Docs: adds a note about issubclass usage in common_issues.rst

[sobolevn]

It was really not clear for some users that you can use issubclass as well.

[sobolevn]

@terencehonles oh! Thanks! I make this typo all the time! 😆

[hauntsaninja]

Thanks for all your work to improve mypy docs. I'm not sure this is the right section: I don't think issubclass is a common solution and I'd like to try and keep the common issues page focussed.

Also, looking more carefully, I believe mypy does now narrow for type(o) is int, so this whole section feels out of date...

Anyway, I think what I'd want on the "Common issues" page is a description of how to use assert and cast to convince mypy that something is of a given type.

Maybe we can add a new top level section on narrowing with detailed examples like in this PR (that we could then link to from common issues). I remember us struggling to find a home for the excellent TypeGuard documentation you wrote, a narrowing section could be a good place for that too.

[sobolevn]

Anyway, I think what I'd want on the "Common issues" page is a description of how to use assert and cast to convince mypy that something is of a given type.

Sure! I will change this PR to add cast and assert.

Also, looking more carefully, I believe mypy does now narrow for type(o) is int, so this whole section feels out of date...

Yes, it works for type(o) is ... case. But, it does not work for the case in my example:

def f(o: object) -> None:
    if issubclass(type(o), MyCalcMeta): 
        reveal_type(type(o))  # Revealed type is "Type[builtins.object]"

Should I open a new issue to track this?

Maybe we can add a new top level section on narrowing with detailed examples like in this PR (that we could then link to from common issues)

Sounds good to me! I will create one more PR later.

[sobolevn]

Moreover, current mypy docs here are outdated:

1. This statement:

Mypy can’t infer the type of o after the type() check because it only knows about isinstance() (and the latter is better style anyway)

This is not true anymore.

 def f(o: object) -> None:
       if type(o) is int:
           o = cast(int, o)
           g(o + 1)    # This would be an error without the cast
           ...
       else:
           ...

In this example cast is no longer required. The revealed type of o is Revealed type is "builtins.int".

I am going to change this example completely. I will find some really useful usecases for cast() that cannot be modeled with isinstance, issubclass, or with custom typeguards.

[sobolevn]

I will add a link to the new page here in the next PR

[sobolevn]

I will add a link here as well in the next PR

[hauntsaninja]

Awesome, this looks great!

[sobolevn]

@hauntsaninja thanks a lot for your help! 👍

I will return with the next PR somewhere this week.