Re: -funsafe-loop-optimizations

[Geoffrey Keating <geoffk at geoffk dot org>]

dewar@gnat.com (Robert Dewar) writes:

> > (because it is infinite if n is ~0).  So they aren't a "presumed
> > mistake".  I think the warning would only be useful for people who were
> > considering turning on the matching optimization.
> 
> But how useful? I would guess that virtually 100% of such warnings would
> be false positives, in the sense that they are not warning of situations
> in which the standard semantics would be fine. Let's take the example:
> 
> > void foo (unsigned int n)
> > {
> >   for (i = 0; i <= n; i++)
> >     bar ();
> > }
> 
> I simply don't believe that anyone would write this code, expecting that
> in the case where n was the max value, an infinite loop would occur, and
> that's just what the programmer wants.

They might not expect an infinite loop, but they might very well
expect that the last case gets executed; for instance, if bar does

if (a[i] == i)
  exit (0);

and 'a' happens to be an array with exactly 256 elements, and 'i' is
of type 'unsigned char'.

IMO, this kind of thing is a really slippery slope.  There are
probably hundreds of optimisations that are much easier to do if you
just assume some apparently-trivial condition like "this procedure
terminates".