Sugar shell crashes when threads used.

[cscott]

Sugar never calls gtk.gdk.threads_init(), so any attempt to use threads in code called by the sugar shell (control panel code in particular) causes sugar to crash -- python throws an assertion about its GILlock.

The fix is very simple: in /usr/share/sugar/shell/main.py, right after:

pygtk.require('2.0')
import gtk
import gobject

add the line:

gtk.gdk.threads_init()

Voila! No more crashes.

Found this bug while writing the 'activity update' sugar control panel, which uses threads to keep the UI responsive while it is doing network operations in the background to update activity version status.

[mstone]

Was Sugar _intentionally_ not calling threads_init()?

[marco]

Replying to mstone:

Was Sugar _intentionally_ not calling threads_init()?

Well, you usually call threads_init when you are using threads, and the sugar shell has not been using them so far. I guess the cp panel Scott is talking about is not part of sugar yet?

[cscott]

No one can call threads_init after sugar starts, so if anyone installs a control panel which needs threads (which I do), sugar needs to properly init the threads at the start.

The control panel deliberately has a modular design which lets people slot in external control panels. Sugar should support their use of threads.

[tomeu]

We actively eliminated use of threads in sugar because it used to cause pygtk to wake up 10 times per second. I think that both fedora and ubuntu have applied a patch about that to both python and pygtk, so I think we can add it back. We should test again if we are causing unnecessary wakeups that may interfere with battery conservation.

[tomeu]

|TestCase|

In a build generated by a pilgrim version without the fix linked below, check that Sugar starts up correctly.

http://dev.laptop.org/git?p=users/cscott/pilgrim;a=commitdiff;h=205c907270a07836f29709cf1d13257d56d4a846

(please improve this test case)

[cscott]

If you can tell me when sugar packages with this patch will show up in joyride (either temporarily via the public_rpms mechanism or via koji), I can have the pilgrim patch removed.

cjb, as our resident power guru, could you double check that pygtk in current joyride builds (which have a patch to enable threads) isn't causing excessive wakeups?

[erikos]

Ok we will add it to the next build and will tell you when to remove from pilgrim. A test from cjb about wakeups would be nice indeed.

[erikos]

This was not pushed yet - did now: http://dev.laptop.org/git?p=sugar;a=commit;h=9902b4562ea87f0c56700a287d199f746daee1f4

[erikos]

The pilgrim patch has been removed and the control panel starts up fine in joyride >= 2268