WIP: Meepgeom refactoring

[smartalecH]

Closes #1649.

[stevengj]

Might be good to make a PR just for the refactoring, and then a separate PR for the gradient.

[smartalecH]

Here's a summary of the proposed changes:

• Added new functions make_geom_epsilon(), init_libctl(), andset_materials_from_geom_epsilon().
• Moved geom_epsilon class to meepgeom.hpp (previously local to the meepgeom.cpp only) so that I could pass around geom_epsilon objects in the overloaded set_materials_from_geom_epsilon() and make_geom_epsilon().
• In order to move things to the meepgeom header file, I had to modify the name of the symmetric matrix routines so they wouldn't clash with mpb.cpp, which imports meepgeom.hpp and mpb.h, both of which have symmetric matrix functions (I couldn't get the namespaces to work properly).

All of my code compiles, but something weird is going on with the swig generated code -- particularly with the new set_materials_from_geom_epsilon() and make_geom_epsilon() functions.

Before I spend too much time debugging, can I get some feedback as to whether this is the best way to refactor things (I'm sure it's not...)?

[stevengj]

This looks fine.

[smartalecH]

Almost all of the tests are passing now.

There is an issue with symmetric prism objects and subpixel smoothing. For example, two of the python tests (test_mode_coeffs.py and test_bend_flux.py) use a prism object to draw a simple, straight waveguide. Ideally, the waveguide should look like this:

Instead, it looks like this:

Notice that x>0 looks different than x<0. In fact, it seems that subpixel smoothing is not working for x<0 (for this particular prism object). If I turn subpixel smoothing off, I get:

So we know the issue is with subpixel smoothing and Prism objects. But not all prism objects, because the other prism object in the tests (a simple bend) looks fine Actually, this prism object looks off too (again, when x<0):

Any thoughts?
(Also, not sure why it won't compile using CI... I'll look into that after I debug this subpixel smoothing issue)

[codecov-commenter]

Codecov Report

Merging #1754 (fbd795f) into master (d41b0a4) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #1754      +/-   ##
==========================================
+ Coverage   74.41%   74.43%   +0.02%     
==========================================
  Files          13       13              
  Lines        4581     4589       +8     
==========================================
+ Hits         3409     3416       +7     
- Misses       1172     1173       +1     

Impacted Files	Coverage Δ
python/simulation.py	76.72% <100.00%> (-0.06%)	⬇️
python/adjoint/objective.py	91.76% <0.00%> (+0.85%)	⬆️

[smartalecH]

Oof seems like the above issue was due to an outdated version of libctl on my machine.

Should be good for review/merging now.

[stevengj]

LGTM.