[mangod12]

Summary

Concurrent calls to execute() on a shared SQLite connection can obtain the same cached sqlite3_stmt before SQLite marks it busy. The cursor releases the GIL around SQLite operations, leaving a race where concurrent cursors can reset or use the same statement and crash the process.

Reserve cached statements before releasing the GIL. If a cached statement is already reserved, create an uncached statement for the concurrent execution. Release the reservation when the statement is reset.

Add a regression test that runs concurrent parameterized queries on a shared connection and verifies the returned BLOB contents.

Fixes gh-160699.

Verification

• Windows debug build succeeded.
• PCbuild/amd64/python_d.exe -m unittest test.test_sqlite3.test_dbapi: 211 passed, 2 skipped.
• Tools/patchcheck/patchcheck.py: passed; NEWS fragment recognized.

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[mangod12]

CI for PR #160706 fails on all three macOS jobs in the new test_concurrent_queries_on_shared_connection: it segfaults inside sqlite3_prepare_v2 while threads share the connection. I pushed cf13208 to explicitly request SQLITE_OPEN_FULLMUTEX whenever the linked SQLite build has mutex support, while keeping the statement reservation fix. SQLite documents that flag as selecting serialized mode and notes sqlite3_threadsafe() alone does not report runtime mode: https://sqlite.org/threadsafe.html

The new CI run will tell us whether that addresses the macOS failure. I could not build CPython locally here because the Windows C toolchain is unavailable.

[mangod12]

@erlend-aasland The latest CI run for 76eb2d3 failed on macOS Intel with a hosted-runner communication error; GitHub’s annotation reports that the runner lost communication with the server, and the detailed job log is unavailable. Could you rerun that job when convenient? The other macOS jobs passed.

[mangod12]

@erlend-aasland A separate blocker surfaced in the Read the Docs preview for build 35421294 (commit 76eb2d3): it fetched refs/pull/160706/head, then failed before the docs build with fatal: reference is not a tree when checking out that commit. Could you check the preview’s commit/ref resolution or trigger a fresh preview? The failure appears to happen before Sphinx runs.

[bedevere-app]

Thank you for your interest in contributing to Python. This is a community-driven project, and we want it to be a welcoming place to learn and collaborate.

Contributors are responsible for reviewing and understanding what they submit, checking its claims, and being able to explain changes in their own words. This helps reviewers focus on the contribution and leaves more time to support other contributors.

This contribution appears to have been authored with AI assistance. Please explain your contribution in your own words and describe how you checked its correctness.

Please read our guidelines for using AI tools.

[mangod12]

Yes it was done with help of ai. And thanks for the reply.

The issue as i understood was that concurrent execution calls on a shared connection can pick the same cached SQLite statement before sqlite3 mark it busy. Both threads can then try to use that same statement.

This fix reserves that statement with a flag ( in_use) so other cursor gets an uncached one.
Also used fullmutex so that sql lite processes on the shared connection are in order.

For regression i ran concurrent queries from 4 threads sharing connection and returned blob data were checked also that subprocesses end cleanly.

For testing i had issues initially then i installed c toolchain and ran it with help of ai.