bug fix from peter.wainwright@ieee.org

Amazingly, I have discovered a potentially serious bug in the routines
which perform edge collapse in the BDS_Mesh.

Owing to this, it is possible for a point to collapse onto a linear
boundary creating slivers (triangles with collinear points).

The problem is that the sides of the triangle, a and b, are not
recalculated after the point is moved.  Thus, area_final and area_init
are always equal!  So the test for slivers is defeated.

Mesh/BDS.cpp

@@ -1004,7 +1004,7 @@ static bool test_move_point_parametric_quad(BDS_Point *p, double u, double v, BD
   double ori_final1 = gmsh::orient2d(pa, pb, pc);
   double ori_final2 = gmsh::orient2d(pc, pd, pa);
   // allow to move a point when a triangle was flat
-  return ori_init1*ori_final1 > 0 && ori_init2*ori_final2 > 0 ;
+  return (ori_init1 * ori_final1 > 0) && (ori_init2 * ori_final2 > 0);
 }
 
 static bool test_move_point_parametric_triangle(BDS_Point *p, double u, double v, BDS_Face *t)
@@ -1042,6 +1042,9 @@ static bool test_move_point_parametric_triangle(BDS_Point *p, double u, double v
   else
     return false;
 
+  a[0] = pb[0] - pa[0]; a[1] = pb[1] - pa[1];
+  b[0] = pc[0] - pa[0]; b[1] = pc[1] - pa[1];
+  
   double area_final = fabs(a[0] * b[1] - a[1] * b[0]);
   if(area_final < 0.1 * area_init) return false;
   double ori_final = gmsh::orient2d(pa, pb, pc);

doc/TODO.txt

-$Id: TODO.txt,v 1.14 2009-01-07 10:59:20 geuzaine Exp $
+$Id: TODO.txt,v 1.15 2009-01-18 11:44:36 geuzaine Exp $
+
+********************************************************************
+
+cleanup API for orientation of edges/faces
 
 ********************************************************************