SwathDefinition.update_hash() raise error after slicing the swath object

[ghiggi]

Code Sample, a minimal, complete, and verifiable piece of code

Calling SwathDefinition.update_hash() after having sliced the objects raises an error.

import numpy as np 
from pyresample import SwathDefinition 
dtype = "float32" # "float64" # "int32"
lons = np.ones((2,100), dtype=dtype)
lats = np.ones((2,100), dtype=dtype)

# Create dummy SwathDef 
swath_def = SwathDefinition(lons=lons, lats=lats)
print(swath_def.lats.data.contiguous) # True
swath_def.update_hash() # Value Error 

# Slice SwathDef
swath_def_subset = swath_def[:, slice(0,5)]
print(swath_def_subset.lats.data.contiguous) # False

# Try to update the hash
swath_def_subset.update_hash() # Value Error 

Problem description

This error does not allow parallelized computations (using dask delayed) over blocks of SwathDefinition objects.
The following approach fails because update_hash() is called by dask:

import dask 
@dask.delayed
def delayed_fun(swath_def):
    # whatever 
    pass 
    return None 

list_slices = [slice(0,3), slice(2,5)]
results = []
for slc in list_slices:
    swath_def_block = swath_def[:, slc]
    result = delayed_fun(swath_def_block)
    results.append(result)
   
out = dask.compute(*results) # FAILS
results[0].compute()               # FAILS 

Expected Output

SwathDefinition.update_hash() should not raise an error when the lats and lons arrays are not contiguous (i.e. after having sliced the objects).
The problem does not occur with AreaDefinition objects.

Traceback

Traceback (most recent call last):

  File ~/anaconda3/envs/satpy39/lib/python3.9/site-packages/pyresample/geometry.py:668 in get_array_hashable
    return arr.name.encode('utf-8')  # dask array

AttributeError: 'numpy.ndarray' object has no attribute 'name'


During handling of the above exception, another exception occurred:

Traceback (most recent call last):

  Input In [145] in <cell line: 13>
    swath_def_subset.update_hash() # Value Error

  File ~/anaconda3/envs/satpy39/lib/python3.9/site-packages/pyresample/geometry.py:122 in update_hash
    existing_hash.update(get_array_hashable(self.lons))

  File ~/anaconda3/envs/satpy39/lib/python3.9/site-packages/pyresample/geometry.py:670 in get_array_hashable
    return np.asarray(arr).view(np.uint8)  # np array

ValueError: To change to a dtype of a different size, the array must be C-contiguous

[djhoese]

So the main problem seems to be the conversion to uint8. This makes sense since the sliced array would not have a nice even striding when converted from individual 4-byte elements (32-bit floats) to groups of 4 uint8 elements. @mraspaud do you remember why we needed to convert to np.uint8? Does the numpy buffer protocol implementation not allow for buffer access to floats?

For reference: https://docs.python.org/3/c-api/buffer.html