katana_public_api_client.models_pydantic._mapped_shim¶
katana_public_api_client.models_pydantic._mapped_shim
¶
Runtime identity shim for sqlalchemy.orm.Mapped.
The generated Cached* classes wrap each field type in Mapped[T] so
type checkers see CachedX.field as InstrumentedAttribute[T] (which
exposes .in_/.is_/.desc/.ilike). At runtime, however, SQLModel +
pydantic 2.13 reject Mapped[T] as a field type — pydantic's schema
generator can't unwrap it, so class definition raises
PydanticSchemaGenerationError.
This shim resolves the conflict by being a runtime identity version of
Mapped: at type-check time, Mapped resolves to the real
sqlalchemy.orm.Mapped; at runtime, Mapped[T] returns T itself
via __class_getitem__. The generated Annotated[Mapped[T], Field(...)]
shape therefore evaluates to Annotated[T, Field(...)] at runtime —
identical to the pre-shim behavior — while the type checker sees the
descriptor-bearing Mapped[T] shape.
Why a shim instead of natively typing fields as Mapped[T]: SQLModel
0.0.38 + pydantic 2.13 don't yet support that form, and the workarounds
(arbitrary_types_allowed=True plus per-field sa_type= overrides)
re-implement SQLModel's existing type inference at every field. The shim
is a 10-line solution that disappears once SQLModel ships native
Mapped[T] support.