bug for very high dim

Spatial_sorting/doc_tex/Spatial_sorting/spatial_sorting.tex

@@ -23,7 +23,7 @@ to be able to combine the good randomized complexity and the
 good effects of locality \cite{acr-icb-03}.
 
 
-The only predicates used by this package are comparisons between coordinates,
+The predicates used by this package are comparisons between coordinates,
 thus there is no robustness issues involved here, for example to choose the
 arithmetic of the kernel.
 
@@ -129,10 +129,16 @@ similar manner and we get also a suitable ordering of the points
 \cgal\ provides Hilbert sorting for points in 2D, 3D and higher dimensions,
 in the middle and the median policies.
 
-The median policy seems interesting in practice and his invariant to
+The middle policy is easier to analyze, and is interesting in practice
-scaling, while the middle policy is easier to analyze. Most
+for well distributed set of points in small dimension (if the number
-theoretical results are using the middle policy
+of points is really smaller than $2^d$).
-\cite{acr-icb-03,other-references}..
+The median policy should be prefered for high dimension or if 
+the point set distribution is not regular (or unknown).
+Since the median policy cannot be much worse than the middle
+policy, while the converse can happen, the median policy is the
+default behavior.
+Most theoretical results are using the middle policy
+\cite{acr-icb-03,other-references}.
 
 
 This other example illustrates the use of the two different policies

Spatial_sorting/include/CGAL/Hilbert_sort_median_d.h

@@ -90,7 +90,6 @@ public:
        }
      }
 
-
      std::vector<RandomAccessIterator> places(nb_splits +1);
      std::vector<int>                  dir   (nb_splits +1);
      places[0]=begin;
@@ -144,16 +143,15 @@ public:
       two_to_dim = 1;
       Starting_position start(_dimension);
 
-
+      int N=end-begin;N*=2;
+      for (int i=0; i<_dimension; ++i) 	start[i]=false; 	// we start below in all coordinates
       for (int i=0; i<_dimension; ++i) {
-	start[i]=false; 	// we start below in all coordinates
 	two_to_dim *= 2;        // compute 2^_dimension
-	if (two_to_dim*2 <= 0) {
+	N/=2;
-	  CGAL_assertion(end-begin < two_to_dim);//too many points in such dim
+	if (N==0) break;  // not many points, this number of dimension is enough
-	  break;
-	}
       }
 
+
       // we start with  direction 0;
       sort (begin, end, start, 0);
     }

Spatial_sorting/test/Spatial_sorting/test_hilbert.cpp

@@ -342,6 +342,39 @@ int main ()
 
         std::cout << "no points lost." << std::endl;
     }
+    {
+      int dim = 10;
+      std::cout << "Testing "<<dim<<"D (middle policy): Generating "<<nb_points_d<<" random points... " << std::flush;
+
+        std::vector<Point> v;
+        v.reserve (nb_points_d);
+
+        CGAL::Random_points_in_cube_d<Point> gen (dim, 1.0, random);
+
+        for (int i = 0; i < nb_points_d; ++i)
+            v.push_back (*gen++);
+
+        std::cout << "done." << std::endl;
+
+        std::vector<Point> v2 (v);
+
+        std::cout << "            Sorting points ...    " << std::flush;
+
+	cost.reset();cost.start();
+        CGAL::hilbert_sort (v.begin(), v.end(),
+			    CGAL::Hilbert_sort_middle_policy());
+	cost.stop();
+
+        std::cout << "done in "<<cost.time()<<"seconds." << std::endl;
+
+        std::cout << "            Checking...          " << std::flush;
+
+        std::sort (v.begin(),  v.end(), Kd().less_lexicographically_d_object());
+        std::sort (v2.begin(), v2.end(),Kd().less_lexicographically_d_object());
+        assert(v == v2);
+
+        std::cout << "no points lost." << std::endl;
+    }
     {
       int dim=5;
       std::cout << "Testing "<<dim<<"D: Generating "<<small_nb_points_d<<" random points... " << std::flush;