[Bug c++/109653] New: [[clang::fallthrough]] recognised up to g++ v12 not not g++ v13

[Bug c++/109653] New: [[clang::fallthrough]] recognised up to g++ v12 not not g++ v13

[gcc-bugs at aitchison dot me.uk]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109653

            Bug ID: 109653
           Summary: [[clang::fallthrough]] recognised up to g++ v12 not
                    not g++ v13
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gcc-bugs at aitchison dot me.uk
  Target Milestone: ---

Created attachment 54945
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54945&action=edit
Test alternate spelling clang::fallthrough

g++ v12 and earlier ignore the line
    [[clang::fallthrough]];
but v13 complains:
warning: attributes at the beginning of statement are ignored [-Wattributes]

This is a common alternative spelling of
  [[gnu::fallthrough]];
and
  [[fallthrough]];
so it is a pity that when g++ supports the feature in its default mode (ie
without specifying  a --std) it starts objecting to code which on some other
compilers enables the functionality.

[Bug c++/109653] [[clang::fallthrough]] recognised up to g++ v12 not not g++ v13

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109653

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> ---
The warning started with r13-3149
"c++: Improve handling of foreigner namespace attributes"

[Bug c++/109653] [[clang::fallthrough]] recognised up to g++ v12 not not g++ v13

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109653

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Hmm, __has_cpp_attribute never returns 0 for unknown attributes

#if defined __has_cpp_attribute 
#  if __has_cpp_attribute (fakeattribute)
#    error errorout
#  endif
#endif

I thought it should.

[Bug c++/109653] [[clang::fallthrough]] recognised up to g++ v12 not not g++ v13

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109653

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #2)
> Hmm, __has_cpp_attribute never returns 0 for unknown attributes
> 
> #if defined __has_cpp_attribute 
> #  if __has_cpp_attribute (fakeattribute)
> #    error errorout
> #  endif
> #endif
> 
> I thought it should.

Never mind, it does.

This is how you are supposed to use attributes:
#if defined __has_cpp_attribute 
#  if __has_cpp_attribute (clang::fallthrough)
#   define f [[clang::fallthrough]]
#  elif __has_cpp_attribute (gnu::fallthrough)
#   define f [[gnu::fallthrough]]
#  elif  __has_cpp_attribute (fallthrough)
#   define f [[fallthrough]]
# endif
#endif

#ifndef f
#define f
#endif


#include <cstdio>

int main (int argc, char* argv[])
{
    switch(argc)
    {
    case 0:
        printf("%s\n", argv[1]);
        f;
    default:
        printf("%d args\n", argc);
    }

    return 0;
}

[Bug c++/109653] [[clang::fallthrough]] recognised up to g++ v12 not not g++ v13

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109653

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
That change was completely intentional, previously for some of the well known
attribute names from random other namespaces we were handling them like the gnu
or standard ones, which is obviously wrong, we don't really know anything about
some other vendor wants to treat them.
Generally, one can use e.g. -Wno-attributes=clang:: to avoid warnings about
clang:: namespace attributes, but in this case it warns anyway.  Maybe we
should filter away the -Wno-attributes= in the statement cases as well, rather
than just in the more usual attribute spots.