Generalize replace() function

Ideas
21

You’re not allowed to use P.kwargs in an annotation without P.args, so it will be hard for us to provide a precise signature for this function at typeshed in the way you suggest above. (FWIW, I agree that this is an awesome new feature, and don’t think that the fact that it’s difficult to have accurate type annotations for it is in any way an argument against adding the feature.)

1 Like
22

I hope it didn’t sound like I was suggesting that!

Just curious, but the dependency of one type variable on another is okay?

def replace[T: SupportsReplace[P], **P]

I wasn’t sure if that would work.

23

It didn’t – I was just clarifying that it also was not my opinion either :wink:

Er, yeah, that’s illegal too, good point :stuck_out_tongue:

1 Like
24

I proposed adding SupportsReplace protocol to typing here: Incorrect rendering of `__replace__` method in `copy` docs: consider adding `typing.SupportsReplace`? · Issue #109961 · python/cpython · GitHub :slight_smile:

**kwargs would have to be Any (for now at least).

1 Like
25

copy.replace is great, however I keep writing:

    def __replace__(self, **kwargs: Any) -> Self:  # Allow replacements.
        new = copy.deepcopy(self)
        for key, value in kwargs.items():
            setattr(new, key, value)
        return new

Why not add something like this to object?

1 Like
26
  1. It only works for mutable objects.
  2. You used deepcopy(), but in many case you don’t need to make copies of all attributes (they can even be uncopiable) or you only need to make a level 1 copy of an attribute. This cannot be generalized. deepcopy() is too rought tool.
  3. replace() usually does not allow setting arbitrary attributes.
  4. Constructor can have side effect, and you want this side effect happens with values specified as the keyword arguments of replace() rather than with original values.
2 Likes
27

@storchaka Yes to all your points, but that doesn’t prevent a useful default from being provided. The documentation could list all your points as reasons to override.

28

You could just use dataclasses if you want useful defaults for most things.

1 Like
29

@NeilGirdhar Yes, I use dataclasses a lot and pydantic, but there are still times when a normal class is a good choice.