Re: [committed] Fix -Wmissing-attributes warning in libstdc++ (PR libstdc++/89402)

[Martin Sebor <msebor at gmail dot com>]

On 2/21/19 6:10 PM, Jonathan Wakely wrote:
> On 20/02/19 09:07 +0100, Jakub Jelinek wrote:
> > Hi!
> >
> > The following patch fixes a warning when building on targets where long
> > double switched from being the same as double to something wider,
> > like e.g. powerpc*-linux, alpha*-linux, s390*-linux etc.
> >
> > Fixed thusly, bootstrapped/regtested on {x86_64,i686,s390x}-linux and
> > bootstrapped on powerpc64le-linux, on both s390x and powerpc64le it 
> > removed
> > all
> > ../../../../libstdc++-v3/src/c++98/compatibility-ldbl.cc:77:17: 
> > warning: 'void _ZNKSt4hashIeEclEe()' specifies less restrictive 
> > attribute than its target 'std::size_t 
> > std::tr1::hash<_Tp>::operator()(_Tp) const [with _Tp = long double]': 
> > 'pure' [-Wmissing-attributes]
> > warnings from the build log.  Preapproved by Jonathan in the PR, 
> > committed
> > to trunk.
>
> Now I see:
>
> /home/jwakely/src/gcc/libstdc++-v3/src/c++98/compatibility-ldbl.cc:78:65: warning: 
> ‘pure’ attribute on function returning ‘void’ [-Wattributes]
>    78 |   _GLIBCXX_PURE __attribute__((alias ("_ZNKSt3tr14hashIeEclEe")));
>       |                                                                 ^

Are the attributes on this alias ever considered in calls to it?
If not (and assuming there's a way to tell, which I think there
should be but haven't verified it) the warning could be relaxed
to only trigger for those that are.  I think those would be
either aliases declared in headers, or those with the TREE_USED
bit set.

Martin

> We can disable that with a diagnostic pragma.
>
>
>
> > 2019-02-20  Jakub Jelinek  <jakub@redhat.com>
> >
> >     PR libstdc++/89402
> >     * src/c++98/compatibility-ldbl.cc (_ZNKSt4hashIeEclEe): Add
> >     _GLIBCXX_PURE to the alias declaration.
> >
> > --- libstdc++-v3/src/c++98/compatibility-ldbl.cc.jj    2019-01-01 
> > 12:39:41.530606161 +0100
> > +++ libstdc++-v3/src/c++98/compatibility-ldbl.cc    2019-02-19 
> > 16:12:52.402123217 +0100
> > @@ -75,6 +75,6 @@ namespace std _GLIBCXX_VISIBILITY(defaul
> > // and std::hash<long double>::operator()
> > // are the same, no need to duplicate them.
> > extern "C" void _ZNKSt4hashIeEclEe (void)
> > -  __attribute__((alias ("_ZNKSt3tr14hashIeEclEe")));
> > +  _GLIBCXX_PURE __attribute__((alias ("_ZNKSt3tr14hashIeEclEe")));
> >
> > #endif
> >
> >
> >     Jakub