> No, we should already be calling _bfd_archive_close_and_cleanup via 
> 
> > --- a/bfd/opncls.c
> > +++ b/bfd/opncls.c
> > @@ -719,6 +719,17 @@ bfd_close (bfd *abfd)
> >    if (! BFD_SEND (abfd, _close_and_cleanup, (abfd)))
> 
> this call.  The problem is in coff-rs6000.c (and coff64-rs6000.c)
> where the bfd_target vector just uses bfd_true for close_and_cleanup.

Attached is a patch that implements your suggestion. Tested on ppc-aix
using AdaCore's GDB testsuite.

While at it, the powermac XCOFF backend also seemed to be using
the same code that eventually leads to the dangling pointer, and so
I fixed it as well. It's untested, but seems straightforward enough.

OK to commit?

Thanks,
-- 
Joel