Re: [libgfortran] Minor tweak to support Compaq Tru64

[Roger Sayle <roger at eyesopen dot com>]

Hi Andrew,
On Sun, 25 Jul 2004, Andrew Pinski wrote:
> See <http://gcc.gnu.org/ml/fortran/2004-05/msg00234.html>
> and PR 15930, PR 16137 which shows what issues are for some
> other targets with (and without) this patch.

The good news is that my patch is an improvement of the situation,
or atleast no worse.  PR libgfortran/15930 concerns IRIX, whose
header files provide both isnan and isinf, but require fixincludes
hackery to make them work.  My patch here does no harm (and I can
even see a way to fix the problems with David's proposed patch;
the trick is to use sizeof to decide whether to call _isnan, _isnanf
or isnanl, which can be done without multiple evaluations).

PR libgfortran/16137 concerns Solaris and the symbols required to
link libgfortran.  Interestingly, Solaris is like Tru64 in that it
provides an "isnan" but not an "isinf", so my patch results in one
less unresolved symbol for this PR.  Unfortunately, libgfortran
still requires the C99 float forms of the math library functions,
i.e. sqrtf, expf, logf, etc...  The correct fix here is to do something
similar to libstdc++-v3's math stubs [it might even be possible to
use the same source files].

In summary, IRIX is just as broken (or soon to be just as fixed), Solaris
is one symbol closer to building, but with this patch libgfortran works
on Compaq Tru64 (and presumably other OSF/1 targets).

Roger
--