Writing non-contiguous views of multidimensional arrays produces incorrect data

[ResStump]

Description:
When writing a view of a three-dimensional (or higher) array to an HDF5 file, if the view is non-contiguous, the data is not written correctly. Reading the data back results in an incorrect array.

Minimal Working Example:

using HDF5

arr1 = reshape(collect(1:8), 2, 2, 2)

h5open("test.h5", "w") do file
    file["arr"] = @view arr1[:, 1, :]
end

arr2 = h5read("test.h5", "arr")

println(arr2 == arr1[:, 1, :]) # false
println(arr2 == arr1[:, :, 1]) # true

Expected Behavior:
The data written to "test.h5" should match arr1[:, 1, :]. Or if views with non-contiguous arrays are not supported an error should be raised.

Actual Behavior:
Instead of writing the 4 numbers in the view, the first 4 numbers of arr1 are written.

Additional Context:
This issue seems to occur only when writing non-contiguous views of multidimensional arrays to an HDF5 file. Writing contiguous slices (e.g., arr1[:, :, 1]) works correctly.
If the first index is fixed (e.g. arr1[1, :, :]) then an error is raised (as expected).

System Information:
HDF5.jl version: 0.17.2
Julia version: 1.11.3
OS: Ubuntu 22.04.5 LTS (x86_64)

I'm looking forward to your thoughts on this.

[mkitti]

Yes, there are probably too many methods that take AbstractArray arguments which probably should check assumptions.