[gdb-7.0] One week to the gdb-7.0 release!

[gdb-7.0] One week to the gdb-7.0 release!

[Joel Brobecker]

Hello,

Just a quick update and reminder that the gdb-7.0 release is scheduled
to be made in a week from now. We promised we would try to review all
unreviewed patches that were sent before branch time, which was Sept 16.
If you're concerned about the status of your patch, please send *me*
a ping with a URL of the patch in question, and I'll try my best to
take a look (but please do ping C++ patches on the gdb-patches list,
however, as I don't have any experience in this area).

Please note that the push for reviews is an effort to clear the backlog
more than an effort to get them all in 7.0. At this point, only selected
patches that are safe and/or fix a critical issue will be included in
the branch.

Cheers,
-- 
Joel

RE: [gdb-7.0] One week to the gdb-7.0 release!

[Anirban Sinha]

Hello all;

So are we going to see a new gdb release tomorrow?

Cheers,

Ani


>-----Original Message-----
>From: Joel Brobecker [mailto:brobecker@adacore.com]
>Sent: Wednesday, September 23, 2009 3:09 PM
>To: gdb@sourceware.org
>Subject: [gdb-7.0] One week to the gdb-7.0 release!
>
>Hello,
>
>Just a quick update and reminder that the gdb-7.0 release is scheduled
>to be made in a week from now. We promised we would try to review all
>unreviewed patches that were sent before branch time, which was Sept
16.
>If you're concerned about the status of your patch, please send *me*
>a ping with a URL of the patch in question, and I'll try my best to
>take a look (but please do ping C++ patches on the gdb-patches list,
>however, as I don't have any experience in this area).
>
>Please note that the push for reviews is an effort to clear the backlog
>more than an effort to get them all in 7.0. At this point, only
selected
>patches that are safe and/or fix a critical issue will be included in
>the branch.
>
>Cheers,
>--
>Joel

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Joel Brobecker]

> So are we going to see a new gdb release tomorrow?

We've had some last-second instability on the branch, which we should
now have taken care of.  Unless we find some evidence of the contrary,
we're now thinking that it would be more prudent to extend the testing
period for a few more days, probably until Monday or Tuesday next week.
And instead of making a release tomorrow, we'll make another pre-release
for everyone to test and enjoy.  The pre-release should be very close
to the actual release.

-- 
Joel

RE: [gdb-7.0] One week to the gdb-7.0 release!

[Anirban Sinha]

>We've had some last-second instability on the branch, which we should
>now have taken care of.  Unless we find some evidence of the contrary,
>we're now thinking that it would be more prudent to extend the testing
>period for a few more days, probably until Monday or Tuesday next week.

Sigh ...

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Jack Howarth]

On Tue, Sep 29, 2009 at 05:44:18PM -0700, Anirban Sinha wrote:
> >We've had some last-second instability on the branch, which we should
> >now have taken care of.  Unless we find some evidence of the contrary,
> >we're now thinking that it would be more prudent to extend the testing
> >period for a few more days, probably until Monday or Tuesday next week.
> 
> Sigh ...

   I haven't tested the gdb 7.0 branch yet, but gdb trunk seems to be
totally broken now on darwin even with the proposed patch...

http://sourceware.org/ml/gdb-patches/2009-09/msg00430.html

Now gdb errors out earlier and claims that darwin executables
aren't valid executable files. Is anyone else seeing that on
darwin?
           Jack
ps This is from './dist/bin/gdb ./a.out'.

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Joel Brobecker]

>    I haven't tested the gdb 7.0 branch yet, but gdb trunk seems to be
> totally broken now on darwin even with the proposed patch...
> 
> http://sourceware.org/ml/gdb-patches/2009-09/msg00430.html
> 
> Now gdb errors out earlier and claims that darwin executables
> aren't valid executable files. Is anyone else seeing that on
> darwin?

Again, you really have to qualify your statement with which version
of Darwin you are using.

I believe that you are using Snow Leopard whereas things to be OK
on Leopard.  The problem on Snow Leopard seems to be coming from
the fact that the method for giving GDB the priviledges it needs
in order to debug the program have changed.  AdaCore is planning on
looking at that, eventually, but we just won't have the resources
to do so for at least a few weeks.

-- 
Joel

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Jack Howarth]

On Wed, Sep 30, 2009 at 09:29:49AM -0700, Joel Brobecker wrote:
> >    I haven't tested the gdb 7.0 branch yet, but gdb trunk seems to be
> > totally broken now on darwin even with the proposed patch...
> > 
> > http://sourceware.org/ml/gdb-patches/2009-09/msg00430.html
> > 
> > Now gdb errors out earlier and claims that darwin executables
> > aren't valid executable files. Is anyone else seeing that on
> > darwin?
> 
> Again, you really have to qualify your statement with which version
> of Darwin you are using.
> 
> I believe that you are using Snow Leopard whereas things to be OK
> on Leopard.  The problem on Snow Leopard seems to be coming from
> the fact that the method for giving GDB the priviledges it needs
> in order to debug the program have changed.  AdaCore is planning on
> looking at that, eventually, but we just won't have the resources
> to do so for at least a few weeks.
> 
> -- 
> Joel

Joel,
   This is under darwin10, however I am seeing a totally new error
than before. A week ago gdb would report errors about opening the
mach port. Now it seems to not even recognize the binary as being
valid. So unless the error code in gdb was change, it seems to be
a new bug. I'll post the exact error message tonight.
                 Jack

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Paul Pluzhnikov]

On Wed, Sep 30, 2009 at 9:29 AM, Joel Brobecker <brobecker@adacore.com> wrote:

> I believe that you are using Snow Leopard whereas things to be OK
> on Leopard. The problem on Snow Leopard seems to be coming from
> the fact that the method for giving GDB the priviledges it needs
> in order to debug the program have changed.  AdaCore is planning on
> looking at that, eventually, but we just won't have the resources
> to do so for at least a few weeks.

Clarification: the "new method" (code signing?) is required on newer
(latest?) releases of Leopard as well: I have *not* upgraded to Snow Leopard,
but I still see the problem.

-- 
Paul Pluzhnikov

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Jack Howarth]

On Wed, Sep 30, 2009 at 10:08:41AM -0700, Paul Pluzhnikov wrote:
> On Wed, Sep 30, 2009 at 9:29 AM, Joel Brobecker <brobecker@adacore.com> wrote:
> 
> > I believe that you are using Snow Leopard whereas things to be OK
> > on Leopard. The problem on Snow Leopard seems to be coming from
> > the fact that the method for giving GDB the priviledges it needs
> > in order to debug the program have changed.  AdaCore is planning on
> > looking at that, eventually, but we just won't have the resources
> > to do so for at least a few weeks.
> 
> Clarification: the "new method" (code signing?) is required on newer
> (latest?) releases of Leopard as well: I have *not* upgraded to Snow Leopard,
> but I still see the problem.
> 
> -- 
> Paul Pluzhnikov

Paul,
   I assume 10.5.8 and Snow Leopard share the same problem. I used to
see the error...

Unable to find Mach task port for process-id 39624: (os/kern) failure (0x5).
 (please check gdb is setgid procmod)

Now in gdb trunk, with the proposed darwin patch, I don't even get that
far. When passing an executable to gdb, it complains that the binary isn't
a valid executable. Is that what you are seeing under 10.5.8? 
            Jack
ps Actually I would have thought this issue requiring signing of gdb
would have been coupled more to a new Xcode release than to a new
Mac OS X release.

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Paul Pluzhnikov]

On Wed, Sep 30, 2009 at 10:31 AM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:

> I assume 10.5.8 and Snow Leopard share the same problem. I used to
> see the error...

Quite possible. I don't have Snow Leopard, so I don't know what happens
there.

> Unable to find Mach task port for process-id 39624: (os/kern) failure (0x5).
>   (please check gdb is setgid procmod)

Yes, that is exactly the error I see on Leopard (unless I run gdb as root).

> Now in gdb trunk, with the proposed darwin patch, I don't even get that
> far. When passing an executable to gdb, it complains that the binary isn't
> a valid executable. Is that what you are seeing under 10.5.8?

I don't normally do any development on MacOS -- I only got dragged into it
because my patch broke it :-(

I have not tested the latest trunk, nor 7.0 branch, after I've "fixed"
the find_pc_section breakage. I will try trunk tonight, just to confirm that
I can replicate what you see (and maybe fix it if I manage to understand
what's happening).

> ps Actually I would have thought this issue requiring signing of gdb
> would have been coupled more to a new Xcode release than to a new
> Mac OS X release.

Why would you think that?

The kernel denies GDB access to debug port; I don't see what this would
have to do with Xcode.

Cheers,
-- 
Paul Pluzhnikov

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Jack Howarth]

On Wed, Sep 30, 2009 at 10:42:12AM -0700, Paul Pluzhnikov wrote:
> On Wed, Sep 30, 2009 at 10:31 AM, Jack Howarth <howarth@bromo.med.uc.edu> wrote:
> 
> > I assume 10.5.8 and Snow Leopard share the same problem. I used to
> > see the error...
> 
> Quite possible. I don't have Snow Leopard, so I don't know what happens
> there.
> 

I have a spare drive with Leopard (10.5.8) on it so I can double
check that gdb cvs shows the same behavior.

> The kernel denies GDB access to debug port; I don't see what this would
> have to do with Xcode.

I was just wondering if the behavior was any different from Xcode 3.1.2
to Xcode 3.1.3.

Re: [gdb-7.0] One week to the gdb-7.0 release!

[Joel Brobecker]

> Clarification: the "new method" (code signing?) is required on newer
> (latest?) releases of Leopard as well: I have *not* upgraded to Snow Leopard,
> but I still see the problem.

Aya-eek! This is just lovely :-(.

-- 
Joel

Re: gdb seems to be broken on darwin 10.5.8 and later

[Jonas Maebe]

On 30 Sep 2009, at 18:20, Jack Howarth wrote:

> Now gdb errors out earlier and claims that darwin executables
> aren't valid executable files. Is anyone else seeing that on
> darwin?

Since you mentioned later that you are on 10.6, might the problem be  
that previously you only tested gdb on programs compiled under 10.5.x  
or with "-mmacosx-version-min 10.5" (or earlier), while you are now  
trying with a program compiled on and for 10.6? Apple added/changed  
some things to the Mach-O object format in 10.6 (just like in 10.5),  
and e.g. (some?) binaries compiled for 10.6 are not recognised as  
valid executables at all on 10.5 or earlier. gdb might be confused for  
similar reasons.

You may want to try compiling your program with "-mmacosx-version-min  
10.5" and check again.


Jonas

Re: gdb seems to be broken on darwin 10.5.8 and later

[Jack Howarth]

On Thu, Oct 01, 2009 at 01:29:28PM +0200, Jonas Maebe wrote:
>
> On 30 Sep 2009, at 18:20, Jack Howarth wrote:
>
>> Now gdb errors out earlier and claims that darwin executables
>> aren't valid executable files. Is anyone else seeing that on
>> darwin?
>
> Since you mentioned later that you are on 10.6, might the problem be  
> that previously you only tested gdb on programs compiled under 10.5.x or 
> with "-mmacosx-version-min 10.5" (or earlier), while you are now trying 
> with a program compiled on and for 10.6? Apple added/changed some things 
> to the Mach-O object format in 10.6 (just like in 10.5), and e.g. (some?) 
> binaries compiled for 10.6 are not recognised as valid executables at all 
> on 10.5 or earlier. gdb might be confused for similar reasons.
>
> You may want to try compiling your program with "-mmacosx-version-min  
> 10.5" and check again.
>
>
> Jonas

Actually tonight's gdb cvs built on x86_64-apple-darwin10 seems to work
fairly well with the proposed macho changes as long as I run as root.
Targeting binaries for 10.6 doesn't seem to be a problem. The only issue
I have run into so far is with binaries built at -O3 with gcc trunk.
With Apple's gcc-4.2 compiler, if I compile...

gcc -m64 -g -O3 himenoBMTxpa.c

break points are set normally...

(gdb) break 4
Breakpoint 1 at 0x100001b14: file himenoBMTxpa.c, line 4.

but with binaries built with gcc 4.5, I find that this doesn't work...

(gdb) break 4
Breakpoint 1 at 0x100002832: file himenoBMTxpa.c, line 70. (4 locations)

The problem disappears if I regress the compile back to -O2 with gcc 4.5.
Is this problem being seen with gdb under linux?
             Jack

Re: gdb seems to be broken on darwin 10.5.8 and later

[Joel Brobecker]

> gcc -m64 -g -O3 himenoBMTxpa.c

Debugging code optimized with GCC is a real challenge, even at -O1
sometimes. That's usually because the compile emitted some debugging
info that's not precise enough.

-- 
Joel

Re: gdb seems to be broken on darwin 10.5.8 and later

[Jack Howarth]

On Fri, Oct 02, 2009 at 08:40:01AM -0700, Joel Brobecker wrote:
> > gcc -m64 -g -O3 himenoBMTxpa.c
> 
> Debugging code optimized with GCC is a real challenge, even at -O1
> sometimes. That's usually because the compile emitted some debugging
> info that's not precise enough.
> 
> -- 
> Joel

Joel,
  I was just curious if this was being seen on linux as well or just
on darwin. FYI, the source code for this example is at...

http://accc.riken.jp/HPC/HimenoBMT/compile_op_e.html

if anyone else wants to try it on linux with current gcc trunk.
             Jack

Re: gdb seems to be broken on darwin 10.5.8 and later

[Joel Brobecker]

> I was just curious if this was being seen on linux as well or just
> on darwin. FYI, the source code for this example is at...

The general issue of debugging optimized code from GCC is independent
of the platform. There are some projects out there that try to improve
the situation, I'm not really familiar with the GCC project to tell
you more (Alexander Oliva was involved in one of these efforts, I think).

-- 
Joel