-
-
Notifications
You must be signed in to change notification settings - Fork 2.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
What to do about setters of a different type than their property? #3004
Comments
In other words "with arbitrary setters, you can have an lvalue be of a different type than its corresponding rvalue" |
IIRC we debated a similar issues when descriptors were added and decided that it's a perverse style that we just won't support. If you really have this you can add |
Right now, you have to |
What does a discrepancy between |
One real-life case where I encountered this a few times now is a normalizing property like this: from typing import Set, Iterable
class Foo:
def __init__(self) -> None:
self._foo = set() # type: Set[int]
@property
def foo(self) -> Set[int]:
return self._foo
@foo.setter
def foo(self, v: Iterable[int]) -> None:
self._foo = set(v)
Foo().foo = [1, 2, 3] I like this implementation, because it allows a user of the cleverly named class Using the same type in the getter and setter would complicate the life of its users. |
I disagree that this is "perverse" -- contravariance in the setter is one example, e.g. where the getter returns from typing import Collection, Set
class X():
@property
def hello(self) -> Set[str]:
return {"x", "y"}
@hello.setter
def hello(self, value: Collection[str]) -> None:
pass
x = X()
x.hello = ["1", "2", "3"] % mypy --version
mypy 0.701
% mypy mypy3004.py
mypy3004.py:13: error: Incompatible types in assignment (expression has type "List[str]", variable has type "Set[str]") In my current case, I'm being even stricter in the getter, returning |
Descriptors actually support different types as one would expect. However, properties were implemented before descriptors (using some heavy special-casing) so they don't support this. I think this is a valid feature to support. I have seen this a lot recently in internal code bases, mostly in context similar to example by @srittau (canonical representation). This however may be tricky to implement (because of the special-casing I mentioned). Maybe a better strategy would be to just make property a regular descriptor in mypy (I believe we already have an issue for that), then this will be supported automatically. |
FTR, the main issue about properties is #220 |
Here's another example that I don't find terribly perverse. Using class A:
def __init__(self, label=None):
# type: (Optional[str]) -> None
self._user_label = None # type: Optional[str]
self.label = label
@property
def label(self):
# type: () -> str
return self._user_label or self.default_label()
@label.setter
def label(self, value):
# type: (Optional[str]) -> None
self._user_label = value
def default_label(self):
# type: () -> str
return self.__class__.__name__ |
I am going to remove the "needs discussion" label. IMO it is now pretty clear we should support this, the only discussion is what is the best way to implement this. |
perhaps I'm missing sth, but doesn't that mean you can get class X: pass
class Foo:
@property
def foo(self) -> int:
...
@foo.setter
def foo(self, o: Union[X, int]) -> None:
...
foo = Foo()
x = X()
foo.bar = x
assert foo.bar != x I wonder if there are cases where this makes sense (e.g. where equality still holds - perhaps this can work in the Iterable/Set example), but others where I also feel it seems odd e.g. using My request would be for there to be a flag to turn this on, and it to be off by default |
I think this is the problem I am encountering now with mypy 0.781. Here's my distilled example: from datetime import timedelta
from typing import Union
Interval = Union[timedelta, int]
class Foo:
def __init__(self):
self._x = timedelta(seconds=15)
@property
def x(self) -> timedelta:
return self._x
@x.setter
def x(self, delta: Interval) -> None:
if isinstance(delta, timedelta):
self.x = delta
else:
self.x = timedelta(seconds=delta)
foo = Foo()
foo.x = 7
foo.x = timedelta(seconds=8) And the error I'm getting:
The idea behind the Seems like this is the same problem described in this issue. I can understand that mypy may not be able to infer this. |
@joelberkeley-pio, what you're asking for is a Python feature, not a MyPy feature. This is part of the whole point of properties -- to divorce the getters and setters from exposing bare values of members, so we don't need to have For example, this is also valid Python (and, apart from the contrived simplification, actually used in some libraries): from typing import Union
import requests
class RemoteObject:
def __init__(self, object_id: str):
self.object_id = object_id
@property
def value(self) -> float:
return float(requests.get(f"https://example.com/{self.object_id}/value").text)
@value.setter
def value(self, new_value: Union[int, float]) -> None:
requests.post(f"https://example.com/{self.object_id}/value",
data=str(new_value).encode("utf-8")) Anything could happen on the remote server between the time the value is set and retrieved. There is absolutely no expectation that the values should be identical. |
For the record, this can be done at least in the class Example:
def __init__(self, value=None):
self._value = frozenset()
type(self).value.fset(value)
@property
def value(self) -> frozenset[int]:
return self._value
@value.setter
def value(self, value: Iterable[float, int] | None) -> None:
if value is None:
self._value = frozenset()
else:
self._value = frozenset(int(_) for _ in value) That said, I would argue that in this case it should be |
As a counter-example to the "perverse" comment, let me provide a use case of why this is needed. Pandas dataframes have a More details here. |
I was surprised to run into this issue and then read Guido's message calling such functionality "perverse". There is also precedent: for example, My use case is a vector library where a user might, say, zero-initialize a component of a 2D vector:
These vectors are themselves vectorized, i.e., the "x" component is a slice of a larger data structure where the setter performs an implicit conversion that MyPy sadly cannot type-check. The alternative of having to put explicit casts everywhere is surprisingly ugly. So this is all to say that I think the "perverse" comment is not seeing the big picture. |
Thank you @wjakob (and @amgcc) -- I think that the initial "perverse" comments have softened and that there is general consensus now that different getter and setter types (like in Typescript and other languages) are desirable both in numeric programming and in other areas. What's needed to move this issue from Green (Open) to Purple (Merged) are (1) people who know the typing system of MyPy and Pyright and other systems who can make implementations, and (2) technical writers who can explain the changes/situation to others. Given the huge goodwill of the community that would come from getting these changes in, are there people who would volunteer for either (1) or (2)? |
This is intended to support contravariant setters (e.g. fget returning a concrete container, but fset taking any collection), for the case that mypy eventually supports them (see: python/mypy#3004)
Delving into the code a bit, it looks like the important bit is This line finds the type of the variable being assigned to. Line 2902 in bcb3747
Down here expands member assignment. Lines 2987 to 2990 in bcb3747
Here are some examples of the current implementation. Normal variableclass Test:
val: int
t = Test()
t.val = 2.5
Custom descriptorfrom __future__ import annotations
from typing import Any, overload
class MyInt:
@overload
def __get__(self, obj: None, objtype: None) -> MyInt: ...
@overload
def __get__(self, obj: object, objtype: type[object]) -> int: ...
def __get__(self, obj: Any, objtype: Any = None) -> int | MyInt:
return 1
def __set__(self, obj: Any, value: int | float) -> None:
pass
class Test:
val = MyInt()
t = Test()
t.val = 2.5
Propertyclass Test:
@property
def val(self) -> int:
return 1
@val.setter
def val(self, val: int | float) -> None:
pass
t = Test()
t.val = 2.5
I would have thought it would be more correct if Edit: Here is the part where it pulls the getter return value. Lines 314 to 318 in bcb3747
Edit2: I think if |
mypy doesn't use the type of setters as of 1.9.0 (see python/mypy#3004), but I think it's still good to have these be accurate (maybe the other type checkers work better here).
mypy doesn't use the type of setters as of 1.9.0 (see python/mypy#3004), but I think it's still good to have these be accurate (maybe the other type checkers work better here).
mypy doesn't use the type of setters as of 1.9.0 (see python/mypy#3004), but I think it's still good to have these be accurate (maybe the other type checkers work better here).
mypy doesn't use the type of setters as of 1.9.0 (see python/mypy#3004), but I think it's still good to have these be accurate (maybe the other type checkers work better here). mypy's recommendation is to use `# type: ignore` comments if setter types don't match getters, which you see when setting no_cache to True.
- Ignore None type in setter for descriptor_list to enable optional input validation and an optional None. - See python/mypy#3004 as reference.
Cache-Control no-transform directive is a boolean no-transform has no arguments as a request or response directive (RFC 9111). Prior to this fix, cc.no_transform would return None whether the directive is present or not. Cache-Control min-fresh directive requires argument The type for this property is `int | None`, so getting `"*"` for a malformed directive is surprising. I think dropping the empty value here is better than fixing the type. Fix CacheControl getter type stubs - cache_control_property with type=bool never return None - some non-bool types were marked as returning bool instead of str - max_stale can return "*" in addition to int or None Reflect immutability of RequestCacheControl in type stubs Fix CacheControl setter type stubs mypy doesn't use the type of setters as of 1.9.0 (see python/mypy#3004), but I think it's still good to have these be accurate (maybe the other type checkers work better here). mypy's recommendation is to use `# type: ignore` comments if setter types don't match getters, which you see when setting no_cache to True. Support must-understand response directive
Consider this code:
Whatever we decide, I'm happy to build a PR; I have this code loaded into my head.
The text was updated successfully, but these errors were encountered: