Propagate chunk_layout to Python meep.Simulation object

[bencbartlett]

If a meep.Simulation object is created without specifying chunk_layout, then the default logic for choosing chunk partitionings is invoked when sim.init_sim() is called. However, the chunk_layout attribute on the Python meep.Simulation object previously was not updated and would still read as None. This pull request adds a bp_to_py_bp() function to convert the C++ object into a Python meep.BinaryPartition object and updates the sim.chunk_layout attribute (and for consistency, sim.num_chunks), in sim._init_structure(). (Fixes #1648)

[codecov-commenter]

Codecov Report

Merging #1673 (43b3795) into master (89c9349) will increase coverage by 0.08%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1673      +/-   ##
==========================================
+ Coverage   73.20%   73.28%   +0.08%     
==========================================
  Files          13       13              
  Lines        4515     4522       +7     
==========================================
+ Hits         3305     3314       +9     
+ Misses       1210     1208       -2     

Impacted Files	Coverage Δ
python/simulation.py	76.17% <100.00%> (+0.16%)	⬆️

[oskooi]

I think the failing test test_chunk_layout.py is due to a bug in the way the process IDs of the leaves are always assigned a value of -1:

meep/src/structure.cpp

Line 69 in 89c9349

if (n == 1) { return std::unique_ptr<binary_partition>(new binary_partition(-1)); }

The reason this bug is a little hard to track down is because if the user does not pass in a chunk_layout parameter to the Simulation object and Meep therefore falls back on using its internal cell-partition scheme, the proc of each chunk is set correctly but the corresponding proc_id of the leaves from the binary tree is never updated to this same value:

meep/src/structure.cpp

Lines 133 to 141 in 89c9349

	for (size_t i = 0, stop = chunk_volumes.size(); i < stop; ++i) {
	const int proc = (!_bp) ? i * count_processors() / chunk_volumes.size() : ids[i] % count_processors();
	for (int j = 0; j < num_effort_volumes; ++j) {
	grid_volume vc;
	if (chunk_volumes[i].intersect_with(effort_volumes[j], &vc)) {
	chunks[num_chunks] = new structure_chunk(vc, v, Courant, proc);
	br.apply(this, chunks[num_chunks++]);
	}
	}

What should happen is the proc_id of each leaf/chunk should be set to proc before returning the binary_partition object bp back to Python.