[Bug c++/99677] New: [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[Bug c++/99677] New: [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

            Bug ID: 99677
           Summary: [[gnu::const]] attribute does not prevent dead global
                    variable gets emitted.
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: unlvsur at live dot com
  Target Milestone: ---

https://godbolt.org/z/Toncas

The code tries to import functions from DLLs in the global.

Clang does not generate the nt_create_file function pointer, GCC does. Even at
-O0, clang emits no code. It only emits code any function actually uses it.

That is super annoying for Windows since we have no way to invoke Windows NT
syscalls without emitting dead code with GCC.

[Bug c++/99677] [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

--- Comment #1 from cqwrteur <unlvsur at live dot com> ---
Here is the GCC. it emits code of dead global variables
https://godbolt.org/z/oWddaf

[Bug c++/99677] [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

--- Comment #2 from cqwrteur <unlvsur at live dot com> ---
Even with [[gnu::pure]] clang does not emit code either.
https://godbolt.org/z/e4sG6r

We can see it emits code only when we do not mark the functions are pure.
https://godbolt.org/z/s3cEK3

That is not the case for GCC.

[[gnu::pure]] is a GCC attribute but clang does things better. So sad :(

[Bug c++/99677] [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I think this is a dup of bug 99456.

[Bug c++/99677] [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

--- Comment #4 from cqwrteur <unlvsur at live dot com> ---
(In reply to Andrew Pinski from comment #3)
> I think this is a dup of bug 99456.

I do not think so because this has nothing to do with constexpr.

better @jakub to see whether he has some solutions.

[Bug c++/99677] [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think Andrew is correct.  Note we can't elide "unused" not statically
initialized variables (since the initialization is seen as use).

[Bug c++/99677] [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[unlvsur at live dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

--- Comment #6 from cqwrteur <unlvsur at live dot com> ---
(In reply to Richard Biener from comment #5)
> I think Andrew is correct.  Note we can't elide "unused" not statically
> initialized variables (since the initialization is seen as use).

but here the result is always the same from importing functions from ntdll.dll
or kernel32.dll

[Bug c++/99677] [[gnu::const]] attribute does not prevent dead global variable gets emitted.

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99677

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

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

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
As the functions aren't constexpr, they aren't even considered during C++ FE
constexpr processing.  gnu::const means something completely different.
I bet we optimize away static const variables only written (even with
non-constant initializers) and never address taken or otherwise used, this PR
seems to ask to handle inline const (with the C++ FE specific behavior for
them) in such cases too.
I'm not sure it is valid if the vars are odr-used somewhere and the odr-uses
are later optimized away, I think emitting the inline vars in that case is part
of the ABI.  And the not odr-used cases which are only used by their
initialization is yet another question.