Writing strided numpy arrays

[MortenHannemose]

There seems to be an issue when writing numpy arrays with non-standard strides. Consider the following example of me writing a multi-channel image

import numpy as np
shape = (4,4)

channels_np = np.zeros((*shape, 10), dtype=np.float32)
channels_np += np.arange(10)
channels = {}
for i in range(channels_np.shape[-1]):
    channels[f"chan{i}"] = channels_np[:, :, i]

print("chan1 before writing\n", channels["chan1"])
header = {"compression" : OpenEXR.RLE_COMPRESSION, "type" : OpenEXR.scanlineimage}
with OpenEXR.File(header, channels) as outfile:
    outfile.write("tmp.exr")

with OpenEXR.File("tmp.exr") as infile:
    print("chan1 read from file\n", infile.channels()["chan1"].pixels)

This prints

chan1 before writing
 [[1. 1. 1. 1.]
 [1. 1. 1. 1.]
 [1. 1. 1. 1.]
 [1. 1. 1. 1.]]
chan1 read from file
 [[1. 2. 3. 4.]
 [5. 6. 7. 8.]
 [9. 0. 1. 2.]
 [3. 4. 5. 6.]]

If I .copy() the arrays before adding them to the channels dict, this goes away, but it took me a long time to track down the error.
Ideally it should perform this copy internally if necessary, or at least warn the user that the strides are not supported.

[cary-ilm]

Apologies for the lack of notice in the documentation, there is indeed an assumption that the numpy array has standard stride. I'll look into what it would take to relax that requirement, and/or update the documentation.