fix(mistral): aggregate streamed tool call deltas by index - #4594

Description

MistralModel.stream aggregated streamed tool-call deltas in a dict keyed by tool_call.id (strands-py/src/strands/models/mistral.py:538 before this change). Mistral sends a tool call's id and name only on the first delta; continuation deltas carry only the index and an arguments fragment, and the mistralai SDK fills the missing id with its default "null" (ToolCall.id: Optional[str] = "null" in mistralai 2.x). Every continuation fragment therefore landed in a spurious "null" bucket:

• the toolUse block carrying the real id was emitted with only the first delta's empty arguments, so the event loop parsed its input as {} (strands/event_loop/streaming.py:301-302) and the tool executed without its arguments
• a second block with name="" / toolUseId="null" carried the fragments and was dropped as an incomplete tool use (strands/event_loop/streaming.py:318)
• with parallel tool calls, fragments from different tools merged into that single "null" bucket

Keying the dict by tool_call.index keeps each tool's fragments under the delta that carries its id and name, matching the OpenAI (openai.py:780), LiteLLM (litellm.py:518), SageMaker (sagemaker.py:475), and Writer (writer.py:441) providers.

Related Issues

None — no existing issue found for this bug.

Documentation PR

Not needed — no public API change; internal aggregation fix only.

Type of Change

Bug fix

Testing

Two regression tests in tests/strands/models/test_mistral.py stream (1) a single fragmented tool call and (2) two parallel tool calls, with continuation deltas carrying id="null" as the mistralai SDK fills in. Both fail on the previous id-keyed aggregation and pass with this change.

Run from strands-py/ (project venv, Python 3.12):

• .venv/bin/python -m pytest tests/strands/models/test_mistral.py -k fragmented -q → 2 passed, 59 deselected (on unpatched main: 2 failed, 59 deselected)
• .venv/bin/python -m pytest tests/strands/models/test_mistral.py -q → 61 passed
• .venv/bin/python -m pytest tests/ -q → 6973 passed, 31 skipped
• .venv/bin/ruff format --check → 727 files already formatted
• .venv/bin/ruff check → All checks passed!
• .venv/bin/mypy ./src ./tests_typing → blocked locally by a pre-existing environment mismatch: the venv's installed smithy_* packages use Python 3.12-only syntax while the repo pins python_version = "3.10", and the failure is identical on unmodified main. With --python-version 3.12 it completes over 327 files with only a pre-existing unused-ignore in _exception_notes.py (an artifact of overriding the target version); this change adds no mypy errors.

hatch run prepare itself could not run in this environment (hatch env creation fails building a native dependency), so the constituent checks above were run directly.

• I ran hatch run prepare

Checklist

• I have read the CONTRIBUTING document
• I have reviewed and understand every line of code in this PR, including any generated by AI tools, and I can explain why it works
• My change is focused and reasonably small; I have split unrelated work into separate PRs
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.