Re: [PATCH] bitops: avoid clang shift-count-overflow warnings

[Nick Desaulniers]

On Tue, May 5, 2020 at 8:33 AM Arnd Bergmann <arnd@xxxxxxxx> wrote:
>
> On Tue, May 5, 2020 at 4:08 PM Andy Shevchenko
> <andy.shevchenko@xxxxxxxxx> wrote:
> > On Tue, May 5, 2020 at 4:58 PM Arnd Bergmann <arnd@xxxxxxxx> wrote:
> > > Clang normally does not warn about certain issues in inline functions when
> > > it only happens in an eliminated code path. However if something else
> > > goes wrong, it does tend to complain about the definition of hweight_long()
> > > on 32-bit targets:
> >
> > Shouldn't it be fixed in CLang?
> >
> > > include/linux/bitops.h:75:41: error: shift count >= width of type [-Werror,-Wshift-count-overflow]
> > >         return sizeof(w) == 4 ? hweight32(w) : hweight64(w);
> > >                                                ^~~~~~~~~~~~
> >
> > sizeof(w) is compile-time constant. It can easily drop the second part
> > without even looking at it.
> >
> > > Adding an explicit cast to __u64 avoids that warning and makes it easier
> > > to read other output.
> >
> > Looks like papering over the real issue.
>
> I'm not sure if there is anything to be done about it in clang, since it
> always does syntactic analysis before dead-code elimination by design.

That's pretty much it.  We had a patch to Clang to use delayed
diagnostics to delay emitting the warning in case the AST node was
dropped, but it wasn't accepted in code review.

>
> It is a bit odd though that it only prints the warning sometimes, but

Sometimes?

> I suspect this is also something that works as designed. Maybe someone
> on the clang-built-linux list knows more about the background.


-- 
Thanks,
~Nick Desaulniers