On Mon, Mar 18, 2019 at 10:31 PM Jeff Law <l...@redhat.com> wrote: > > On 3/18/19 1:03 PM, Martin Sebor wrote: > > I the -Warray-bounds enhancement committed at the beginning > > of the GCC 9 window I tried to correctly handle offsets in > > MEM_REFs in the [max, min] kind of a range after converting > > from sizetype to ptrdiff_type, but I did it wrong. The bug > > results in false positives in some unusual use cases that > > I didn't consider at the time. > > > > The attached patch removes this incorrect handling and uses > > the conservative anti-range handling for these cases instead. > > > > Martin > > > > PS Is there some technical reason why pointer offsets are > > represented as the unsigned sizetype when they can be signed? > I'm not aware of a conscious decision to treat them as unsigned or a > particular target need to do so. It's most likely a historical accident.
That historical accident included treating those unsigned types as sign-extending ... But yes, changing sizetype to ssizetype wherever we use it in offset context would be a cleanup I guess. But IIRC Bin tried this and the fallout is (as usual) non-trivial to fix... Richard. > > > > > gcc-89720.diff > > > > PR tree-optimization/89720 - Spurious -Warray-bounds warning on a range > > with max < min > > > > gcc/ChangeLog: > > > > PR tree-optimization/89720 > > * tree-vrp.c (vrp_prop::check_mem_ref): Treat range with max < min > > more conservatively, the same as anti-range. > > > > gcc/testsuite/ChangeLog: > > > > PR tree-optimization/89720 > > * gcc.dg/Warray-bounds-42.c: New test. > OK > jeff
