Re: [C,C++,Ada,Java,F77] opt/5076 and opt/2847

[Richard Henderson <rth at redhat dot com>]

On Wed, Jan 30, 2002 at 10:28:34AM -0800, Mark Mitchell wrote:
> Why are we getting the wrong answer in expand_end_loop?

  while (test == 0)
    {
      inc ();
      if (iNbr == 0)
        break;
      else

The (iNbr == 0) test looks like part of the conditional since
it jumps to the end of the loop.  That leaves us with just

        {
          inc ();
          iNbr--;
        }
      test = 1;

as the body of the loop, which has too few calls to inc().

> It seems that the simplest thing might be just to notice how
> it's broken, and perhaps conservatively punt at that point if
> it's a case we cannot handle?

How do you tell the difference between a call+test in the condition
and a call+test+break in the body?  I don't think we can. 

The only way I could think is by having a marker.


r~