1. What is the efficiency gain from ordered keys?
(In general, the more restrictive your semantics are, the more freedom you
have about implementing them, and hence the more efficient you can be.
So this doesn't make sense to me.
Ordering keys may happen to be faster as an _implementation_ strategy in
some situations and given some access patterns, and in that case the
implementation should certainly use such a representation. But I do not
think forcing semantics is a good performance argument.)
2. It is possible to materialize arbitrary orderings on-demand by
indexing with an appropriately ordered array of keys. I do not think
this is a good argument for making the data inherently ordered.
-E
On Tue, 1 Feb 2022, Raul Miller wrote:
On Mon, Jan 31, 2022 at 9:16 PM Elijah Stone <[email protected]> wrote:
I think not; keys should not be ordered.
I think I need to disagree here.
(1) K (and Q) have shown that you get great efficiencies with ordered keys.
(2) Also, arrays require ordered data for meaningful operations.
(3) Lack of order tends to introduce needlessly arbitrary distinctions.
Now... we might want to support multiple concepts of order, but that's
relatively straightforward.
Thanks,
--
Raul
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
