The code seems to be derived from the LAMMPS COUPLE library. The original COUPLE code
allowed using a subset of the global processors for the coupled code. Some fractions
of that code remained, but on their own don't make sense anymore. Since no additional
colors are assigned to processors, MPI_Comm_split effectively just duplicates the
global communicator, which can be easily done using MPI_Comm_dup.
The second simplification is that the code tried to limit IO to MPI rank 0. The filename
of the input script was read in by one MPI rank and then broadcasted to all other ranks.
While from the outside this seems to make sense from an MPI programmer standpoint, it does
not take the OpenFOAM implementation into account. IODictionary is already multi-processor
aware. Reading operations are done only on the master processor. This means the dictionary
is already in memory for each MPI processor at this point in time and lookup() is an
in-memory operation.
This avoids irritating warnings of wmake about it not finding certain headers
(e.g. system headers). Similar things were necessary in some headers of the
LIGGGHTS code. While these did not cause compilation errors, they messed up
the compilation output with hundrets of false alarms. Not very helpful if
you're trying to find out what went wrong.
One special case is the change from RASModel.H to turbulenceModel.H in
cfdemCloud.H. Between major OpenFOAM versions there was a change of how
turbulence models were implemented. While the CFDEMcoupling code uses
preprocessors to allow compatibility, the wmkdepend utility is too simple
to understand this. It does not evaluate #ifdef constructs and just looks
for #include lines. Therefore it tries to access RASModel.H in newer
versions of OpenFOAM, which no longer exists.
To circumvent this issue, both includes were changed from "" to <> syntax.
This suppresses the warning and compiles, since the OpenFOAM header is in
the include path anyway.
