RFA[threads]: Fork event updates, part the thirteenth

RFA[threads]: Fork event updates, part the thirteenth

[Daniel Jacobowitz]

Now is where it starts to get interesting.  Michael, I mentioned this patch
to you at lunch last week.  If you take a short-lived program, run it, and
detach it, and run it again, you'll see the exit of the _previous_ copy. 
Then GDB gets hopelessly confused.  I have a testcase for this which I'll
post in a moment.

The reason it's included here is that that's essentially what happens if you
are using "set follow-fork-mode child".  We detach from the parent, which
exits, confusing GDB.

Is this OK?

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

2002-12-15  Daniel Jacobowitz  <drow@mvista.com>

	* lin-lwp.c (child_wait): Ignore exit statuses for processes other
	than inferior_ptid.
	(lin_lwp_wait): Ignore exit statuses for unknown LWPs.

Index: lin-lwp.c
===================================================================
RCS file: /cvs/src/src/gdb/lin-lwp.c,v
retrieving revision 1.39
diff -u -p -r1.39 lin-lwp.c
--- lin-lwp.c	9 Dec 2002 18:41:42 -0000	1.39
+++ lin-lwp.c	15 Dec 2002 21:16:34 -0000
@@ -964,6 +964,14 @@ child_wait (ptid_t ptid, struct target_w
 	pid = waitpid (GET_PID (ptid), &status, __WCLONE);
       save_errno = errno;
 
+      /* Make sure we don't report an event for the exit of the
+	 original program, if we've detached from it.  */
+      if (pid != -1 && ! WIFSTOPPED (status) && pid != GET_PID (inferior_ptid))
+	{
+	  pid = -1;
+	  save_errno = EINTR;
+	}
+
       clear_sigio_trap ();
       clear_sigint_trap ();
     }
@@ -1091,6 +1099,17 @@ lin_lwp_wait (ptid_t ptid, struct target
 	  gdb_assert (pid == -1 || lwpid == pid);
 
 	  lp = find_lwp_pid (pid_to_ptid (lwpid));
+
+	  /* Make sure we don't report an event for the exit of an LWP not in
+	     our list, i.e.  not part of the current process.  This can happen
+	     if we detach from a program we original forked and then it
+	     exits.  */
+	  if (! WIFSTOPPED (status) && ! lp)
+	    {
+	      status = 0;
+	      continue;
+	    }
+
 	  if (! lp)
 	    {
 	      lp = add_lwp (BUILD_LWP (lwpid, GET_PID (inferior_ptid)));

Re: RFA[threads]: Fork event updates, part the thirteenth

[Michael Snyder]

Daniel Jacobowitz wrote:
> 
> Now is where it starts to get interesting.  Michael, I mentioned this patch
> to you at lunch last week.  If you take a short-lived program, run it, and
> detach it, and run it again, you'll see the exit of the _previous_ copy.
> Then GDB gets hopelessly confused.  I have a testcase for this which I'll
> post in a moment.
> 
> The reason it's included here is that that's essentially what happens if you
> are using "set follow-fork-mode child".  We detach from the parent, which
> exits, confusing GDB.
> 
> Is this OK?

Hi Dan, 

Please excuse the delay.  This seems OK.  In child_wait, 
would it be possible to add a check to see if the exiting
process is in our lwp list?

Michael

> 
> --
> Daniel Jacobowitz
> MontaVista Software                         Debian GNU/Linux Developer
> 
> 2002-12-15  Daniel Jacobowitz  <drow@mvista.com>
> 
>         * lin-lwp.c (child_wait): Ignore exit statuses for processes other
>         than inferior_ptid.
>         (lin_lwp_wait): Ignore exit statuses for unknown LWPs.
> 
> Index: lin-lwp.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/lin-lwp.c,v
> retrieving revision 1.39
> diff -u -p -r1.39 lin-lwp.c
> --- lin-lwp.c   9 Dec 2002 18:41:42 -0000       1.39
> +++ lin-lwp.c   15 Dec 2002 21:16:34 -0000
> @@ -964,6 +964,14 @@ child_wait (ptid_t ptid, struct target_w
>         pid = waitpid (GET_PID (ptid), &status, __WCLONE);
>        save_errno = errno;
> 
> +      /* Make sure we don't report an event for the exit of the
> +        original program, if we've detached from it.  */
> +      if (pid != -1 && ! WIFSTOPPED (status) && pid != GET_PID (inferior_ptid))
> +       {
> +         pid = -1;
> +         save_errno = EINTR;
> +       }
> +
>        clear_sigio_trap ();
>        clear_sigint_trap ();
>      }
> @@ -1091,6 +1099,17 @@ lin_lwp_wait (ptid_t ptid, struct target
>           gdb_assert (pid == -1 || lwpid == pid);
> 
>           lp = find_lwp_pid (pid_to_ptid (lwpid));
> +
> +         /* Make sure we don't report an event for the exit of an LWP not in
> +            our list, i.e.  not part of the current process.  This can happen
> +            if we detach from a program we original forked and then it
> +            exits.  */
> +         if (! WIFSTOPPED (status) && ! lp)
> +           {
> +             status = 0;
> +             continue;
> +           }
> +
>           if (! lp)
>             {
>               lp = add_lwp (BUILD_LWP (lwpid, GET_PID (inferior_ptid)));

Re: RFA[threads]: Fork event updates, part the thirteenth

[Daniel Jacobowitz]

On Mon, Jan 06, 2003 at 03:53:52PM -0800, Michael Snyder wrote:
> Daniel Jacobowitz wrote:
> > 
> > Now is where it starts to get interesting.  Michael, I mentioned this patch
> > to you at lunch last week.  If you take a short-lived program, run it, and
> > detach it, and run it again, you'll see the exit of the _previous_ copy.
> > Then GDB gets hopelessly confused.  I have a testcase for this which I'll
> > post in a moment.
> > 
> > The reason it's included here is that that's essentially what happens if you
> > are using "set follow-fork-mode child".  We detach from the parent, which
> > exits, confusing GDB.
> > 
> > Is this OK?
> 
> Hi Dan, 
> 
> Please excuse the delay.  This seems OK.  In child_wait, 
> would it be possible to add a check to see if the exiting
> process is in our lwp list?

I _think_ that child_wait will never be called if there is anything in
the LWP list; if we have LWPs, we'll have pushed thread_db onto the
stack, and we'll go to lin_lwp_wait instead if thre are any LWPs.  But
I'm sleepy, so I may be missing something; I'll sit on this and look at
it again tomorrow :)

Thanks.

> > 2002-12-15  Daniel Jacobowitz  <drow@mvista.com>
> > 
> >         * lin-lwp.c (child_wait): Ignore exit statuses for processes other
> >         than inferior_ptid.
> >         (lin_lwp_wait): Ignore exit statuses for unknown LWPs.
> > 
> > Index: lin-lwp.c
> > ===================================================================
> > RCS file: /cvs/src/src/gdb/lin-lwp.c,v
> > retrieving revision 1.39
> > diff -u -p -r1.39 lin-lwp.c
> > --- lin-lwp.c   9 Dec 2002 18:41:42 -0000       1.39
> > +++ lin-lwp.c   15 Dec 2002 21:16:34 -0000
> > @@ -964,6 +964,14 @@ child_wait (ptid_t ptid, struct target_w
> >         pid = waitpid (GET_PID (ptid), &status, __WCLONE);
> >        save_errno = errno;
> > 
> > +      /* Make sure we don't report an event for the exit of the
> > +        original program, if we've detached from it.  */
> > +      if (pid != -1 && ! WIFSTOPPED (status) && pid != GET_PID (inferior_ptid))
> > +       {
> > +         pid = -1;
> > +         save_errno = EINTR;
> > +       }
> > +
> >        clear_sigio_trap ();
> >        clear_sigint_trap ();
> >      }
> > @@ -1091,6 +1099,17 @@ lin_lwp_wait (ptid_t ptid, struct target
> >           gdb_assert (pid == -1 || lwpid == pid);
> > 
> >           lp = find_lwp_pid (pid_to_ptid (lwpid));
> > +
> > +         /* Make sure we don't report an event for the exit of an LWP not in
> > +            our list, i.e.  not part of the current process.  This can happen
> > +            if we detach from a program we original forked and then it
> > +            exits.  */
> > +         if (! WIFSTOPPED (status) && ! lp)
> > +           {
> > +             status = 0;
> > +             continue;
> > +           }
> > +
> >           if (! lp)
> >             {
> >               lp = add_lwp (BUILD_LWP (lwpid, GET_PID (inferior_ptid)));
> 

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: RFA[threads]: Fork event updates, part the thirteenth

[Daniel Jacobowitz]

On Mon, Jan 06, 2003 at 07:50:55PM -0500, Daniel Jacobowitz wrote:
> On Mon, Jan 06, 2003 at 03:53:52PM -0800, Michael Snyder wrote:
> > Daniel Jacobowitz wrote:
> > > 
> > > Now is where it starts to get interesting.  Michael, I mentioned this patch
> > > to you at lunch last week.  If you take a short-lived program, run it, and
> > > detach it, and run it again, you'll see the exit of the _previous_ copy.
> > > Then GDB gets hopelessly confused.  I have a testcase for this which I'll
> > > post in a moment.
> > > 
> > > The reason it's included here is that that's essentially what happens if you
> > > are using "set follow-fork-mode child".  We detach from the parent, which
> > > exits, confusing GDB.
> > > 
> > > Is this OK?
> > 
> > Hi Dan, 
> > 
> > Please excuse the delay.  This seems OK.  In child_wait, 
> > would it be possible to add a check to see if the exiting
> > process is in our lwp list?
> 
> I _think_ that child_wait will never be called if there is anything in
> the LWP list; if we have LWPs, we'll have pushed thread_db onto the
> stack, and we'll go to lin_lwp_wait instead if thre are any LWPs.  But
> I'm sleepy, so I may be missing something; I'll sit on this and look at
> it again tomorrow :)
> 
> Thanks.

Having convinced myself of this, I've checked in the patch as-is. 
Thanks.  Now for the testcase it fixes.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer