On Fri, Dec 17, 2021 at 12:26:56AM +0530, Siddhesh Poyarekar wrote:
> On 12/16/21 21:19, Jakub Jelinek wrote:
> > Yes, but please fix up formatting, wholeval should go below old_wholeval.
> > Though, perhaps it would be better to:
> >
> > if (tree_int_cst_compare (oldval, val))
> > return true;
> >
> > gcc_checking_assert (tree_int_cst_compare (old_wholeval, wholeval) == 0);
> > return false;
> >
> > Ok with that change.
> >
> > Jakub
> >
>
> I thought about this some more and I think I'm wrong in assuming that
> wholeval won't change in subsequent passes; my tests show I'm wrong too.
> I'm now testing with the condition fixed up to:
>
> return (tree_int_cst_compare (oldval, val) != 0
> || tree_int_cst_compare (old_wholeval, wholeval) != 0);
>
> as you had suggested earlier and will push it if it passes bootstrap and a
> clean testsuite run. I'm doing a ubsan bootstrap too because it's been
> quite useful in catching corner cases in __bos before.
Ok.
Jakub
