The problem occurs because coverage is run in-process using coverage.start() (here), but the test runs under a subprocess and is unobservable to coverage.

Here are some approaches to fixing it.

Keep coverage in-process. Also run manage.py in-process.

Three variations on this:

Use exec() to run manage.py in the same interpreter. This has a bad security smell but it's hard to say if there's really any difference from ...
Import manage.py and run main() (here). Because manage.py is hard-coded to read its arguments from sys.argv, this approach would require manipulation of that variable.
Do the stuff manage.py does. Something like this:

from django.core.management import execute_from_command_line
execute_from_command_line(["/path/to/manage.py", "test", ...])

Don't run coverage in-process. Call manage.py as designed

In this case, we'd simply replace [sys.executable] (here) with ["coverage", "run"]. The test execution code would need to be cleaned up.

I think all of these have their plusses and minuses. This is my first time here and I'm just working on a weekend project so I don't know the backstory on the current implementation. I like the execute_from_command_line() approach for simplicity.

Support Django Test Runs with Coverage #24199

Description

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions