|
From: Patrick J. L. <lop...@gm...> - 2017-04-25 19:07:56
|
On Tue, Apr 25, 2017 at 11:42 AM, Julian Seward <js...@ac...> wrote: > > I thought it was the default now. Isn't it? It certainly should be. > > But I think that's not the issue. Memcheck isn't complaining about > bad addresses here. It's complaining that the resulting branch is > on undefined data -- and that undefined data exists exactly because > the partial-loads machinery paints the data acquired from the > invalid areas as undefined. So the partial-loads machinery is > working perfectly. IIUC, that is. Yes, my mistake. I should have read more carefully. This looks very hard to fix in general, since Valgrind is correct that the branch depends on invalid data. Setting partial-loads-ok=no would not help; it would just move the complaint to the memory access itself. Maybe try compiling the application with "-fno-builtin-strcmp"? - Pat |