On Fri, Oct 1, 2010 at 18:33, Barry Smith <bsmith at mcs.anl.gov> wrote: >>>>> On Sep 30, 2010, at 10:50 PM, Brad Aagaard wrote: >>>>> >>>>>> We are having trouble building petsc-dev w/ML on the CIG buildbots. ML >>>>>> is configured and built but the test for using the ML library fails due >>>>>> to >>>>>> >>>>>> #error "SEEK_SET is #defined but must not be for the C++ binding of MPI >>>>>> >>>>>> even though we have used -DMPICH_IGNORE_CXX_SEEK to avoid this error. >>>>>> The logs are available at http://www.geodynamics.org:8009/?project=PETSc >>>>>> (second column x86_64 linux). >>>>>> >>>>>> Any insight you can provide on what might be the cause of this error >>>>>> would be appreciated.
This is an old version of MPICH (obvious from configure.log, not sure why it's ignoring MPICH_IGNORE_CXX_SEEK, maybe it's vendor-modified?). Current versions of MPICH have a workaround that is similar to the workaround that has been in Open MPI for a few years. These *_IGNORE_CXX_SEEK macros are not necessary these days (but I still think they are a good idea, anyone who uses MPI::SEEK_SET instead of MPI_SEEK_SET is begging for trouble). Jed
