homepage

[Nick]
> Another idea we may want to revisit is the PyManagedBuffer concept,
> which would be a true PyObject that existed solely to simplify sharing
> of Py_buffer structs.
> If memoryview used such an object internally, then copying and slicing
> would be quite simple - just INCREF the managed buffer instance without
> letting the original source object know anything was going on.
I think this is the nicest solution since memoryview would then always 
have a proper base object. Do I understand correctly that PyManagedBuffer
should only handle 1-dimensional objects?
There is an additional point about slicing and sub-views:
I think slicing (esp. multidimensional slicing) would be greatly simplified
if we added a requirement for the *exporting* object to provide a sliced
view. (The same applies to sub-views, also see source comments below [1]).
For example, an exporting object could provide a sliced view by adding 
a getslicedbufferproc to PyBufferProcs:
int PyObject_GetSlicedBuffer(PyObject *obj, Py_buffer *view, int flags, PyObject *key);
By "sliced view" I mean that the exporting object changes buf, shape and 
strides. There are several advantages:
 
 o The invariant that all allocated memory in the buffer belongs
 to the exporting object remains intact.
 o The responsibility for creating correct multidimensional sliced views
 is shifted to the implementor of the exporting object.
[1] memoryobject.c: 
 
 /* XXX There should be an API to create a subbuffer */
 /* XXX: This needs to be fixed so it actually returns a sub-view */