Re: [PATCH] mm/gfp: Add kernel-doc for gfp_t

[Nick Desaulniers <ndesaulniers@google.com>]

On Fri, Feb 19, 2021 at 2:15 PM Miguel Ojeda
<miguel.ojeda.sandonis@gmail.com> wrote:
>
> On Fri, Feb 19, 2021 at 10:45 PM 'Nick Desaulniers' via Clang Built
> Linux <clang-built-linux@googlegroups.com> wrote:
> >
> > That said, I'm fine disabling this warning; there's a separate error
> > for redefining a typedef to a different underlying type.  That's
> > what's useful IMO, this one really is not.
> >
> > This warning doesn't really provide any value to us in the kernel; I
> > would guess the intent was to be helpful to code expected to be
> > portable across different -std=*
>
> It seems it would also be useful to sport unintended cases, e.g.:
>
>   - Collisions on short identifiers (that by chance typedef to the same type).
>   - Copy-pasting and forgetting to remove the original definition
> (i.e. it should have be cut-pasting instead).
>   - Double inclusion of headers (with missing or broken #ifdef guards).

(There is a separate warning flag for broken header guards,
-Wheader-guard:
https://github.com/ClangBuiltLinux/linux/issues?q=is%3Aissue+label%3A-Wheader-guard+is%3Aclosed)

What happens should the kernel move to gnu11, as discussed once GCC
5.1+ becomes the minimum supported version for all arches?
https://lore.kernel.org/lkml/CAHk-=whnKkj5CSbj-uG_MVVUsPZ6ppd_MFhZf_kpXDkh2MAVRA@mail.gmail.com/

Then the warning will not fire, since it's only really meant to help C
code be portable between -std=c11.

(Another change to clang could be to move this flag into the
-Wpedantic group, which is only really meant for from guarding against
the use of non-ISO C functionality, but that still would require
disabling the warning for older but supported versions of clang).

>
> Those may be providing value in the kernel. In particular, if we don't
> see any warning at the moment, it means those cases are not happening
> now anywhere, so we would be weakening things.
>
> Having said that, I don't see the original patch, so perhaps I am
> missing something.

https://lore.kernel.org/linux-mm/20210215204909.3824509-1-willy@infradead.org/

-- 
Thanks,
~Nick Desaulniers