Re: [PATCH] include/linux/compiler*.h: make compiler-*.h mutually exclusive

[Nick Desaulniers]

On Thu, Aug 23, 2018 at 2:19 PM Joe Perches <joe@xxxxxxxxxxx> wrote:
>
> On Thu, 2018-08-23 at 14:03 -0700, Nick Desaulniers wrote:
> > One reply for a bunch of the various threads, to keep the number of emails down:
> >
> > On Wed, Aug 22, 2018 at 5:20 PM Joe Perches <joe@xxxxxxxxxxx> wrote:
> > > On Wed, 2018-08-22 at 16:37 -0700, Nick Desaulniers wrote:
> > > > +/* Compiler specific macros. */
> > > >  #ifdef __clang__
> > > >  #include <linux/compiler-clang.h>
> > >
> > > probably better as
> > >
> > > #if defined(__clang)
> > >
> > > to match the style of the #elif defined()s below it
> >
> > Hi Joe,
> > Thanks for the feedback. I always appreciate it.  If you have some
> > cleanups, want to send them to me, and I'll bundle them up for a PR?
> > I'm ok with that change.
> >
> > > > +#ifdef __GNUC_STDC_INLINE__
> > > > +# define __gnu_inline        __attribute__((gnu_inline))
> > > > +#else
> > > > +# define __gnu_inline
> > > > +#endif
> > >
> > > Perhaps __gnu_inline should be in compiler-gcc and this
> > > should use
> > >
> > > #ifndef __gnu_inline
> > > #define __gnu_inline
> > > #endif
> >
> > Not this case; it's how we get gnu89 semantics for `extern inline` is
> > not compiler specific (therefor should not go in a compiler specific
> > header).
>
> It's not possible to know that compilers support what
> __attribute__((<foo>)) and at what version that support
> exists unless it is specified somewhere.

__has_attribute:
https://clang.llvm.org/docs/LanguageExtensions.html#has-attribute

The release notes of GCC-5 mention __has_attribute.
https://gcc.gnu.org/gcc-5/changes.html

The point of feature detection is that it _doesn't matter_ what
version that support exists.  Either it does and you can use it, or it
doesn't and you can decide whether to stop compiling or there's a
valid work around.

Feature detection should be preferred to explicit version checks
except in 2 specific cases:
1. It's not possible to properly perform feature detection.  Language
features should not be added unless it's possible to safely check for
them.
2. A very specific version of a very specific compiler is broken and
needs to be explicitly guarded against.

> As far as I can tell,  gnu_inline is not recognized by clang.
>
> https://clang.llvm.org/docs/AttributeReference.html

If that was the case, I would not have added it in commit d03db2bc26f0
("compiler-gcc.h: Add __attribute__((gnu_inline)) to all inline
declarations").
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d03db2bc26f0e4a6849ad649a09c9c73fccdc656

Docs can sometimes fall behind, the lone source of truth is the source code.
https://github.com/llvm-mirror/clang/search?q=gnu_inline&unscoped_q=gnu_inline

Godbolt is also incredibly helpful for testing various compiler versions:
https://godbolt.org/z/uMJ-mo


https://reviews.llvm.org/D51190

-- 
Thanks,
~Nick Desaulniers