[petsc-dev] [petsc-maint] problem installing petsc 3.4

[Satish Balay]

On Wed, 16 Apr 2014, Jed Brown wrote:

> Satish Balay <balay at mcs.anl.gov> writes:
> 
> > On Wed, 16 Apr 2014, Jed Brown wrote:
> >
> >> "Borazjani, Iman" <iman at buffalo.edu> writes:
> >> 
> >> > Hi Jed,
> >> >
> >> > I tried both ways and still get a compiler error, but this time I think 
> >> > it is the -g option:
> >> >    CLINKER 
> >> > /share/apps/petsc-petsc-7cd361a5361c/linux-gnu-debug/lib/libpetsc.so.3.04.4
> >> > /usr/bin/ld: 
> >> > /share/apps/open64/lib/gcc-lib/x86_64-open64-linux/4.2.4/libopen64rt.a(cacheinfo.o): 
> >> > relocation R_X86_64_32S against `a local symbol' can not be used when 
> >> > making a shared object; recompile with -fPIC
> >> > /share/apps/open64/lib/gcc-lib/x86_64-open64-linux/4.2.4/libopen64rt.a: 
> >> > could not read symbols: Bad value
> >> 
> >> Looks like this compiler claims to be able to link with shared
> >> libraries, but in fact cannot (since no shared version of libopen64rt.a
> >> is available).  Configure --with-shared-libraries=0 to get a static
> >> build of PETSc.
> >
> > There was a discussion about this on petsc-maint a couple of years ago with a reference to
> >
> > http://devgurus.amd.com/thread/159277
> >
> > i.e the compiler links to libopen64rt_shared.a when -shared options is specified.
> >
> > But we detect the compiler libraries libraries without using '-shared'
> >
> >
> > Workarround discussed at that time: 
> >
> > --with-shared-libraries=1 --with-clib-autodetect=0 --with-fortranlib-autodetect=0 LIBS='-lfortran -lmv -lffio'
> >
> > [but we mignt need --with-cxx-autodetect=0 aswell [with -lstdc++ in LIBS]
> 
> Would we have more or less failures if we used the Fortran linker and
> passed -lstdc++ when needed (with no other autodetect)?  I hate
> second-guessing the compilers.

Whenever we make such assumptions - there is always a corner case
where something else fails.

[the current issue is an example of one such assumption.]

Satish

> 
> >> 
> >> This is something we should be able to detect in our configure tests.
> >> 
>