Summary

Removes unreachable dead code in _streaming.py that was caused by a logical impossibility in the conditional checks.

Fixes #2796

Problem

The check sse.event == "error" was inside a block that requires sse.event.startswith("thread."):

if sse.event and sse.event.startswith("thread."):
    # ...
    if sse.event == "error":  # UNREACHABLE: "error" doesn't start with "thread."
        raise APIError(...)

Since "error" does not start with "thread.", this condition can never be true, making the entire error handling block dead code.

Solution

Restructured the code to:

1. Parse JSON data first (common operation for both code paths)
2. Check for errors in the data payload (unified error handling)
3. Then handle the thread.* event special case for the yield

This removes ~28 lines of dead code while preserving all intended functionality. The error handling now works correctly for all SSE events, not just theoretically for thread.* events.

Changes

• src/openai/_streaming.py: Fixed both Stream (sync) and AsyncStream (async) classes

Test plan

• Verified the logic is equivalent (error handling still checks data.get("error"))
• Verified thread.* events still get wrapped with {"data": data, "event": sse.event}
• Verified non-thread events still yield data directly

🤖 Generated with Claude Code