I analyzed the problem in PR port-sparc64/11506 a bit and came to the
conclusion that it's a bad volatile cast in the ksh source.

But then additional toolchain lossage strikes, and I couldn't reproduce it
with a small example. So here the real code:

Index: jobs.c
===================================================================
RCS file: /cvsroot/basesrc/bin/ksh/jobs.c,v
retrieving revision 1.6
diff -r1.6 jobs.c
1176a1177,1184
> 	volatile int * jstate = &j->state;
> 
> /*
> 	{
> 	   char buf[20];
> 	   sprintf(buf, "%p", jstate);
> 	}
> */
1188c1196
< 	while ((volatile int) j->state == PRUNNING
---
> 	while (*jstate == PRUNNING
1190c1198
< 		    && (volatile int) j->state == PSTOPPED))
---
> 		    && *jstate == PSTOPPED))


I would expect this to work. It crashes with a SIGSEGV, because "jstate" is
zero (according to gdb).

If I uncomment the meaningless sprintf block in above patch, it works.

Any ideas? I certainly don't want to commit that sprintf block ;-)


Martin

P.S.: Just pure speculation, but maybe the "sometimes ^C'ing make does not stop
sub-makes" maybe a similar (volatile cast related) effect in our make (but I
didn't look at the source yet).