remove integer split interface?

[cwsmith]

Now that the 'fractional' splitter interface exists:

EnGPar/partition/engpar_input.h

Lines 25 to 28 in d87959a

	Input* createSplitInput(agi::Ngraph* g, MPI_Comm smallComm,
	MPI_Comm largeComm, bool isPartOfSmall,
	double tolerance,
	agi::etype adj_type = 0, agi::part_t* others = NULL);

I'm thinking that we should replace the integer splitter interface with a local splitter interface. I think this is the only use case where the integer splitter provides a different functionality than the fractional splitter (assuming that becomes the global splitter interface). For the local splitter we will assume that, at the extreme, the user will load one input part per node and the splitFactor will be at most the number of threads per node. Thus, we'd like every splitfactor mpi process to load an input part.

[diamog]

I worked out the idea. The fortran code still passes, but can you check it, its my first edits to that. We should collapse some of the tests to global and local splits because I think we have duplicates now.

[diamog]

Creating an Initial Partition page has been updated for the new input APIs

[cwsmith]

I think we should remove/change the second argument to the split function since the input object now controls the action - i.e., passing GLOBAL_PARMETIS and a local split object should fail.

EnGPar/partition/engpar_split.cpp

Line 7 in 94de40d

void split(Input* input, SPLIT_METHOD method) {

Maybe make PARMETIS the default until we support other partitioners (geometric, our own, etc.) and remove the existing LOCAL_PARMETIS and GLOBAL_PARMETIS args.