RFC: hacky fix for PR 12406

RFC: hacky fix for PR 12406

[Tom Tromey]

I would like to get some feedback on this.

PR 12406 concerns what is printed when stop-on-solib-events is set:

    (gdb) set stop-on-solib-events 1
    (gdb) r
    Starting program: /tmp/q 
    Stopped due to shared library event

What is missing here is any indication of what library was loaded.

This patch changes the output to:

    (gdb) set stop-on-solib-events 1
    (gdb) r
    Starting program: /tmp/q 
    [Inferior loaded library /lib64/ld-linux-x86-64.so.2]
    Stopped due to shared library event

I could not find a good way to get the information about what library
was loaded from the library-loading code to the bpstat_print code.  This
approach seems like a hack to me -- so I was wondering if anyone has a
better suggestion.

This regtests fine on x86-64 F15.

Tom

2012-01-11  Tom Tromey  <tromey@redhat.com>

	PR symtab/12406:
	* solib.c (update_solib_list): Print message if
	stop_on_solib_events is set.

diff --git a/gdb/solib.c b/gdb/solib.c
index f591785..66e2f94 100644
--- a/gdb/solib.c
+++ b/gdb/solib.c
@@ -810,6 +810,11 @@ update_solib_list (int from_tty, struct target_ops *target)
 			       _("Error while mapping shared "
 				 "library sections:\n"));
 
+	  if (stop_on_solib_events)
+	    ui_out_message (current_uiout, 0,
+			    _("[Inferior loaded library %s]\n"),
+			    i->so_name);
+
 	  /* Notify any observer that the shared object has been
 	     loaded now that we've added it to GDB's tables.  */
 	  observer_notify_solib_loaded (i);

Re: RFC: hacky fix for PR 12406

[Jan Kratochvil]

On Wed, 11 Jan 2012 19:55:30 +0100, Tom Tromey wrote:
> This patch changes the output to:
> 
>     (gdb) set stop-on-solib-events 1
>     (gdb) r
>     Starting program: /tmp/q 
>     [Inferior loaded library /lib64/ld-linux-x86-64.so.2]
>     Stopped due to shared library event

I find it OK, also an unload would be good to notify there.  There will still
be some no-operations stops without Gary's STAP-probe-in-ld.so.


Thanks,
Jan

Re: RFC: hacky fix for PR 12406

[Joel Brobecker]

> This patch changes the output to:
> 
>     (gdb) set stop-on-solib-events 1
>     (gdb) r
>     Starting program: /tmp/q 
>     [Inferior loaded library /lib64/ld-linux-x86-64.so.2]
>     Stopped due to shared library event
> 
> I could not find a good way to get the information about what library
> was loaded from the library-loading code to the bpstat_print code.  This
> approach seems like a hack to me -- so I was wondering if anyone has a
> better suggestion.

Yeah, it doesn't look super great, but it is better than nothing, IMO.

It looks a lot easier to implement this on platforms that use
TARGET_WAITKIND_LOADED events, because we could associate the library
name to the target_waitstatus value. For other platforms where this
is based on a breakpoint rather than a kernel event, on the other
hand... I'm not even convinced that we'll get one breakpoint per
shared library being mapped. Particularly at startup time, I can
see us hitting one breakpoint once all SOs have been mapped, and
not before. If it were the case, attaching one shared-library event
per breakpoint event does not seem to make sense.

An idea that crossed my mind: Add a flag to each entry in our SO
list. Everytime we stop, we use that flag to determine which entries
are new. We can probably use something like the normal-stop hook to
set the flag just before giving the prompt back to the user. Would
that work? It would support the situation I mentioned above where
we get one breakpoint event for multiple shared libraries...

-- 
Joel

Re: RFC: hacky fix for PR 12406

[Tom Tromey]

>>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:

Joel> An idea that crossed my mind: Add a flag to each entry in our SO
Joel> list. Everytime we stop, we use that flag to determine which entries
Joel> are new. We can probably use something like the normal-stop hook to
Joel> set the flag just before giving the prompt back to the user. Would
Joel> that work? It would support the situation I mentioned above where
Joel> we get one breakpoint event for multiple shared libraries...

Thanks for the idea.  I will give it a try.

Tom> [Inferior loaded library /lib64/ld-linux-x86-64.so.2]
Tom> Stopped due to shared library event

Gary> Could this be a separate option, trace-solib-events maybe?  It seems
Gary> like this printing might also be useful when stop-on-solib-events is
Gary> off.

Gary> Also, as Jan said, it would be nice to have unload notifications too.

This comment inspired me to take another look at the whole problem.

I actually do not much like stop-on-solib-events.  I think 'catch'
commands are better, because they offer the user more control: commands
at the stop point, "silent", conditions.

So now I am thinking that, while I may still add the notification in
some form, I will also resurrect "catch load" and "catch unload",
something like "catch load [REGEX] [if ...]"

Tom

Re: RFC: hacky fix for PR 12406

[Gary Benson]

Tom Tromey wrote:
> >>>>> "Joel" == Joel Brobecker <brobecker@adacore.com> writes:
> 
> Joel> An idea that crossed my mind: Add a flag to each entry in our
> Joel> SO list. Everytime we stop, we use that flag to determine
> Joel> which entries are new. We can probably use something like the
> Joel> normal-stop hook to set the flag just before giving the prompt
> Joel> back to the user. Would that work? It would support the
> Joel> situation I mentioned above where we get one breakpoint event
> Joel> for multiple shared libraries...
> 
> Thanks for the idea.  I will give it a try.
> 
> Tom> [Inferior loaded library /lib64/ld-linux-x86-64.so.2]
> Tom> Stopped due to shared library event
> 
> Gary> Could this be a separate option, trace-solib-events maybe?
> Gary> It seems like this printing might also be useful when
> Gary> stop-on-solib-events is off.
> 
> Gary> Also, as Jan said, it would be nice to have unload
> Gary> notifications too.
> 
> This comment inspired me to take another look at the whole problem.
> 
> I actually do not much like stop-on-solib-events.  I think 'catch'
> commands are better, because they offer the user more control:
> commands at the stop point, "silent", conditions.
> 
> So now I am thinking that, while I may still add the notification in
> some form, I will also resurrect "catch load" and "catch unload",
> something like "catch load [REGEX] [if ...]"

Sounds nice!

Cheers,
Gary

-- 
http://gbenson.net/

Re: RFC: hacky fix for PR 12406

[Gary Benson]

Tom Tromey wrote:
> I would like to get some feedback on this.
> 
> PR 12406 concerns what is printed when stop-on-solib-events is set:
> 
>     (gdb) set stop-on-solib-events 1
>     (gdb) r
>     Starting program: /tmp/q 
>     Stopped due to shared library event
> 
> What is missing here is any indication of what library was loaded.
> 
> This patch changes the output to:
> 
>     (gdb) set stop-on-solib-events 1
>     (gdb) r
>     Starting program: /tmp/q 
>     [Inferior loaded library /lib64/ld-linux-x86-64.so.2]
>     Stopped due to shared library event

Could this be a separate option, trace-solib-events maybe?  It seems
like this printing might also be useful when stop-on-solib-events is
off.

Also, as Jan said, it would be nice to have unload notifications too.

> I could not find a good way to get the information about what
> library was loaded from the library-loading code to the bpstat_print
> code.  This approach seems like a hack to me -- so I was wondering
> if anyone has a better suggestion.

I like Joel's idea of a flag, it's what I'd try.  You would probably
need to do something like that for unload notifications anyway.

Cheers,
Gary

-- 
http://gbenson.net/