Invalid Python in test cases

[spkersten]

Many test cases (>100) contain invalid Python code like this,

a = A()

class A: pass

where a name is referenced before it is bound.

This impedes solving #686.

[JukkaL]

Yeah, this is unfortunate.

[JukkaL]

Many of these cases can likely be found using the test data parser used for running test cases and regular expressions.

[ArgentFalcon]

I'm looking into this

[ArgentFalcon]

Do we want blanket fixes over all tests so that names are only referenced after they are bound?
Or do we want to leave some tests incorrect so that we can validate that mypy correctly errors for them?

Also, @spkersten, can you point me to your changes that enable mypy to detect this issue?

[gvanrossum]

I'd define a few separate tests that check for various kinds of forward references (e.g. module, class, function, nested class, type variable, type alias, other kind of variable). Though it's possible that mypy needs to be fixed first to actually error out on them. It's also possible that we should fix the tests without fixing this in mypy (maybe fixing this in mypy is tricky, and of course it must still support forward references from type comments or type annotations in string literals).

[ArgentFalcon]

Going by @gvanrossum comment, it seems that we want to make a separate set of tests to test forward references. Therefore I will move forward on my quest to fix all mis-referenced tests I can find

[ArgentFalcon]

So right now I'm writing a test suite that tries to exec all the test cases and report their failure.
My next task will be blacklisting the test-files that shouldn't be run (testing the pyi stuff), as well as
the files/test cases that should intentionally fail

[hauntsaninja]

yup, still open, feel free to work on it. see https://github.com/python/mypy/blob/master/CONTRIBUTING.md#first-time-contributors

You do not need to ask for permission to work on any of these issues. Just fix the issue yourself, try to add a unit test and open a pull request.

[JukkaL]

I think that this was addressed in #14166.