Create model using a function signature

[antoniomdk]

Feature Request

Hi everyone, first things first, thank you for this fantastic tool you're developing.

I think it would be interesting to create a model dynamically based on a function signature.
In my case, I have some kind of function runners: functions that receive a function and
call it adding decorators, etc. I've created the following snippet to create a model from
a function signature. If you're interested in this feature, I can create a PR for it.

Output of python -c "import pydantic.utils; print(pydantic.utils.version_info())":

             pydantic version: 1.4
            pydantic compiled: False
                 install path: /Users/antonio/Documents/TFG-AntonioMolner/env/lib/python3.7/site-packages/pydantic
               python version: 3.7.7 (default, Mar 10 2020, 15:43:33)  [Clang 11.0.0 (clang-1100.0.33.17)]
                     platform: Darwin-19.2.0-x86_64-i386-64bit
     optional deps. installed: []

import pydantic

def create_model_from_signature(func: Callable,
                                model_name: str,
                                base_model: Type[BaseModel] = BaseModel):
    args, _, varkw, defaults, kwonlyargs, kwonlydefaults, annotations = getfullargspec(func)
    defaults = defaults or []
    args = args or []

    non_default_args = len(args) - len(defaults)
    defaults = (...,) * non_default_args + defaults

    keyword_only_params = {param: kwonlydefaults.get(param, Any) for param in kwonlyargs}
    params = {param: (annotations.get(param, Any), default) for param, default in zip(args, defaults)}

    class Config:
        extra = 'allow'

    # Allow extra params if there is a **kwargs parameter in the function signature
    config = Config if varkw else None

    return create_model(
        model_name,
        **params,
        **keyword_only_params,
        __base__=base_model,
        __config__=config,
    )
...

Example

In[12] m = create_model_from_signature(open, 'MyModel', BaseModel)

In[13] m.__fields__

Out[14]: 
{'file': ModelField(name='file', type=Optional[Any], required=True),
 'mode': ModelField(name='mode', type=Optional[Any], required=False, default='r'),
 'buffering': ModelField(name='buffering', type=Optional[Any], required=False, default=-1),
 'encoding': ModelField(name='encoding', type=Optional[Any], required=False, default=None),
 'errors': ModelField(name='errors', type=Optional[Any], required=False, default=None),
 'newline': ModelField(name='newline', type=Optional[Any], required=False, default=None),
 'closefd': ModelField(name='closefd', type=Optional[Any], required=False, default=True),
 'opener': ModelField(name='opener', type=Optional[Any], required=False, default=None)}

[antoniomdk]

I've realized there is a validate_arguments method, I'm closing this issue.

[danielbraun89]

validate_arguments serve a different use case. Personally I don't understand why not going forward with the original idea...

[marsninja]

This is a great idea @antoniomdk , though I agree with @danielbraun89 , I'm trying to build an automated function signature to base model to fastAPI request body and would need to extract the base model from function signatures. validate_arguments doesn't actually provide the base model it uses internally it appears, unless I've missed something.

[marsninja]

NVM, I just noticed:

https://github.com/samuelcolvin/pydantic/blob/5bc1fcfb92aace50f18bf4ddcc915ef23e1c7431/pydantic/decorator.py#L45

I can use this as the model, didn't see it in the documentation, LMK if putting in a PR in for the docs if useful.