fix: preserve nontransparent live resumption

[he-yufeng]

Summary

Fixes #5675.

BaseLlmFlow.run_live always set session_resumption.transparent = True when reconnecting with a saved live-session handle. That overwrote an explicit transparent=False config, which is the configuration needed for direct Gemini API live-session resumption because GoogleLLM.connect rejects transparent resumption on the Gemini API backend.

This keeps the existing default behavior for reconnects that have no session resumption config yet, but preserves a caller-provided non-transparent config.

Validation

python -m pytest tests/unittests/flows/llm_flows/test_base_llm_flow.py -q
# 29 passed

pyink --check src/google/adk/flows/llm_flows/base_llm_flow.py tests/unittests/flows/llm_flows/test_base_llm_flow.py
isort --check-only src/google/adk/flows/llm_flows/base_llm_flow.py tests/unittests/flows/llm_flows/test_base_llm_flow.py
python -m py_compile src/google/adk/flows/llm_flows/base_llm_flow.py tests/unittests/flows/llm_flows/test_base_llm_flow.py
git diff --check

[rohityan]

Hi @he-yufeng ,Thank you for your contribution! We appreciate you taking the time to submit this pull request. Please fix formatting errors by running autoformat.sh

[he-yufeng]

Rebased onto current upstream/main and re-ran the focused checks locally.

Validation:

PYTHONUTF8=1 .\.venv\Scripts\python.exe -m pytest tests\unittests\flows\llm_flows\test_base_llm_flow.py -q --basetemp .tmp\pytest-5676
# 29 passed, 1 warning

PYTHONUTF8=1 .\.venv\Scripts\python.exe -m pyink src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py --check
PYTHONUTF8=1 .\.venv\Scripts\python.exe -m py_compile src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py
git diff --check

The previous failure was the pre-commit formatting check; this branch is now a single commit on current main.

[he-yufeng]

Rebased this onto current main and reran the focused checks.

Validation:

PYTHONUTF8=1 .\.venv\Scripts\python.exe -m pytest tests\unittests\flows\llm_flows\test_base_llm_flow.py -q --basetemp .tmp\pytest-5676-rebase
# 31 passed, 1 warning

PYTHONUTF8=1 .\.venv\Scripts\python.exe -m pyink src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py --check
PYTHONUTF8=1 .\.venv\Scripts\python.exe -m py_compile src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py
git diff --check upstream/main..HEAD

[he-yufeng]

Rebased onto current main again; no code changes beyond the rebase.\n\nValidation on Windows:\n- uv run --no-sync pytest tests\unittests\flows\llm_flows\test_base_llm_flow.py -q --basetemp .tmp\pytest-5676-20260524 -p no:cacheprovider -> 31 passed, 1 existing experimental warning\n- uv run --no-sync pyink --check src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py\n- uv run --no-sync python -m py_compile src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py\n- git diff --check upstream/main..HEAD

[he-yufeng]

Rebased onto current main and force-pushed fe58e74. Local validation: py_compile passed; focused live reconnection regression test passed; pyink, isort, and git diff --check passed.

[rohityan]

Hi @he-yufeng , could you please resolve the branch conflicts.

[he-yufeng]

Rebased onto current main and resolved the live resumption conflict. I kept upstream's Vertex-only transparent=True behavior and preserved this branch's existing-session-resumption regression coverage. Validation on Windows: uv run --no-sync pytest tests\unittests\flows\llm_flows\test_base_llm_flow.py -q --basetemp .tmp\pytest-5676-20260530 -p no:cacheprovider -> 35 passed, 5 warnings; uv run --no-sync pyink src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py --check; uv run --no-sync python -m py_compile src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py; git diff --check upstream/main..HEAD.

[he-yufeng]

Rebased onto current upstream/main again and fixed the pyink formatting drift in the touched test file.

Validation on Windows:

PYTHONUTF8=1 uv run --no-sync pytest tests\unittests\flows\llm_flows\test_base_llm_flow.py -q --basetemp .tmp\pytest-5676-20260604 -p no:cacheprovider
# 38 passed, 5 warnings

PYTHONUTF8=1 uv run --no-sync pyink src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py --check
PYTHONUTF8=1 uv run --no-sync python -m py_compile src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py
git diff --check upstream/main..HEAD

[he-yufeng]

Rebased this branch onto current upstream/main and force-pushed the cleaned head.

Validation on Windows:

python -m py_compile src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py
PYTHONUTF8=1 PYTHONPATH=src C:\dev\GITHUB-clean\adk-python\.venv\Scripts\python.exe -m pytest tests\unittests\flows\llm_flows\test_base_llm_flow.py -q --basetemp .tmp\pytest-5676-20260607 -p no:cacheprovider
# 38 passed, 5 warnings
PYTHONUTF8=1 uv run --no-sync pyink src\google\adk\flows\llm_flows\base_llm_flow.py tests\unittests\flows\llm_flows\test_base_llm_flow.py --check
# 2 files would be left unchanged
git diff --check upstream/main..HEAD

[adk-bot]

I have completed a thorough, read-only analysis of Pull Request #5676 according to the established ADK triage principles.

The details have been compiled into a premium PR Analysis Report. Please review the artifact here:

• analysis_results.md

Key Highlights of the Triage

1. Google CLA Verification: Verified successfully. The contributor @he-yufeng has signed the Google CLA (cla/google status is SUCCESS).
2. Problem & Root Cause: Fixes Issue #5675, which previously blocked live-session resumption on the direct Gemini API backend. BaseLlmFlow.run_live unconditionally overrode retry connections with transparent=True, causing a crash on the direct Gemini API backend since transparent resumption is only supported by Vertex AI.
3. Elegant Implementation: Correctly scopes the transparent=True fallback only to the Vertex AI backend on line 544-555 of base_llm_flow.py.
4. Testing & Validation: High-quality regression tests were added in test_base_llm_flow.py verifying behavior parity between both backends.
5. Recommendation: Approve and Merge. No open architectural nits or style violations were found.