Opened 8 years ago

Closed 8 years ago

Last modified 7 years ago

#40417 closed defect (fixed)

arpack @3.1.3: remove openmpi from default variants

Reported by: tenomoto (Takeshi Enomoto) Owned by: mamoll (Mark Moll)
Priority: Normal Milestone:
Component: ports Version: 2.2.0
Keywords: Cc: michaelld (Michael Dickens), cooljeanius (Eric Gallager), dstrubbe (David Strubbe)
Port: arpack

Description

A novice user reported a failure of octave-devel. I found that it was because of arpack. When installing octave-devel from scratch, a user implicitly or explicitly specifies one of the fortran compilers (g95/gcc4x). However, g95 or gcc4x variant conflicts with openmpi, which is one of the default variants.

I assume that those who need PARPACK are power users. They can serially install openmpi with on of the fortran compilers, then arpack +openmpi and octave-devel with one of the fortran compilers.

For those who just needs octave-devel, the serial installation would be a hurdle and it is difficult for them to understand why installation is unsuccessful. Therefore I suggest to remove openmpi from the default variants of arpack.

Change History (7)

comment:1 Changed 8 years ago by cooljeanius (Eric Gallager)

Cc: egall@… added

Cc Me!

comment:2 Changed 8 years ago by cooljeanius (Eric Gallager)

Arpack's +openmpi default variant is also impeding my progress on my work on a port for Scilab (#29595). Currently I'm just taking the easy route and just disabling arpack support, but I'd probably be able to more easily add arpack support if the arpack port didn't have +openmpi as a default variant.

comment:3 Changed 8 years ago by mamoll (Mark Moll)

There really is no good solution. The slepc port depends on arpack and petsc. For this port PARPACK is necessary. Perhaps this is more of a power user port and there are more octave users. I am ok with removing +openmpi as a default variant. takeshi, do you want to add yourself as maintainer and make the change?

comment:4 Changed 8 years ago by tenomoto (Takeshi Enomoto)

A solution could be use portgroup active_variants to specify openmpi variant in slepc and petsc. Another approach could be separate parpack from arpack as a subport of arpack and slepc and petsc depend on parpack. Although I am not familiar with the build process of arpack and how slepc and petsc need parpack, I imagine the latter would be cleaner. I have too may ports to maintain: how about adding openmaintainer?

comment:5 Changed 8 years ago by mamoll (Mark Moll)

Resolution: fixed
Status: newclosed

Fixed in r110931.

comment:6 Changed 8 years ago by dstrubbe (David Strubbe)

Cc: dstrubbe@… added

Cc Me!

comment:7 Changed 7 years ago by dstrubbe (David Strubbe)

Cc: dstrubbe@… added; dstrubbe@… removed
Note: See TracTickets for help on using tickets.