[clang] [clang][dataflow] Fix an issue with `Environment::getResultObjectLocation()`. (PR #75483)

[Gábor Horváth via cfe-commits]

================
@@ -726,27 +726,69 @@ void Environment::setStorageLocation(const Expr &E, StorageLocation &Loc) {
   // so allow these as an exception.
   assert(E.isGLValue() ||
          E.getType()->isSpecificBuiltinType(BuiltinType::BuiltinFn));
-  setStorageLocationInternal(E, Loc);
+  const Expr &CanonE = ignoreCFGOmittedNodes(E);
+  assert(!ExprToLoc.contains(&CanonE));
+  ExprToLoc[&CanonE] = &Loc;
 }
 
 StorageLocation *Environment::getStorageLocation(const Expr &E) const {
   // See comment in `setStorageLocation()`.
   assert(E.isGLValue() ||
          E.getType()->isSpecificBuiltinType(BuiltinType::BuiltinFn));
-  return getStorageLocationInternal(E);
+  auto It = ExprToLoc.find(&ignoreCFGOmittedNodes(E));
+  return It == ExprToLoc.end() ? nullptr : &*It->second;
+}
+
+// Returns whether a prvalue of record type is the one that originally
+// constructs the object (i.e. it doesn't propagate it from one of its
+// children).
+static bool isOriginalRecordConstructor(const Expr &RecordPRValue) {
----------------
Xazax-hun wrote:

I am wondering whether it is better to enumerate the "transparent" AST nodes where we need to do the propagation.
My questions are:
* Roughly how many such transparent nodes do we have?
* In case we miss something, is it less desirable to erroneously propagate something or to erroneously create a new PRValue?


https://github.com/llvm/llvm-project/pull/75483