[clang] [analyzer] Fix unary/binary op support for SMT symbolic execution (PR #205078)

[Balázs Benics via cfe-commits]

================
@@ -342,6 +342,30 @@ class SMTConv {
                     Ctx.getTypeSize(FromTy));
   }
 
+  static inline llvm::SMTExprRef convertToBoolExpr(llvm::SMTSolverRef &Solver,
+                                                   ASTContext &Ctx,
+                                                   const llvm::SMTExprRef &Exp,
+                                                   QualType Ty) {
+    if (Ty->isBooleanType())
+      return Exp;
+
+    if (Ty->isRealFloatingType()) {
+      llvm::APFloat Zero =
+          llvm::APFloat::getZero(Ctx.getFloatTypeSemantics(Ty));
+      return fromFloatBinOp(Solver, Exp, BO_NE, Solver->mkFloat(Zero));
+    }
+
+    if (Ty->isIntegralOrEnumerationType() || Ty->isAnyPointerType() ||
+        Ty->isBlockPointerType() || Ty->isReferenceType()) {
+      return fromBinOp(
+          Solver, Exp, BO_NE,
+          Solver->mkBitvector(llvm::APSInt("0"), Ctx.getTypeSize(Ty)),
----------------
steakhal wrote:

I think we could do better than parsing "0" and constructing an APSInt from it. Isn't there some static method that would create you an APSInt of zero with the right bitwidth?

https://github.com/llvm/llvm-project/pull/205078