Fix for REUSE_DB problems

[alexjg]

I had a look into the specific problem on my machine. The essential problem is that several db backends cache their underlying connection. The nose test runner detects whether a new database is needed by calling cursor() on the backend after renaming the NAME field of the settings dictionary which the backend uses, this fails if the database does not exist, allowing nose to decide to create the database.

Unfortunately some backends cache their underlying connection objects so if any code connects to the database (which hasn't been renamed by django-nose yet) before nose does then any subsequent cursor() calls will succeed (as they use the cached connection) and nose is unable to detect that the database exists. The attached code fixes that.

Fixes #76

[erikrose]

If this really fixes it, you've made my day! :-) Unfortunately, I've lost the ability to reproduce it. Do you happen to have a reduced (or even non-reduced) test case?

[alexjg]

Well it's solving the problem for me but unfortunately the code base is a clients website. I'll try and extract a configuration which reproduces the problem in the next few days.

[dcramer]

FWIW this is how we solve these things.

for conn in connections.all(): conn.close

[alexjg]

Yeah that seems like a much more sensible approach.

[akaihola]

@dcramer Where did you do the conn.close() calls?

On an old project which runs on Django 1.3.1, I just can't figure out why django-nose creates and syncdbs a test_myproject database, but tries to run the tests on myproject. I've tried different combinations of NAME and TEST_NAME in settings.DATABASES but haven't found any working combination.