No type inference for indexed expressions

[pkch]

I'm sure there's a limit to how far type inference can go, and it's reasonable that it doesn't evaluate the result of applying an index:

from typing import *

def f(x: Dict[str, Union[int, str]], k: str) -> None:
    if isinstance(x[k], int):
        x[k] + 1  # type error: Unsupported operand types for + ("Union[int, str]" and "int")

But maybe an example in the docs would be helpful? A suggestion could be given to either rewrite it like this:

def f(x: Dict[str, Union[int, str]], k: str) -> None:
    v = x[k]
    # note: both x[k] (in the if expression and in the body) must be replaced with a simple variable
    if isinstance(v, int):
        v + 1  # no type error

or to just mark it as # type: ignore.

[refi64]

I'm not sure how things are here, but I know over in Crystal (a type-inferred language), it's the same way, and it's entirely intentional; in theory, in a multi-threaded program, x[k] could change between the two points.

[gvanrossum]

If you can find a suitable spot in the docs to add this a PR would be appreciated!

[pkch]

@kirbyfan64:I think in mypy case, this is only because inference complexity grows really fast once you move away from simple variables. It's not because of thread safety: after all, it works fine if you replace x[k] with a global variable, which can be changed from a different thread.

[gvanrossum]

Yeah, mypy has an implicit rule that we don't worry about threading issues like that (that would be an entirely different kind of static checking algorithm).

…

On Wed, Mar 15, 2017 at 9:13 AM, pkch ***@***.***> wrote: @kirbyfan64 <https://github.com/kirbyfan64>:I think in mypy case, this is only because inference complexity grows really fast once you move away from simple variables. It's not because of thread safety: after all, it works fine if you replace x[k] with a global variable, which can be changed from a different thread. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#3006 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACwrMo-ZpgdN8rs8cP5v44pNpjroDJZOks5rmA47gaJpZM4Mdovq> .

-- --Guido van Rossum (python.org/~guido)

[miedzinski]

Duplicate of #2458.