package.py says that arpack depends on MPI, but configure says that it depends on lapack

[yurivict]

https://github.com/davydden/spack/blob/ca97d8c928eaa84eb25c3ecf2e4c86856458ac32/var/spack/repos/builtin/packages/dealii/package.py

configure passes when MPI is off, but not when LAPACK is off.

Which one is correct?

[davydden]

Can you please elaborate what is the problem with configure of deal.ii?

EDIT: from your post, I am also not sure whether you have a problem with arpack or deal.II.

p.s. Note that Spack is a separate project that some of deal.ii developers use.

[davydden]

also please check README and installation instructions at https://www.dealii.org/developer/index.html / https://www.dealii.org/developer/readme.html

[yurivict]

Can you please elaborate what is the problem with configure of deal.ii?

There is an obvious discrepancy. ARPACK seems to really depend on LAPACK, not MPI, unless there is also a bug in cmake scripts. I just wanted to confirm. In the FreeBSD port dependency between options needs to be specified.

[davydden]

arpack depends on LAPACK unconditionally, it may also depends on MPI in that case you also build parallel Arpack (pArpack) part of it. But that's NOT required by deal.II, deal.II can distinguish between Arpack or Arpack + pArpack.

p.s. A side note, I strongly recommend using arpack-ng, which have several important fixes like opencollab/arpack-ng#85 and opencollab/arpack-ng#21

There is an obvious discrepancy.

p.p.s. In Spack we always build Arpack with MPI, but that's just a convenience decision solely for Spack.

[tamiko]

I think this question has been resolved let's close the issue for the time being. 😺