[gmx-developers] GCC Versions

Roland Schulz roland at utk.edu
Wed Dec 17 18:59:17 CET 2008


On Wed, Dec 17, 2008 at 3:01 AM, David van der Spoel
<spoel at xray.bmc.uu.se> wrote:
> Roland Schulz wrote:
>>
>> Hi,
>>
>> regarding the warning (/*WARNING: do not use the gcc 4.1.x set of
>> compilers.) */on the download page:
>>
>> Do we have a small test case which shows the error in gcc 4.1? Or do we
>> only have the reports of people that gromacs does "strange" things with
>> 4.1.x? Are all 4.1 affected?
>>
>> What about 4.2.x and 4.3.x?
>>
>> As I said before: I would be happy to (help to) set up an automatic build
>> and test system so we better know which compiler and platforms have
>> problems. We could use http://buildbot.net/trac
>
> It seems that new platforms also give problems, whether these are due to
> compilers remains to see. Problems also show up in other programs than
> mdrun. Running make distcheck verifies the distribution based on a CVS tree,
> and running the test set under different environments would be a first step
> to check the grompp/mdrun/pdb2gmx codes.
>
> It seems that the real problem is to do this on many different machines, and
> how does buildbot help with this?

I think there are two things:

We would need more test cases to really see when things go wrong. As
you said there are reports on the mailinglist that 4.1 gives problems.
But the tests works. I just tested that on gcc 4.1.2 20071124 (Red Hat
4.1.2-42). Only the pdb2gmx tests don't run, but they don't work yet
with Gromacs 4.x, right?

The other thing is what you say about the many platforms. When we have
tests, which really show the problems, we can have a website showing
which compiler/platforms give compiler errors or have wrong test
results. Of course we would need the required machines, but I think we
could find enough volunteers to cover enough platforms.

Roland


-- 
ORNL/UT Center for Molecular Biophysics cmb.ornl.gov



More information about the gromacs.org_gmx-developers mailing list