On Wed, May 25, 2005 at 02:09:12PM -0400, Richard Kenner wrote:
> I may be mistaken, but didn't GCC miscompile the Ada front end because
> of this VRP bug here: http://gcc.gnu.org/PR21332 ?
>
> I'm not sure if that was it, but there was a VRP item on my plate from a
> few weeks ago since I claimed that it was an Ada front-end bug and not
> a VRP bug. Unfortunately, I got distracted and then was away. I will make
> a real attempt to look at it in the next few days.
>
The VRP bug you remember is already worked around by
2005-05-02 Diego Novillo <[EMAIL PROTECTED]>
* tree-vrp.c (set_value_range_to_nonnull): Declare inline.
(set_value_range_to_null): Likewise.
(set_value_range_to_varying): New function. Use it in all
call sites that used to call set_value_range (vr, VR_VARYING, ...).
(extract_range_from_assert): If the integral type has a
super-type, and LIMIT is not within MIN and MAX values of that
type, set the resulting range to VR_VARYING.
As soon as the Ada FE bug is fixed, we can remove the
work-around.
PR21332 is fixed by a local patch I'm working on and by a patch
from Steven Bosscher that I approved last Friday (though I
believe he still hasn't committed yet).
The other bug related to Ada is PR21029. It needs an scev
maintainer to look at. I've already pinged Sebastian and Dan,
but I'm not sure if they've had the time to look at it yet.
Diego.
