I don't understand. Are you suggesting that lock somehow relinquishes ownership of mutex before new (&lock) char[sizeof lock] ends its lifetime by reusing its storage to hold an array of characters?

There are two objects of type std::lock_guard<std::mutex> in this program. One that owns mutex during its entire lifetime, and one that owns dummy_mutex during its entire lifetime. Both objects occupy the same storage but at different times.

Thank you for your quick response, I'm happy to clarify

Are you suggesting that lock somehow relinquishes ownership of mutex before new (&lock) char[sizeof lock] ends its lifetime by reusing its storage to hold an array of characters?

No, that'd be (more or less) impossible.

Let me preface this by saying it's entirely possible I'm misreading / misunderstanding but to me [thread.lock.guard] p1 implies that it should release the mutex when its lifetime ends. But in this case the mutex is not unlocked.

"I check my email frequently while I am at work" implies nothing about my email-checking frequency when I'm at home. I might not check at all, or check rarely, or check frequently. None of those would contradict the fact that I check frequently while I am at work.

Similarly, "A lock_guard object maintains ownership of a lockable object throughout the lock_guard object's lifetime" says nothing about what a lock_guard object may or may not do outside its lifetime. A zombie lock_guard continuing to own a lockable object may not be ideal, but it still doesn't contradict a statement about the behavior of living lock_guard objects.