[SUMO-Integration] Delaunay triangulation doesn't finish

[bcoueraud87]

• Current date : 03/04/2023
• Compiler : MSVC
• Commit hash : c27f9139445aa34d1f55ba0830207819f039cc14
• Commit date : Tue Mar 21 12:02:29 2023 +0100
• Branch : SUMO-Integration

I can't load into JuPedSim the geometry that describes some large pedestrian network, as the Delaunay triangulation that takes place in CGAL doesn't finish.

Some context:

• The polygon comes from GEOS and contains holes.
• The joint WKT file pedestrianNetwork_Cpp.txt loads in visdbg but arrays with flattened coordinates (contained in pedestrian_network_connected_component_0.txt) passed to JuPedSim give the problem above.
• The code that produces this flattened arrays of coordinates and the calls to the JuPedSim API are visible here, function preparePolygonForJPS.
• But this code is similar to what visdbg does.
• Going into Release mode doesn't solve the issue.
• The infinite triangulation appears after GeometryBuilder::Build starts, and then time is lost in various CGAL calls, like (for example) CGAL::Surface_sweep, etc.
• This code works for networks of smaller size.

pedestrian_network_connected_component_0.txt pedestrianNetwork_Cpp.txt

[Ozaq]

@bcoueraud87 I cannot reproduce the issue you see. I converted the attached files (See attachment of this comment). pedestrianNetwork_Cpp.txt loads fine, pedestrian_network_connected_component_0.txt (corrected.txt) is not a simple polygon.

pedestrianNetwork_Cpp.txt

corrected.txt

[bcoueraud87]

I uploaded two "corrected" files that wouldn't bring any confusion. The coordinates are now in the same order for both files. Note that the second one is volontarily missing the GEOMETRYCOLLECTION string at its beginning, and that for each polygon in the first one, the last point has been removed to make CGAL happy.

[Ozaq]

Underlying issue was the wrong dependency under windows (old version of poly2tri)