[petsc-dev] Configure problem on Titan

[Satish Balay]

Mark, - I had already responded to the issue you had.

[To work arround this misconfiguration of ACML] You might want to
explicitly set LD_LIBRARY_PATH to the location of acml library and
then run configure.

Satish

>>>>>>>>>

Date: Wed, 11 Mar 2015 23:40:11 -0500
From: Satish Balay <balay at mcs.anl.gov>
Reply-To: petsc-dev <petsc-dev at mcs.anl.gov>
To: Mark Adams <mfadams at lbl.gov>
Cc: petsc-dev <petsc-dev at mcs.anl.gov>
Subject: Re: [petsc-dev] Configure problem on Titan

>>>
Executing: cc  -o /tmp/scratch/petsc-sY83lc/config.setCompilers/conftest
/tmp/scratch/petsc-sY83lc/config.setCompilers/conftest.o
Executing: /tmp/scratch/petsc-sY83lc/config.setCompilers/conftest
ERROR while running executable: Could not execute "/tmp/scratch/petsc-sY83lc/config.setCompilers/conftest":
/tmp/scratch/petsc-sY83lc/config.setCompilers/conftest: error while loading shared libraries: libacml.so: cannot open shared
object file: No such file or directory
<<<

Presumably ACML module is not setup correctly - as its not adding the location to LD_LIBRARY_PATH
[or shoud it be CRAY_LD_LIBRARY_PATH?]


<<<<<<<<<

On Thu, 12 Mar 2015, Mark Adams wrote:

> Just checking, do I have something to try/do?
> Mark
> 
> On Thu, Mar 12, 2015 at 5:59 AM, Jed Brown <jed at jedbrown.org> wrote:
> 
> > Barry Smith <bsmith at mcs.anl.gov> writes:
> > > Anything is possible, if one based ./configure on "what could go wrong
> > with it" then ./configure couldn't really do anything.
> >
> > When we talk about failure rates, we also have to include how expensive
> > the failure mode is.  If the cost is that everything goes through and
> > maybe even some examples seem to run correctly, but subtle things (like
> > MatElemental) cause data corruption, it's going to be a really painful
> > debugging session.  We won't be able to reproduce it locally because we
> > link correctly.  Really expensive both in human time and in reputation
> > (depending on the user).
> >
> > Is it still better for configure to guess instead of asking the user?
> > Maybe, though I hate to say it.  The user might get -lstdc++ wrong as
> > much or more than configure.
> >
>