J3/06-273

Date: 10 August 2006
To: J3
From: Dick Hendrickson
Subject: Macros, kinds, integration, and feature creep
References: J3/06-007

One of the major driving forces behind the intelligent macro
feature was the desire to have some sort of template
available. Then generic entities, like Bessel functions,
which merely differ in arguments kinds and have essentially
the same body can be easily written. The current macro is a
little weak in this.

Note 3.15 (page 39) shows the contortions needed to generate
functions that cover all of the real kinds. It's not
immediately obvious what this does, nor that it covers all
of the kinds available (and if Malcolm's comment on the bit
intrinsic is correct, it stops way short of being complete).

I propose that we add a new family of intrinsic functions
--NEXT_REAL_KIND(k), etc. They return the next processor
supported kind number greater than k, or -1 if k is greater
than or equal to the maximum. Then the example could be
written as:
i = -1
MACRO DO !yes, we'll have to do this also
i = next_real_kind(i)
macro if (i < 0 ) macro exit ! yes again
function i_square_range_%%i(a) result (r)
...

I think that's easier to understand, less error prone, and
more in the spirit of template programming.

The functions have moderate general use. We could add
them to chapter 13, or we could prefix them with "macro_"
and make them local syntax things to the macro facility.