Robustness issue with StraightSkeleton and inexact construction kernel about sfcgal HOT 12 CLOSED

\cc @vmora, @mhugo

from sfcgal.

Simplified input still reproducing the robustness issue:

POLYGON((1686350.74365082 5402217.7661505,1684030.810026 5402682.520688,1684000 5402672.31399816,1683763.966259 5402588.179891,1681281.134295 5403807.881624,1683915.838219 5402676.741864,1684000 5402711.41802501,1684020.618162 5402719.913077,1687524.6533522 5404490.5663652,1687854.358551 5403919.598884,1686350.74365082 5402217.7661505))

from sfcgal.

Filed upstream here: CGAL/cgal#123

from sfcgal.

I guess I could implement the polygon translation in SFCGAL, but it'd be best done after pr #85 is merged

from sfcgal.

Translating a Polygon_with_holes_2 is revealing harder than I hoped for. It'd be a pity to clone the input Geometry just for the sake of translating, am I missing an easy way to do that ?

from sfcgal.

It's probably not the issue but worth double checking the vertex ordering. I forget which it wants but I was getting back 0,0 points. Reversing the direction fixed that. If you're only getting that one point that's wrong it might be an totally different issue.

from sfcgal.

Single point out, and fixed by augmenting the available bits for computation (shifting to origin)

from sfcgal.

It looks like the problematic vertex joins two aligned segments. If it's really the case, then maybe joining aligned ring segment would fix things.

I don't mind gaining precision/robustness by moving vtx near origin, but this alignment, considering the definition of a straight skeleton bugs me a little.

from sfcgal.

from sfcgal.

Here's the HEXWKB format for the offending polygon input as I just found out that parsing it from WKT makes the problem go away:

010300002091080000010000000A00000063022B0A53C139419EFCD3EF4E9B54412DDD5DCF3EB23941C4F352A10E9C54410400000020B23941C08B18140C9C544116E9B9D592A739410DA9B7E75A9B54413B28612281A7394113876CF8279D54413C8595D6CBB1394125B37A2F0D9C54410400000020B23941F9EBC0DA159C544168DD3F9E34B2394184DA6FFA179C5441524F641C54C13941C3874722D79D544163022B0A53C139419EFCD3EF4E9B5441

Any way to have SFCGAL parse WKB for writing the testcase ?

from sfcgal.

about the aligned segments idea, this simple polygon does not trigger the problem:
POLYGON((0 0, 10 0, 10 10, 5 10, 0 10, 0 0))

from sfcgal.

Also passing the input polygon trough ST_Simplify() with a tolerance of 1e-10 does not remove the "almost collinear" vertex. A tolerance of 1e-9 does, fixing the problem.

from sfcgal.