--- Comment #4 from pinskia at gcc dot gnu dot org 2006-09-08 04:31 ---
Reduced testcase:
extern void dynreplace_trampoline(void);
extern void dynreplace_trampoline_endlabel(void);
int dynreplace_add_trampoline(void)
{
unsigned long trampoline_code[(((unsigned long)
(&(dynreplace_tramp
--- Comment #5 from pinskia at gcc dot gnu dot org 2006-09-08 04:33 ---
Confirmed, note this is semi undefined code. Really this is a big weird as we
are taking the difference between two function pointers. I think the ICE with
tree_low_cst has been fixed but the ICE with loc_descripto
--- Comment #6 from pinskia at gcc dot gnu dot org 2006-09-08 04:44 ---
This is just like PR 14492 now.
Janis,
Could you do a regression hunt on when the reduced testcase in comment #4
started to pass without any options?
Thanks,
Andrew
--
pinskia at gcc dot gnu dot org changed:
--- Comment #7 from janis at gcc dot gnu dot org 2006-09-08 19:11 ---
A regression hunt on powerpc-linux, using the reduced testcase from comment #4
with no options, identified this patch for which that test starts compiling
cleanly:
http://gcc.gnu.org/viewcvs?view=rev&rev=116450