Type Alias inside class unsupported

[nipunn1313]

Bug Report

When there's a type alias inside a class, it's not considered to be a type by mypy.
pyright appears to support this.

To Reproduce

https://mypy-play.net/?mypy=latest&python=3.10&gist=509f8043f236ccc7e178dd56f526eaf1

import typing

T = typing.NewType('T', int)
class C:
    V = T
x: T = T(5)
y: C.V = C.V(5)

Expected Behavior

No errors

Actual Behavior

main.py:7: error: Variable "main.C.V" is not valid as a type
main.py:7: note: See https://mypy.readthedocs.io/en/stable/common_issues.html#variables-vs-type-aliases
Found 1 error in 1 file (checked 1 source file)

[nipunn1313]

I found that

import typing

T = typing.NewType('T', int)
class C:
    V1 = T
    V2: typing.TypeAlias = T
x: T = T(5)
y1: C.V1 = C.V1(5)
y2: C.V2 = C.V2(5)

I found that y2 works but y1 doesn't. This is consistent with this comment - https://mail.python.org/archives/list/[email protected]/message/QNJEVUIHGCO6V4RYKBQK7FSRYHZZGWNH/ - on PEP613

[hauntsaninja]

Yeah, I think this is by design and PEP 613 is the appropriate way forward, see

mypy/mypy/semanal.py

Line 2591 in 3d82f6f

# Fourth rule (special case): Non-subscripted right hand side creates a variable

[nipunn1313]

PEP613 appears to explicitly ban type aliases in nested scope (https://www.python.org/dev/peps/pep-0613/#scope-restrictions)

Though the wording around it makes it seem to me that it meant to declare that implicit type aliases were banned in nested scope - and explicit type aliases are to be preferred (AKA V: TypeAlias = T). If this is the case, then perhaps we should modify the wording on PEP613 in the scope restriction section.

[hauntsaninja]

Oh interesting. I think mypy's current behaviour is good and don't see a reason explicit type aliases in nested scope should be banned by PEP. Going through mypy issues today it seems like this is not an uncommon use case.

I'd be happy with "only use explicit type aliases in nested scope" as PEP guidance. I guess pyre or whatever type checker can choose not to support type aliases in nested scope, but I don't think PEP should prohibit mypy from recognising this. Maybe worth emailing typing-sig about?

[erictraut]

FWIW, I interpreted the PEP to mean that an explicit type alias wasn't allowed in a nested scope, so pyright generates an error for this condition. Conversely, I allow implicit type aliases in nested scopes, since PEP 484 doesn't appear to disallow this. So if we want to clarify in the PEP that explicit type aliases are allowed in nested scopes, I'd be happy to make that change in pyright's logic.

[nipunn1313]

IMO, explicit type alias / explicit variable should be allowed at any scope

V: TypeAlias = T  # Alias
V: Type[T] = T  # Variable

For the implicit ones - some ideas

1. Could leave it up to the type checker.
2. Could make it explicit in the PEP613
   a) global scope -> Alias, nested scope -> variable
   b) Always treat as Alias (any scope)

IMO, 2b makes the most sense to me (pyright's current behavior). It seems really rare to me that a type gets assigned to a variable other than an alias. 2a is closer to mypy's current behavior.

(1) allows for flexibility of either, which might be a good starting point. I'll try to email typing-sig! Thanks for commenting.