Funny behaviour with Union[Decimal, int]

[admirabilis]

Considering the following program:

from decimal import Decimal
from typing import Union

def foo(bar: Union[Decimal, int]):
    print(Decimal(1) + bar)
    print(bar + Decimal(1))

Why does mypy complain in the second print()?

$ mypy foo.py
foo.py: note: In function "foo":
foo.py:6: error: Unsupported operand types for + ("Union[Decimal, int]" and "Decimal")

I'm using mypy 0.3.1 on Python 3.5.1, Ubuntu 16.04.

[gvanrossum]

I'm not sure. I can repro this with the latest mypy from the repo.

[refi64]

@gvanrossum Is this a typeshed bug? Note that Decimal.__add__ is defined with rhs as _Decimal (or Union[Decimal, int]), but __radd__ expects just a normal int.

[gvanrossum]

But I can't repro it without using a Union.

[refi64]

@gvanrossum Exactly, because __radd__ expects just an int, not a union. Here's an example outside of Decimal:

from typing import Union

class Dec:
    def __add__(self, x: Union['Dec', int]) -> 'Dec': pass # This accepts another Dec.
    def __radd__(self, x: int) -> 'Dec': pass # But this does NOT!

def foo(bar: Union[Dec, int]):
    d = None # type: Dec
    print(d + bar)
    print(bar + d) # Error, since Dec.__radd__ wants `int` for its lhs, but `bar` is `Union[Dec, int]`

[gvanrossum]

But at runtime, `__radd__` would not be invoked with a Decimal argument -- so its declaration feels correct.

[refi64]

So then the general issue is binary operators, I guess. It seems Mypy sees this:

x + y # where x's type is Union[A,B], and y's type is C

and tries to look for either Union[A,B].__add__(C) or C.__radd__, but it should instead look for both A and B separately. So it would insteas check for one of the following:

• A.__add__(C) and B.__add__(C)
• C.__radd__(A) and C.__radd__(B)

This is mostly a guess, though...

[JukkaL]

@kirbyfan64 I believe that you are right. The logic for checking binary operations should probably decompose union types early and process each union item independently (and finally combine the results, possibly resulting in another union type).

[gvanrossum]

Maybe #1264 is relevant? (That's holding up the stubs for fractions.py in typeshed, for example.)

[JukkaL]

That may be relevant, but this may also be a separate issue/bug.

[ilevkivskyi]

FWIW this particular example works correctly on master (due to a recent operator overhaul by @Michael0x2a ). We have a separate issue for general union destructuring in operator checks.

[ilevkivskyi]

Just for reference, the main issue is #2128