Failure on n-gons with even n

[dmitryplatonov]

I feel it's due to parallel edges in such n-gons, but not completely sure.
Made simple test generator: http://pastebin.com/ULwQdgSw
git rev-parse --short HEAD
6c9c25f
Output before failure with n=100
add line 49 : 370 377
add line 50 : 377 384
add line 51 : 384 390
WARNING empty solution set!!
WARNING in_region_filter() results in empty solution set!!
WARNING t_filter() results in empty solution set!!
solution edge: (-1.39563e-05, -0.000444096)1 - (5.52635e-17, -8.75288e-18)1
solution edge: 3206[1]{1} -[2]- 3208[1]{0}
s1= LineSite: (-0.898224, 0.0565115) - (-0.9, 1.10218e-16)(k=-1)
s2= LineSite: (0.898224, -0.0565115) - (0.9, 0)(k=-1)
s3= LineSite: (-0.898224, -0.0565115) - (-0.9, 1.10218e-16)
Running solvers again:
LLLPARASolver.
s1 : -0.999507 0.0314108 -0.899556 -1
s2 : 0.999507 -0.0314108 -0.899556 -1
s3 : 0.999507 0.0314108 0.899556 1
bisector: -0.999507 0.0314108 5.55112e-17
Solution: t=0.899556 (-0.9, -28.6385) k3=1
LLLPARASolver.
s1 : -0.999507 0.0314108 -0.899556 -1
s2 : 0.999507 -0.0314108 -0.899556 -1
s3 : 0.999507 0.0314108 0.899556 1
bisector: -0.999507 0.0314108 5.55112e-17
Solution: t=0.899556 (-2.77693e-17, -2.6509e-15) k3=-1
No solutions found by solvers!
VertexPositioner::desperate_solution()
edge: (-1.39563e-05, -0.000444096) - (5.52635e-17, -8.75288e-18)
dist(): 0.899556 - 0.899556
WARNING: Returning desperate solution:
(0.0305228, 0.971251) t=0.899556 k3=-1 e_err=0.97173
s1_err= 0.404023
s2_err= 0.404023
s3_err= 0.445509
solution_on_edge() ERROR err= 0.97173
solution edge: 3206[1]{1} -[2]- 3208[1]{0}
edge: 3206(t=0.899556) - 3208(t=0.899556)
edge: (-1.39563e-05, -0.000444096) - (5.52635e-17, -8.75288e-18)
solution: (0.0305228, 0.971251) t=0.899556
python: /home/shadowjack/projects/openvoronoi/src/vertex_positioner.cpp:98: ovd::solvers::Solution ovd::VertexPositioner::position(ovd::HEEdge, ovd::Site*): Assertion `solution_on_edge(sl)' failed.
Aborted

[aewallin]

Thanks! I will add this as a test for use with ctest when I have time.
The issue of numeric robustness and exceptions probably requires some more thought and two or three layers of more 'desperate'/brute-force algorithms to fall back on in hard cases.

[aewallin]

I have added this now in python_examples/issues
I will look at the solver - hopefully soon!