proper support of __get_pydantic_core_schema__

[jbkoh]

Summary

I'd like to dynamically add non-public methods to my Enums

What is the feature request for?

The core library

The Problem

I'm mixing up betterproto.Enum with pydantic classes for easier conversion between my database objects and protobuf messages. To use pydantic, I have to add validator based on this instruction. Thus, I decided to dynamically add a class method to a generated betterproto.Enum. However, betterproto.enum does not allow it by this

Are there any other parts of the code affected by adding a custom function __get_pydantic_core_schema__ to my generated Enum?

The Ideal Solution

I'd like __set_attr__ to accept new attributes if the new attributes are private (i.e.,starting with __) I can contribute if this direction is acceptable.

The Current Solution

If no other solutions, I will fork this repo and just remove the constraint. I don't like to fork for a hack.

[Gobot1234]

Would it be more useful to just add support in the root Enum for this? Can you not serialise an Enum without a custom implementation even though it's just a custom int?

[jbkoh]

I actually misread the original guideline from Pydantic and just found that I don't need to modify betterproto, even though if we can add such a handler to the codegen step to betterproto.Enum, I can avoid some headaches. Can we add a custom handler in the generation step as a configuration? I guess there are many other cases to "augment" the generated classes with some helper functions.

In other words, yes, it'd be useful to add this to the root Enum. I just don't know if betterproto users would like a dependency to pydantic. I was dumb that betterproto already has a dependency to it.

I didn't fully get the other question.

Can you not serialise an Enum without a custom implementation even though it's just a custom int?

[Gobot1234]

For the other question I was mainly just surprised that pydantic can't do this already

[jbkoh]

While I wass trying to add this to Enum class, I found that --custom_opt=pydantic_dataclasses already exists and there __get_pydantic_core_schema__ is inserted as found in tests/output_betterproto_pydantic/enum/__init__.py. However, when I activated that option in my codegen, that pydantic function wasn't added. Any idea what I should look at?

Following is the command I used:

          protoc \
            --plugin=protoc-gen-custom=${betterproto_loc} \
            --custom_opt=pydantic_dataclasses \
            --python_out=packages/python \
            --experimental_allow_proto3_optional \
            --python_betterproto_out=packages/python \
            --proto_path=./ $(find myproject -iname "*.proto")

[jbkoh]

Found that I was using a wrong option. Somehow test code uses --custom_opt and the README says --python_betterproto_opt. It's all good now. Closing.

[jbkoh]

@Gobot1234

I actually need more help than what it has right now because it currently just treats betterproto.Enum as int at the Pydantic side. The currently generated code snippet is

    @classmethod
    def __get_pydantic_core_schema__(cls, _source_type, _handler):
        from pydantic_core import core_schema
        return core_schema.int_schema(ge=0)

and binds an enum field as an int. However, I want to preserve the custom enum field. I actually needed one in the original link, something like below.

        from pydantic_core import core_schema
        def validate(value: int) -> "MyCustomEnum":
            return cls(value)

        # Return the schema for validation and serialization
        return core_schema.chain_schema(
            [
                core_schema.int_schema(ge=0),  # Validate as a string first
                core_schema.no_info_plain_validator_function(validate),  # Custom validation
            ]
        )

It seems to be desired in general, and I'm curious if we can update the generation snippet to this.

[jbkoh]

This PR works for me, though I can add more tests to it:
#651