Re: [PATCH v3] Support glibc multiple namespace extension

[Kevin Buettner <kevinb@redhat.com>]

On Fri, 25 Feb 2022 16:53:38 +0000
"Metzger, Markus T" <markus.t.metzger@intel.com> wrote:

> Hello Kevin,
> 
> >This is indeed wrong and leads to
> >
> >      if (li->l_prev != prev_lm)
> >	{
> >	  warning (_("Corrupted shared library list: %s != %s"),
> >		   paddress (target_gdbarch (), prev_lm),
> >		   paddress (target_gdbarch (), li->l_prev));
> >	  return 0;
> >	}
> >
> >in svr4_read_so_list().  Assigning prev_lm to zero avoided the warning for
> >the first library in another namespace.  I added more test cases to detect
> >this reliably.
> >
> >I'll try organizing SVr4 DSOs in per-namespace lists.  
> 
> That works.  During testing I noticed
> 
> 	Invalid cast.
> 	warning: Probes-based dynamic linker interface failed.
> 	Reverting to original interface.
> 
> on dlcose().  This can also be reproduced with gdb.base/unload.exp
> and upstream GDB.  I have not looked into this, yet.
> 
> This is not detected by the test system and I'm not sure whether we
> actually want to consider falling back to the old interface a test fail.
> The invalid cast error we probably want to detect.
> 
> I added support for detecting this invalid cast and the corrupted
> library list warning in gdb_continue_to_breakpoint.
> 
> I still need to extend the library-list-svr4 XML to cover namespaces
> before I send an updated patch.  I'll be out the next week so this
> will take a while.
> 
> I currently check the XML version and put everything into a special
> namespace on the GDB side.  We cannot use incremental updates
> this way but we're backwards compatible.  The final version will
> support both the new format with namespaces and incremental
> updates and the old format without.

This sounds like good progress!

I look forward to seeing your next patch...

Kevin