Skip to content

Arrays

Name What it does
Shape the dimensions of an array field
DType the element type of an array field
SupportsDLPack an object a DLPack library can import; a field annotated with it reads back sharedbox's own array
register_array_type makes fields of another array type read back through that library's from_dlpack

Shape dataclass 

Shape(*dims: int)

Dimensions of an array field, fixed when the class is defined.

It goes in the annotation, as in Annotated[numpy.ndarray, Shape(480, 640), DType("uint8")].

Raises:

Type Description
ValueError

Unless there are 1 to 8 dimensions, each an int of at least 1.

DType dataclass 

DType(dtype: object)

Element type of an array field.

It takes a name as numpy spells it ("float32", "bfloat16", "complex64", "bool"), a numpy.dtype or scalar type, or a torch.dtype.

Raises:

Type Description
ValueError

For an element type DLPack cannot describe.

SupportsDLPack

Bases: Protocol

An object another library can import through DLPack, such as numpy.from_dlpack takes.

A field annotated with it reads back sharedbox's own array object, which implements it.

register_array_type 

register_array_type(array_type: type, from_dlpack: Callable[[SupportsDLPack], Any]) -> None

Read fields annotated with array_type back through from_dlpack.

from_dlpack receives the array a read copied out of the box and returns one of array_type. Register before defining a class that names the type; numpy and torch need no registration.