Properly handle prob=1.0f due to precision conversion.

src/QMCHamiltonians/NonLocalTOperator.cpp

@@ -83,6 +83,14 @@ NonLocalTOperator::NonLocalTOperator(const TmoveKind non_local_move_option,
 
 const NonLocalData* NonLocalTOperator::selectMove(RealType prob, const std::vector<NonLocalData>& txy)
 {
+  // Although prob is required to be [0, 1), the received value can still be 1 when there is precision conversion.
+  // For example, when the caller value is the largest value smaller than 1.0 in double precision,
+  // the callee (RealType=float) can still receive 1.0f after an implicit conversion.
+  // Here we adjust the value of prob to be slightly smaller than 1.
+  if (prob == RealType(1))
+    prob = std::nextafter(RealType(1), RealType(0));
+  assert(prob >= 0 && prob < 1);
+
   // txy_scan_[0] = 1.0, txy_scan_[i>0] = txy_scan_[i-1] + txy[i-1].Weight (modified)
   txy_scan_.resize(txy.size());
   RealType wgt_t = 1.0;

src/QMCHamiltonians/tests/test_NonLocalTOperator.cpp

@@ -53,8 +53,8 @@ TEST_CASE("NonLocalTOperator", "[hamiltonian]")
   REQUIRE(select5 != nullptr);
   CHECK(select5->PID == 2);
 
-  RealType near_one = std::nextafter(RealType(1), RealType(0));
-  CHECK(near_one < RealType(1));
+  double near_one = std::nextafter(1.0, 0.0);
+  CHECK(near_one < 1.0);
   auto select6 = t_op.selectMove(near_one, Txy);
   REQUIRE(select6 != nullptr);
   CHECK(select6->PID == 2);