[petsc-users] PETSc with Julia Binary Builder

[Kozdon, Jeremy (CIV)]

So it seems that if I rollback some of the compiler updates the current build recipe works, so it seems that the upstream dependency that sets the compiler libraries needs fixing.

Thanks for all your help. Once I get this sorted, I will look into some of the other build weirdness that folks have pointed out.

> On Jul 2, 2021, at 9:03 AM, Satish Balay <balay at mcs.anl.gov> wrote:
> 
> On Fri, 2 Jul 2021, Kozdon, Jeremy (CIV) wrote:
> 
>> Thanks for all the feedback!
>> 
>> Digging a bit deeper in the dependencies and it seems that the compiler have been updated but MPICH has not been rebuilt since then. Wondering if this is causing some of the issues. Going to try to manually rebuilt MPICH to see if that help.
> 
> BTW: I noticed:
> 
>> /workspace/destdir/lib/libstdc++.so: undefined reference to `__divmoddi4 at GCC_7.0.0'
> 
> --prefix=/workspace/destdir
> --with-blaslapack-lib=/workspace/destdir/lib/libopenblas.so
> --with-mpi-include=/workspace/destdir/include
> 
> So /workspace/destdir has a dupliate compile install - that is conflicting with /opt/bin/i686-linux-gnu-libgfortran3-cxx11/c++?
> 
> And what compiler is used to build blas/mpi in this location?
> 
> And the 'updated compiler' you refer to is in /workspace/destdir - and you shouldn't be using /opt/bin/i686-linux-gnu-libgfortran3-cxx11 ?
> 
> 
>> 
>>> On Jul 2, 2021, at 8:35 AM, Satish Balay via petsc-users <petsc-users at mcs.anl.gov> wrote:
>>> 
>>> So yeah - the compiler install is likely broken. Something to try is --with-cxx=0
>> 
>> For wrapping in Julia I would assume that the Fortran, Python, or c++ interfaces are not needed. Python is already off, is much lost by disabling Fortran and cxx in this case?
> 
> Likely not - but then you get fortran dependency from blas [-lgfortran - which can be easily specified to configure - if needed]. And if building PETSc with some externalpackages - like mumps/hypre - that require fortran or c++ compilers.
> 
>> 
>> Also, I happened to just be looking at the PETSc website and saw that note about not sending install questions for the users email list. My bad! Sorry all.
> 
> Its fine to send install issues here. That recommendation [use petsc-maint] is primarily to avoid flooding mailing list users with huge configure.log attachments.
> 
> Satish
>