Description
Feature
One thing that bothered me for a long time is getattr
and its very broad return type even for simplest cases:
from typing import Literal
class Some:
def __init__(self) -> None:
self.a = 1
s = Some()
reveal_type(s.a) # Revealed type is "builtins.int"
reveal_type(getattr(s, 'a')) # Revealed type is "Any"
reveal_type(getattr(s, 'b')) # Revealed type is "Any"
reveal_type(getattr(s, 'a', None)) # Revealed type is "Union[Any, None]"
a: Literal['a']
reveal_type(getattr(s, a)) # Revealed type is "Any"
ab: Literal['a', 'b']
reveal_type(getattr(s, ab)) # Revealed type is "Any"
So, I propose to add better support for it to mypy
.
The same example, after the support is added:
from typing import Literal
class Some:
def __init__(self) -> None:
self.a = 1
s = Some()
reveal_type(s.a) # Revealed type is "builtins.int"
reveal_type(getattr(s, 'a')) # Revealed type is "builtins.int"
reveal_type(getattr(s, 'b')) # error: "Some" has no attribute "b"
reveal_type(getattr(s, 'a', None)) # Revealed type is "builtins.int"
a: Literal['a']
reveal_type(getattr(s, a)) # Revealed type is "builtins.int"
ab: Literal['a', 'b']
reveal_type(getattr(s, ab)) # error: "Some" has no attribute "b"
reveal_type(getattr(s, ab, None)) # Revealed type is "builtins.int | None"
I have even written a custom plugin for it at some point.
Why?
This pattern is quite popular among Python developers. geattr(obj, 'string_key', None)
is used for compatibility, optional dependencies, etc. That's something you can find in almost any project.
Technical details
I am going to utilize existing checkmember
logic.
I even plan to support attribute plugins, where possible, because a.x
and getattr(a, 'x')
are semantically identical.
For now only str literals and Literal
str types are going to be supported.
Related issues and problems
I am calling this issue "meta", because there are several very related problems I want to mention / solve here:
Next steps
I will start with a PR with a code prototype in a couple of days.