Related Issues

• fixes google_genai: streamed parallel function calls are dropped (tool call index restarts in every chunk) #4121

Proposed Changes:

Gemini sends each parallel function call in its own stream chunk. _convert_google_chunk_to_streaming_chunk numbered tool calls per chunk, so every call got ToolCallDelta(index=0), and _convert_streaming_chunks_to_chat_message merged them into one call with concatenated names and arguments. That call failed to parse and was dropped, so a streamed reply with parallel calls had no tool calls at all.

• _convert_google_chunk_to_streaming_chunk takes a keyword-only tool_call_index_offset (default 0), and tool call indices start from it.
• The sync and async stream handlers count tool calls across the stream and pass that count, next to the existing tool_calls_seen.

Calls that share a chunk are numbered as before.

How did you test it?

• New tests in tests/test_chat_generator.py: test_run_streaming_parallel_tool_calls_in_separate_chunks and test_run_async_streaming_parallel_tool_calls_in_separate_chunks. They use hand-built chunks shaped like a Gemini 3 stream (text, call, call, empty text with STOP). Both fail on main (no tool calls) and pass with the change.
• tests/test_chat_generator.py + tests/test_chat_generator_utils.py (-m "not integration"): 85 passed.
• ruff check / ruff format --check clean.

Notes for the reviewer

Function call ids exist on Gemini 3 but not on older models, so the fix keys on a running index rather than on the id.

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.

This PR was made with Claude Code (Anthropic); the unit tests above were run on main and on this branch. No live Gemini calls were made for this change.

🤖 Generated with Claude Code