bug in splitting 2d cell with cylindrical coordinates and z-mirror symmetry into multiple chunks

[oskooi]

1499 seems to have revealed a latent bug when splitting a 2d cell with cylindrical coordinates and z-mirror symmetry into multiple chunks. There are two separate tests in tests/symmetry.cpp which involve a cylindrical cell with z-mirror symmetry: test_cyl_metal_mirror and test_cyl_metal_mirror_nonlinear. Both of these tests fail whenever the cell size is changed slightly from the existing volcyl(1.0, 1.0, a) to e.g. volcyl(1.0, 1.1, a) or volcyl(1.0, 0.8, a), or alternatively the resolution a is changed slightly from 8 to e.g. 9 (for test_cyl_metal_mirror) or from 16 to e.g. 17 (for test_cyl_metal_mirror_nonlinear). The test failure is also independent of the chunk division: it does not matter whether the cell is split in either the r or z directions.

It is just a coincidence that the current choice of the cell size and resolution in these tests cause them both to pass. As a result, these two tests should be disabled until this bug is resolved.

[stevengj]

Does it depend on whether the cell is an even or odd number of pixels in the mirror direction?

[oskooi]

Yes, it seems that whenever the size of the chunk in the z-direction with no mirror symmetry is an odd number of pixels the test fails and whenever the size is an even number of pixels, it passes.