tech-pkg archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: gfortran




I started a similar discussion a couple years ago. The response was that g95 is the only thing that works on many platforms and pkgsrc is committed to portability.

The good news is that a "native" gfortran is detected by pkgsrc, so it won't have to use g95. I always install gfortran from Yum on our CentOS systems before boostrapping pkgsrc.

I actually use a script to automate the setup:

    http://acadix.biz/pkgsrc.php

That said, I wonder if one of the GCC packages could be used instead of g95 on platforms where they are known to work, such as NetBSD. This would lead to much better performance of Fortran apps as well as enabling packages that simply won't build with g95 anymore. I think this should be fairly simple to code into the pkgsrc base, although it will add some issues for Fortran package maintainers.

I've seen improvements in the GCC packages the past couple of years. They build cleanly on CentOS now, which has been my savior for code with C++11 features. ( CentOS 6 uses GCC 4.4.7, which doesn't fully support C++11. ) While others in the HPC business are struggling with caveman GCC installations, I just add a GCC_REQD line to my package.

As you said, g95 is getting very obsolete, so something is going to have to happen with Fortran support soon.

Regards,

   Jason

On 09/08/2016 06:47, coypu%SDF.ORG@localhost wrote:
g95 is the default fortran compiler in pkgsrc.
gfortran is a fork of it from 2003 that is more widely used and actively
maintained.

issues with g95 is that it relies on older gcc which doesn't compile on
some platforms:

pkg/51402 [non-critical/medium]:
         pkgsrc-2016Q2/lang/g95 does not build on NetBSD7.0.1/arm7hf
	(raspberrypi2)

pkg/51198 [serious/medium]:
         lang/g95 fails to build on Linux

pkg/49977 [critical/medium]:
         lang/g95 fails to build on OS X Yosemite

pkg/44547 [non-critical/medium]:
         pkgsrc build failure for g95 on NetBSD/cobalt

It is also a dead end as it is not maintained, and as far as I can tell
- no serious fortran users use it.
Its merits is that it is faster to compile, and more well-tested on
pkgsrc.

gfortran requires relying on pkgsrc GCC, which has previously been
extremely problematic. I hope some of those are corrected now.

some of tests are failing, but this is also true e.g. for testing base
gcc on netbsd against gcc tests, so I'm not sure about the meaning of
it.

what are the issues that remain for flipping to switch to gfortran?
I'd like to address them.

there will always be the issue 'new gcc takes an eternity to compile'
(it really does), maybe I can create a gfortran-only package?



Home | Main Index | Thread Index | Old Index