empty selection evaluates dask.delayed #10211

mhungen · 2025-04-09T06:55:31Z

mhungen
Apr 9, 2025

Hi, I don't understand the behaviour of xarray + dask when I call compute on an empty selection.

import dask.array
import numpy as np
import xarray as xr

call_count = 0


def loader_function(url: str):
    global call_count
    call_count += 1
    print(f'{url=} has been called')
    return np.zeros((1, 10, 10), dtype=np.uint8)


da_src = xr.DataArray(
    data=dask.array.from_delayed(dask.delayed(loader_function)('/url0'), shape=(1, 10, 10), dtype=np.uint8),
    dims=['time', 'y', 'x'],
    coords={
        'time': [np.datetime64('2025-04-08')],
    }
)

da_empty = da_src.sel(time=slice('2025-04-06', '2025-04-07'), drop=True)

assert da_empty.time.size == 0
assert call_count == 0

da_empty.compute()

assert call_count == 0, 'unexpected call of loader_function'

Why is the loader_function called in this case?

If the selection is not empty, everything works as expected:

call_count = 0

da_src0 = xr.DataArray(
    data=dask.array.from_delayed(dask.delayed(loader_function)('/url0'), shape=(1, 10, 10), dtype=np.uint8),
    dims=['time', 'y', 'x'],
    coords={
        'time': [np.datetime64('2025-04-08')],
    }
)

da_src1 = xr.DataArray(
    data=dask.array.zeros((1, 10, 10)),
    dims=['time', 'y', 'x'],
    coords={
        'time': [np.datetime64('2025-04-07')],
    }
)

da_src3 = xr.concat([da_src1, da_src0], dim='time')

da_not_empty = da_src3.sel(time=slice('2025-04-07', '2025-04-07'))

assert da_not_empty.time.size == 1
assert call_count == 0

da_not_empty.compute()

assert call_count == 0, 'unexpected call of loader_function'

In this example, loader_function is never called, which is what I would expect.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

empty selection evaluates dask.delayed #10211

Uh oh!

{{title}}

Uh oh!

Replies: 0 comments

Select a reply

Uh oh!

Uh oh!

empty selection evaluates dask.delayed #10211

Uh oh!

mhungen Apr 9, 2025

Replies: 0 comments

mhungen
Apr 9, 2025