Type annotation parsing warning for Pydantic BaseModel private attributes

[DawnEver]

Problem Description

When using pdoc to generate documentation for Pydantic BaseModel classes, warnings appear about parsing type annotations for __pydantic_private__ and __pydantic_extra__ attributes. The warning indicates that a 'function' object is not subscriptable when trying to parse the type annotations dict[str, Any] | None.

This occurs even with simple Pydantic model definitions and affects the documentation generation process.

Steps to reproduce the behavior:

1. Install pdoc and pydantic
2. Create a Python file with the following content:

from pydantic import BaseModel

class Foo(BaseModel):
    """Test model"""
    bar: int

class Bar(BaseModel):
    """Another test model"""
    foo: Foo

3. Run pdoc foobar.py, warning appears:

Warn: Error parsing type annotation dict[str, Any] | None for pydantic.main.BaseModel.__pydantic_private__: 'function' object is 
not subscriptable(.../lib/python3.13/site-packages/pdoc/doc_types.py:123)
Warn: Error parsing type annotation dict[str, Any] | None for pydantic.main.BaseModel.__pydantic_extra__: 'function' object is not subscriptable(.../lib/python3.13/site-packages/pdoc/doc_types.py:123)

System Information

pdoc: 15.0.1
Python: 3.13.2
Platform: macOS-10.16-x86_64-i386-64bit-Mach-O
pydantic: 2.10.6

[mhils]

I think pydantic is mainstream enough that it's worth to support this - contributions welcome!

[jinnovation]

I notice that "normal" annotations like dict[str, Any] | None in pydantic.BaseModel-inherited classes get converted into ForwardRefs as part of pdoc.doc_types._eval_type(). This is, itself, related to the annotation getting passed into _eval_type() as a string rather than as the type-annotation type itself, e.g. <class 'types.UnionType'>.

Notably, this does not happen with, for example, with vanilla Python dataclasses.

Could be worth looking into why the annotations for Pydantic classes get handled as though they were strings.

[jinnovation]

I note also the following:

1. Both of the problematic fields that @DawnEver cites are dicts
2. There is this very relevant-sounding disclaimer in Pydantic code regarding dict being in the namespace of BaseModel -- namely, as a deprecated dict() method

Rewriting both of the offending annotations as typing.Dict[str, Any] rather than the current dict[str, Any] seems to do the trick. This suggests that the fix here should be a PR to Pydantic itself.