Fix config behavior with plain dataclass

[sydney-runkle]

Fix #8693

Selected Reviewer: @hramezani

[cloudflare-pages]

Deploying pydantic-docs with    Cloudflare Pages

Latest commit:	90d6324
Status:	✅  Deploy successful!
Preview URL:	https://a23b477c.pydantic-docs2.pages.dev
Branch Preview URL:	https://fix-ta-with-dataclass-config.pydantic-docs2.pages.dev

View logs

[codspeed-hq]

CodSpeed Performance Report

Merging #9150 will not alter performance

_{Comparing fix_ta_with_dataclass_config (90d6324) with main (02ee0f9)}

Summary

✅ 13 untouched benchmarks

[sydney-runkle]

Please review

[davidhewitt]

Something seems strange here. We seem to have lots of tests in test_dataclasses.py which use config. Why are those passing?

[sydney-runkle]

@davidhewitt,

I believe it's because we're only testing validation focused settings, and the __pydantic_config__ is pulled off of a class when schema is generated, see

pydantic/pydantic/_internal/_generate_schema.py

Line 1231 in 02ee0f9

config: ConfigDict | None = get_attribute_from_bases(typed_dict_cls, '__pydantic_config__')

for example.

But in this case, we also make sure to pass said config to the constructors for SchemaValidator and SchemaSerializer

[sydney-runkle]

That being said, I think we could clean this up a bit, I can create an issue for that.

[davidhewitt]

Right I see. So I don't think this is the right solution, because as you say the config is already pulled from the type into the schema.

I think actually if there's a bug, it's probably in core. See https://github.com/pydantic/pydantic-core/blob/e73b2d1edc5c3c062f1e0b381cda29d6ed6e5901/src/serializers/type_serializers/dataclass.rs#L83, where we do pull the config out of the schema to build a dataclass serializer. But why is that config not applying?

[sydney-runkle]

Good question. 30 mins of digging didn't help me much, but I'd be happy to look into this further in core. I still think that we might want this change though, as the schema serializer is what's used in accordance with the relevant settings here?

[davidhewitt]

I think that in all cases the schema should already contain the config fetched here; in effect doing this seems to me like it plasters over a core bug by sending the same config in twice. So I'd rather we wait to understand the interaction in core before considering this further.