fix(openai): report max_tokens when a Responses function call is cut off - #4139

Description

When the OpenAI Responses API returns response.incomplete with incomplete_details.reason == "max_output_tokens" after a function_call output item has started, OpenAIResponsesModel reported stopReason: "tool_use" instead of "max_tokens". The event loop then executed the truncated tool call (with input: {} because the partial JSON does not parse) instead of running recover_message_on_max_tokens_reached and raising MaxTokensReachedException.

Root cause: tool_calls is populated on response.output_item.added, i.e. before any arguments have streamed, and the finish-reason mapping gave tool_calls precedence over the length stop reason. So any function call in flight when the response was cut off won over length.

Changes:

• openai_responses.py: check stop_reason == "length" before tool_calls when computing finish_reason. This matches OpenAIModel (Chat Completions), which forwards finish_reason == "length" regardless of accumulated tool-call deltas.
• test_openai_responses.py: add a regression test for a function call cut off by max_output_tokens, asserting max_tokens is reported and tool_use is not.

Note: #3419 contains the same two-line reordering as part of a much larger change. This PR isolates the stop-reason fix so it can land on its own; the resulting code in that block is identical, so there is no conflict in outcome.

The TypeScript Responses adapter has the same precedence (responses-adapter.ts, stopReason = 'toolUse' after maxTokens was set). I left it out to keep this PR focused and can follow up separately.

Related Issues

Fixes #4135

Documentation PR

No documentation change needed.

Type of Change

Bug fix

Testing

• pytest tests/strands/models/test_openai_responses.py: 137 passed (136 existing + 1 new).

• New test fails on main without the source change and passes with it.

• pytest tests/strands/event_loop: 173 passed.

• ruff format --check, ruff check, mypy on the two changed files: clean.

• Full tests/ run: 5174 passed; the 2 failures in test_mcp_client.py::test_mcp_client_client_info_passed also fail on unmodified upstream/main in my environment, and the collection errors are optional-dependency modules (a2a, writer, cedar) not installed in my env.

• Repro from [BUG] OpenAIResponsesModel reports tool_use instead of max_tokens when a function call is cut off by max_output_tokens #4135 now prints messageStop: [{'messageStop': {'stopReason': 'max_tokens'}}] and the agent raises MaxTokensReachedException instead of invoking the tool with empty input.

• I ran hatch run prepare (hatch is not installed locally; I ran the equivalent ruff/mypy/pytest commands above)

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.