Dangling false positive if the the owner is also moved in the initializer.

[hokein]

This issue is identified in #112751.

// case1
namespace std {
template<typename T>
struct unique_ptr {
  T &operator*();
  T *get() const [[clang::lifetimebound]];
};
} // namespace std

struct X {
  X(std::unique_ptr<int> up) :
    pointer(up.get()), owner(std::move(up)) {}

  int *pointer;
  std::unique_ptr<int> owner;
};

When we add the clang::lifetimebound annotation to unique_ptr::get(), clang emits a dangling-field warning for the pointer(up.get()) member initializer. This warning is a false positive in this context, as the owner member is moved as part of the initialization, retaining ownership.

Another example occurs in designated-initializer cases:

// case2
struct X {
   int *pointer;
   std::unique_ptr<int> owner;
};

X func(std::unique_ptr<int> up) {
   return {
      .pointer = up.get(),
      .owner = std::move(up)
   };
}

Fixing these false positives is hard because it would require tracking dependencies between expressions, which is beyond the capabilities of the current statement-local analysis.

[hokein]

Interestingly, clang will emit the diagnostic on case 2 even the get() is not annotated with lifetimebound, https://godbolt.org/z/vYeG9Eb3Y.

The GSL analysis implicitly tracks the *this object for some hard-coded GSL owner methods (vector.get()), the case 1 have been filtered out here, but it is just for the GSL code path.