[clang] [analyzer] Reduce constraint on modulo with small concrete range (PR #65448)

[Balazs Benics via cfe-commits]

================
@@ -1824,6 +1835,94 @@ RangeSet SymbolicRangeInferrer::VisitBinaryOperator<BO_Rem>(Range LHS,
   return {RangeFactory, ValueFactory.getValue(Min), ValueFactory.getValue(Max)};
 }
 
+RangeSet SymbolicRangeInferrer::handleConcreteModulo(Range LHS,
+                                                     llvm::APSInt Modulo,
+                                                     QualType T) {
+  APSIntType ResultType = ValueFactory.getAPSIntType(T);
+  llvm::APSInt Zero = ResultType.getZeroValue();
+  llvm::APSInt One = ResultType.getValue(1);
+
+  if (Modulo == Zero)
+    return RangeFactory.getEmptySet();
+  if (Modulo < 0)
+    Modulo = -Modulo;
+
+  auto ComputeModuloN = [&](llvm::APSInt From, llvm::APSInt To,
+                            llvm::APSInt N) -> RangeSet {
+    assert(N > Zero && "Non-positive N!");
+    bool NonNegative = From >= Zero;
+    assert(NonNegative == (To >= Zero) && "Signedness mismatch!");
+
+    if (From > To)
+      return RangeFactory.getEmptySet();
+
+    llvm::APSInt N1 = N - One;
----------------
steakhal wrote:

APSInt cannot wrap, right? So if N was the minimum representable on the same number of bits, it would extend the number of bits used for the representation, which might be different from the bitwidth of `Zero` and friends.
This could lead to constructing Ranges with APSInts of different widths, which is unexpected.

https://github.com/llvm/llvm-project/pull/65448