[Bug c/103960] New: Clang's -Wunknown-attributes is more useful than -Wattributes

[Bug c/103960] New: Clang's -Wunknown-attributes is more useful than -Wattributes

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103960

            Bug ID: 103960
           Summary: Clang's -Wunknown-attributes is more useful than
                    -Wattributes
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: diagnostic
          Severity: enhancement
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: redi at gcc dot gnu.org
  Target Milestone: ---

(This is really c-family not c)

GCC's -Wattributes is used for both completely unknown attributes (e.g. the
user has a typo, or they use a new attribute on an old GCC) and for misplaced
attributes (e.g. a function attribute on a variable). This leaves users
wondering what they've done wrong.

For example, GCC 9 prints this because 'access' was added in GCC 10:

f.c:5:1: warning: ‘access’ attribute directive ignored [-Wattributes]
    5 | __attribute__ ((access (read_only, 1))) bool my_func (const MyStruct *
const self);
      | ^~~~~~~~~~~~~

This doesn't tell the user *why* it was ignored. Did they just use the wrong
grammar? You have to be quite familiar with GCC to know that using it in the
wrong place would give a different diagnostic:

f.c:5:1: warning: ‘access’ attribute only applies to function types
[-Wattributes]
    5 | bool my_func (const MyStruct * const __attribute__ ((access (read_only,
1)))  self);
      | ^~~~

Apart from the caret being in the wrong place, this is pretty good. But for
unknown attributes our diagnostic is vague about what the problem is.

Clang's -Wunknown-attributes is more explicit:

f.c:5:17: warning: unknown attribute 'access' ignored [-Wunknown-attributes]
__attribute__ ((access (read_only, 1))) bool my_func (const MyStruct * const
self);
                ^~~~~~~~~~~~~~~~~~~~~

This tells you clearly that clang simply doesn't recognize the 'access'
attribute.

Could we just add the word "unknown" to the diagnostic for unrecognised
attributes?

[Bug c/103960] Clang's -Wunknown-attributes is more useful than -Wattributes

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103960

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So it looks like the warning is coming from the front-end though and not from
c-family :).

I noticed this because take:
struct MyStruct;
[[gg (read_only, 1)]] int my_func (const struct MyStruct * const self);

----- CUT ----
The C front-end produces:
<source>:2:1: warning: 'gg' attribute ignored [-Wattributes]

While the C++ front-end produces:
<source>:2:70: warning: 'gg' attribute directive ignored [-Wattributes]

[Bug c/103960] Clang's -Wunknown-attributes is more useful than -Wattributes

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103960

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2022-01-10
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
But yes I agree GCC's warning should be much better.

[Bug c/103960] Clang's -Wunknown-attributes is more useful than -Wattributes

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103960

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=64332

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note the location issue is PR 64332.

[Bug c/103960] Clang's -Wunknown-attributes is more useful than -Wattributes

[avi at scylladb dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103960

Avi Kivity <avi at scylladb dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |avi at scylladb dot com

--- Comment #4 from Avi Kivity <avi at scylladb dot com> ---
It would be nice to have -fignore-attribute-namespace=clang, so we can ignore
[[clang::very_useful_attribute(blah)]] on gcc and
[[gnu::another_good_thing(foo)]] on clang, instead of having to wrap everything
with macros.

(complicated a little by clang recognizing some attributes in the gnu
namespace).

[Bug c/103960] Clang's -Wunknown-attributes is more useful than -Wattributes

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103960

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
-Wno-attributes=clang:: ?

[Bug c/103960] Clang's -Wunknown-attributes is more useful than -Wattributes

[avi at scylladb dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103960

--- Comment #6 from Avi Kivity <avi at scylladb dot com> ---
Ah! thanks.