Type inference regression involving filter(None, ...) #8093

JukkaL · 2019-12-05T16:58:46Z

This example started generating errors after #8077:

from typing import Optional, TypeVar, Iterable, Iterator, List

_T = TypeVar('_T')

def filter(__function: None, __iterable: Iterable[Optional[_T]]) -> List[_T]: ...

x: Optional[str]

# Need type annotation for 'pagelets'
# List item 0 has incompatible type "Optional[str]"; expected "None"
pagelets = filter(None, [x])

The issue also happens with Python 2 filter(), at least.

cc @ilevkivskyi

The text was updated successfully, but these errors were encountered:

Fixes #8093 The fix is kind of simple, we really need to keep all `ErasedType`s intact to preserve the existing type inference logic, but it is also long because I need to thread the flag down like dozen function calls. The issue itself btw highlight how fragile is current type inference logic (note that `expand_type()` implicitly calls `make_simplified_union()` too, but I think that one doesn't need to be updated because it shouldn't be called with any erased components).

JukkaL added bug mypy got something wrong priority-0-high false-positive mypy gave an error on correct code labels Dec 5, 2019

ilevkivskyi mentioned this issue Dec 5, 2019

Don't simplify out ErasedType from unions during type inference #8095

Merged

ilevkivskyi closed this as completed in #8095 Dec 5, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Type inference regression involving filter(None, ...) #8093

Type inference regression involving filter(None, ...) #8093

JukkaL commented Dec 5, 2019

Type inference regression involving filter(None, ...) #8093

Type inference regression involving filter(None, ...) #8093

Comments

JukkaL commented Dec 5, 2019