On 2026-08-15 Bruno Haible wrote:
> Indeed, I see a few test failures in the test cases, still (despite
> the no_sanitize ("function") attributes). Fixing them through the
> attached patch.Thanks! > But some other test failures will remain: > - jit/test-cache, > - test-list-c++, test-map-c++, test-omap-c++, test-oset-c++. > > The reason is that C is not only used for application programming — > for which the above options "-flto -fsanitize=cfi -fvisibility=hidden" > make sense [except for GNOME / Qt and their "signals"] — but also for > system programming, where treating functions of different prototypes > in the same way is commonplace (think of libffi or libffcall), or for > generics (where C++ programmers like to have 10 different code > instantiations for 10 different pointer types, but C programmers don't > like that). Maybe there are solutions to those (maybe the nocf_check attribute), but fixing them doesn't sound urgent. Currently Linux on x86-64 supports shadow stack (backward edge CFI) for userspace, but indirect branch tracking (forward edge CFI) isn't supported for userspace yet.[1] [1] https://lkml.org/lkml/2026/5/17/734 -- Lasse Collin
