numpy quad-precision float dtype

[d-v-b]

there's a numpy-compatible quad-precision float defined in this package. This issue tracks the creation of a Zarr data type to support saving these quad-precision float values to Zarr.

[SwayamInSync]

Hi @d-v-b, I’m Swayam (maintainer of numpy-quaddtype). Happy to help if needed.

[d-v-b]

hi Swayam! if you or anyone else is interested in taking on this project, the best place to start might be this example which shows the complete process of defining a new data type. Since the floats are pretty similar, most of the methods wouldn't require overrides: see

zarr-python/src/zarr/core/dtype/npy/float.py

Lines 392 to 425 in 3fa19d2

	class Float64(BaseFloat[np.dtypes.Float64DType, np.float64]):
	"""
	A Zarr data type for arrays containing 64-bit floating point numbers.
	Wraps the [`np.dtypes.Float64DType`][numpy.dtypes.Float64DType] data type. Scalars for this data type are instances
	of [`np.float64`][numpy.float64].
	Attributes
	----------
	dtype_cls : Type[np.dtypes.Float64DType]
	The NumPy dtype class for this data type.
	References
	----------
	This class implements the float64 data type defined in Zarr V2 and V3.
	See the [Zarr V2](https://github.com/zarr-developers/zarr-specs/blob/main/docs/v2/v2.0.rst#data-type-encoding) and [Zarr V3](https://github.com/zarr-developers/zarr-specs/blob/main/docs/v3/data-types/index.rst) specification documents for details.
	"""
	dtype_cls = np.dtypes.Float64DType
	_zarr_v3_name = "float64"
	_zarr_v2_names: ClassVar[tuple[Literal[">f8"], Literal["<f8"]]] = (">f8", "<f8")
	@property
	def item_size(self) -> int:
	"""
	The size of a single scalar in bytes.
	Returns
	-------
	int
	The size of a single scalar in bytes.
	"""
	return 8

[d-v-b]

we would also need a spec added to zarr-extensions, but it could be really short: see https://github.com/zarr-developers/zarr-extensions/tree/main/data-types/complex_bfloat16