type inference depends on method definition order

[matejcik]

With current mypy master and python 3.5.2, the following fails type checking:

class A:
    def hello(self) -> None:
        if self.x:
            self.x = False

    def world(self) -> None:
        self.x = True

poc.py:3: error: Cannot determine type of 'x'

The following passes:

class A:
    def world(self) -> None:
        self.x = True

    def hello(self) -> None:
        if self.x:
            self.x = False

The following also passes:

class A:
    def hello(self) -> None:
        if self.x:
            self.x = False  # type: bool

    def world(self) -> None:
        self.x = True

despite the fact that at the line where I'm putting the annotation, it's already redundant. However, annotating the usage in world() doesn't help.

(Aside: I suspect the underlying reason for this is the same as why mypy won't infer types for None variables set in __init__ ? That is a major paint point, it would be super nice if that was resolved.)

[wiml]

Is it possible this is the more general case of issue #1727 ?

[ilevkivskyi]

No, this is a problem with deferred nodes (multi-pass type checking). However taking into account two simple workarounds, I would say this is low priority.

[ilevkivskyi]

This actually shows two errors on master:

main:3: error: Cannot determine type of 'x'
main:7: error: Incompatible types in assignment (expression has type "bool", variable has type "bool")

This is quite confusing so raising priority to normal. This may indicate a problem in the new semantic analyzer.