[REBOL] Re: Tupleware
From: holger:rebol at: 22-Feb-2001 10:41
On Thu, Feb 22, 2001 at 12:32:21PM -0500, [Sanghabum--aol--com] wrote:
> Thanks for the information. I still think that, with a bit of imagination,
> the tuple could become a much more usable datatype.
>
> Given that a tuple doesn't handle IPv6 addresses (which have 8 parts, each 1
> 16-bit number, usually expressed in hex, with a colon as a separator, and
> with shorthand forms for omitting zeroes) perhaps there is a case for Carl to
> sit down and rethink or extend the tuple concept.
Actually to do IPv6 addresses right you would need considerably more than
16 bytes. You would also need to store the scope identifier and, depending on
use, perhaps a netmask length.
There is a reason why tuples are limited the way they are: REBOL has two
classes of values, those which fit into a "value slot", i.e. which are completely
copied when a block containing those values is copied, and those which require
additional storage subject to garbage collection, i.e. which have reference
semantics when a block containing those values is copied.
Simple
types such as integer!, logic! and all words fall into the first
category. Series, ports, objects, functions etc. fall into the second
category. Tuples have a limited length and value space because they
are considered simple types. This improves efficiency because it means that
color values and IP addresses don't have to be garbage collected.
If you need value sequences of arbitrary length then you have to use some type
of series, probably a block of integers or a binary. IPv6 addresses will probably
be a type by themselves, both because of the tuple limitation and because having
a separate type allows them to be distinguished from other types by the scanner.
--
Holger Kruse
[holger--rebol--com]