There was a more egregoius one which got Linus further pissed off with GCC, which was due to a 'dereference' that would not trap, but still deleted a later null check (because e.g. int *foo = &bar->baz is basically just calculating an offset to bar, and so will not fail at runtime, but it is still a dereference according to the abstract machine and so is undefined if bar is NULL). I think the risk of something like that is why it's still disabled.