chore(llm): Backend Fallback Logic Tests

[justin-tahara]

Description

Introducing new tests to make sure we are covering fallback extraction paths and prevent regressions in tool-call reconstruction.

How Has This Been Tested?

Ran the tests locally

Additional Options

• [Required] I have considered whether this PR needs to be cherry-picked to the latest beta branch.
• [Optional] Override Linear Check

---

Summary by cubic

Add unit and integration tests for fallback tool-call extraction when tool choice is REQUIRED, including deep_research mode and cases where tool-call JSON is embedded in assistant text. Update chat test helpers to pass deep_research in send_message and disconnect, and cover extraction from answer/reasoning, no-op when already attempted or tool_calls exist, and not-extractable cases.

^{Written for commit b1eae31. Summary will update on new commits.}

[greptile-apps]

Greptile Overview

Greptile Summary

This PR adds comprehensive test coverage for the backend fallback logic that extracts tool calls from LLM response text when the model doesn't natively support structured tool calling. The changes introduce both unit tests and integration tests to validate the fallback extraction mechanism.

Key Changes:

• Added 5 new unit tests in test_llm_loop.py covering edge cases for _try_fallback_tool_extraction function (already attempted, answer extraction, reasoning fallback, no extraction possible, existing tool calls)
• Added 1 integration test validating the end-to-end fallback extraction flow in deep research mode
• Extended test utilities with deep_research parameter to support the new test scenario

The tests are well-structured, have clear assertions, and provide good coverage of the fallback extraction logic to prevent regressions.

Confidence Score: 5/5

• This PR is safe to merge with no risk
• This is a test-only PR that adds comprehensive coverage for existing fallback logic. No production code changes. All tests follow established patterns, have proper assertions, and cover edge cases thoroughly. No custom instruction violations detected.
• No files require special attention

Important Files Changed

Filename	Overview
backend/tests/integration/common_utils/managers/chat.py	Added deep_research parameter to test utilities for supporting reasoning fallback tests
backend/tests/integration/tests/llm_workflows/test_mock_llm_tool_calls.py	Added integration test validating fallback tool extraction from reasoning text in deep research mode
backend/tests/unit/onyx/chat/test_llm_loop.py	Added comprehensive unit tests for _try_fallback_tool_extraction covering all edge cases

[cubic-dev-ai]

No issues found across 3 files

[github-actions]

Preview Deployment

Status	Preview	Commit	Updated
✅	https://onyx-preview-4jwe806hf-danswer.vercel.app	1826b66	2026-02-12 00:12:55 UTC