fix(flows): Drop orphaned FRs instead of poisoning sessions

[a2105z]

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Closes: contents request-processor (_rearrange_events_for_latest_function_response) raises ValueError on an orphaned function response, permanently poisoning the session — raise happens before before_model_callback, so no user hook can intercept it #6582

Problem:
If a session ever persisted a function response whose matching function call was absent from the branch-filtered history, _rearrange_events_for_latest_function_response raised ValueError while building llm_request.contents.

That raise happens in request preprocessing — before before_model_callback — so no user hook can heal it. Every later turn replays the same history and re-raises, permanently poisoning the session.

Solution:
When the trailing event carries function responses with no matching function call in history, drop those orphaned response event(s), log a warning, and continue rearranging / assembling contents from the surviving history. The model simply never sees a result it never asked for, and the session stays usable.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

Updated / new tests in tests/unittests/flows/llm_flows/test_contents_function.py:

• test_orphaned_function_response_is_dropped_not_raised — former raise expectation replaced with drop behavior
• test_orphaned_fr_after_valid_history_keeps_session_usable — issue contents request-processor (_rearrange_events_for_latest_function_response) raises ValueError on an orphaned function response, permanently poisoning the session — raise happens before before_model_callback, so no user hook can intercept it #6582 history shape (valid FC/FR + answer + trailing orphan)
• test_multiple_trailing_orphaned_frs_are_all_dropped — consecutive orphans
• test_runner_continues_after_orphaned_fr_in_session — full Runner.run_async path; model is called and orphan is absent from prompt

pytest results (local):

pytest tests/unittests/flows/llm_flows/test_contents_function.py -v
# 10 passed

pytest tests/unittests/flows/llm_flows/test_contents_function.py \
       tests/unittests/flows/llm_flows/test_contents.py \
       tests/unittests/apps/test_compaction_runner_e2e.py -q
# 45 passed

pytest tests/unittests/flows/llm_flows/ -q
# 485 passed

Manual / repro verification:

• Reproduced contents request-processor (_rearrange_events_for_latest_function_response) raises ValueError on an orphaned function response, permanently poisoning the session — raise happens before before_model_callback, so no user hook can intercept it #6582 against current main with the issue's contents-processor repro:
  • Before: ValueError: No function call event found for function responses ids: {'orphan-1'}
  • After: no raise; contents built from surviving history (hi → FC/FR → answer), orphan omitted

Pre-commit:

pre-commit run --files \
  src/google/adk/flows/llm_flows/contents.py \
  tests/unittests/flows/llm_flows/test_contents_function.py
# all hooks passed

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

Additional context

Files changed:

• src/google/adk/flows/llm_flows/contents.py — drop orphaned trailing FRs instead of raising
• tests/unittests/flows/llm_flows/test_contents_function.py — regression + runner coverage

Companion producer-side issues (e.g. progressive SSE leaving orphaned FRs) are separate; this change hardens the consumption path so any orphaned FR no longer kills the session.

Exact commands run

pytest tests/unittests/flows/llm_flows/test_contents_function.py -v
# 10 passed

pytest tests/unittests/flows/llm_flows/test_contents_function.py \
       tests/unittests/flows/llm_flows/test_contents.py \
       tests/unittests/apps/test_compaction_runner_e2e.py -q
# 45 passed

pytest tests/unittests/flows/llm_flows/ -q
# 485 passed

pre-commit run --files \
  src/google/adk/flows/llm_flows/contents.py \
  tests/unittests/flows/llm_flows/test_contents_function.py
# all hooks passed

[a2105z]

Hi @GWeale — hoping you might have a minute when free. You’ve touched contents.py a lot recently, so I wanted to check this with you.

This is a fix for #6582: an orphaned trailing function response used to raise during contents assembly (before any callback), which permanently poisoned the session. The change drops those unmatched FRs, logs a warning, and continues so later turns can still run.

I tried to keep the blast radius small and covered it with contents-processor + full Runner tests. Happy to adjust if you’d rather surface this differently — just say the word. Thanks so much for any feedback.

[a2105z]

Hi @DeanChensj — looping you in as well for a review when you have a moment.

This is the consumption-side companion to the mixed-turn session-poisoning work: orphaned trailing function responses were raising during contents assembly and leaving the session unusable. The change drops those unmatched FRs and continues.

Happy to tweak anything that looks off — thanks so much for taking a look.