[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
This is a concrete idea for revising the shape representation. At
present we have
* (shape bound ...) to make arrays that represent shapes
* (array-shape arr) to access the shape of the array
(array-rank arr) to access the number of dimensions of arr
(make-array shape . obj)
(array shape obj ...)
(share-array arr shape proc)
and some language about it not being alright to mutate a shape.
Now I propose to replace (array-shape arr) with
(array-start arr d) for the start index of arr along dimension d
(array-end arr d) for the end index of arr along dimension d
and specify only that a (shape bound ...) is suitable input to
make-array, array and share-array - in particular, it need not be an
array. (It occurred to me that "begin" is not a noun, hence "start".
These better be short.)
People can easily write array-shape themselves, with the practical
advantage that the result will be newly allocated and mutable,
together with array->shape to create the corresponding shape object.
When people don't need the shape as a data structure, they can write
shorter and more transparent code with array-start and array-end.
Implementors can represent shapes in arrays more efficiently.