Passing `xarray_reduce` parameter `dtype` not doing anything

[claytharrison]

When I try passing a dtype to xarray_reduce, the output is still uint64 for some reason:

import xarray as xr
from flox.xarray import xarray_reduce

ds = xr.Dataset(
    {
        "counts": ("time", np.array([0,2,1,0,1])),
        "group": ("time", np.array([1,1,1,2,2]))
    },
)
xarray_reduce(ds["counts"],
                ds["group"],
                expected_groups=(np.array([1,2]),),
                func="sum",
                dtype="uint8")

<xarray.DataArray 'counts' (group: 2)> Size: 16B
array([3, 1], dtype=uint64)
Coordinates:
  * group    (group) int64 16B 1 2

This happens whether I use "uint8" as a string or np.uint8.

Of course, I can just use .astype after the fact, but I expect this is an allowed parameter for some reason or another?

Python 3.12.5
flox               0.9.10
xarray             2024.7.0

[dcherian]

Yes, I'm following numpy here:

np.sum(ds.counts.data).dtype  # np.uint64

https://numpy.org/doc/stable/reference/generated/numpy.sum.html

The dtype of a is used by default unless a has an integer dtype of less precision than the default platform integer. In that case, if a is signed then the platform integer is used while if a is unsigned then an unsigned integer of the same precision as the platform integer is used.

[claytharrison]

Yes but if I pass a dtype argument explicitly, then it should do something, right?

>>> np.sum(np.array([1, 2, 3], dtype=np.int8)).dtype
dtype('int64')
>>> np.sum(np.array([1, 2, 3], dtype=np.int8), dtype=np.int8).dtype
dtype('int8')

[dcherian]

haha, yes, totally misinterpreted the docs there.