Support pinning of IPythonBuffer

[slozier]

Currently the IPythonBuffer interface only provides access to the buffer via a Span. Adding IPythonBuffer.AsMemory<byte>() would enable some scenarios such as ctypes._CData.from_buffer which requires(?) pinning of the memory for use in unmanaged code.

@BCSharp since you implemented IPythonBuffer, any thoughts or concerns?

[BCSharp]

I did not have the opportunity to check ctypes._CData.from_buffer in depth, but from what I remember about IPythonBuffer, there is no inherent reason to use Span over Memory. It all boils down to a trade-off: usingSpan allows more sources but puts more constraints on the consumers, using Memory requires the source to be on the heap but allows more consuming cases.

I started with Span and was curious to see if we hit the wall somewhere. It seems that you have found the case where Span is not good enough. If so, Memory is the way to go; there is no need to keep Span around in the API if it gets replaced by Memory.

[BCSharp]

After dusting off this issue I now remember why IPythonBuffer provides access only by span: it allows access to both managed and unmanaged memory. I expect unmanaged buffers being created by numpy, pillow, and the like.

[slozier]

I can rename this issue to something like "Support pinning of IPythonBuffer" or something of the sort if you think AsMemory is not the way to go (I think I had run into some annoyances when I initially tried to run with it so I'm fine with going with another solution). Or we can close this issue and make a new one with a better description of the issue. The array module already does something of the sort (exposes its address when you call buffer_info) so it might offer some insight.

[BCSharp]

I can rename this issue to something like "Support pinning of IPythonBuffer"

Sounds good.