fix(responses): honor tool_choice when parsing tool calls from text

[xianlubird]

Summary

Non-streaming /v1/responses can turn literal <tool_call> examples into function calls even when the request disables tool calling. Gate the text fallback on the presence of an actual function definition and tool_choice not being "none".

Overview

Requests with no callable functions, including empty namespaces, preserve assistant text verbatim. Requests with enabled functions keep the existing text fallback.

Details

Check both top-level functions and namespace members. Cover disabled functions, absent tools, and empty namespaces, as well as an enabled auto choice and namespace restoration. The streaming implementation is unchanged.

Where should the reviewer start?

lib/llm/src/protocols/openai/responses/mod.rs: the fallback gate in chat_completion_to_response and the adjacent text-tool-call regression tests.

Validation

• Confirmed the original disabled-tool regression fails before the fix.
• cargo test -p dynamo-llm --no-default-features --lib protocols::openai::responses
• Rustfmt and applicable pre-commit checks.

Related Issues

• Confirmed — no related issue

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[dynamo-ops]

/ok to test 4e37196

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 6a0aa318-9331-41d5-8182-9e4498d4db05

📥 Commits

Reviewing files that changed from the base of the PR and between d481e51 and 4e37196.

📒 Files selected for processing (2)

• lib/llm/src/protocols/openai/responses/mod.rs
• lib/llm/src/protocols/openai/responses/stream_converter.rs

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

---

Walkthrough

The OpenAI response converters now preserve textual <tool_call> markup when tools are absent or disabled. Tests verify parsing for enabled choices and exact streamed text preservation when tool selection is disabled.

Changes

Tool-call handling

Layer / File(s)	Summary
Conditional tool-call parsing lib/llm/src/protocols/openai/responses/mod.rs	chat_completion_to_response preserves literal markup when tools are absent, empty, or set to none. It parses markup for enabled tool-choice modes.
Conversion regression coverage lib/llm/src/protocols/openai/responses/mod.rs, lib/llm/src/protocols/openai/responses/stream_converter.rs	Tests cover disabled parsing, enabled choices, absence of function-call argument events, and exact streamed text reconstruction.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~15 minutes

Merge Risk: ⚪ Minimal · up to 4e371

The conversion behavior and its regression coverage align with the intended tool-disabled and tool-enabled handling, with no actionable merge risk identified.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 8 functions across 2 files.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly and concisely describes the main change: honoring tool_choice when parsing text-based tool calls in responses.
Description check	✅ Passed	The description explains the issue, fix, affected behavior, reviewer starting point, validation steps, and required related-issues status. It provides the information required by the template.

---

_{Comment @coderabbitai help to get the list of available commands.}

[rmccorm4]

LGTM other than this comment: #14843 (comment)

[dynamo-review-agent]

Previously reported defects still present:

• Original discussion: An empty namespace leaves the request's outer tools array nonempty, so the new gate parses literal <tool_call> text even though request conversion flattens that namespace to zero callable tools. Preserve the text unless at least one top-level or namespace function is present.
• Original discussion: The raised empty-namespace defect is still present: has_enabled_tools only checks that the outer tools array is nonempty, so an empty namespace with tool_choice: auto would still enable textual tool-call parsing even though request conversion flattens it to zero callable function tools.
• Original discussion: Verified: a nonempty tools array containing an empty namespace passes the new check and enables text fallback parsing, while request conversion flattens it to no backend tools. Literal <tool_call> text is therefore still converted into function-call output for this supported request shape.
• Original discussion: Verified: a nonempty tools array containing only an empty namespace passes has_enabled_tools, while request conversion flattens it to no backend tools. Literal <tool_call> text is therefore still converted into function calls for this disabled-tool case.

[dynamo-ops]

/ok to test ae63b5d

[dynamo-ops]

/ok to test eaa6350

[dynamo-ops]

/ok to test f86db4d

[dynamo-ops]

/ok to test c72643c

[dynamo-ops]

/ok to test 09b4bdd