Brice.

into mpicc and friends.

created as I didn't realize subcfg.m4 already existed).

configure.in and assumes that the mpicc and other scripts have a
definition for opalibdir (this is the installed lib path after "make
install", but an internal path with the build directory before that).

specified in the host file.

binding settings in the host file.

to use.

 it to mpicc and friends, pkgconfig files, etc.

was happening because the rest of MPICH2 is not setting WRAPPER flags
correctly.

and friends are automatically exported down to subconfigures. If a
sub-configure chooses not to use whatever is exported by the
upper-level configure, it can reset the flags to the user-specified
ones using a new macro.

instead of waiting till the end of the application.

environment variable settings.

not available or doesn't work correctly.

[svn-r6027] using MPIU_ versions of malloc and free in uthash.h to fix compile error with certain --enable-g options

[svn-r6026] Removing debug stmts that were included during merging the changes from async progress branch

timeouts in such cases.

use the INTERNALS_CFLAGS and friends set by the rest of MPICH2.

The problem mainly occurs when the rest of MPICH2 is setting the
internal flags to link to a shared library, and Hydra is not able to
see where the shared library is located. This information
(LD_LIBRARY_PATH) can be passed to the MPI processes by setting the
environment through mpiexec, but the only way to pass this information
to mpiexec and friends is by setting it in the default shell path.

Thanks to Brice Goglin for reporting the error.

to wait for the exit status. Fixes the timeout test segmentation
faults in the nightly tests.

the last PG irrespective of which PG the command was for.

on. Suppose the bootstrap server registers an fd, which later gets
deregistered by the callback function, and gets registered back in by
a different component. The bootstrap server would wrongly believe that
the original fd it registered is still active, and would try to wait
on it. The callback function needs to remove it from the bootstrap
server's fd list when it closes a socket.

and destroyed.

reuses the fd value, our search for the process structure is accurate.

command (used in multi-spawn commands).

processes. With this fix, some of the spawn programs seem to be
working. Some more bugs to go.