[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.