Re: Optimzed vs. readable code

[Joe Buck <Joe dot Buck at synopsys dot COM>]

On Wed, Nov 30, 2005 at 06:14:44PM +0100, Pierre THIERRY wrote:
> I don't see the point. I repeat: i would not have even proposed the
> modification if I wasn't expecting any decent compiler to do the
> following optimization:
> 
> 	void foo(Iterator first, Iterator last)
> 	{
> 		Iterator current = first;
> 		for ( ; current != last; ++current)
> 			do_something(current);
> 		}
> 	}
> 
> becomes:
> 
> 	void foo(Iterator first, Iterator last)
> 	{
> 		for ( ; first != last; ++first)
> 			do_something(first);
> 		}
> 	}

Then gcc is not a decent compiler.  Actually, I suspect that no conforming
C++ compiler is allowed to be "decent" in the general case: if the
constructor or destructor of Iterator has a side effect, the optimization
you suggest would change the order of side effects.  There are cases
where this is allowed (elimination of copy constructors), but I don't
think that this is such a case.