[PATCH] [analyzer][Review request] Improved checker lists.

[Jordan Rose]

Okay, let's look at this example:

  class A {
  public:
    virtual void f();
  };
  
  class B : public A {
  public:
    void f();
  };
  
  void test() {
    A *a = new A;
    new(a) B;
    a->f(); // warn
  }

Both `A` and `B` have `f()` methods, so it's okay to invoke `f()` on an instance of type `B` (whether through an `A*` or a `B*`). //This// code is perfectly acceptable:

  void test() {
    A *a = new A;
    a->~A();
    static_assert(sizeof(A) == sizeof(B) && alignof(A) == alignof(B),
                  "different allocation requirements");
    new(a) B;
    a->f(); // okay, calls B::f.
  }

In fact, this would be okay even if `A::f` weren't virtual.

I think the interesting thing to check is whether you are reinitializing an object whose destructor hasn't been called, and if there are any references to sub-objects used after the destructor has been called. Simply accessed members of the new object through the existing pointer is not an error.

http://reviews.llvm.org/D3457