Infer function parameter types using annotations

[ringohoffman]

Type of Changes

	Type
✓	🐛 Bug fix
✓	✨ New feature

Description

Fixes pylint-dev/pylint#4813
Fixes pylint-dev/pylint#8781

Type annotations are not currently used for inferencing, which means that, for example, functions with parameter typed as a logging.Logger would not receive relevant pylint warnings such as logging-fstring-interpolation.

To make use of these type annotations, we now modify infer_assign to add yield additional Instances if an AssignName is being processed whose parent is an Arguments with annotations for the AssignName.

cc: @Pierre-Sassoulas

[codecov]

Codecov Report

Merging #2216 (3ae1618) into main (eabc643) will increase coverage by 0.03%.
The diff coverage is 100.00%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2216      +/-   ##
==========================================
+ Coverage   92.71%   92.75%   +0.03%     
==========================================
  Files          94       94              
  Lines       10837    10847      +10     
==========================================
+ Hits        10048    10061      +13     
+ Misses        789      786       -3     

Flag	Coverage Δ
linux	92.48% <100.00%> (+<0.01%)	⬆️
pypy	88.00% <100.00%> (+0.02%)	⬆️
windows	92.34% <100.00%> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
astroid/inference.py	94.67% <100.00%> (+0.09%)	⬆️

... and 1 file with indirect coverage changes

[DanielNoord]

The issue with this approach is that we have no way of knowing that the annotations are correct. Personally I wouldn't like pylint to assume all annotations are correct and think this should be handled with some sort of flag (--consider-annotations-are-correct).
Just turning this on by default doesn't seem like the best solution to me.

I'm wondering what other maintainers thinks of this.

[ringohoffman]

I think this should be handled with some sort of flag (--consider-annotations-are-correct).

I support this. If we could assume hinted return types were correct, I would think we could save a lot of time on inferencing, assuming we could skip manual inferencing.

Personally I wouldn't like pylint to assume all annotations are correct

I understand that astroid was developed to infer types before type hints were introduced to Python. However, trusting type hints is what most type checkers do. Having to infer the return types even on standard library functions would seem like a major reason that pylint takes, for example, ~25 minutes to run on pytorch/torch.

[Pierre-Sassoulas]

We need to be careful here, using inference and not trusting annotations is the differentiating factor that makes pylint find issues that other tools don't in partially typed code base.

On the other hand it would be nice to be able to detect that we're in a fully typed code base where we should trust the typing. This is not going to be trivial, unless we ask users to tell us via options. It seems possible to take typing in builtin into account without any drawbacks and without detecting anything or adding an option. (Because we know that typing are truthful and reliable in builtin)