mirror of
https://github.com/alibaba/higress.git
synced 2026-03-21 03:27:37 +08:00
Instead of promoting reasoning to content inline per-chunk (which would emit reasoning as content prematurely if real content arrives later), the streaming path now buffers reasoning content and strips it from chunks. On the last chunk, if no content was ever seen, the buffered reasoning is flushed as a single content chunk. Also moves tests into test/openai.go TestOpenAI suite and adds MockHttpContext for provider-level streaming tests. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>