Re: IA64 libc.so always depends on libunwind?

[David Mosberger <davidm at napali dot hpl dot hp dot com>]

>>>>> On Wed, 7 Apr 2004 07:13:09 +0200, Jakub Jelinek <jakub@redhat.com> said:

  Jakub> On Tue, Apr 06, 2004 at 05:22:43PM -0700, David Mosberger
  Jakub> wrote:
  >> >>>>> On Tue, 6 Apr 2004 17:15:13 -0700, "H. J. Lu"
  >> <hjl@lucon.org> said:

  HJ> The same is true for libgcc_s.so. But we link in libgcc_s.so
  HJ> only when the executable needs exception-handling support.
  >>  Yes, but libgcc_eh.a is the archive version of libgcc_s.so, so
  >> you _do_ link in the unwinder into each and every binary.  Except
  >> that with libunwind, we don't want libunwind as part of
  >> libgcc_eh.a.

  Jakub> No, we don't.  I think that libgcc_eh.a is completely
  Jakub> unnecessary during libc.so link.

It wasn't when I tried last time.  IIRC, the cleanup handlers
would end up calling _Unwind_Resume().

  Jakub> Unlike -lgcc_eh which is an archive only and thus not linked
  Jakub> in if not needed, -lunwind is also a shared lib and thus
  Jakub> linked even when not needed.  So, either both -lgcc_eh and
  Jakub> -lunwind should be nuked, or --as-needed -lunwind
  Jakub> --no-as-needed added instead of -lunwind.

--as-needed?  Cool, I didn't know about that option.

	--david