Improve handling of GroupedMetadata

[Zac-HD]

Fixes #3980, closes #3981.

Motivated by pydantic/pydantic#4682 (comment); supporting nested-GroupedMetadata and short-circuiting if such an item contains an st.SearchStrategy will allow for extensive control purely from downstream code.

[Viicos]

Just for my understanding: is this addition (supporting nested grouped metadata) related to pydantic/pydantic#4682 (comment), or is it just something you added as an enhancement?

[Zac-HD]

Unrelated bugfix, really.

[Viicos]

For the needs of the Pydantic plugin, this looks good! thx

[Viicos]

Sorry I misread some bits of the PR, I'm just a bit worried about defining an __iter__ method that yields hypothesis strategies on the Pydantic types because:

• It breaks type checking, as GroupedMetadata.__iter__ is typed as Iterator[BaseMetadata], so yielding a strategy results in type checking error.
• It feels a bit weird to have an hypothesis integration implemented on the type class directly, especially if we need to do:

@_dataclasses.dataclass(**_internal_dataclass.slots_true)
class UuidVersion(annotated_types.GroupedMetadata):
    """A field metadata class to indicate a [UUID](https://docs.python.org/3/library/uuid.html) version."""

    uuid_version: Literal[1, 3, 4, 5]

    def __get_pydantic_json_schema__(
        self, core_schema: core_schema.CoreSchema, handler: GetJsonSchemaHandler
    ) -> JsonSchemaValue:
        ...

    def __get_pydantic_core_schema__(self, source: Any, handler: GetCoreSchemaHandler) -> core_schema.CoreSchema:
        ...

    def __iter__(self):
        if "hypothesis" in sys.modules:
            yield some_strategy
        else:
            # What to do? Raise an error?

However, it seems pretty common to have this pattern, e.g. defining __rich_repr__.

I'll start the implementation on the Pydantic side, and see if they are ok with it!

[Zac-HD]

It breaks type checking, as GroupedMetadata.__iter__ is typed as Iterator[BaseMetadata], so yielding a strategy results in type checking error.

The readme types it as Iterator[object], and says "libraries consuming annotated-types constraints should check for GroupedMetadata and unpack it by iterating over the object and treating the results as if they had been "unpacked" in the Annotated type".

So actually I have two followups:

1. fix the annotation in the definition of GroupedMetadata to match the docs (in annotated-types)
2. add another clause to handle Unpack the same way as GroupedMetadata (in Hypothesis)

[Viicos]

Ah yes sorry was looking at the actual type definition indeed!

[dvir478]

I'm partner dvir478 is working on issue #3986

[Esra-Al]

I am Esra-Al and working on issue #3986