Re: [PATCH?] GDB HEAD (partly) broken for GNU/Hurd

[Thomas Schwinge]

Hello!

On Mon, Oct 13, 2008 at 07:35:34PM +0100, Pedro Alves wrote:
> Thanks Thomas :-)  One thing I asked myself was, if gnu-nat.c couldn't be 
> using
> the port's id as thread ids instead of a locally auto-generated number.  Maybe
> the thread id of the main thread would be preserved across execs this way, 
> but,
> this is off-scope for now.

I can't tell without having a detailed look at that -- for which I
absolutely don't have the time at the moment, unfortunately.  I made this
a TODO item:
<http://www.bddebian.com/~wiki/hurd/open_issues/gdb_thread_ids/>


> On Monday 13 October 2008 17:40:25, Ulrich Weigand wrote:
> > Hmm.  It would appear that "set exec-wrapper" is currently broken with
> > the gnu-nat.c target, right?
> 
> Yeah, it appears so.  Don't know if it's possible to get rid of the local
> pending execs handling in gnu-nat.c.  An alternative would be to make
> pending_execs a property of inferior.h:`struct inferior' instead of of
> gnu-nat.c:`struct inf'.

Another TODO item:
<http://www.bddebian.com/~wiki/hurd/open_issues/gdb_pending_execs/>


> (I also notice that when going through the shell in non-stop
> mode, it would be more correct to resume all threads --- we
> don't want non-stop and its scheduler-locking to apply to the shell.
> Basically, non-stop should be off if there are pending execs.
> This was an existing issue, and doesn't affect linux today, so I'll
> just ignore that for now, as it needs more tweaking to fix.)

And non-stop mode isn't even supported with gnu-nat.c -- yet another TODO
item: <http://www.bddebian.com/~wiki/hurd/open_issues/gdb_non-stop_mode/>


> What do you guys think?  Thomas, could you try the attached patch
> on the Hurd, please?  I just gave it a spin on x86-64-unknown-linux-gnu
> without regressions.

> 2008-10-13  Pedro Alves  <pedro@codesourcery.com>
> 
>       * fork-child.c (startup_inferior): Only set threads not-executing
>       after getting all the pending execs.  On TARGET_WAITKIND_IGNORE,
>       keep waiting, don't resume.  On all other cases but
>       TARGET_WAITKIND_SIGNALLED and TARGET_WAITKIND_EXITED, switch to
>       the event ptid.

Fine on Hurd as well, thanks!


Regards,
 Thomas

signature.asc
Description: Digital signature