whoopsie process is leaking memory

The process keeps growing, now at 1.5gb, attaching the strace for 5 seconds.

Status changed to 'Confirmed' because the bug affects multiple users.

==12580== LEAK SUMMARY:
==12580== definitely lost: 10,060,860 bytes in 9,559 blocks
==12580== indirectly lost: 240 bytes in 10 blocks
==12580== possibly lost: 104,174,131 bytes in 151,412 blocks
==12580== still reachable: 416,562 bytes in 1,273 blocks
==12580== suppressed: 0 bytes in 0 blocks

During last 14 hours the process gained 100 Mb. Attaching valgrind output (installed debug symbols for glib and rebuilt whoopsie to get unstripped version).

Attaching /var/crash per ev's request

The main leak there seems to be:

==12580== 10,059,776 bytes in 9,557 blocks are definitely lost in loss record 1,071 of 1,074
==12580== at 0x4C2B7B2: realloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
==12580== by 0x541DAF6: g_realloc (gmem.c:224)
==12580== by 0x53ECC67: g_ptr_array_maybe_expand (garray.c:1093)
==12580== by 0x53EDC22: g_ptr_array_add (garray.c:1350)
==12580== by 0x4EC77C7: read_netlink_messages (gnetworkmonitornetlink.c:237)
==12580== by 0x4E9F7F5: socket_source_dispatch (gsocket.c:3168)
==12580== by 0x5417C99: g_main_context_dispatch (gmain.c:2515)
==12580== by 0x541805F: g_main_context_iterate.isra.23 (gmain.c:3123)
==12580== by 0x5418459: g_main_loop_run (gmain.c:3317)
==12580== by 0x402EA8: main (whoopsie.c:934)

The bug has been fixed upstream:
http://git.gnome.org/browse/glib/commit/?id=17e95c59c9f7b1cd023aabb3645c70b9f1f37577

Setting "fix commited" for quantal since that's going to be included in the next version, we will also include the fix in the glib SRU that should be uploaded next week to precise (update to the current stable version)

Indeed. And apologies, I hadn't seen your reply to this bug and went ahead and uploaded it to -proposed.

I've rejected the current glib2.0 in the queue so that the full upstream stable SRU can go through more easily.

This is a huge diff, not just "a few minor changes" IMO. Its sort of shocking that glib adds so much in a a few "patch" releases. I know some of these things are for non-Linux platforms, but its really quite hard to separate them out.

If this is just a one line fix, why aren't we just cherry picking the one line fix for this bug? Thats typically how we control regression potential. I'm digging through the diff (34,000 lines!) but at this point I can't really justify the time to review it all.

Can somebody update the test case to include a comprehensive test plan so we can get enough test coverage, or re-upload with just the fix for this bug?

Clint:

you can see the details of the commits on http://git.gnome.org/browse/glib/log/?h=glib-2-32

summary of the .news:

* GApplication: can now have a NULL application ID

* g_clear_object: fix warnings when using it on C++ (due to lack of
  ability to implicitly cast void*)

* GDBus:
 - add our own implementation of the message bus for use on Windows only
 - fix up a few bugs that use of this bus uncovered in GDBus
 - escape nonce files in dbus addressess (think 'c:\')
 - support initial underscores in dbus codegen namespace (for private)

* Fix misdetection of GNUstep as Cocoa (for the MacOS GSettings backend)

* make sure configure fails if AC_CHECK_ALIGNOF cannot detect the alignment

* GAppInfo: overwrite the DISPLAY only if it is set in the launch context

* glib/tests/date: force US locale running the GDateTime tests

* GSocketControlMessage: Don't warn about unknown messages

* Resources:
 - fix broken use of GVDB on big endian machines
 - set a 'display name' so that pretty file names appear in Gtk CSS
   warning messages

* GMainContext:
 - block child sources when blocking the parent
 - introduce more testcases for child sources

* Bug fixed:
 619026 avoid warning in gutils.h when using gcc with -Wconversion
 669260 Open/Save dialog hangs waiting for data with libsoup
 671249 GApplication: Allow a null application_id?
 672786 goa-daemon: action in notification doesn't work
 673409 g_resource_lookup_data may return stale data pointer
 674172 glib-2.32.1 misdetects GNUstep as Cocoa
 674345 cssprovider: Make sure to print out file name in css warnings
 674483 broken configure results when cross-compiling with gcc >= 4.5

* Bug fixes:
 Correct a typo in the ELF configure check
 Reset LIBS after the ELF configure check
 Fix GIO/GObject Visual C++ projects
 gio: fix error handling in async case of GProxy...
 GConverterInputStream: fix an infinite loop...

* Translation updates

------

The new version is shipped in Debian testing and fedora, a big part of the diff is the dbus server added for win32 (it's not used on linux)

Testing ... there is nothing specific to test out of the fact that the desktop keeps running normally

Clint, on the diff when filtering out "noise":

$ debdiff --exclude *win32* --exclude tests* --exclude Makefile.in --exclude po --exclude docs --exclude ChangeLog --exclude configure --exclude *generated* glib2.0_2.32.1-0ubuntu2.dsc glib2.0_2.32.3-0ubuntu1.dsc | diffstat
...
 INSTALL | 4
 NEWS | 75
 README | 2
 configure.ac | 16
 debian/changelog | 12
 debian/control | 2
 debian/control.in | 2
 debian/patches/git_powerpc_gresources.patch | 28
 debian/patches/gnetworkmonitor_dont_leak_networks.patch | 21
 debian/patches/series | 2
 gio/Makefile.am | 25
 gio/dbus-daemon.xml | 76
 gio/gapplication.c | 57
 gio/gapplicationimpl-dbus.c | 6
 gio/gconverterinputstream.c | 12
 gio/gdbus-2.0/codegen/config.py | 2
 gio/gdbus-2.0/codegen/utils.py | 7
 gio/gdbusaddress.c | 306 ++
 gio/gdbusdaemon.c | 1752 ++++++++++++++++
 gio/gdbusdaemon.h | 19
 gio/gdbusprivate.c | 42
 gio/gdbusserver.c | 5
 gio/gdesktopappinfo.c | 3
 gio/gio.rc | 8
 gio/gioenums.h | 9
 gio/giomodule-priv.h | 4
 gio/giomodule.c | 6
 gio/gproxyaddressenumerator.c | 95
 gio/gresource.c | 5
 gio/gresourcefile.c | 1
 gio/gsocketcontrolmessage.c | 11
 glib/glib.rc | 8
 glib/gmain.c | 35
 glib/gutils.h | 2
 gmodule/gmodule.rc | 8
 gobject/gobject.h | 4
 gobject/gobject.rc | 8
 gthread/gthread.rc | 8
 38 files changed, 2543 insertions(+), 145 deletions(-)

so the diff is around 2690 lines, 1771 of those being the gdbusdaemon for win32 use, it means the remaining diff is around 900 lines of actual changes, it's not trivial but far from the 34k you listed

Excerpts from Sebastien Bacher's message of 2012-05-31 08:54:24 UTC:
> Clint, on the diff when filtering out "noise":
>
> $ debdiff --exclude *win32* --exclude tests* --exclude Makefile.in --exclude po --exclude docs --exclude ChangeLog --exclude configure --exclude *generated* glib2.0_2.32.1-0ubuntu2.dsc glib2.0_2.32.3-0ubuntu1.dsc | diffstat
> ...
> INSTALL | 4
> NEWS | 75
> README | 2
> configure.ac | 16
> debian/changelog | 12
> debian/control | 2
> debian/control.in | 2
> debian/patches/git_powerpc_gresources.patch | 28
> debian/patches/gnetworkmonitor_dont_leak_networks.patch | 21
> debian/patches/series | 2
> gio/Makefile.am | 25
> gio/dbus-daemon.xml | 76
> gio/gapplication.c | 57
> gio/gapplicationimpl-dbus.c | 6
> gio/gconverterinputstream.c | 12
> gio/gdbus-2.0/codegen/config.py | 2
> gio/gdbus-2.0/codegen/utils.py | 7
> gio/gdbusaddress.c | 306 ++
> gio/gdbusdaemon.c | 1752 ++++++++++++++++
> gio/gdbusdaemon.h | 19
> gio/gdbusprivate.c | 42
> gio/gdbusserver.c | 5
> gio/gdesktopappinfo.c | 3
> gio/gio.rc | 8
> gio/gioenums.h | 9
> gio/giomodule-priv.h | 4
> gio/giomodule.c | 6
> gio/gproxyaddressenumerator.c | 95
> gio/gresource.c | 5
> gio/gresourcefile.c | 1
> gio/gsocketcontrolmessage.c | 11
> glib/glib.rc | 8
> glib/gmain.c | 35
> glib/gutils.h | 2
> gmodule/gmodule.rc | 8
> gobject/gobject.h | 4
> gobject/gobject.rc | 8
> gthread/gthread.rc | 8
> 38 files changed, 2543 insertions(+), 145 deletions(-)
>
> so the diff is around 2690 lines, 1771 of those being the gdbusdaemon
> for win32 use, it means the remaining diff is around 900 lines of actual
> changes, it's not trivial but far from the 34k you listed
>

Agreed. Tha...

summary of the changes

small bug fixes, including (which we would probably backport):
* "GSocketControlMessage: Don't warn about unknown messages"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=9b1a9ed4ce326d4ebfca68b2a98ddb4d8110e5e7

* "gmain: block child sources when blocking the parent"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=b1d447bd664bcbc607811c7920b67856d4f551e6
* "resourcefile: Set display name"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=ad35c6cae1ad065bc46968b1169c18e624dd8ada

* "Support initial underscores in dbus codegen namespace"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=5051aaca68eacf280ce48d1500a19cccf2125fc8

* "gdbus: Escape nonce files in dbus addressess"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=27352d96e28cf9e65bce12e2250a679261872aae

* "Fix g_clear_object macro with C++ compilers"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=d92736ab032bb650b1d115042e2a0ff965175940

* "GApplication: allow null application_id"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=7e9306e84e3cfc9b9e43991bd6d1978a77463349

* "gio: fix error handling in async case of GProxyAddressEnumerator"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=b62b5a3d8c67e3245e6d642207d07cafe3943fc2

* "GConverterInputStream: fix an infinite loop when fill_buffer returns an error"
http://git.gnome.org/browse/glib/commit/?h=glib-2-32&id=6799e4a73a93c4c6c2e3fbfbd44f05e5a5e26bcf

translations updates: 18
win32 specific fixes: 9
tests cases improvements: 3
change we already backported: 1
build system fixes (including other platforms): 8

The previous summary is basically the diff between the versions (add the patch Evan backported), what strategy the SRU team suggest to get that update in?

Hello Roman, or anyone else affected,

Accepted glib2.0 into precise-proposed. The package will build now and be available in a few hours. Please test and give feedback here. See https://wiki.ubuntu.com/Testing/EnableProposed for documentation how to enable and use -proposed. Thank you in advance!

Left the same system running with the new glib package running overnight - whoopsie is at 3Mb RES after 11 hours (earlier it would go to 100Mb and beyond).

This bug was fixed in the package glib2.0 - 2.32.3-0ubuntu1

---------------
glib2.0 (2.32.3-0ubuntu1) precise-proposed; urgency=low

  * New upstream version
  * debian/patches/git_powerpc_gresources.patch:
    - dropped, the fix in the new version

  [ Evan Dandrea ]
  * gnetworkmonitor_dont_leak_networks.patch: Do not leak the list of
    networks in GNetworkMonitor. Thanks Dan Winship (LP: #998521)!
 -- Sebastien Bacher <email address hidden> Tue, 29 May 2012 17:53:18 +0200

the fix is in the current glib, and was not a whoopsie issue