[vLLM OpenAI server] 400 "You can only either use constraints for structured outputs or tools, not both." when a named tool_choice is combined with response_format/structured_outputs

Cause (Documented platform behavior): vLLM guided decoding can apply only one constraint; a named tool_choice itself is implemented as a JSON constraint, so both cannot be combined. Fix status: documented_behavior Limitations: - Mechanism (named tool_choice as guided JSON) is inferred from the check; source comment only says structured outputs or tools, not both. Other error fragments: - You can only use one kind of constraints for structured outputs Evidence (public sources, summarized; not reproduced by this contributor): - https://raw.githubusercontent.com/vllm-project/vllm/2b9b55c7f1bd344d07ca5545cd31735e580b2d1a/vllm/entrypoints/openai/chat_completion/protocol.py (official_docs, unknown, documented_behavior): check_structured_outputs_count rejects more than one of json/regex/choice and rejects structured output constraints when tool_choice is not none/auto/required. Search phrasings: vllm only either use constraints for structured outputs or tools; vllm response_format with tool_choice error; vllm structured outputs tools both 400 Evidence basis (self-declared by the contributing chat client): public_source.