assert vs numpy.testing.assert_ in testing #8583Description
A few times the idea of using direct assert statements in tests rather than assert_ from numpy.testing has come up. IIUC pytest guarantees that assert statements are executed.
Some advantages of assert :
- Simplicity. No need for imports.
- Improved tracebacks.
pytesthas put work into makingasserterror messages readable for diagnosing problems. - Long-term consistency (?). I have seen a proliferation of
assertstatements in other packages, so that seems to be the zeitgeist. - Lowering contribution/review burden. Given the increasing use of
assert, the probability of contributors using it assuming it's acceptable increases, and allowing its use is one fewer thing we have to nitpick.
Some advantages of numpy.testing.assert_
- Internal consistency (near-/medium-term). We have historically used
assert_and have 1561 of instances according togit grep "assert_(" | wc -l. - Guaranteed support.
numpy.testingcan be guaranteed to work as long asnumpydoes, whereasassertrelies onpytestmaintenance and/orpytest-compatibleassertsupport by any future testing framework.
There are probably others, feel free to comment and/or edit this top comment to keep track.
My +1 is for allowing assert if people want to use it. I have begun moving a few of my other repos to follow this pattern and so far have found it beneficial.