| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This way we avoid having a lot of code in a static (and
unmaintainable) library. The s60main static library now currently has
only one task: to call main().
To move the initialization into QtGui also meant a change in how the
S60 framework is created, because we can no longer use the trick
where we create and start the the S60 event loop and then have the
framework call us back to start main(). The initialization now
follows the creation and destruction of QApplication, which is a lot
more in line with how other platforms do it.
Since S60 doesn't support creating the environment, and *then*
starting it (both are executed by the same call), we had to open up
the S60 framework construction classes and just mirror what they do.
This means that after QApplication construction is done, the S60
framework is initialized, but nothing will run yet and control will
return to main(), where the user can start the event loop himself.
One of the quirks of this approach is that the construction of the
S60 framework makes a new cleanup stack. This means that any active
traps will not be active anymore, and leaving without setting a new
trap will most likely panic. This shouldn't be a problem for us,
since Qt is never supposed to leave, but it means that if anyone uses
the cleanup stack without setting a new trap, they will receive a
panic.
It was considered to add a trap mark in QApplication construction and
then removing it on destruction, but it was dropped because leaving
from main() is still undefined (even if the old cleanup stack would
be restored in the destructor, we wouldn't be able to stop the
exception from unwinding the stack, and the cleanup stack would then
be unbalanced).
RevBy: Jason Barron
RevBy: Janne Anttila
AutoTest: QWidget passed with same failure count
|
