Is self-referencing type alias supported?

[2AtM7]

Consider the following self-referencing type alias, defined for JSON-able objects (see this post for reference):

Json = dict[str, 'Json'] | list['Json'] | str | int | float | bool | None

I am kinda new to Pydantic, so not pretty sure if this is supported. But apparently the following code snippet did not work for me. I'm using Python 3.10.2 with pydantic 1.10.7.

from pydantic import BaseModel


Json = dict[str, 'Json'] | list['Json'] | str | int | float | bool | None


class Foo(BaseModel):
    str_attr: str
    int_attr: int
    json_attr: Json


Foo(str_attr='abc', int_attr=123, json_attr={'key': [1, False, None]})  # Should work fine
Foo(str_attr='abc', int_attr=123, json_attr={1: 2})  # Should fail the validation

The following error was thrown:

Traceback (most recent call last):
  File "/path/to/test.py", line 7, in <module>
    class Foo(BaseModel):
  File "pydantic/main.py", line 286, in pydantic.main.ModelMetaclass.__new__
  File "pydantic/main.py", line 808, in pydantic.main.BaseModel.__try_update_forward_refs__
  File "pydantic/typing.py", line 553, in pydantic.typing.update_model_forward_refs
    """
  File "pydantic/typing.py", line 528, in pydantic.typing.update_field_forward_refs
    arg = _type_check(parameters, f"{self} requires a single type.")
  File "pydantic/typing.py", line 528, in pydantic.typing.update_field_forward_refs
    arg = _type_check(parameters, f"{self} requires a single type.")
  File "pydantic/typing.py", line 528, in pydantic.typing.update_field_forward_refs
    arg = _type_check(parameters, f"{self} requires a single type.")
  [Previous line repeated 966 more times]
  File "pydantic/typing.py", line 524, in pydantic.typing.update_field_forward_refs
    """Optional type.
  File "pydantic/fields.py", line 552, in pydantic.fields.ModelField.prepare
  File "pydantic/fields.py", line 663, in pydantic.fields.ModelField._type_analysis
  File "pydantic/fields.py", line 808, in pydantic.fields.ModelField._create_sub_type
  File "pydantic/fields.py", line 436, in pydantic.fields.ModelField.__init__
  File "pydantic/fields.py", line 552, in pydantic.fields.ModelField.prepare
  File "pydantic/fields.py", line 724, in pydantic.fields.ModelField._type_analysis
  File "pydantic/fields.py", line 808, in pydantic.fields.ModelField._create_sub_type
  File "pydantic/fields.py", line 436, in pydantic.fields.ModelField.__init__
  File "pydantic/fields.py", line 557, in pydantic.fields.ModelField.prepare
  File "pydantic/fields.py", line 834, in pydantic.fields.ModelField.populate_validators
  File "pydantic/class_validators.py", line 282, in pydantic.class_validators.prep_validators
  File "pydantic/class_validators.py", line 256, in pydantic.class_validators.make_generic_validator
  File "/usr/local/lib/python3.10/inspect.py", line 3247, in signature
    return Signature.from_callable(obj, follow_wrapped=follow_wrapped,
  File "/usr/local/lib/python3.10/inspect.py", line 2995, in from_callable
    return _signature_from_callable(obj, sigcls=cls,
  File "/usr/local/lib/python3.10/inspect.py", line 2456, in _signature_from_callable
    return _signature_from_function(sigcls, obj,
  File "/usr/local/lib/python3.10/inspect.py", line 2321, in _signature_from_function
    parameters.append(Parameter(name, annotation=annotation,
  File "/usr/local/lib/python3.10/inspect.py", line 2632, in __init__
    self._kind = _ParameterKind(kind)
  File "/usr/local/lib/python3.10/enum.py", line 385, in __call__
    return cls.__new__(cls, value)
RecursionError: maximum recursion depth exceeded while calling a Python object

Seems like it hits the max recursion depth of Python when trying to deal with the Json type alias. Am I missing the right way to do it, or self-referencing type aliases are not yet supported by Pydantic? If the latter, then this could turn into a feature request.

Thanks for the help in advance!

[dmontagu]

The short answer is — not yet. But it's something we do want to add support for.

Once the type alias syntax introduced in this PEP https://peps.python.org/pep-0695/#generic-type-alias is introduced, I think that will be the obvious way to handle it, though that will require python 3.12+.

Also, it looks like there is active work toward backporting the TypeAliasType into typing_extensions, and once it is added, I think that will be the obvious way to support it in earlier python versions. See python/typing_extensions#159 and python/typing_extensions#160 for reference.

For what it's worth, I think the pydantic implementation will probably look a lot like this: ef5dd26 which was part of an attempt I gave up on after subsequently reworking it to try to automatically infer the alias name, which was misguided.)

So basically, stay tuned!

[dmontagu]

Also, see #5171 for a tracking issue