The WinRT NetworkStatusChanged callback may happen during or slightly
before we unregister our token, which we usually follow up by destroying
the object. So we have to avoid potentially doing work on a deallocated
object.
Do this using the old QPointer-trick. Neither me nor reporter can
reproduce it locally, so this is only a best-measure.
Further problems may be that the storage for the lambda has already
been destroyed and repurposed, in which case the pointer may be valid,
but junk, which would lead to another crash. But this is unavoidable as
long as MS does not synchronize callbacks with (un)registering new
callbacks. To attempt combatting this we hold our own lock around
unregistration and the "meat" of the callback.
Pick-to: 6.4 6.4.1
Fixes: QTBUG-108218
Change-Id: Iacf8b8f458cca3152ff395e9a38e8df193534f46
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
7898de4258