banshee fails to run on arm

[Oliver Grawert]

Binary package hint: banshee

currently banshee dies on startup on armel

[Oliver Grawert (ogra)]

david sugar is currently investigating the gstreamer vorbis "Invalid number format" error seen in the log so we can make sure banshee is not dying because of a missing feature in the foundation. he will follow up with info he found here.

[David Sugar (dyfet-deactivatedaccount)]

I appologize for not commenting earlier, I had been battling other hw issues that kept me from testing this properly. So far, I have not been able to get Banshee to come up far enough to even hit this failure point on my present hw, so I had started digging more into the code to understand why this occurs.

[David Sugar (dyfet-deactivatedaccount)]

I had not reproduced the original issue as reported here, as I had instead encountered a different issue that kills banshee startup before this point which appeared to be a race condition/locking issue. I also have not been able to test other than with xnest, which may be why I see this race.

[David Sugar (dyfet-deactivatedaccount)]

I have attached what I now get today during startup, which is also different...

[David Sugar (dyfet-deactivatedaccount)]

The problem as originally reported clearly no longer exists. The current segfault/crash I find is a new and different problem actually that occurs shortly after Banshee has started up. I also have had banshee simply freeze. Each seem tracable to low level mono runtime issues that are armel specific. I think I want to change the status of this bug to triaged and/or open a new bug on mono runtime, but I will wait for feedback before doing so.

[Oliver Grawert (ogra)]

could you please attach the output of running "banshee --debug" in a terminal again from a freshly installed image so we can compare the two logs ? i still see the very same issue on a freah alpha5 install with all upgrades.

[Paul Larson (pwlars)]

The one I'm getting looks very much like the one you posted called banshee-debug.txt. Calling this one banshee-pwl1.txt

[Oliver Grawert (ogra)]

hrm, i really dont get why i get different errors using the very same images and identical hardware here ...

[David Sugar (dyfet-deactivatedaccount)]

Oliver, thats why I suggested splitting the bug into two bugs :).

Paul, /usr/bin/banshee is of course a script that executes a banshee-1.exe through /usr/bin/mono. If you do run the command under gdb, you will get some native trace. Using udeb did not let me see any symbols in the callback except one for a native arm exception handling routine in mono. I think lool's suggestion of separately rebuilding mono locally (and maybe a few other things) with debug symbols is the way to go with this, and what I was in the middle of doing when I took Oliver's suggestion of rebuilding my environment under a more current (daily) image...

[David Sugar (dyfet-deactivatedaccount)]

I believe this bug, or at least the version of it I was experiencing, could be related to the toolchain exception handling bug, and that hence mono needs to also be rebuilt with the new toolchain.

[Steve Langasek (vorlon)]

Is this bug still in progress for karmic?

[Loïc Minier (lool)]

I could start banshee over ssh -X to a marvell dove Y1 board running latest karmic.

Could someone please try again on a board with working sound?

[Loïc Minier (lool)]

After clicking on a menu, I got:
[Warn 13:03:10.465] Migrating Internet Radio Stations - Directory '/home/lool/.config/banshee/plugins/stations/user' not found. (in `mscorlib')
  at System.IO.Directory.GetFileSystemEntries (System.String path, System.String searchPattern, FileAttributes mask, FileAttributes attrs) [0x00000]
  at System.IO.Directory.GetFiles (System.String path, System.String searchPattern) [0x00000]
  at Banshee.InternetRadio.XspfMigrator.Migrate () [0x00000]
[Info 13:03:13.387] nereid Client Started
Stacktrace:

Native stacktrace:

        [0xb]
        [(nil)]

Debug info from gdb:

hang gdb was hung again, as in bug #452066.

[Loïc Minier (lool)]

@Steve: no, I don't think this bug is on track for fixing in karmic unless it's already fixed with the toolchain changes (seems not).

The bug was initially milestoned due to banshee being a candidate default media player, but it's still going to be rhythmbox this cycle.

[Loïc Minier (lool)]

Wontfix in karmic I fear

[Loïc Minier (lool)]

Confirming there's an issue with running banshee; it does startup though.

[Paul Larson (pwlars)]

@loic - I tried this on my b2.5 with sound working, and banshee still has the same problem.

[David Sugar (dyfet-deactivatedaccount)]

@loic - I have to report I see the same on my b2.5 as I originally reported, too, with the most current updates.

[Mohamed Amine Ilidrissi (ilidrissi.amine)]

Is this bug still relevant? Please test on Lucid or Maverick if it can be reproduced.

[Oliver Grawert (ogra)]

yes, please see Bug 607291

[Mohamed Amine Ilidrissi (ilidrissi.amine)]

Confirming this, then.

[Gabriel Burt (gabaug)]

Oliver, in terms of ARM issues, bug 607291 only references bug 619981, which doesn't seem to provide justification for this bug remaining open. Can anybody still reproduce these native segfaults? And if so, can they install GStreamer, GLib, and Gtk+ debug packages to get a good native stack trace? Banshee 1.4.3 is quite old as this point, as is the version of Mono it was probably running on.

[Jani Monoses (jani)]

Sorry, I had mistakenly commented on the wrong bug all along, thinking there's only one banshee arm issue filed.

I find it is not crashing with banshee installed on an ubuntu-minimal image.
Once I install gtk2-engines-murrine and rerun it crashes.

[Ricardo Salveti (rsalveti)]

Tested on Pandaboard, running Natty with an up-to-date rootfs.

While running with both CPUs activated I get random crashes when starting Banshee, and sometimes it even starts well, but breaks later on (when adding music, like bug 619981).

Tested on Panda with "nosmp", and was able to start and use Banshee for more than 20 times, without any crash, so this seems that Mono or Banshee is not thread safe in this case, causing the issues (it also works fine if I shutdown one CPU at runtime).

[Tobin Davis (gruemaster)]

Ricardo's data should be added to bug 619981. This bug was against an older version on a single core system only. Not the same bug.