dataclass_transform alias does not work with python version 3.11

[RobertCraigie]

Describe the bug

Pyright does not use the given alias for the synthesised __init__ method if pythonVersion is set to 3.11.

To Reproduce

from pydantic import BaseModel, Field

class Model(BaseModel):
    prettyName: str = Field(alias='name')

print(Model(name='foo'))
reveal_type(Model.__init__)

With this pyproject.toml file:

[tool.pyright]
pythonVersion = "3.11"
typeCheckingMode = "strict"

And the third party library used, pydantic==1.10.2.

This is what pyright outputs:

test.py
  test.py:8:13 - error: No parameter named "name" (reportGeneralTypeIssues)
  test.py:9:13 - information: Type of "Model.__init__" is "(self: Model, *, prettyName: str = Field(alias='name')) -> None"
1 error, 0 warnings, 1 information 

It should be noted that I also encountered this error with pythonVersion being left unset in the config file.

Expected behavior

Pyright should not error and the revealed type should be:

(self: Model, *, name: str) -> None

VS Code extension or command-line

Command-line, v.1.1.269.

[erictraut]

This is a bug in pydantic's use of dataclass_transform. Prior to the finalization of PEP 681, we changed the name of one of the parameters from field_descriptors to field_specifiers. This was based on feedback. Pydantic is still using the old (obsolete) name in its code.

@dataclass_transform(kw_only_default=True, field_descriptors=(Field, FieldInfo))

It needs to be updated to match the PEP. Please file a bug with the pydantic maintainers. In the meantime, you can make the change manually in your copy of pydantic sources if you want this to work.

I'll also see if there's a short-term workaround that I can make in pyright to help people who are using the current pydantic version.

[erictraut]

The maintainer of pydantic confirmed that this will be fixed in a dot release within the next several days. pydantic/pydantic#4500