On Tue, Oct 28, 2003 at 03:25:44PM -1000, Shiro Kawai wrote:
> The ideal resolution, seems to me, to have two srfis submitted
> together, one for a generic collection srfi and the other for
> a dictionary srfi.   The former just mention a dictionary obeys
> generic collection attributes, but leaves the concrete API and
> implementation to the latter.
>

They naturally wouldn't be submitted exactly together, theres no reason
to delay 44 that long, but I can see a coupling established through a
forward looking statement in 44, or delaying 44 until the dictionary
SRFI enters draft and gets a number.

> Srfi-34 is referred in this discussion as a sort of "abstract"
> srfi that left some concrete parts to srfi-35 and srfi-36.
> As far as I remember, however, those three srfis are submitted
> altogether (I think it was initially one srfi, then splitted
> to three).  Doing that made people easy to understand how
> abstract description related to concrete implementation.

As I'm sure you're aware, there was a similar rationale here for not
specifying future collections, because the set is so large.  The problem
was hard enough but significant enough to use this SRFI to lay some
groundwork for anticipated future expansion.

Apart from that, the differences between the 34-36 approach would only
be in the timing.  I see no disadvantage in releasing 44 with a nod to
(47?) the dictionary SRFI, and plenty of advantages (people could
immediately begin on set collections, numeric vectors, etc).

	Scott