feat: avoid dict coercion for subclasses

[PrettyWood]

Change Summary

When using a subclass of dict as type of a field, the shape of the field was still considered SHAPE_DICT. When a value was validated, it was hence coerced to dict here
Now we set the shape to SHAPE_MAPPING in those cases and keep target class

Related issue number

closes #3122

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]

please review

[JacobHayes]

deleted bad suggestion

Maybe:

        elif issubclass(origin, (dict, Dict)):

?

further investigation

Hmm the origin is dict or origin is Dict does fix the conversion to dict, but I still get the AttributeError: name for the subclass in the example in #3122. 🤔 Maybe that's a separate problem though.

...

One additional piece of info: it looks like type(Base.__fields__["b"].outer_type_) is <class 'types.GenericAlias'>, but type(Sub.__fields__["b"].outer_type_) is <class 'dict'>. 🤔

It looks like maybe .outer_type_ is getting replaced with some other dictionary instance somehow in subclass creation?

(Pdb) m.__fields__["b"].outer_type_.keys()
dict_keys([4336358208, 4340970304, 4341193536, 4341045632, 4341212736, 4340612832, 4341193344, 4341212480, 4341239232, 4341200384, 4341200448, 4340970688, 4341193856])
(Pdb) m.__fields__["b"].outer_type_.values()
*** AttributeError: name

Update:

This PR does seem to fix the frozendict[str, int] -> dict conversion for the base class (yay, thanks)! However, there is a bug with the upstream frozendict package (prompted by the deepcopy of __fields__ on subclass creation):

>>> from frozendict import frozendict
>>> from copy import copy, deepcopy
>>>
>>> deepcopy(frozendict)
<class 'frozendict.core.frozendict'>
>>> deepcopy(frozendict[str, int])
{}

I'll open a separate ticket there.

[PrettyWood]

Hi @JacobHayes
Thanks for the review
So yes what didn't work before is

m = Base(a=frozendict(a=1), b=frozendict(b='1'))
assert isinstance(m.a, frozendict)
assert isinstance(m.b, frozendict)
assert m.b['b'] == 1  # coerced

This is now resolved because the frozendict is not coerced in dict.
The 'name' issue is something else indeed.

EDIT: Great that you found the root cause! I let you see on frozendict side or monkeypatch it

[JacobHayes]

In case you're interested, this escalated to a cPython bug with types.GenericAlias: https://bugs.python.org/issue45167

[JacobHayes]

@PrettyWood do you know if this might make it into the next release? 🙏

[PrettyWood]

It should

[samuelcolvin]

otherwise looks good, just not clear what happens in these cases

[samuelcolvin]

Suggested change

	b: MyDict[str, int]
	b: MyDict[str, int]
	c: Dict[str, int]
	d: Mapping[str, int]

What happens to these types with the input being a MyDict?

[PrettyWood]

for c coercion is done (as currently expected with pydantic) and for d no coercion is done. I added them in the test

[samuelcolvin]

please update

[PrettyWood]

Please review

[samuelcolvin]

awesome, thank you.