Protection parameter of Filtered_predicate2

Filtered_kernel/include/CGAL/Filtered_predicate.h

@@ -37,6 +37,10 @@ namespace CGAL {
 //   not, or we let all this up to the compiler optimizer to figure out ?
 // - Some caching could be done at the Point_2 level.
 
+// Protection is undocumented and currently always true, meaning that it
+// assumes a default rounding mode of round-to-nearest. false would correspond
+// to a default of round-towards-infinity, so interval arithmetic does not
+// require protection but regular code may.
 
 template <class EP, class AP, class C2E, class C2A, bool Protection = true>
 class Filtered_predicate

NewKernel_d/include/CGAL/NewKernel_d/Filtered_predicate2.h

@@ -39,7 +39,11 @@ namespace CGAL {
 //   not, or we let all this up to the compiler optimizer to figure out ?
 // - Some caching could be done at the Point_2 level.
 
-// FIXME: understand and document what the parameter 'Protection' means exactly
+// Protection has a different meaning than in Filtered_predicate, it says
+// whether we need to set the rounding mode: some predicates only do
+// comparisons and don't need it. Probably this should be done inside this
+// class, based on Uses_no_arithmetic, but I have some doubts about C2A,
+// converting a long long to Interval_nt requires protection.
 
 template <class K, class EP, class AP, class C2E, class C2A, bool Protection = true>
 class Filtered_predicate2
@@ -88,7 +92,6 @@ public:
       catch (Uncertain_conversion_exception&) {}
     }
     CGAL_BRANCH_PROFILER_BRANCH(tmp);
-    Protect_FPU_rounding<!Protection> p(CGAL_FE_TONEAREST);
     return ep(c2e(std::forward<Args>(args))...);
   }
 };