GDB and Java

GDB and Java

[David Carlton]

Tom, I just committed a large GDB patch that, among other things,
touches the dynamic class symbol part of jv-lang.c.  Can you give
current mainline a whirl to make sure that I didn't screw anything up?
These changes should make that code more stable in the long term, but
there's always the possibility that I missed something in the short
term.

David Carlton
carlton@math.stanford.edu

Re: GDB and Java

[Tom Tromey]

>>>>> "David" == David Carlton <carlton@math.stanford.edu> writes:

David> Tom, I just committed a large GDB patch that, among other
David> things, touches the dynamic class symbol part of jv-lang.c.
David> Can you give current mainline a whirl to make sure that I
David> didn't screw anything up?

I'll probably have to do a fair amount of java debugging next week.
I'll try it out then.

Tom

Re: GDB and Java

[Tom Tromey]

>>>>> "David" == David Carlton <carlton@math.stanford.edu> writes:

David> Tom, I just committed a large GDB patch that, among other things,
David> touches the dynamic class symbol part of jv-lang.c.  Can you give
David> current mainline a whirl to make sure that I didn't screw anything up?

I said I'd do this a while back, and I wanted to let you know I didn't
forget -- but also that I can't easily do it.  I'm using RHL 9, and my
understanding is that the NPTL patches aren't in main line gdb yet.
At least, when I tried debugging with the cvs gdb a couple weeks back,
I had no end of troubles.  So, sorry about that.

Tom

Re: GDB and Java

[Daniel Jacobowitz]

On Fri, Jul 18, 2003 at 09:59:13AM -0600, Tom Tromey wrote:
> >>>>> "David" == David Carlton <carlton@math.stanford.edu> writes:
> 
> David> Tom, I just committed a large GDB patch that, among other things,
> David> touches the dynamic class symbol part of jv-lang.c.  Can you give
> David> current mainline a whirl to make sure that I didn't screw anything up?
> 
> I said I'd do this a while back, and I wanted to let you know I didn't
> forget -- but also that I can't easily do it.  I'm using RHL 9, and my
> understanding is that the NPTL patches aren't in main line gdb yet.
> At least, when I tried debugging with the cvs gdb a couple weeks back,
> I had no end of troubles.  So, sorry about that.

It should, mostly, work now.  I'm not sure what the remaining issues
are, but the NPTL patches were added.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: GDB and Java

[Tom Tromey]

>>>>> "Daniel" == Daniel Jacobowitz <drow@mvista.com> writes:

Daniel> It should, mostly, work now.  I'm not sure what the remaining issues
Daniel> are, but the NPTL patches were added.

Ok, bad info on my part.  Thanks.  I'm updating and rebuilding cvs
now; I'll report problems if I find them.

Tom

Re: GDB and Java

[Elena Zannoni]

Daniel Jacobowitz writes:
 > On Fri, Jul 18, 2003 at 09:59:13AM -0600, Tom Tromey wrote:
 > > >>>>> "David" == David Carlton <carlton@math.stanford.edu> writes:
 > > 
 > > David> Tom, I just committed a large GDB patch that, among other things,
 > > David> touches the dynamic class symbol part of jv-lang.c.  Can you give
 > > David> current mainline a whirl to make sure that I didn't screw anything up?
 > > 
 > > I said I'd do this a while back, and I wanted to let you know I didn't
 > > forget -- but also that I can't easily do it.  I'm using RHL 9, and my
 > > understanding is that the NPTL patches aren't in main line gdb yet.
 > > At least, when I tried debugging with the cvs gdb a couple weeks back,
 > > I had no end of troubles.  So, sorry about that.
 > 
 > It should, mostly, work now.  I'm not sure what the remaining issues
 > are, but the NPTL patches were added.
 > 

Yes it works fine. I am using it now on RHL.

elena


 > -- 
 > Daniel Jacobowitz
 > MontaVista Software                         Debian GNU/Linux Developer

Re: GDB and Java

[Tom Tromey]

>>>>> "David" == David Carlton <carlton@math.stanford.edu> writes:
David> Tom, I just committed a large GDB patch that, among other things,
David> touches the dynamic class symbol part of jv-lang.c.  Can you give
David> current mainline a whirl to make sure that I didn't screw anything up?

Ok, I've tried cvs head today.  I'm debugging a large java application
that consists of about 90 shared libraries (not including libgcj).

I can consistently make gdb crash like this:

../../src/gdb/dwarf2-frame.c:1032: internal-error: sect_index_text not initialized
A problem internal to GDB has been detected,
further debugging may prove unreliable.


This happens in response to fairly innocuous things like "next" or
"bt".  It doesn't happen in response to every "next" or "bt" command,
and I don't know how to trigger it.  Still, I haven't managed to have
a non-crashing "real" debugging session.  I see this same crash with
an internal gdb I've been using for the last couple days, frequently
right when I've finally gotten to the point I'm actually interested in
(of course :-).


My theory is that the objfile created in jv-lang.c never has
sect_index_text set to a value other than -1.  I haven't tried to
verify this yet.

If anybody has a suggestion for a fix, no matter how hacky, I'd love
to hear it.  Debugging this application -- and hence, fixing gdb if
necessary -- is a priority for me right now.

Tom

Re: GDB and Java

[David Carlton]

On 18 Jul 2003 10:28:54 -0600, Tom Tromey <tromey@redhat.com> said:

> Ok, I've tried cvs head today.  I'm debugging a large java application
> that consists of about 90 shared libraries (not including libgcj).

> I can consistently make gdb crash like this:

> ../../src/gdb/dwarf2-frame.c:1032: internal-error: sect_index_text not initialized
> A problem internal to GDB has been detected,
> further debugging may prove unreliable.

This is probably the same unfortunate interaction between anonymous
objfiles and the new DWARF 2 frame stuff that Daniel noted in
conjunction with another patch of mine.  I think
<http://sources.redhat.com/ml/gdb-patches/2003-07/msg00264.html> is
supposed to cure it, so give that a try.  Is that patch, or a variant,
supposed to go into 6.0?  This argues that it had better go in.

David Carlton
carlton@kealia.com

Re: GDB and Java

[Daniel Jacobowitz]

On Fri, Jul 18, 2003 at 09:43:48AM -0700, David Carlton wrote:
> On 18 Jul 2003 10:28:54 -0600, Tom Tromey <tromey@redhat.com> said:
> 
> > Ok, I've tried cvs head today.  I'm debugging a large java application
> > that consists of about 90 shared libraries (not including libgcj).
> 
> > I can consistently make gdb crash like this:
> 
> > ../../src/gdb/dwarf2-frame.c:1032: internal-error: sect_index_text not initialized
> > A problem internal to GDB has been detected,
> > further debugging may prove unreliable.
> 
> This is probably the same unfortunate interaction between anonymous
> objfiles and the new DWARF 2 frame stuff that Daniel noted in
> conjunction with another patch of mine.  I think
> <http://sources.redhat.com/ml/gdb-patches/2003-07/msg00264.html> is
> supposed to cure it, so give that a try.  Is that patch, or a variant,
> supposed to go into 6.0?  This argues that it had better go in.

I'm not sure if it's for the same issue or not, but that patch
definitely fixes Cygwin.  Mark, everyone seems happy with that patch;
would you check it in on both branches?

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: GDB and Java

[Andrew Cagney]

> 
> I'm not sure if it's for the same issue or not, but that patch
> definitely fixes Cygwin.  Mark, everyone seems happy with that patch;
> would you check it in on both branches?

Might want to start with the mainline then do a pullup once the 
`damange' is known.

Re: GDB and Java

[Tom Tromey]

>>>>> "David" == David Carlton <carlton@kealia.com> writes:

David> I think
David> <http://sources.redhat.com/ml/gdb-patches/2003-07/msg00264.html>
David> is supposed to cure it, so give that a try.

I tried this patch but I still get the same crash.
It might have delayed the crash a little, but it is hard to tell.

I'll send more info after I debug gdb a bit.

Tom

Re: GDB and Java

[Tom Tromey]

Tom> I'll send more info after I debug gdb a bit.

It is definitely an anonymous objfile causing the crash.

I added this just before the use of SECT_OFF_TEXT in
dwarf2_frame_find_fde:

      if (objfile->sect_index_text == -1)
	continue;

With this gdb stops crashing.  It still isn't all that usable --
sometimes "next" will act like "cont".  I'm sure this change is pretty
bogus...

I can debug ok if I just remember to set a breakpoint whenever I
would usually type "n".  Pretty painful though.

Tom

Re: GDB and Java

[Kevin Buettner]

On Jul 18, 12:28pm, Tom Tromey wrote:

> With this gdb stops crashing.  It still isn't all that usable --
> sometimes "next" will act like "cont".  I'm sure this change is pretty
> bogus...
> 
> I can debug ok if I just remember to set a breakpoint whenever I
> would usually type "n".  Pretty painful though.

I think this might be a different bug.  See PR1237.

Kevin