[petsc-dev] installing external packages as shared?

[Barry Smith]

On Sep 8, 2014, at 5:32 PM, Jed Brown <jed at jedbrown.org> wrote:

> Barry Smith <bsmith at mcs.anl.gov> writes:
>>  3) A simple and dependable way to install many dependencies of PETSc
>>  :-) and other HPC numerical software libraries
> 
> If it's a general way to install software, then why does it depend on
> PETSc?

  Because it is for dependencies of PETSc, obviously :-)

>  PETSc and PETSC_ARCH are currently an integral part of the way
> this is distributed, but would the role you envision make sense if PETSc
> was not the center of the universe?

   Very good question and central to the question of the IDEAS project. Making it slightly less PETSc centric but at the same time focused on HPC numerical computing and not a general packaging system (which is too difficult).

  One of the reasons I am going through a large reorg of —download now is to see what can be done to make it less PETSc centric without making it a complete packaging system.


  Barry

> 
> Why can't one add packages (especially those that can be --download'ed,
> but on which PETSc does not depend) without reconfiguring PETSc?
> 
>>   In my mind the —prefix should (if used) generally be a path unique
>>   to a particular build of PETSc (that can be removed with a single
>>   rm) and NOT standard locations like /usr/local where they get mixed
>>   up with packages installed in other ways. We could even add a
>>   warning :-), that no one will read.
>