Restore lost streaming reveal-emit + add e2e Phase 15 cadence test - #55
Merged
Conversation
…content Spec-Ref: helix-specs@b6c43b8a9:001895_i-thought-we-flushed
The existing e2e suite only asserts on the final response after
message_completed, by which point flush_streaming_text has run and
content is fully drained. That left a regression class invisible:
streaming patches from external_websocket_sync that show up in stale
chunks (or only at the Stopped re-send) instead of arriving live.
Phase 15 sends a long prose prompt with no tool calls and records
(timestamp, content_length) for every assistant message_added on the
phase-15 thread between thread_created and message_completed. Three
asserts run at validation:
1. >= 5 assistant message_added events (catches "almost no streaming
events arrive — content shows up in a single end-of-turn burst").
2. Longest gap between consecutive message_added events <= 20s
(catches multi-minute pauses while content sits in the streaming
buffer with no EntryUpdated re-emit).
3. >= 30% of the final content length is observed by the midpoint of
the streaming duration (catches "all content arrives at the end").
Bumps test-server timeout to 360s/agent and the run_e2e.sh default
to 300s/agent to accommodate the extra ~30–60s for Phase 15. Also
bumps the validateStore expected counts from 4→5 sessions and 7→8
completed interactions per round to account for the new phase.
This phase fails reliably without the companion fix that re-adds
`cx.emit(AcpThreadEvent::EntryUpdated(entries_len - 1))` after the
streaming-reveal drain in acp_thread.rs.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Initial loose threshold (>=5 samples) was too permissive: in a controlled e2e the without-fix path still yields ~31 message_added events for a ~2.5 KB streaming response (one per LLM chunk), whereas the with-fix path yields ~58 events (chunk-arrival emissions PLUS streaming-reveal drain emissions). Both passed minSamples=5. Calibrated against zed-agent + claude-sonnet-4-5 (verified locally): with cherry-pick: 59 samples ✅ (passes minSamples=40) without cherry-pick: 31 samples ❌ FAIL: only 31 ... need >= 40 Bumping the threshold to 40 sits comfortably between the two with margin for LLM/throttle variance. If a future legitimate change to STREAMING_THROTTLE_INTERVAL or chunk batching shifts the baseline, the calibration comment in the code points at where to recheck. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Merged
2 tasks
lukemarsden
added a commit
to helixml/helix
that referenced
this pull request
May 11, 2026
Was pointing at the pre-merge branch tip (41a62b2e61). Now that helixml/zed#55 merged, update to the merge commit on main (cd4e279d80) so CI builds against helixml/zed/main rather than a deleted branch ref. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This was referenced Jun 12, 2026
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.
Summary
Cherry-picks
edbbb5a5e8("Emit EntryUpdated after streaming-reveal drain so WS sync sees fresh content") back onto main, and adds an e2e regression test that asserts streaming patches arrive incrementally so this fix doesn't get silently dropped again.Why this is needed
User report: streaming content appears in real time in Zed's UI but takes a long time to show up in Helix.
The fix originally landed via Helix PR helixml/helix#2296 on May 6 (Helix bumped
ZED_COMMITtoedbbb5a5e8). On May 8 the cancel_current_turn work merged and bumpedZED_COMMITtofe8f4f4e3f, which forked off a Zed branch that does NOT includeedbbb5a5e8— silently losing the fix. The May 11 upstream merge (#54 / helixml/helix#2403) didn't pick it back up either.Without the fix:
push_assistant_content_block_with_indentemitsEntryUpdated(idx)and stashes the text instreaming_text_buffer.pending.external_websocket_syncreadsmarkdown.source()viacontent_only(cx)— but the new text is still inpending, so it reads stale content.start_streaming_revealtask drainspendinginto the markdown entity but does NOT emitEntryUpdated.push_chunkfiresEntryUpdated. Between chunks the text sits invisible.In a controlled e2e against zed-agent + claude-sonnet-4-5 streaming a ~2.8 KB prose response, this halves the number of
message_addedevents Helix observes (≈58 with the fix, ≈31 without).Changes
crates/acp_thread/src/acp_thread.rs(1 file, 12 inserts) — cherry-pick ofedbbb5a5e8. Addscx.emit(AcpThreadEvent::EntryUpdated(entries_len - 1))after the drain instart_streaming_reveal.crates/external_websocket_sync/e2e-test/helix-ws-test-server/main.go+run_e2e.sh— new Phase 15 ("Streaming patches arrive incrementally"). Sends a 400-word plain-prose prompt, records(timestamp, content_length)for every assistantmessage_addedon the phase-15 thread, and asserts:>= 40message_added events arrive (calibrated ~58 with fix vs ~31 without — comfortably between with margin for LLM/throttle variance).<= 20s.>= 30%of the final content is observed by the streaming midpoint.Bumps the test-server timeout to 360s/agent and the
run_e2e.shdefault to 300s/agent to accommodate the extra ~30s for Phase 15.Verification
Verified locally: rebuilt Zed with and without the cherry-pick and ran
E2E_AGENTS=zed-agent ./run_docker_e2e.shagainst each.Companion Helix PR
helixml/helix# bumps
ZED_COMMITto this branch's head.Merge order (per
helixml/helixCLAUDE.md)Merge this Zed PR first, then the Helix PR.
🤖 Generated with Claude Code