On Tue, 18 Jul 2006, Norman Gray wrote:
> On 2006 Jul 18 , at 01.12, Tim Jenness wrote:
>
> >On Tue, 18 Jul 2006, Starlink Software wrote:
> >
> > > thirdparty/fsf/autoconf/lib/autoconf/fortran.m4
> >
> >This reminds me, shouldn't the patch to starconf recently to check for the
> >return type of a real function (as used by CNF, AST and SLA now) really be
> >part of the standard fortran autoconf macros rather than a special
> >starlink-specific macro? It's not at all starlink specific. Is the feeling
> >that the test itself is not really a common requirement and we don't want
> >macro-bloat in the standard autoconf?
>
> That sounds plausible. Where is the change, though: it's not leaping out at
> me from the fortran.m4 CVS log.
>
> Reasonably generic tests like that I tended to put in the add-fpp-support
> branch; Peter has tended to add new tests to the trunk. The latter probably
> makes more sense, since that way they can be promptly tested, and then some or
> all of them moved to the add-fpp-support branch. I'll probably do that when
> I'm preparing the patch submission.
>
> I've not consciously avoided bloat.
>
> Peter: do the macros you've added have corresponding entries in the autoconf
> documentation?
Hi Norman,
I added these changes as the starconf macro STAR_CNF_F2C_COMPATIBLE, so no
(I thought I'd moved some of the comments to SSN/78, but that turns out to
be wishful thinking).
The test itself is a little less grand than the name suggests, as the
check is for the return type of REAL functions, not whether the compiler
has -ff2c actually enabled. This can be fooled on 32bit platforms where
the return type seems to have no effect (which is why it wasn't an issue).
Given the test isn't infallible, I judged it not worthy of promotion to
autoconf (and in fact initially wanted to keep it private to CNF, but we
have various copies of f77.h).
The interesting section of g77 is:
info g77 "Other compilers"
if you want to know more.
Cheers,
Peter.
|