[clang] [mutation analyzer] support mutation analysis for pointee (PR #118593)

[Congcong Cai via cfe-commits]

================
@@ -112,6 +115,57 @@ AST_MATCHER_P(Stmt, canResolveToExpr, const Stmt *, Inner) {
   return canExprResolveTo(Exp, Target);
 }
 
+// use class member to store data can reduce stack usage to avoid stack overflow
+// when recursive call.
+class ExprPointeeResolve {
+  const Expr *T;
+
+  bool resolveExpr(const Expr *E) {
+    if (E == nullptr)
+      return false;
+    if (E == T)
+      return true;
+
+    if (const auto *BO = dyn_cast<BinaryOperator>(E)) {
+      if (BO->isAdditiveOp())
+        return (resolveExpr(BO->getLHS()) || resolveExpr(BO->getRHS()));
+      if (BO->isCommaOp())
+        return resolveExpr(BO->getRHS());
+      return false;
+    }
+
+    if (const auto *PE = dyn_cast<ParenExpr>(E))
+      return resolveExpr(PE->getSubExpr());
+
+    if (const auto *ICE = dyn_cast<ImplicitCastExpr>(E)) {
+      const CastKind kind = ICE->getCastKind();
+      if (kind == CK_LValueToRValue || kind == CK_DerivedToBase ||
+          kind == CK_UncheckedDerivedToBase)
+        return resolveExpr(ICE->getSubExpr());
+      return false;
+    }
+
----------------
HerrCai0907 wrote:

`ExplicitCastExpr` should be treat as different case instead of directly resolvable.
Because explicit cast only need to fulfill requirement of case itself.

e.g. 
In `dynamicCast<Derived*>(BasePtr)`, we don't need to analyze whether the whole expr is mutable. Here we only need to check the constness compatibility between `decltype(BasePtr)` and `Derived*`. Which is already checked in cast part of `findPointeeToNonConst`

https://github.com/llvm/llvm-project/pull/118593