Two things changes:
a) the actual crash cause in on_client_readonly_changed(), to test whether
the 'unit' is valid at all before dereferencing it
b) the callback doesn't seem to be disconnected at all, thus added a disconnect
of it on the source removal as well
The b) might be enough, but just to be sure I included also a).
https://bugzilla.gnome.org/show_bug.cgi?id=751211
It was caused by missing view events remove notifications when replacing
ongoing ECalClientView. The previously known (ViewData::lost_components)
were not notified as removed, thus made incorrect counts in the mini-calendar.