Okay, so <https://docs.python.org/dev/c-api/buffer.html#c.PyBUF_WRITABLE> says writability must be consistent. As far as I can see, there is no similar requirement for contiguity. So in theory PyBUF_FULL_RO could produce a discontiguous buffer when PyBUF_WRITABLE would have produced a contiguous one. But I guess this is rather unlikely, so Eryksun’s approach might be good enough.

I am starting to wonder why the current memory view hack should even be necessary. It seems like a memory view created by PyMemoryView_FromBuffer() will eventually do the equivalent of PyBuffer_Release(), except that it leaks a reference to the underlying Python object. This was also brought up at <https://bugs.python.org/issue15903#msg170658>, but the question is not resolved that I can see.

Possibly also related is Issue 15821, but I don’t know enough details or history to be sure.