This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH 1/3] gdbserver/IPA: Export some functions via global function pointers.
- From: "Ulrich Weigand" <uweigand at de dot ibm dot com>
- To: koriakin at 0x04 dot net (Marcin KoÅcielnicki)
- Cc: gdb-patches at sourceware dot org
- Date: Wed, 30 Mar 2016 13:32:39 +0200 (CEST)
- Subject: Re: [PATCH 1/3] gdbserver/IPA: Export some functions via global function pointers.
- Authentication-results: sourceware.org; auth=none
Marcin KoÅ?cielnicki wrote:
> On 29/03/16 20:08, Ulrich Weigand wrote:
> > Marcin KoÅ?cielnicki wrote:
> >> On 14/03/16 18:49, Ulrich Weigand wrote:
> >>> The more I think about it, the more I tend to agree that your
> >>> proposal is actually the best solution. I'd still like to give
> >>> it a couple of days to give others a chance to comment as well ...
> >>
> >> Alright, so what should we do about this issue?
> >
> > Since nobody came up with a better idea, and since your patch doesn't
> > actually preclude anybody from implementing any better idea they might
> > come up later (since it doesn't actually change anything in the
> > gdbserver protocol), I'd say we just go with your patch for now.
>
> Very well, then. For this to be actually useful for powerpc64, I'll
> also need an ack on the other patch
> (https://sourceware.org/ml/gdb-patches/2016-03/msg00201.html).
This looks to be resolved now.
> > However, there does seem to be one issue: your patch changes the
> > interface between gdbserver and the in-process agent in an incompatible
> > way. Binaries with an old IPA built in will no longer work with a
> > new gdbserver, since it will will expect exported symbols like
> > gdb_collect_ptr, which the old binary doesn't export.
> >
> > I think it would be preferable to implement a backward-compatible
> > way where gdbserver checks for the new symbol, and if it isn't
> > present, falls back to the old symbol.
>
> Alright, I can do that, though I seem to recall we don't care about
> gdbserver/IPA interface compatibility (and IPA is always built as
> shared, so there's no concern about an executable with old version built
> in).
And this turns out to be not necessary after all, see the recent
mail by Pedro. Sorry for the confusion.
I think the patch should be OK now.
Thanks,
Ulrich
--
Dr. Ulrich Weigand
GNU/Linux compilers and toolchain
Ulrich.Weigand@de.ibm.com