alessandrofelder
commented
1 month ago I've created this as a draft PR because I think this should be fixed, but maybe this PR should be merged and then an issue created to improve the test? This PR doesn't introduce the problem.
I don't mind either way. A new issue is probably cleaner (but also more overhead). One thing maybe worth trying (that I think would be a good idea across BrainGlobe) is to monkeypatch Path.home() for all tests when not running on CI. It looks like there's a few constant variables that need to be adapted for the patching to work. This way
- we test a slightly more real-life case on CI (
Path.home()is not patched, but that's fine, because the CI does not have a user set up.) - we locally keep
~./brainglobe/(user data) and~/.brainglobe-tests/.brainglobe/(tests interacting with fake user data) always completely separate.
adamltyson
This PR closes #297 by removing the slow pytest marker.
I changed the test a bit, but it's still not ideal, because it changes the default brainglobe config file. It changes it back, but that's not great.
I assumed I could get around this by setting
monkeypatch.setenv("BRAINGLOBE_CONFIG_DIR", str(tmp_path)), this however doesn't seem to set the config dir temporarily. I don't know if I'm monkeypatching this incorrectly, or there's a bug in the atlas api itself.I've created this as a draft PR because I think this should be fixed, but maybe this PR should be merged and then an issue created to improve the test? This PR doesn't introduce the problem.