Create ConfDict using json-mode model dump

[PierreGtch]

Hi @jrapin @kingjr,

Super cool library :D

I was playing a bit, but my model had some non-simple types (like date time.timedelta or pydantic.ImportString) that are not compatible with exec.ConfDict' s to_yaml and to_uid methods.

See, before fix:

exca/test_utils.py:417: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
exca/confdict.py:200: in to_yaml
    out: str = _yaml.safe_dump(_to_simplified_dict(self), sort_keys=True)
../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/__init__.py:269: in safe_dump
    return dump_all([data], stream, Dumper=SafeDumper, **kwds)
../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/__init__.py:241: in dump_all
    dumper.represent(data)
../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/representer.py:27: in represent
    node = self.represent_data(data)
../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/representer.py:48: in represent_data
    node = self.yaml_representers[data_types[0]](self, data)
../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/representer.py:207: in represent_dict
    return self.represent_mapping('tag:yaml.org,2002:map', data)
../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/representer.py:118: in represent_mapping
    node_value = self.represent_data(item_value)
../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/representer.py:58: in represent_data
    node = self.yaml_representers[None](self, data)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <yaml.dumper.SafeDumper object at 0x157555030>
data = datetime.timedelta(seconds=60)

    def represent_undefined(self, data):
>       raise RepresenterError("cannot represent an object", data)
E       yaml.representer.RepresenterError: ('cannot represent an object', datetime.timedelta(seconds=60))

../../miniforge3/envs/exca/lib/python3.10/site-packages/yaml/representer.py:231: RepresenterError

This PR sets model_dump(..., mode="json"), which uses pydantic's logic to convert many types to JSON-compatible ones.

All the current tests pass. Could this change cause issues at non-tested places?

[kingjr]

Would type casting work? e.g.

from pydantic import BaseModel, field_validator
from datetime import timedelta

class MyModel(BaseModel):
    duration: str

    @field_validator("duration", mode="before")
    @classmethod
    def validate_duration(cls, value):
        if isinstance(value, timedelta):
            return str(value)
        raise ValueError("duration must be a timedelta")

MyModel(duration=timedelta(days=1, hours=2, minutes=30)).duration

[PierreGtch]

@kingjr this type casting would work but it would mean re-implementing many functionalities already present in pydantic.

For example, this is quite handy:

import pydantic
import math
import datetime

class MyModel(pydantic.BaseModel):
    x: pydantic.ImportString
    y: datetime.timedelta

# Call with strings:
MyModel(x="math.sqrt", y="PT1M")
# MyModel(x=<built-in function sqrt>, y=datetime.timedelta(seconds=60))

# Or with the actual objects:
MyModel(x=math.sqrt, y=datetime.timedelta(minutes=1))
# MyModel(x=<built-in function sqrt>, y=datetime.timedelta(seconds=60))

# Converted back to strings to be JSON-compatible:
MyModel(x=math.sqrt, y=datetime.timedelta(minutes=1)).model_dump(mode='json’)
# {'x': 'math.sqrt', 'y': 'PT1M'}

See all the types natively supported by pydantic for such casting to json-compatible formats:

• https://docs.pydantic.dev/latest/api/standard_library_types/
• https://docs.pydantic.dev/latest/api/types/
• https://docs.pydantic.dev/latest/api/networks/

[jrapin]

Hi @PierreGtch
Thanks for the contribution, I was not aware of this option and that's quite handy indeed + I may be able to remove some special yaml handling of Path type in confdict.py.