bzr should use apport to report errors

On Fri, 2009-06-19 at 05:48 +0000, Martin Pool wrote:
> Public bug reported:
>
> bzr should when possible use https://launchpad.net/apport to
> semiautomatically file bugs into Launchpad, rather than asking the user
> to do it manually.

I wrote a patch to do this some years back and it got landed. I think it
may have been reverted.

-Rob

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Martin Pool wrote:
> Public bug reported:
>
> bzr should when possible use https://launchpad.net/apport to
> semiautomatically file bugs into Launchpad, rather than asking the user
> to do it manually.
>
> ** Affects: bzr
> Importance: Medium
> Status: Confirmed
>

Agreed. When we tried it in the past, we got random crashes *from*
apport, so we disabled support.

apport has probably matured since then, and we should try to use it if
it is available.

John
=:->

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAko7pigACgkQJdeBCYSNAAOo8gCfVmK0hvmheTfA1EH8p2RkCV5u
A8sAnR0oI2J3h3fJ6T6btST7lBNXr4nc
=/7fT
-----END PGP SIGNATURE-----

I definitely recommend giving this a second look. I don't know what problems you ran into, but if there are bug reports with details, please point them out here and we can try to get them fixed.

apport adds a lot of value to bug reports these days, and is getting better all the time. In Karmic, it now has the capability to ask the user questions about their bug report before it even hits Launchpad, which gives you the chance to collect more information from the system as well.

On Fri, 2009-06-19 at 14:52 +0000, John A Meinel wrote:
>
> Agreed. When we tried it in the past, we got random crashes *from*
> apport, so we disabled support.

I thought it was on non-ubuntu platforms that the errors occured?

-Rob

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Robert Collins wrote:
> On Fri, 2009-06-19 at 14:52 +0000, John A Meinel wrote:
>> Agreed. When we tried it in the past, we got random crashes *from*
>> apport, so we disabled support.
>
> I thought it was on non-ubuntu platforms that the errors occured?
>
> -Rob
>

Absolutely could be. I just remember the apport bug reporting causing
more problems, enough that we removed support.

John
=:->

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkpA8DMACgkQJdeBCYSNAAPgPgCfYMNzgu/uYozAverujlceu+TG
92gAnj4mFJFDSHu2bKOveAX3qXU29DAh
=rp/Q
-----END PGP SIGNATURE-----

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

John A Meinel wrote:
> Robert Collins wrote:
>> On Fri, 2009-06-19 at 14:52 +0000, John A Meinel wrote:
>>> Agreed. When we tried it in the past, we got random crashes *from*
>>> apport, so we disabled support.
>> I thought it was on non-ubuntu platforms that the errors occured?
>
>> -Rob
>
>
> Absolutely could be. I just remember the apport bug reporting causing
> more problems, enough that we removed support.
>
> John
> =:->
>

Here are the threads that I dug up. The conversation was back in Nov 2007:

http://news.gmane.org/find-root.php?message_id=%3c1162946743.8532.18.camel%40localhost.localdomain%3e
http://news.gmane.org/find-root.php?message_id=%3c455D4FDE.6020601%40utoronto.ca%3e
http://news.gmane.org/find-root.php?message_id=%3c47359BBD.9030809%40arbash%2dmeinel.com%3e

essentially it was:

1) Robert merged a changed to start supporting apport
2) This ended up causing test suite failures for people who *had* apport
available, but not on PQM because it did not.
3) Aaron wanted a way to completely disable it. He was concerned about
failures in apport that would mask the original bzr failure.

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkpA9VkACgkQJdeBCYSNAAOHvgCgw07MVboFupQFexqiImGkpf5a
YGYAoIh8u6Oh7GHUg5c7nc+qPkNvB6vN
=ka8H
-----END PGP SIGNATURE-----

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

John A Meinel wrote:
> 1) Robert merged a changed to start supporting apport
> 2) This ended up causing test suite failures for people who *had* apport
> available, but not on PQM because it did not.
> 3) Aaron wanted a way to completely disable it. He was concerned about
> failures in apport that would mask the original bzr failure.

Thanks for digging that up. I'd completely forgotten the details.

Personally, my general experience with apport has been terrible. It
seems to just go into an infinite loop most of the time and I have to
kill it.

But if you want to have another go at it, that's fine with me, as long
as it doesn't cause problems in the long term.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkpgBMAACgkQ0F+nu1YWqI3Z6QCcCODEScjQmUQgfZTYoifHzTEF
yO0AnAyEfj51s0yWfVinffaabqoa8Ssy
=DKDs
-----END PGP SIGNATURE-----

On Fri, Jul 17, 2009 at 04:57:39AM -0000, Aaron Bentley wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> John A Meinel wrote:
> > 1) Robert merged a changed to start supporting apport
> > 2) This ended up causing test suite failures for people who *had* apport
> > available, but not on PQM because it did not.
> > 3) Aaron wanted a way to completely disable it. He was concerned about
> > failures in apport that would mask the original bzr failure.
>
> Thanks for digging that up. I'd completely forgotten the details.
>
> Personally, my general experience with apport has been terrible. It
> seems to just go into an infinite loop most of the time and I have to
> kill it.

We successfully receive several thousand bug reports per month from apport,
so the problems you're experiencing must be something more specific than it
going into an infinite loop most of the time.

Could you file a bug on apport describing the problem in more detail? The
recommended method to do so is to run "ubuntu-bug apport", which uses apport
itself to file the bug, so perhaps you could try to reproduce the problem at
the same time. :-)

--
 - mdz

Landed for 2.0

For a followon bug 389329 to suggest something better than "bzr crashed" descriptions and bug 416720 for including the log.

Very glad to see this implemented. Thanks, Martin!

Another important followon is bug 391015, to get an apport package hook in place for bzr. This means that when an Ubuntu user reports a bug using "ubuntu-bug bzr", you have the opportunity to collect relevant information from the the system (or the user, by asking interactive questions if you so choose). This same code will be run if bzr crashes.

I haven't looked at your implementation of the fix for this bug, but if you're using the apport API directly to attach debug information, you might consider moving this code into a package hook so that it's run for normal bug reports as well (not only crashes).

Also bug 417881 to turn it off for selftest...

2009/8/21 Matt Zimmerman <email address hidden>:
> Another important followon is bug 391015, to get an apport package hook
> in place for bzr.  This means that when an Ubuntu user reports a bug
> using "ubuntu-bug bzr", you have the opportunity to collect relevant
> information from the the system (or the user, by asking interactive
> questions if you so choose).  This same code will be run if bzr crashes.
>
> I haven't looked at your implementation of the fix for this bug, but if
> you're using the apport API directly to attach debug information, you
> might consider moving this code into a package hook so that it's run for
> normal bug reports as well (not only crashes).

We are using the api directly. I agree that covering both ways would be good.

Most of the interesting information we want to capture (what plugins
are loaded, what Python extensions are loaded, what debug flags are
set) can only be directly observed from inside the process, before it
exits. These wouldn't be applicable if it's run with 'ubuntu-bug bzr'
or after a hard crash.

We also want to make sure that apport is active even on platforms that
don't use it system-wide.

I'm sure these are resolvable. At the moment it seems to me we might
need both a package hook and some code in bzrlib.

--
Martin <http://launchpad.net/~mbp/>

I tried using ubuntu-bug to send the crash file to Launchpad, but that seems to be blocked by bug 422881 in Jaunty. It will probably work in Karmic. (Though we'll have to see if it correctly files against the bzr package or project or neither.)

On Wed, Sep 09, 2009 at 06:29:22AM -0000, Martin Pool wrote:
> I tried using ubuntu-bug to send the crash file to Launchpad, but that
> seems to be blocked by bug 422881 in Jaunty. It will probably work in
> Karmic. (Though we'll have to see if it correctly files against the bzr
> package or project or neither.)

The normal way to send the crash to Launchpad is to click on the "!" icon
that update-notifier displays when it notices a crash file.

The manual method is to run apport-gtk or apport-cli on the crash file.

ubuntu-bug isn't really intended for this purpose, though it looks like
Martin has made it do something intelligent if pointed at a crash file.
Doing it the normal way should work fine in Jaunty, though.

--
 - mdz