Re: [PATCH] detect incompatible aliases (PR c/81854)

[Martin Sebor <msebor at gmail dot com>]

On 08/18/2017 10:48 AM, Jonathan Wakely wrote:
> On 18/08/17 08:54 -0600, Martin Sebor wrote:
> > On 08/18/2017 07:10 AM, Jonathan Wakely wrote:
> > > On 17/08/17 21:21 -0600, Martin Sebor wrote:
> > > > Joseph, while looking into implementing enhancement your request
> > > > pr81824 I noticed that GCC silently accepts incompatible alias
> > > > declarations (pr81854) so as sort of a proof-concept for the
> > > > former I enhanced the checking already done for other kinds of
> > > > incompatibilities to also detect those mentioned in the latter
> > > > bug.  Attached is this patch, tested on x85_64-linux.
> > > >
> > > > Jonathan, the patch requires suppressing the warning in libstdc++
> > > > compatibility symbol definitions in compatibility.cc.  I couldn't
> > > > find a way to do it without the suppression but I'd be happy to
> > > > try again if you have an idea for how.
> > >
> > > Doing it that way is fine, but ...
> > >
> > > > diff --git a/libstdc++-v3/src/c++98/compatibility.cc
> > > > b/libstdc++-v3/src/c++98/compatibility.cc
> > > > index 381f4c4..5f56b9e 100644
> > > > --- a/libstdc++-v3/src/c++98/compatibility.cc
> > > > +++ b/libstdc++-v3/src/c++98/compatibility.cc
> > > > @@ -213,6 +213,11 @@
> > > > _ZNSt19istreambuf_iteratorIcSt11char_traitsIcEEppEv
> > > > _ZNSt19istreambuf_iteratorIwSt11char_traitsIwEEppEv
> > > > */
> > > >
> > > > +// Disable warning about declaring aliases between functions with
> > > > +// incompatible types.
> > > > +#pragma GCC diagnostic push
> > > > +#pragma GCC diagnostic ignored "-Wattributes"
> > > > +
> > >
> > > Could this be moved closer to the point where it's needed?
> > >
> > > It's not needed until after line 361, right?
> >
> > Sure.  The other possibility that I forgot to mention is to
> > declare the alias without a prototype, which in C++ looks
> > like this:
> >
> >  void foo (...);
> >
> > The patch would then look like this.  Do you have a preference
> > between these two approaches?
>
> If this doesn't change the generated code, but avoids the warnings
> then I think I prefer this. i.e. fix the code, not just suppress the
> warnings.

It's the same as calling a function without a prototype in C.
I rebuilt libstdc++ with this change and reran the test suite
with no unexpected failures so I'll go ahead and commit this
change instead.

To be clear, though, this is just a suppression mechanism not
unlike a cast.  The ideal solution would be to declare the
aliases to have the right type, e.g., using __typeof__ or
decltype.  I just couldn't find a way to make it work with
the macros.

Martin

> > Martin
> >
> > diff --git a/libstdc++-v3/src/c++98/compatibility.cc
> > b/libstdc++-v3/src/c++98/compatibility.cc
> > index 381f4c4..b49a5ca 100644
> > --- a/libstdc++-v3/src/c++98/compatibility.cc
> > +++ b/libstdc++-v3/src/c++98/compatibility.cc
> > @@ -367,13 +367,13 @@ _GLIBCXX_END_NAMESPACE_VERSION
> >
> > #define _GLIBCXX_3_4_SYMVER(XXname, name) \
> >    extern "C" void \
> > -   _X##name() \
> > +   _X##name(...) \
> >    __attribute__ ((alias(#XXname))); \
> >    asm (".symver " "_X" #name "," #name "@GLIBCXX_3.4");
> >
> > #define _GLIBCXX_3_4_5_SYMVER(XXname, name) \
> >    extern "C" void \
> > -   _Y##name() \
> > +   _Y##name(...) \
> >    __attribute__ ((alias(#XXname))); \
> >    asm (".symver " "_Y" #name  "," #name "@@GLIBCXX_3.4.5");