[MacPorts] #39352: arpack @3.1.2 -openmpi cannot build without mpi

[MacPorts]

#39352: arpack @3.1.2 -openmpi cannot build without mpi
------------------------+--------------------------------
 Reporter:  dstrubbe@…  |      Owner:  macports-tickets@…
     Type:  defect      |     Status:  new
 Priority:  Normal      |  Milestone:
Component:  ports       |    Version:  2.1.3
 Keywords:  haspatch    |       Port:  arpack
------------------------+--------------------------------
 configure always receives --enable-mpi, even when neither +openmpi or
 +mpich was set, and -openmpi was set to remove the default +openmpi. As a
 result, configure will fail if mpi is not installed. Instead, --enable-mpi
 should be set only with +openmpi or +mpich. Then, -openmpi can build
 successfully, just without PARPACK.

 I also noticed some other improvements that can be made:
 * The description and variants do not make clear that PARPACK is included
 too (used to be in variants but was removed at some point, not clear why).
 * Rather than patching configure for atlas (as discussed in #34695), it
 seems wiser to specify blas explicitly. In fact, if you do that, the
 lapack argument is not needed. Similarly, specifying blas explicitly for
 accelerate makes it no longer the case that +accelerate has no effect if
 atlas is installed, as was stated in a comment in the Portfile.
 * ARPACK doesn't seem to use threads. Therefore, it seems better to use
 the sequential -lsatlas rather than the threaded -ltatlas, or at least to
 offer the user a choice about threads. The discussion in #34695 didn't
 seem to have any reason for choosing -ltatlas.
 * The path dependency on mpicc seems unwise: the code actually uses
 mpif77, not mpicc (there is no C code), so I think it is better to have a
 dependency on mpif77.
 * If the reason there is no universal variant is because of openmpi, as
 stated in the comment, then it seems that should apply only with +openmpi,
 i.e. by a variant conflict.
 * Removed some trailing spaces, as warned by nitpick.

 The attached Portfile diff addresses these issues. Let me know if you have
 questions or doubts about any of this.
 (I also added a comment about why LDFLAGS is being defined, since it looks
 like it is just the default.)

-- 
Ticket URL: <https://trac.macports.org/ticket/39352>
MacPorts <http://www.macports.org/>
Ports system for OS X