Richard Biener wrote:
> On Mon, 26 Jan 2015, Jakub Jelinek wrote: > > Then it probably should be ok. I'm really afraid of emitting more warnings > > with such high false positive rate now. > > As the patch also mitigates some of the code bloat we get with > the complete peeling (regression against 4.7) I have installed it. > It's also the easiest vehicle to verify range-info is not broken > by passes between vrp1 and vrp2. You could make warnings appear only for warn_array_bounds > 1 if there are concerns about false positives. For what it's worth, I tested the old version of both patches on one of my projects (mostly numerical algorithms) and it did not produce additional warnings. I really appreciate all improvements in this area. Martin