[feiiiiii5]

Link to Issue or Description of Change

Problem: The CLI/Web SQLite session factory extracts only the URI path. A ?mode=ro database is opened writable, and an encoded filename can open a different database.

Solution: Pass the original URI to SqliteSessionService, whose existing parser already handles path decoding and SQLite query options. Keep the no-path in-memory case.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally. See the matrix notes below.

The two new tests use real SQLite files through the registry. Both fail on main (read-only writes succeed; the encoded path misses the saved session). Registry, service-factory and session tests: 866 passed, 1 xfailed. Pre-commit and targeted mypy pass.

Manual End-to-End (E2E) Tests:

Used create_session_service_from_options with temporary databases: the encoded URI reads the saved count=1; the plain URI with ?mode=ro reads it and rejects a write with OperationalError: attempt to write a readonly database.

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. See below.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

Additional context

Full tox matrix

Python 3.11: 23,598 passed; one dependency-version self-check failed because my initial local lock selected OTel 1.45.1. After updating the ignored local lock to the required 1.45.2, that check and the related suites passed (867 passed, 1 xfailed).

Python 3.12: 23,589 passed; one untouched telemetry timing test measured 0.161867s against a 0.10±0.05s bound. The same node then passed independently on both main and this branch with the same dependencies; its test and implementation files are identical between them.

Python 3.13: 23,590 passed, 116 skipped, 12 xfailed, 2 xpassed; tox passed.

Python 3.14: 23,592 passed, 114 skipped, 12 xfailed, 2 xpassed; tox passed.

No tracked dependency or lock files changed.

[its-amann]

passing the raw URI through still breaks once the encoded path and an option show up together: sqlite:///sessions%20%231.db?mode=ro turns into file:sessions #1.db?mode=ro in _parse_db_path, sqlite reads the # as the end of the path, so it quietly creates and opens a new writable db called sessions (the session lookup returns None and mode=ro is gone). re-encoding the path there fixes it, and _sqlite_memory_service._parse_db_path needs the same line:

--- a/src/google/adk/sessions/sqlite_session_service.py
+++ b/src/google/adk/sessions/sqlite_session_service.py
@@ -24,6 +24,7 @@ import os
 import sqlite3
 from typing import Any
 from typing import cast
+from urllib.parse import quote
 from urllib.parse import unquote
 from urllib.parse import urlparse

@@ -177,7 +178,10 @@ def _parse_db_path(db_path: str) -> tuple[str, str, bool]:

   if parsed.query:
     # sqlite3 only treats the filename as a URI when it starts with `file:`.
-    return normalized_path, f"file:{normalized_path}?{parsed.query}", True
+    # Re-encode the decoded path, or a `#` or `?` in the filename would end
+    # the path early and drop the query options.
+    connect_uri = f"file:{quote(normalized_path)}?{parsed.query}"
+    return normalized_path, connect_uri, True

   return normalized_path, normalized_path, False

[feiiiiii5]

Done in bb24f03: both _parse_db_path implementations now build the connect URI as file:{quote(normalized_path)}?{query}, so #, ? and a literal % in the decoded filename no longer truncate the path or drop mode=ro. The regressions are extended across reserved characters (#, ?, %23, +&= ü), relative and absolute paths, and no-query / ro / rw modes.

Verified on the branch: the two affected suites (tests/unittests/cli/test_service_registry.py, tests/unittests/memory/test_sqlite_memory_service.py) are 36 failed / 164 passed with the parent source for those two files swapped in, and 200 passed at bb24f03 (Python 3.12). Not run: the full tox matrix and Windows.

[its-amann]

thanks for applying my fix in bb24f03! could you add a Co-authored-by: Aman Agnihotri <amanagnihotri902@gmail.com> trailer to that commit so it counts for me too

[feiiiiii5]

Added the Co-authored-by: Aman Agnihotri <amanagnihotri902@gmail.com> trailer to the tip commit and updated the branch head to a66dda73. The PR should pick up the new head and rerun checks.

[feiiiiii5]

Heads-up: after adding the trailer, cla/google now reports the co-author amanagnihotri902@gmail.com as not covered (the PR author is covered). If that is you, signing at https://cla.developers.google.com/ and then using the check's "New Contributors" rescan link should clear it. I kept the trailer as requested.