support for return_schema on functional serializers

[samuelcolvin]

Companion to pydantic/pydantic-core#646.

[cloudflare-pages]

Deploying with    Cloudflare Pages

Latest commit:	e288232
Status:	✅  Deploy successful!
Preview URL:	https://7d8de78b.pydantic-docs2.pages.dev
Branch Preview URL:	https://support-return-serializer.pydantic-docs2.pages.dev

View logs

[dmontagu]

Would it make sense to change this to be more similar to @computed_field? There, we have
return_type: Any = _UNSET and docstring (which probably needs some updates):

    return_type: optional return for serialization logic to expect when serialising to JSON, if included
            this must be correct, otherwise a `TypeError` is raised.
            If you don't include a return type Any is used, which does runtime introspection to handle arbitrary
            objects.

[dmontagu]

and logic:

        if return_type is _UNSET:
            # try to get it from the type annotation
            res = _typing_extra.get_type_hints(_decorators.unwrap_wrapped_function(f), include_extras=True)
            return_type = res.get('return', _UNSET)
            if return_type is _UNSET:
                raise PydanticUserError(
                    'Computed field is missing return type annotation or specifying `return_type`'
                    ' to the `@computed_field` decorator (e.g. `@computed_field(return_type=int|str)`)',
                    code='model-field-missing-annotation',
                )

[samuelcolvin]

ye, probably good idea.

[dmontagu]

And during schema generation we just do

return_type_schema = self.generate_schema(d.info.return_type)

in pydantic._internal._generate_schema.GenerateSchema._computed_field_schema