Improve unification of conditional expressions

I have this snippet of code

from typing import Any,List

class Node:
    def __init__(self, value:Any, *, children:List['Node']=None) -> None:
        self.value = value
        self.children = [] if children is None else children # type: List[Node]

if I use mypy I get

Tree.py: note: In member "__init__" of class "Node":
Tree.py:10: error: Incompatible types in assignment (expression has type "object", variable has type List[Node])

I think it's something related to the children parameter setted to None, but the problem is... how do I specify that yes, maybe the children parameter can be None, but self.children, by construction is always a List of Nodes?

This looks like the type of the expression `([] if ... else children)`, where children has type `List[Node]`, is inferred as `object` whereas it should really be inferred as `List[Node]`. We should fix this. In the meantime you can work around it by replacing `[]` with `cast(List[Node], [])`. But that's a very verbose solution and mypy should do better.

The if..else expression doesn't propagate the type context (in your example List[Node]) to the child expressions. Propagating the context would let mypy infer type List[Node] for the left operand.

A related way to fix this would be to use the type of the right operand as the type context for the first expression, in case the latter has an incomplete inferred type. This would have the benefit of not requiring a type annotation.

I think that you could also work around the problem by rearranging the expression:

self.children = children if children is not None else []

Well, in this case if the unification of empty list and List[Node] was the latter we wouldn't even need to propagate the context.

…

--Guido (mobile)

I keep running into this. The type of (a if x else b) where a and b don't have the same type often degenerates to object or some other useless type rather than the intersection of the two types. Latest example:

class A: pass
class B:
    def foo(self): pass
class C(A, B): pass  # No error if we switch this to C(B, A)
a = B() if True else C()
a.foo()  # E: "object" has no attribute "foo"

Also, e.g.in Python 2 mode:

float(Decimal(0) if True else 0)  # E: No overload variant of "float" matches argument types [builtins.object]

In Python 3 mode this gives

error: Argument 1 to "float" has incompatible type "object"; expected "Union[SupportsFloat, str, bytes]"

I'm going to have a crack at this.

Improve unification of conditional expressions when the left branch h…

Navigation Menu

Repository navigation

Improve unification of conditional expressions #1094

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Improve unification of conditional expressions #1094

Description

Activity

gvanrossum commented on Dec 22, 2015

JukkaL commented on Dec 28, 2015

gvanrossum commented on Dec 30, 2015

gvanrossum commented on Mar 8, 2016

gvanrossum commented on Apr 15, 2016

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions