Re: try/finally in GNU C

[Mike Stump <mrs at apple dot com>]

On Jun 15, 2006, at 4:47 PM, Andrew Pinski wrote:
> > The front-end in question has stopped working because the traditional
> > setjmp/longjmp translation of try/catch constructs is no longer
> > working correctly with versions of GCC higher than 4.0.

> How is it no longer working?  I don't understand how it could break.

I can.  Just do up the simplest use of __builtin_setjmp and  
__builtin_longjmp and you'll see that there are no control flow  
edges.  People totally broke C++ on non-dw2 systems?!  This also  
caused warnings for variables that should be marked volatile that are  
homed in registers that cross the boundary to not give the `can be  
clobbered' warning.  If there were an edge, it would be put in a  
register and then the warning kicks in.  Without the edge, ssa  
`knows' the variable doesn't change:

foo() {
  int i = 99;
  __builtin_setjmp(A)
  if (i) {
    print i
    --i;
    __builtin_longjump(A);
  }

It used to not infinite loop, now it does.