This repository was archived by the owner on Feb 18, 2026. It is now read-only.
Open
Conversation
This fixes two critical bugs that caused SSE streaming to fail:
## 1. Event Queue Bug (src/providers/streaming.rs)
**Problem**: SseStream was dropping events when multiple SSE events arrived
in the same TCP chunk. Only the first event was emitted, others were discarded.
**Impact**: finish_reason chunks were frequently dropped, causing clients to
hang or error with incomplete responses.
**Fix**: Added event_queue (VecDeque) to buffer all parsed events. Now:
- Parse all complete events from each chunk
- Queue them all
- Emit events one at a time from the queue
- Properly handles multiple events arriving together
## 2. Double SSE Wrapping (src/server/mod.rs)
**Problem**: Server was wrapping provider SSE events (which already have
"event:" and "data:" lines) with axum's Event::default().data(), causing
double wrapping:
Before (broken):
data: event: message_start
data: data: {"type":"message_start",...}
After (correct):
event: message_start
data: {"type":"message_start",...}
**Fix**: Pass through raw SSE bytes from provider without wrapping:
- Use Body::from_stream() instead of Sse::new()
- Set SSE headers manually
- Provider output goes directly to client
Files changed:
- src/providers/streaming.rs: +43/-8 lines
- src/server/mod.rs: +29/-12 lines
…aming Implements full bidirectional tool transformation for OpenAI-compatible providers: ## Non-streaming Tool Calls - Transform OpenAI message.tool_calls[] → Anthropic content[].tool_use - Parse tool arguments from JSON string to structured object - Handle mixed responses (text + tool calls) ## Streaming Tool Calls - Detect and transform tool_calls in OpenAI streaming chunks - Handle incremental tool calls (sent across multiple chunks) - Emit proper Anthropic event sequence: content_block_start → delta → stop - Use StreamTransformState to track state across chunks - Close text blocks before tool blocks ## Tool Message Ordering - Fix parallel tool calls: OpenAI requires assistant(tool_calls) → tool → tool → user - Reorder Anthropic format (single user message with tool_results + text) - Emit tool messages BEFORE user content to satisfy OpenAI ordering ## Stop Reason Mapping - Map OpenAI finish_reason → Anthropic stop_reason correctly: - stop → end_turn - length → max_tokens - tool_calls → tool_use - Applies to both streaming and non-streaming Files changed: - Cargo.toml: Add uuid dependency for streaming tool block IDs - src/providers/openai.rs: Complete tool transformation implementation
Problem: When providers close streams without sending finish_reason chunk, the stream ends without message_stop event, causing Claude Code to hang waiting for continuation. This manifests as: - Tool completes but no automatic continuation - Subagent finishes but requires manual "ok" to proceed Fix: Add stream finalization handler that checks if message_stop was sent. If stream ends without finish_reason: - Close any open content/tool blocks - Send message_delta with end_turn - Send message_stop to properly terminate stream This ensures Claude Code always receives proper stream termination and continues automatically after tools/subagents complete.
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
This PR fixes bugs in SSE streaming and implements full tool calling support for OpenAI-compatible providers.
SSE Streaming Fixes
Event Queue Bug (src/providers/streaming.rs)
Double SSE Wrapping (src/server/mod.rs)
Tool Calling Support
Request Transformation
Response Transformation