Savepoint should not assume an opened connection
|Reported by:||Jeremy Dunck||Owned by:||nobody|
|Component:||Database layer (models, ORM)||Version:||1.0|
|Has patch:||yes||Needs documentation:||no|
|Needs tests:||no||Patch needs improvement:||no|
Using django transaction.savepoint() without having previously used a cursor fails if the backend actually supports savepoints.
e.g. Using postgres:
Python 2.5.2 (r252:60911, Mar 18 2008, 15:01:36) [GCC 4.0.1 (Apple Inc. build 5465)] on darwin Type "help", "copyright", "credits" or "license" for more information. >>> from django.db import transaction >>> transaction.savepoint() Traceback (most recent call last): File "<stdin>", line 1, in <module> File "/Users/jeremydunck/work/django/g-django/django/db/transaction.py", line 188, in savepoint connection._savepoint(sid) File "/Users/jeremydunck/work/django/g-django/django/db/backends/__init__.py", line 43, in _savepoint self.connection.cursor().execute(self.ops.savepoint_create_sql(sid)) AttributeError: 'NoneType' object has no attribute 'cursor'
It seems this problem has existed since savepoints were introduced, but I found it by trying to use the cached_db sessions backend.
In any case, the problem is that the savepoint family of methods on BaseDatabaseWrapper make the assumption that self.connection has been initiailized-- this is not a safe assumption. self.connection is initialized the first time BaseDatabaseWrapper.cursor is called, and that may not have happened before BaseDatabaseWrapper._savepoint*.
In all cases, ._savepoint* goes on to use connection.cursor, so I think it's safe to just use self.cursor() directly rather than self.connection.cursor().
Unfortunately, I'm not sure how to make a test for this, because django's test suite itself uses django.db.connection.
At least it is easy to reproduce. :-)