#25176 closed Cleanup/optimization (fixed)
TestCase: an error in setUpTestData snowballs across the entire test suite
Reported by: | Adam Johnson | Owned by: | nobody |
---|---|---|---|
Component: | Testing framework | Version: | 1.8 |
Severity: | Normal | Keywords: | |
Cc: | Triage Stage: | Accepted | |
Has patch: | yes | Needs documentation: | no |
Needs tests: | no | Patch needs improvement: | yes |
Easy pickings: | no | UI/UX: | no |
Description
If a setUpTestData function fails, it causes an error that breaks the current transaction; this is not undone by TestCase, and thus all following database access in all other tests fails, outputting a lot of ERRORs on a test run, and making it hard to debug that it was only the *first* one that has an actual problem. The atomic instances can be exited cleanly on exception fairly easily inside the TestCase code.
Change History (9)
comment:1 by , 9 years ago
comment:2 by , 9 years ago
Triage Stage: | Unreviewed → Accepted |
---|
comment:3 by , 9 years ago
For the test, could you store the result of connection.in_atomic_block
in a variable and then put the assertion in the test method? I didn't understand what you meant when you said, "you can then trigger tearDownClass even though it shouldn't be run".
comment:4 by , 9 years ago
The test method would never be reached - we're triggering an error in setUpClass
which means the tests don't get run.
The above test patch catches the breakage and allows the tests to continue - in the "broken" world. I'm not sure this is legit or even if it works now it might not be forwards compatible with future edits .
I also experimented with a broken test case with @expectedFailure
wrapped around it but that's not sensitive enough since it allows any exception to be expected. It might be possible to write a custom expectedFailure
decorator that checks it was the custom exception raised during the test case and marks it as a pass/skip/expected failure, but when I briefly looked it seemed a bit hacky.
comment:5 by , 9 years ago
If there is no reasonable way to write a test, let's ship the fix as is. That happens occasionally for changes in the testing infrastructure. The latest iteration of the patch looks good to me.
comment:6 by , 9 years ago
Patch needs improvement: | set |
---|
https://github.com/django/django/pull/5051
I've struggled to add a test for this. I had the idea for something like the below, but the problem is that if you try do any checking in the exception handler in setUpClass you can then trigger tearDownClass even though it shouldn't be run. I don't think there is a sane way of testing it, except from creating tests that run a second set of tests where one has a broken setUpTestData, and then asserting that only the one of the second set failed.