MemoryView slicing contiguity bugfix

[rjtobin]

This addresses #2941. The MemoryView scalar assignment code currently generates a stride 1 (ie. contiguous) loop whenever the destination of the slice is contiguous, regardless of the slice. This isn't quite right, since slices of contiguous arrays of course aren't necessarily contiguous. This patch just adds a check: if any of the indices of the slice is not simply :, then do not assume contiguity (these slices are referred to as "ToughSlices" elsewhere in the code).

One can do better than this, ie. if for a C-contiguous array the slice a[1][:] is contiguous, so the old generated code is correct and the new generated code is a little slower. Not ideal, and maybe is worth adding logic to detect these cases, but this seems a reasonable first-pass fix.

[scoder]

The change looks ok, but tests are missing.
Memoryview declaration tests are in tests/memoryview/memslice.pyx and usage tests are in tests/memoryview/memoryview.pyx.

[scoder]

Hmm, hang on. Shouldn't we rather set is_c_contig and is_f_contig to false in this case?

[rjtobin]

Will add tests shortly.

Hmm, hang on. Shouldn't we rather set is_c_contig and is_f_contig to false in this case?

It seemed that that the MemoryView slice itself doesn't have an is_c_contig or is_f_contig attribute, just the underlying MemoryView (which in this case is c contiguous, so I think it is correctly set to True). The current slice_iter code checks the is_c_contiguous attribute of the underlying MemoryView and then assumes the slice is also contiguous. Maybe I'm missing something though!

[rjtobin]

The change looks ok, but tests are missing.
Memoryview declaration tests are in tests/memoryview/memslice.pyx and usage tests are in tests/memoryview/memoryview.pyx.

I've added a couple of tests to the existing contiguous memview slice test code.

[scoder]

?

[rjtobin]

Apologies, that was quite careless. Removed (and rebased on top of current master).

[scoder]

Thanks!