e384c5a2a6
The _migrate_* startup helpers in core/database.py opened a raw sqlite3.connect() inside a try and called conn.close() as the last statement in that try. If any earlier statement raised (locked DB, unexpected schema, a failed ALTER), close() was skipped and the bare except only logged the error — leaking the connection (file handle + lock) for the lifetime of the process. These migrations run on every startup. Wrap each in the conn = None + try/except/finally pattern already used by _migrate_chat_messages_fts in this same file, so the connection is closed on all exit paths. 25 functions; no change on the success path. Helpers that already close safely are left untouched: _migrate_chat_messages_fts and _migrate_backfill_task_folders (the latter uses SQLAlchemy's engine.connect() context manager). Same bug class as the previously merged DB-connection-leak fix (#64) and the IMAP logout-on-all-paths fix (#1530). |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| atomic_io.py | ||
| auth.py | ||
| constants.py | ||
| database.py | ||
| exceptions.py | ||
| middleware.py | ||
| models.py | ||
| platform_compat.py | ||
| session_manager.py | ||