[PATCH] D48460: [analyzer] Fix invalidation on C++ const methods.

[Artem Dergachev via Phabricator via cfe-commits]

NoQ added a comment.

Well, i guess that's just one of those mildly infuriating aspects of the AST and/or the standard :)

> If references are resolved, why are pointers not?

It's kinda understandable. References are simple aliases to glvalues. When you use a reference, you simply get *the* original glvalue, which is exactly what the AST gives you: a glvalue expression of object type. Here's what `Expr::setType()` says:

  130   void setType(QualType t) {
  131     // In C++, the type of an expression is always adjusted so that it
  132     // will not have reference type (C++ [expr]p6). Use
  133     // QualType::getNonReferenceType() to retrieve the non-reference
  134     // type. Additionally, inspect Expr::isLvalue to determine whether
  135     // an expression that is adjusted in this manner should be
  136     // considered an lvalue.
  137     assert((t.isNull() || !t->isReferenceType()) &&
  138            "Expressions can't have reference type");
  139
  140     TR = t;
  141   }



----

This code is still imperfect because it doesn't take dynamic type of the region into account, which may be more specific than the expression's type. I guess i'll add a FIXME.


Repository:
  rC Clang

https://reviews.llvm.org/D48460