Support Pydantic Serialization for CloudPath

[mattijsdp]

CloudPath's are supported in Pydantic BaseModel's but when you try to serialize them I get a PydanticSerializationError: Unable to serialize unknown type.

I think it would be an easy fix to add. Note that Path serialization is supported.

Reproducible code snippet:

from pydantic import BaseModel
from cloudpathlib import CloudPath

class SomeModel(BaseModel):
    field_a: CloudPath

SomeModel(field_a="s3://bucket/key").model_dump_json()

Fix required, in the below function add a serialization parameter to no_info_after_validator_function: https://github.com/drivendataorg/cloudpathlib/blob/28f1d943078517262d045ae50136edf27a5298a1/cloudpathlib/cloudpath.py#L1587-L1598

i.e.

return core_schema.no_info_after_validator_function( 
   cls.validate, 
   core_schema.any_schema(),
   serialization=core_schema.plain_serializer_function_ser_schema(
        lambda x: str(x),
        return_schema=core_schema.str_schema(),
    ),
)

Easy workaround for now is to add a PlainSerializer:

from pydantic import PlainSerializer, BaseModel
from cloudpathlib import CloudPath
from typing import Annotated

class SomeModel(BaseModel):
    field_a: Annotated[CloudPath, PlainSerializer(lambda x: str(x))]

SomeModel(field_a="s3://bucket/key").model_dump_json()

Let me know if you agree with adding this. Only just started using the library but I think it's a pretty easy fix so can probably help contribute.

[pjbull]

That seems reasonable, thanks @mattijsdp.

One question: will library users still be able to override the str serialization if they want? For example, they may prefer getting the http url or a presigned url at serialization time rather than the string representation. If so, let's add an example to the docs with that as well.

[mattijsdp]

@pjbull yes, you can customise serialization. I added an example to the docs. Let me know what you think.

[pjbull]

Thanks @mattijsdp! This will be in the next release.

[mattijsdp]

Thank you!