(j3.2006) Papers 13-316, 13-317, 13-318

[Van Snyder]

On Wed, 2013-09-11 at 10:33 +0900, Malcolm Cohen wrote:
> >In retrospect, it's not obvious that we need a REDUCE intrinsic
> >function, since (unlike the CO_REDUCE case) it's so easy to write one.
> 
> I do not agree.  SUM might be pretty easy to write, but the general REDUCE not 
> so much.

I wasn't thinking that the general REDUCE is easy to write.  Rather, an
arbitrary REDUCE, for one operation, type, kind type parameters, and
rank, is easy to write.  Even with these restrictions, the co-reduce
case is difficult to write and still get efficiency, or at least that
was Bill's argument that led to including CO_REDUCE in 18508.

I agree it's difficult (errrm, impossible without recourse to C
interoperability, and maybe so even then) to write a general REDUCE that
takes a function argument with arbitrary type, arbitrary type
parameters, and arbitrary rank.