[DDA] Crash when revealing indicators

moving to critical - but leaving as new, we need to get some confirmation / repo'ability

Unity crashes on me all the time, but I can't see any pattern in it. It's definitely not only the indicators. It also happens when switching apps, scrolling in the dash, on the lockscreen, everywhere. So it might be very well related to edge swipes.

Wonder if this is a duplicate of https://bugs.launchpad.net/unity8/+bug/1234940

I'd say yes.

Nope, I just checked. The other is in qtubuntu-android's applicationmanager while this happens in the DirectionalDragArea.

Let's keep hunting for steps to reproduce. I haven't seen unity8 crashing in quite a while! :)

I had two crashes today. None involved the indicators., but at least one involved edge swipes.

Still can't reproduce it though.

Can be easily reproduced by holding the phone in landscape mode and tapping with both thumbs quickly on one edge.

The reason it crashes is because DirectionalDragArea starts a new DragGesture for the second finger even though it is a TouchEnd event. on the next run it can't find that finger any more (as the TouchEnd signalizes the finger was removed) and the logic to process the gesture crashes.

Taking, now that we have more concrete steps to reproduce it. The problem is in code I wrote, so it's only natural for me to fix my own mess.

Thanks Michael for digging into it! Taking from where you left.

Fix committed into lp:unity8 at revision None, scheduled for release in unity8, milestone phone-v1-freeze

This bug was fixed in the package qtubuntu - 0.52+14.04.20131125-0ubuntu1

---------------
qtubuntu (0.52+14.04.20131125-0ubuntu1) trusty; urgency=low

  [ Daniel d'Andrada ]
  * Make no assumptions on pointer ids Code was assuming that it ranged
    from 0 to maximumSimultaneousTouches, but that won't hold in the
    future. Also fixes bogus behavior when tapping with two fingers,
    where you would get TouchBegin twice and the first touch point on
    staying on "pressed" state on the second TouchEvent. (LP: #1228336)

  [ Albert Astals ]
  * Pass QRectF as const reference instead of copying it .
  * Don't include the QtQml catch-all header Include only the ones we
    need .
  * Adapt to the Qt 5.2 way of creating the event dispatcher . (LP:
    #1246498)
  * Link against EGL since we use it It was probably being pulled by a
    dependency before .
  * Remove unused -private . (LP: #1254051)
  * Do not include the catch-all QtQuick include .

  [ Ubuntu daily release ]
  * Automatic snapshot from revision 197
 -- Ubuntu daily release <email address hidden> Mon, 25 Nov 2013 14:32:22 +0000

This bug was fixed in the package unity8 - 7.84+14.04.20131128.2-0ubuntu1

---------------
unity8 (7.84+14.04.20131128.2-0ubuntu1) trusty; urgency=low

  [ Michal Hruby ]
  * Depend on the separate scopes plugin
  * Move the BottomBar* DBus communicator to the Utils plugin.

  [ Michał Sawicz ]
  * Wait for DashHome to be available in tst_Shell. (LP: #1254898)
  * Use plugindir from unity-shell-api.pc.
  * Expect stop in upstart job and raise in case of surfaceflinger. (LP:
    #1239876)

  [ Albert Astals ]
  * Fix time test in Qt 5.2 Make factors an array instead of a
    object/dict Objects/dicts are unordered by definition, it happened
    that Qt 5.0 gave them in the orrder we wanted, but with Qt 5.2 is
    failing, and we don't even need the "key", so array works as well
    :).
  * Test that the dash hswipe is disabled while the inner stuff is
    moving .
  * Skip restMaximizeVisibleAreaMoveUpAndShowHeader, it's causing too
    many failed runs And we are confident it's failing because of the
    suboptimal scenegraph run in 5.0.x.
  * Make Dash::test_show_scope_on_load more robust If we are testing
    showScopeOnLoaded make sure we force a scope reload after we set it,
    otherwise it may just happen that the scope has already been loaded
    and the expectaction that we'll change the list to it is just wrong.
  * Dash renderer signals: No need to pass the model up and down Whoever
    is listening to the signal has access to the item that emits the
    signal and has the model right there accessible if needs it.
  * LVWPH: Fix header going bad when setContentHeight ends up moving the
    viewport How to reproduce the bug easily without the patch: * In the
    Dash Home, search for london * Scroll to the bottom * Start moving
    to the apps scope very slowly * At around 3/4 of the move you'll see
    the header in the home scope went to a bad position * Go back to the
    Dash Home. (LP: #1237942, #1246351)
  * Remove unused AppInfo and VideoInfo files .
  * Kill unused ApplicationsFilterGrid.qml .
  * Unify ScopeView and GenericScopeView .
  * Fix header getting lost as per bug 1245824. (LP: #1245824)
  * Remove unused Time.js and its test .
  * Do not include the QtQml megaheader Include only qqml.h which is
    what we need in these files.
  * Don't do stuff if our parent context is gone We'll be gone soon too
    (and crash probably) so don't do anything. This looks a bit like a
    workaround, wait for 5.2 better painting/dispatching loop to see if
    this is not needed anymore, we find a better way to do it, or we
    decide this is fine.

  [ Lars Uebernickel ]
  * Allow setting different indicator positions for different profiles.

  [ Mirco Müller ]
  * Added checkbox for toggling between echo-modes of password-
    entryfields in ext. snap-decisions.
  * Fixes bug #1200569. (LP: #1200569)

  [ Andrea Cimitan ]
  * Switch to application scope when a dash swipe is taking place and an
    app is on foreground. (LP: #1231996)
  * Shifts wallpaper rendering for greeter lockscreen to be inline with
    shell. (LP: #1231731)
  * Dinamically load the Carousel/Filtergrid with more than 6 items.
    (LP: #1226288, #1234105)
  * R...