Re: ppc: fix for pr31281, ObjC try-catch ICEs 4.2

[Stuart Hastings <stuart at apple dot com>]

On Mar 28, 2007, at 11:22 AM, Andrew Pinski wrote:

> On 3/28/07, Stuart Hastings <stuart@apple.com> wrote:
> > Do you want the testcase?  It's pretty nonsensical:
> >
> > int f(unsigned int i)
> > {
> >   @try { } @catch(id) { }
> >   for (;;)
> >     for (;;)
> >       @try {
> >         if (i)
> >           break;
> >       } @catch(id) { }
> > }
> >
> > If you want the testcase, I'm guessing it would become .../testsuite/
> > objc.dg/pr31281.m (?).
>
> Can you turn this into a quick execute testcase and put it into
> objc/execute/exceptions so that it gets run by both the next and gnu
> runtimes.  It might be better if objc/compile/exceptions is added for
> such compile time testcase also.

After some hacking (IANAOCE, I Am Not An Objective-C Expert :-), I got  
this:

/* From PR31281.  */
extern void abort (void);
int __attribute__((noinline))
f(unsigned int i)
{
  int j, k;
  @try { } @catch(id) { return 13; }
  for (j=0; j<i; j++)
    for (k=0; k<i; k++)
      {
        @try {
          if (i)
            break;
        } @catch(id) { }
        return 9;
      }
  return 0;
}

int
main()
{
  if (f(1))
    abort ();
  return 0 ;
}

This seems to ICE the current 4.2 compiler, compile with Ian's  
refinement of my patch, AND execute without error on my PPC OS X box  
(NeXT runtime).

However, it's kinda weak as an executable testcase; it's pretty  
implausible that either of the non-zero returns will get executed, and  
almost any change alters the CFG enough for it to compile with the  
current 4.2.  (Any suggestions to make it more plausible or robust are  
welcome.)

Assuming it's acceptable, where would you like it?  objc/execute/ 
exceptions/pr31281.m ?

> Thanks,

You're very welcome,

stuart hastings
Apple