More tests for imported modules

[tristanlatr]

Type of Changes

This PR adds more tests, see #1398

Related Issue

Resolves #1398

[tristanlatr]

The way the tests are faillinbg is very weird and I'm unsure if it's linked to my changes.

The following test fails:
tests/unittest_scoped_nodes.py::ClassNodeTest::test_mro_with_factories with a difference in the computed MRO:

E   - ['FinalClass', 'ClassB', 'MixinB', 'ClassA', 'MixinA', 'Base', 'object']
E   + ['FinalClass', 'ClassB', 'MixinB', '', 'ClassA', 'MixinA', '', 'Base', 'object']

I wonder if this is a side effect of calling clear_cache() or it's totally unrelated? Any clue?

[Pierre-Sassoulas]

Ugh, I think you're right the tests might not be independent and rely on the caching between tests. Do you have an idea how to fix the failing test ?

[tristanlatr]

I could restore the cache after the tests ?

[tristanlatr]

Looks like it doesn’t work :/ I must be missing something…

[jacobtylerwalls]

I was just working with astroid's caching mechanism recently and remembered using AstroidCacheSetupMixin to solve test isolation issues, so I just took the liberty to use it here :-)

@tristanlatr @Pierre-Sassoulas do you have thoughts about whether we should merge these tests?

[coveralls]

Pull Request Test Coverage Report for Build 2308678113

• 0 of 0 changed or added relevant lines in 0 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 91.665%

---

Totals
Change from base Build 2303054388:	0.0%
Covered Lines:	9139
Relevant Lines:	9970

---

💛 - Coveralls

[tristanlatr]

Hello @jacobtylerwalls,

Thanks a lot for your input here. If I understand correctly, AstroidCacheSetupMixin clear the cache between each test method, right ?

If all tests passes, I would be happy that these test land in the main branch.

[jacobtylerwalls]

Not between each test method but at the beginning and end of the class. (The reason your test was failing was the same thing I saw and fixed in #1528.) Do we need to clear the cache with every test method? If so we can adjust the behavior of the mixin to use setUp and tearDown instead (probably in another PR).

The failure on 3.11 is unrelated and tracked in #1551

[jacobtylerwalls]

I think we can merge this; if we want to increase the test isolation to clear caches every test method, we can do that in another PR anyway.