LiveServerThreadTest.test_closes_connections() doesn't pass with non-in-memory SQLite

[Chris Jerdonek]

I noticed that LiveServerThreadTest.test_closes_connections() doesn't seem to pass when using non-in-memory SQLite. It's only configured to be skipped ​for in-memory SQLite. It doesn't pass because the test has the ​following line:

self.assertFalse(conn.is_usable())

But SQLite's is_usable() has a hard-coded return value of True:
​https://github.com/django/django/blob/3fa1ed53be370f4b1a94d78b56ff30d23b131623/django/db/backends/sqlite3/base.py#L387-L388

I guess this means that Django's CI doesn't check non-in-memory SQLite. If it's true that Django doesn't check this case, I'm not sure if that means it doesn't need to be fixed.

[Mariusz Felisiak]

I guess this means that Django's CI doesn't check non-in-memory SQLite.

That's True, but I think it's still worth fixing. There is an interesting ​discussion about this test in the original PR. I think we should remove test_closes_connections() from django_test_skips() and use @skipUnlessDBFeature('test_db_allows_multiple_connections') instead.

[Chris Jerdonek]

Just to be totally clear, would @skipUnlessDBFeature('test_db_allows_multiple_connections') mean that the test is or isn't skipped with non-in-memory SQLite?

[Mariusz Felisiak]

Replying to Chris Jerdonek:

Just to be totally clear, would @skipUnlessDBFeature('test_db_allows_multiple_connections') mean that the test is or isn't skipped with non-in-memory SQLite?

It will always be skipped on SQLite, so also with non-in-memory database.

[Chris Jerdonek]

It will always be skipped on SQLite, so also with non-in-memory database.

Okay, that's what I thought. Why do you think the test should always be skipped on SQLite as opposed to making the test work also for non-in-memory SQLite? It seems like making it work would be the better outcome since it would cover more scenarios where the test makes sense.

[Mariusz Felisiak]

Replying to Chris Jerdonek:

Okay, that's what I thought. Why do you think the test should always be skipped on SQLite as opposed to making the test work also for non-in-memory SQLite? It seems like making it work would be the better outcome since it would cover more scenarios where the test makes sense.

Agreed, making the test work would be better, but I'm sure how to do this. I would be happy to review a patch, if you have an idea.

[Chris Jerdonek]

Agreed, thank you. I have some ideas, but not 100% sure it will work out.

[Chris Jerdonek]

Okay, I came up with a solution for getting the test to work with non-in-memory SQLite. The main idea is to use TransactionTestCase instead of TestCase.

[Mariusz Felisiak]

​PR

[Mariusz Felisiak <felisiak.mariusz@…>]

In 5c4c3e2d:

Fixed #32445 -- Fixed LiveServerThreadTest.test_closes_connections() for non-in-memory database on SQLite.