Avoid __dict__ and __weakref__ attributes in AnyUrl and _BaseAddress subclasses

[nuno-andre]

Change Summary

Add an empty __slots__ declaration to AnyUrl and _BaseAddress subclasses to avoid the creation of __dict__ and __weakreef__ attributes. When inheriting from a class with __slots__, child subclasses will get a __dict__ and __weakref__ unless they also define __slots__.

from pydantic import AnyUrl, AnyHttpUrl

class AnyOtherHttpUrl(AnyUrl):
    allowed_schemes = {'http', 'https'}
    __slots__ = ()

print(set(dir(AnyHttpUrl)) - set(dir(AnyUrl)))
#> {'__weakref__', '__dict__'}

print(set(dir(AnyOtherHttpUrl)) - set(dir(AnyUrl)))
#> set()

Related issue number

None.

Checklist

• Unit tests for the changes exist
• Tests pass on CI and coverage remains at 100%
• Documentation reflects the changes where applicable
• changes/<pull request or issue id>-<github username>.md file added describing change
  (see changes/README.md for details)
• My PR is ready to review, please add a comment including the phrase "please review" to assign reviewers

[PrettyWood]

Hi @nuno-andre
What's the matter with having __dict__ and __weakref__?
What problem does it solve?
AFAIR I haven't seen any issue related to this 🤔

[nuno-andre]

Hi @PrettyWood!

One of the __slots__'s purposes is to deny the creation of __dict__ to save memory.

from pydantic import AnyUrl, AnyHttpUrl
from pympler.asizeof import asizeof


class AnyOtherHttpUrl(AnyUrl):
    allowed_schemes = {'http', 'https'}
    __slots__ = ()


a = AnyUrl.__new__(AnyHttpUrl, 'https://example.com')
print(asizeof(a))
#> 224

b = AnyUrl.__new__(AnyOtherHttpUrl, 'https://example.com')
print(asizeof(b))
#> 120

from pydantic import BaseModel


class A(BaseModel):
    url: AnyHttpUrl


class B(BaseModel):
    url: AnyOtherHttpUrl


a = A(url='https://example.com')
print(asizeof(a))
#> 808

b = B(url='https://example.com')
print(asizeof(b))
#> 704

I think this is a pretty straightforward and valuable optimization and also makes a more consistent type model.

[samuelcolvin]

Seems good to me, I can't see how this will break anyone's code. But I'm sure we'll have someone shouting at us about breaking code in a few weeks from some weird usage.

[tl24]

Let the shouting begin... 😃 We had a breaking change as we were reading the slots to be able to clone a RedisDsn and add in a password that was stored in a secret. We worked around by hard-coding the property names, but is there a better way of doing that?

For reference:

def copy_dsn(original, **override_args):
    # object properties don't show up with normal methods,
    # so we need to use the __slots__ property to get field names
    attrs = {n: getattr(original, n) for n in original.__slots__ if getattr(original, n) is not None}
    return type(original)(None, **{**attrs, **override_args})

dsn_with_password = copy_dsn(RedisDsn("redis://localhost:6379/0", password="secret")

[tl24]

I don't know if this would be a frequent use case, but something like copy on BaseModel would be nice:

dsn_with_password = RedisDsn("redis://localhost:6379/0").copy(update={"password":"secret"})