fix: filter empty assistant messages in streaming requests

[bugkeep]

Summary

• Share assistant-message cleanup between non-streaming and streaming OpenAI-compatible requests.
• Filter assistant messages with empty content and no tool calls before streaming requests.
• Treat empty list content like other empty assistant content and normalize empty tool-call assistant content to null.

Fixes #7721

Testing

• uv run pytest tests/test_openai_source.py -k "query_stream_filters_empty_assistant_messages or query_filters_empty_list_assistant_message_without_tool_calls" -q
• uv run pytest tests/test_openai_source.py -q --deselect tests/test_openai_source.py::test_file_uri_to_path_preserves_windows_drive_letter --deselect tests/test_openai_source.py::test_file_uri_to_path_preserves_windows_netloc_drive_letter --deselect tests/test_openai_source.py::test_file_uri_to_path_preserves_remote_netloc_as_unc_path --deselect tests/test_openai_source.py::test_prepare_chat_payload_materializes_context_localhost_file_uri_image_urls
• uv run pytest tests/unit -q
• uv run ruff format .
• uv run ruff check .

Note: The full tests/test_openai_source.py -q run has the same four Windows file URI failures on current upstream astrbotdevs/master; they were excluded in the OpenAI adapter regression run above.

Summary by Sourcery

Normalize and filter empty assistant messages across OpenAI-compatible chat requests, including streaming, to avoid provider errors and ensure spec-compliant payloads.

Bug Fixes:

• Filter assistant messages with empty content and no tool calls before both standard and streaming chat completion requests to prevent strict API errors.
• Normalize assistant messages that have tool calls but empty content (including empty lists) to use null content in accordance with the OpenAI specification.

Tests:

• Add regression tests verifying that streaming queries filter empty assistant messages while preserving valid tool-call messages.
• Add tests ensuring empty-list assistant content without tool calls is removed from requests and that non-empty conversations are preserved.

[gemini-code-assist]

Code Review

This pull request refactors the assistant message cleaning logic in the OpenAI source provider into a dedicated method, _clean_assistant_messages_for_provider, and ensures it is applied to both standard and streaming queries. The logic filters out empty assistant messages without tool calls and normalizes empty content to None when tool calls are present. Additionally, new test cases were added to verify these behaviors. A review comment suggests optimizing the cleaning loop to improve readability and reduce redundant checks.

[gemini-code-assist]

The logic for cleaning assistant messages can be optimized by checking if the content is empty first. This avoids redundant calls to _is_empty_assistant_content and only fetches tool_calls when necessary, improving both readability and performance.

Suggested change

	for idx, msg in enumerate(messages):
	if msg.get("role") == "assistant":
	content = msg.get("content")
	tool_calls = msg.get("tool_calls")
	if not tool_calls and cls._is_empty_assistant_content(content):
	logger.warning(
	f"Filtered empty assistant message at index {idx} "
	"(without tool calls)"
	)
	continue
	if tool_calls and cls._is_empty_assistant_content(content):
	msg["content"] = None
	cleaned_messages.append(msg)
	for idx, msg in enumerate(messages):
	if msg.get("role") == "assistant":
	content = msg.get("content")
	if cls._is_empty_assistant_content(content):
	tool_calls = msg.get("tool_calls")
	if not tool_calls:
	logger.warning(
	f"Filtered empty assistant message at index {idx} "
	"(without tool calls)"
	)
	continue
	msg["content"] = None
	cleaned_messages.append(msg)

[sourcery-ai]

Hey - I've left some high level feedback:

• Consider downgrading the logger.warning in _clean_assistant_messages_for_provider to info or debug, as filtering empty assistant messages is now expected behavior and may otherwise produce noisy logs in normal operation.
• In _clean_assistant_messages_for_provider, you might want to document (or enforce) whether whitespace-only strings should be treated as empty content as well, to avoid provider-specific 400 errors if some backends treat them equivalently to "".

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- Consider downgrading the `logger.warning` in `_clean_assistant_messages_for_provider` to `info` or `debug`, as filtering empty assistant messages is now expected behavior and may otherwise produce noisy logs in normal operation.
- In `_clean_assistant_messages_for_provider`, you might want to document (or enforce) whether whitespace-only strings should be treated as empty content as well, to avoid provider-specific 400 errors if some backends treat them equivalently to `""`.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}