__mempcpy inline/builtin emitting mempcpy ref

[H.J. Lu]

On Tue, Jan 7, 2025 at 7:55 AM Samuel Thibault <samuel.thibault@gnu.org> wrote:
>
> H.J. Lu, le mar. 07 janv. 2025 07:42:12 +0800, a ecrit:
> > On Tue, Jan 7, 2025 at 7:30 AM Samuel Thibault <samuel.thibault@gnu.org> wrote:
> > > In htl/pt-alloc.c, we call __mempcpy(), but in pt-alloc.os we get an
> > > external ref to mempcpy, which then violates symbol exposition rules.
> >
> > Does it only happen in libpthread.a?
>
> I have seen it happen there only yes.
>
> > > Apparently what is happening is that include/string.h #defines __mempcpy
> > > to __builtin_mempcpy, and the compiler doesn't actually inline code and
> > > rather emits a reference to mempcpy. Do we have a proper way to avoid
> > > this?
> > >
> > > #if IS_IN (libc) && !defined SHARED \
> > >   && !defined NO_MEMPCPY_STPCPY_REDIRECT
> > > /* Redirect calls to __builtin_mempcpy and __builtin_stpcpy to call
> > >    __mempcpy and __stpcpy if not inlined.  */
> > > extern __typeof (mempcpy) mempcpy __asm__ ("__mempcpy");
> > >
> > > is done only for IS_IN (libc), should we perhaps drop that piece of the
> >
> > Try this
> >
> > diff --git a/include/string.h b/include/string.h
> > index 1f6aef9f93..b9c41b0eea 100644
> > --- a/include/string.h
> > +++ b/include/string.h
> > @@ -175,7 +175,7 @@ extern __typeof (strnlen) strnlen attribute_hidden;
> >  extern __typeof (strsep) strsep attribute_hidden;
> >  #endif
> >
> > -#if IS_IN (libc) && !defined SHARED \
> > +#if (IS_IN (libc) || IS_IN (libpthread)) && !defined SHARED \
> >    && !defined NO_MEMPCPY_STPCPY_REDIRECT
> >  /* Redirect calls to __builtin_mempcpy and __builtin_stpcpy to call
> >     __mempcpy and __stpcpy if not inlined.  */
>
> That will be working yes, is it fine to commit it?
>

Please submit a formal patch for me after testing.

Thanks.

-- 
H.J.