• H. S. Teoh (30/30) May 16 2012 Lately I've been considering how to generalize multi-dimensional arrays

"H. S. Teoh" <hsteoh quickfur.ath.cx> writes:

Lately I've been considering how to generalize multi-dimensional arrays
along the same lines as std.range.

The motivation for this is that I'm designing a generic interface for
manipulating n-dimensional data, and it seems too implementation-
specific to impose a particular representation of an n-dimensional array
(specifically, a particular ordering of dimensions, say column major or
row major).

I'd like to be able to take arbitrary n-dimensional slices (subarrays)
of a larger array and be able to pass it to the same code as though it
were an independent n-dimensional array, for example.

I also want to be able to take existing data structures, say a linked
list of arrays, or an array of linked lists, and be able to treat it as
though it were a 2D array.

Better yet, the generic code should be able to handle arrays in which
some of the dimensions are infinite (e.g., a finite list of infinite
ranges, or an infinite range of finite lists, or an infinite list of
infinite lists -- programmatically generated, of course).

Also, it should be possible to transpose the array along any pair of
dimensions (e.g., traverse an array of linked-lists in breadth-first
order, i.e., return the first elements of each linked list, then the
second elements, etc.).

How would one go about designing a generic interface that encapsulates
this functionality? Is it possible with the current range definition, or
is something more needed? (Note that I used 2D examples above for
simplicity's sake, but this needs to handle an arbitrary number of
dimensions.)

Any ideas?


T

-- 
Gone Chopin. Bach in a minuet.