[PATCH] elf: Don't set VER_FLG_WEAK

[H.J. Lu]

On Mon, Jan 31, 2022 at 8:48 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> > On Tue, Jan 25, 2022 at 7:07 PM Alan Modra <amodra@gmail.com> wrote:
> >>
> >> On Tue, Jan 25, 2022 at 04:19:53PM -0800, H.J. Lu via Binutils wrote:
> >> > On Solaris, VER_FLG_WEAK indicates a weak version definition with no
> >> > symbols associated with it.  It is used to verify the existence of a
> >> > particular implementation without any symbol references to the weak
> >> > version.  Don't set VER_FLG_WEAK since it is unused in binutils.
> >>
> >> Can you tell me why the presence of this flag is bad?  I don't see it
> >> affecting anything in binutils or glibc.  VER_FLG_WEAK in version refs
> >> does affect resolution of symbol, but I can't see how a VER_FLG_WEAK
> >> in a version def can transfer over to a ref if the def is unused.
> >
> > Florian,  can you share the reason why Linux doesn't want
> > VER_FLG_WEAK on the empty version?
>
> I may have been overly cautious, influenced by the old link editor
> behavior which copied the weak flag of dynamic symbols from the
> definitions to their references.  But it's indeed unclear how this could
> happen here.  If it did, we would lose the coverage check in the glibc
> dynamic loader for that symbol version, I think.  (The loader only
> checks vna_flags in two places, once for printing the flag, and once to
> skip an error report in case of a missing version.)
>
> Do you suggest we should start using weak symbol version definitions in
> glibc, and not worry about it?

Let's use the symbol versions without any symbols in glibc after
glibc 2.35 is released.  It will create symbol versions with
VER_FLG_WEAK.  If it will cause any issues, we can change linker
to remove VER_FLG_WEAK from empty symbol versions.

> Thanks,
> Florian
>


-- 
H.J.