches (avoids per-character markdown.append calls), but no bytes are withheld. This ensures content_only(cx) always returns all received content. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Spec-Ref: helix-specs@b244b04d4:001694_during-streaming-i-still
…eaks The NewEntry handler re-sent ALL preceding entries (from index 0) when a new entry arrived, but this included entries from previous turns. The Go server added these old message_ids to the current interaction's response_entries, causing isolation violations in the store. Apply the same turn-scoping logic already used by the Stopped handler: find the last UserMessage and only re-send entries after it. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Spec-Ref: helix-specs@b244b04d4:001694_during-streaming-i-still
…-streaming-i-still Spec-Ref: helix-specs@15bac144f:001694_during-streaming-i-still
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.
During streaming, I still see truncated sentences. The sentences are almost always truncated prior to a tool call, I think. It's like there's a 100 ms poll or whatever it is inside Zed that somehow doesn't update some sentences some of the time, although when the interaction completes, we do correctly get them all updated. Can you investigate this? I think there's multiple levels here:
🔗 Open in Helix
📋 Spec:
🚀 Built with Helix