Freezegun Kills Pandas about freezegun HOT 9 CLOSED

Same issue here: importing freezegun has uncool side effects. Given the usage as decorator or context manager, this was fully unexpected to me. Can I help fix this?

from freezegun.

The reason freezegun mocks on import is to accomodate those that do from datetime import datetime before freezegun is activated. If you can find a good workaround, I'll be happy to pull it in.

from freezegun.

Hm, if someone imports datetime from datetime before the first import of freezegun (e.g. a web app that imports models on startup and the models import datetime, but freezegun is only imported after, in app.tests), it won’t work anyway, will it? IOW, recommended form of import becomes “import datetime” everywhere, and then import-time monkey-patching is not necessary. My own code would need changes as I always import classes from datetime, but I wouldn’t mind doing that if it allowed me to use freezegun instead of tedious mock.patch + datetime.

from freezegun.

Sorry, my comment was unclear. What works now is if someone imports freezegun, then does from datetime import datetime and then uses the decorator.

When freezegun first started, it didn't do any of this patching on import, but then I got a bunch of complaints. The issue is that people can't always control how third-party software imports datetime.

Sorry that this all kinda sucks. It's my least favorite thing about this library. We could do something like

import freezegun
freezegun.dont_patch()

and have dont_patch revert the patching.

That seems pretty gross to me though. Another idea is to have a different branch that works the old way.

from freezegun.

It's only marginally less gross, but there could also be two different import paths:

from freezegun import freeze_time  # doesn't patch on import
from freezegun.clobbering import freeze_time  # patches on import

from freezegun.

I am not sure that we can find a satisfactory solution. The problem in my case is that I need to control the return value of datetime.now in one module but without overriding the datetime type, as another module is for example checking types to know how to convert objects to JSON. These constraints seem at odds with what freezegun users want (according to the complaints you described earlier). I think I’ll have to write a shortcut function like patch_datetime('module', 'wanted time') using the regular mock module.

from freezegun.

This should be fixed in the newest version.

from freezegun.

Great news! Thank you for considering our use cases.

from freezegun.

Just to be completely clear (in case people find this issue in the future), Freezegun will not work directly with Pandas, but Freezegun now only mocks datetime when it is activated allowing use of Freezegun in non-Panda tests.

from freezegun.