gfortran with Xcode 4.3.2 unreliable

[Jason Swails]

On Sun, Jun 10, 2012 at 7:49 AM, Vincent Habchi <vince at macports.org> wrote:

> On 10 juin 2012, at 09:08, Ryan Schmidt <ryandesign at macports.org> wrote:
>
> > This problem with the gmp port has already been fixed in r94119.
>

> Was there a new test suite applied after this patch? The conclusion we
> must use llvm-gcc42 instead of clang seems to bear a relation with
> something I noticed on Atlas (together with the maintainer), that is
> (purely mathematical) code fails testing. As if the FP code produced by
> clang was somehow buggy.
>

I've been using llvm-gcc42:

Jasons-MacBook-Pro:~ swails$ port installed | grep -i clang
Jasons-MacBook-Pro:~ swails$

Also, after doing a "port selfupdate" and "port upgrade outdated" (and just
installing MacPorts from scratch) resolves the issue.

Before the update:

Jasons-MacBook-Pro:~ swails$ ./a.out
 selected_real_kind(15,307) =           10
 selected_real_kind(6,37) =            8
 selected_real_kind(13,307) =           10

After the update:

Jasons-MacBook-Pro:~ swails$ gfortran test.F90
Jasons-MacBook-Pro:~ swails$ ./a.out
 selected_real_kind(15,307) =            8
 selected_real_kind(6,37) =            4
 selected_real_kind(13,307) =            8

This program was designed to test the behavior of selected_real_kind, and
it now behaves properly.  I can't report regarding clang, but llvm-gcc42
works fine.

Thanks!
Jason

-- 
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.macosforge.org/pipermail/macports-users/attachments/20120611/f2240ae9/attachment.html>