int x = 3, y = 4;
cout << "The value is " << y << endl;
This would warn you that you never used the value 'x', and is a useful warning. But I think they're too overzealous:
int success = doSomethingImportant();
cout << "Did something important. It was "
<< (success ? "successful." :
"unsuccessful.") << endl;
If you don't define the _DEBUG_ON_ symbol, GCC will issue a warning here. And if it's late at night and you're not paying attention, you might make a mistake and move that 'doSomethingImportant()' call inside of the #ifdef, because after all, GCC is warning you that it's not doing anything. And sometime on Saturday, while fixing several hundred warnings, I did precisely that (although in my defense, the real code was a bit longer). It took several hours to figure out why the test suite was failing; if we hadn't had a test suite, we may have accidentally allowed a very buggy piece of software out the door. As it was, it kept me at work late.
This particular warning strikes me as poorly thought out. I agree with the warning when the initialization value is constant, but when the initialization is complex, with the potential for side effects, I believe the warning should not be issued. It certainly would have saved me a literal headache today.
I haven't tried this with the latest GCC yet, by the way; I only have a really old 3.4.5 compiler on my machine. Feel free to contradict my conclusions in a flaming, foaming-at-the-mouth sort of way.