[Bug c++/90670] const& template parameter in a header trips -Wsubobject-linkage

[Bug c++/90670] const& template parameter in a header trips -Wsubobject-linkage

[Kitsune-Ral at users dot sf.net]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90670

Alexey Rusakov <Kitsune-Ral at users dot sf.net> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |Kitsune-Ral at users dot sf.net

--- Comment #1 from Alexey Rusakov <Kitsune-Ral at users dot sf.net> ---
Reproduced with GCC 12.1.1 on Fedora 36.

[Bug c++/90670] const& template parameter in a header trips -Wsubobject-linkage

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90670

Patrick Palka <ppalka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |ppalka at gcc dot gnu.org

--- Comment #2 from Patrick Palka <ppalka at gcc dot gnu.org> ---
The warning seems right to me: the const DummyId has internal linkage, which
makes the type seChildScheduleAccess<DummyId> have internal linkage, which
makes the definition of seGccBugJob (strictly speaking) different in different
TUs.  So the warning is correctly exposing a potential ODR violation.

To fix the ODR violation, you can declare DummyId 'inline'.

[Bug c++/90670] const& template parameter in a header trips -Wsubobject-linkage

[adam1.byrd at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90670

--- Comment #3 from Adam Byrd <adam1.byrd at gmail dot com> ---
Hmm, I reckon that makes sense. Though the wording is confusing. Why is it
"uses the anonymous namespace" instead of something like "has internal
linkage".

I'm not aware of an anonymous namespace being involved here. Is that language
esoterica or a compiler implementation detail?

[Bug c++/90670] const& template parameter in a header trips -Wsubobject-linkage

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90670

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=86491

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Adam Byrd from comment #3)
> I'm not aware of an anonymous namespace being involved here. Is that
> language esoterica or a compiler implementation detail?

Since GCC 13, the diagnostic is:
In file included from t66.cc:1:
t66.h:7:8: warning: ‘seGccBugJob’ has a field ‘seChildScheduleAccess<DummyId>
seGccBugJob::m_test’ whose type has internal linkage [-Wsubobject-linkage]
    7 | struct seGccBugJob
      |        ^~~~~~~~~~~

Basically the diagnostic was not updated for C++11 changes ...