[flang-commits] [flang] [Flang][Runtime] Fix implicit conversion warning when targeting 32bit… (PR #99465)

[via flang-commits]

================
@@ -23,10 +23,16 @@ static RT_API_ATTRS void GetComponentExtents(SubscriptValue (&extents)[maxRank],
     const typeInfo::Component &comp, const Descriptor &derivedInstance) {
   const typeInfo::Value *bounds{comp.bounds()};
   for (int dim{0}; dim < comp.rank(); ++dim) {
-    SubscriptValue lb{bounds[2 * dim].GetValue(&derivedInstance).value_or(0)};
-    SubscriptValue ub{
-        bounds[2 * dim + 1].GetValue(&derivedInstance).value_or(0)};
-    extents[dim] = ub >= lb ? ub - lb + 1 : 0;
+    auto lb = bounds[2 * dim].GetValue(&derivedInstance).value_or(0);
+    auto ub = bounds[2 * dim + 1].GetValue(&derivedInstance).value_or(0);
+    if (ub >= lb) {
+      auto bound_diff = ub - lb;
+      assert(bound_diff <= std::numeric_limits<SubscriptValue>::max() &&
----------------
serge-sans-paille wrote:

I *assume* that for a 32 but target, we should never meet an overflow in subscript computation. But we could imagine targeting a 64 bit target from a 32 bit target (although flang doesn't compile from 32 bit arch at the moment), I guess that's why `TypeParameterValue` is 64 bit wide (?)

https://github.com/llvm/llvm-project/pull/99465