On Tue, Aug 11, 2015 at 1:34 PM, Dan Albert <danalb...@google.com> wrote:
> Yeah, those sound like exactly what we want. Helping people find UB is
> good, but optimizing assuming we've fixed all of the UB isn't something we
> can do.
>
Dan -- that's the situation you're in today.
GCC has done that kind of optimization for *years*.
Consider the following code (simplified to the point that it's a toy,
but...)
void * doFoo ( void *p, size_t sz ) {
std::memcpy(buf, p, sz);
if (p == nullptr)
p = malloc (10);
return p;
}
int main() {
void * q = doFoo(nullptr, 0);
std::cout << q << std::endl;
}
You see the UB there - right? memcpy(buf, null, 0);
Built that with gcc (I used 4.9) - and see what it prints.
At -O3 (on my machine - Mac OS X) it prints "0".
It has removed the check for nullptr, and the malloc.
Interestingly enough, clang completely elides the call to doFoo, and just
calls malloc(10).
-- Marshall
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
