[petsc-dev] Configure problem on Titan

[Barry Smith]

> On Mar 11, 2015, at 9:42 PM, Aron Ahmadia <aron at ahmadia.net> wrote:
> 
> I found the most sane way to do this on the BG/P was to query the compiler wrappers for C/C++/Fortran.  I don't understand why you aren't or are unable to do this within PETSc's configure.

  Aron,

    These actually are the compiler wrappers. Apparently the Cray folks want Fortran users who happen to be linking against some C++ libraries to magically know that they need to also link against some specific system C++ libraries. I do not understand this since Cray could easily add these extra linkages automatically to their compiler wrappers. There seems to be some bias to make interlanguage linking even harder than it has to be.

  Barry

> 
> Of course, we still encounter the crazy errors with shared libraries on our Cray XE6 where if we fail to preload a needed MPI library before importing PETSc we get symbol resolution errors...
> 
> A
> 
> On Wed, Mar 11, 2015 at 10:26 PM, Jed Brown <jed at jedbrown.org> wrote:
> Barry Smith <bsmith at mcs.anl.gov> writes:
> > so that if the basic link fails it automatically tries again here after adding the -lstdc++ library.
> 
> What if you're building with Clang and need -lc++ instead?
>