Life optimization of a C ++ object

Question

I am trying to use RAII to create objects that act on a thread until they are destroyed. For example, I have a thread that is reset every time endl added. Most of the time I want this, but not always. I want to say "not flush on endl", but I also need it to be safe for exceptions. Therefore, I cannot just do stream->NoFlush() to set the class member. Anyway, I want to know this. If I have a code like

 CStreamModifier no_flush; stream->NoFlush(no_flush); // as long as no_flush is in scope I get the behaviour I want ... do some stuff on the stream, without referencing no_flush ... // no_flush goes out of scope here. 

Is the compiler allowed to optimize the no_flush lifetime? For example, it is not used after line 2, but I need it to stay until the end. I really have not heard of optimizations like this, so I think I'm fine, but I would like to make sure.

Answer 1

No, the compiler cannot optimize this. The destructor will be called exactly when the object goes out of scope.

What he can do is optimize his copies if NoFlush accepts a parameter by value, but that doesn't matter.

Copying elision is the only optimization a compiler can do that affects observable behavior.

Answer 2

I would do it like this:

 struct Stream { bool flush = true; // ... }; struct NoFlush { explicit NoFlush(Stream & s) : stream(s) , prev(stream.flush) { stream.flush = false; } ~NoFlush() { stream.flush = prev; } Stream & stream; bool prev; }; 

Now you can use it as follows:

 void foo(T arg, S brg, Stream & stream) { NoFlush _(stream); // do stuff }