dataclass field doesn't change required schema

[ahirner]

Bug

Please complete:

• OS: Mac OS
• Python version import sys; print(sys.version): 3.6.7 with dataclasses backport
• Pydantic version import pydantic; print(pydantic.VERSION): 1.0

It is known that @pydantic.dataclasses.dataclass doesn't fully mimic the BaseModel. However, we'd like use .schema() representation faithfully. I'm wondering what's the best workaround.

Adding a Config for b with required=False doesn't carry over to .schema() either.

from typing import Dict

from dataclasses import field
from pydantic import Field, ValidationError
from pydantic.dataclasses import dataclass

@dataclass
class D1:
    a: int
    # b: Dict[str, int] = {"x":10} # mutable default, not allowed
@dataclass
class D2:
    a: int
    b: Dict[str, int] = Field({"x":10}) # pydantic.Field, unsafe and invalid default
@dataclass        
class D3:
    a: int
    b: Dict[str, int] = field(default_factory=lambda: {"x":10}) # std lib, safe and invalid default
        
class C:
    fields = {"b": { "required": False}} # doesn't carry over!!

@dataclass(config=C)        
class D4:
    a: int
    b: Dict[str, int] = field(default_factory=lambda: {"x":10})

print(D1(10))
try:
    print(D2(10))
except ValidationError as e:
    print(e) # pydantic/dataclasses.py:77

print(D3(10))
req = D3.__pydantic_model__.schema()["required"]
if req != ["a"]:
    print('only ["a"] should be required, is:', req)
    
print()
print(D4.__pydantic_model__.schema()["required"]) # same as D3

Output

D1(a=10)
1 validation error for D2
b
  value is not a valid dict (type=type_error.dict)
D3(a=10, b={'x': 10})
only ["a"] should be required, is: ['a', 'b']

['a', 'b']

[samuelcolvin]

This isn't a bug, but a feature request.

class Config:
    fields = {"b": { "required": False}}

Wouldn't work on a normal model.

I think what you want is better support for the field() function from dataclasses. If so, please describe exactly what parts of field() usage you'd like to support in pydantic. Is it just the default in schema?

[ahirner]

Thanks for clarifying!
Only handling default* in schema is what we need to support. However, it would be nice to have a stable interface for .__pydantic_model__. I'm not sure if this might change.

[samuelcolvin]

__pydantic_model__ won't change, we should document it.

It would be wonderful if you could create a PR for that. 😄

[ahirner]

i'm going for it