Bug#759182: dolfin: FTBFS on mips(el) and sparc: MPI inconsistency

[Aaron M. Ucko]

Source: dolfin
Version: 1.4.0+dfsg-2
Severity: serious
Justification: fails to build from source (but built successfully in the past)

Builds of dolfin on mips, mipsel, and sparc have all been failing:

  In file included from /usr/lib/petscdir/3.4.2/include/petscsys.h:96:0,
                   from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/common/types.h:29,
                   from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/function/Function.h:34,
                   from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/ale/MeshDisplacement.h:28,
                   from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/mesh/Mesh.h:38,
                   from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/mesh/Cell.h:33,
                   from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/mesh/MeshColoring.cpp:32:
  /usr/lib/openmpi/include/mpi.h:319:37: error: conflicting declaration 'typedef struct ompi_communicator_t* MPI_Comm'
   typedef struct ompi_communicator_t *MPI_Comm;
                                       ^
  In file included from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/common/utils.h:33:0,
                   from /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/mesh/MeshColoring.cpp:27:
  /«BUILDDIR»/dolfin-1.4.0+dfsg/dolfin/common/MPI.h:42:13: note: previous declaration as 'typedef int MPI_Comm'
   typedef int MPI_Comm;
               ^

In all three cases, debian/rules tries to disable MPI because these
architectures historically used the LAM implementation, but PETSc
proceeds to include real OpenMPI headers, leading to those errors.

Could you please take alook?

Thanks!