added an e_observer_notify_all_observees() to notify all existing observees of things, such as the creation of a border. This way, an observer can watch for E_EVENT_BORDER_NEW events and register all new borders.
Iconification! Click that iconify button. Then click the middle button on the desktop to get a menu of iconified apps to uniconify. This menu is a decent example of using the observer code to avoid polling or rebuilding more than necessary. Next, maybe instead of completely rebuilding each time something changes, maybe it should dynamically add or remove the respective menu items?
SVN revision: 5854
The enum of events is in observer.h, add any you can think of.
Switched desktops over to observer system (which extends objects).
Added ferite functionality for desktop observers. I will probably make functions to observe most of e's subsystems (views, iconbars, etc).
There is probably a small bit of memory leakage in here (and callbacks) because any objects passed along have to have their refcounts increased (so ferite doesn't automatically clean them up, and i haven't put in code to decrease them yet. It should be fairly small, but I'll work on that.
I also changed a few "sprintf(buf..."'s to "snprintf(buf, PATH_MAX..."'s.
Is this the way to go about preventing buffer overflows?
I think that's it.
SVN revision: 5841
I've changed, this is big, so read this carefully :)
* I've added debugging macros for messages and function call
tracing. Usage:
D("Creating item %i %i %i\n", x, y, z);
Define DEBUG to use the D macro.
D_ENTER;
D_RETURN;
D_RETURN_(x);
These are for call tracing. Use D_RETURN_(x) when returning
something from a function. Define DEBUG_NEST to use this.
* added iconbar header file to Makefile.am
* added proper new()/cleanup() calls for E_Delayed_Action;
* I've completely rewritten the object and observer handling. Bye
bye macros, this was nasty. It'll be hard enough to avoid leaks
with usecounting in C. We now basically have the same system as gtk.
There's a clear separation of observer and object code now.
An E_Object by itself has nothing to do with observing or being
observed, therefore, there are now E_Observers and E_Observees
that are derived from E_Object. IMPORTANT: The cleanup system now
reflects the reference count system, therefore, all ..._free()
calls are now static, because the destructor should never be called explicitly, but implicitly through e_object_unref(). The object handling
now is as follows:
- The cleanup functions clean up everything that is contained in
a struct, but NOT the struct itself. Instead of the final
free() call, they call the destructor of the base class. The
calls will walk up the hierarchy and clean up what's contained in
every struct, and the final e_object_cleanup() will free the
structure itself. E_Delayed_Action is a good example.
- The only calls that influence the reference count are
e_object_ref() and e_object_unref(). If you need to do things
before an object gets destroyed, you can query the use count using
e_object_get_usecount() and check if it's equal to 1. So this:
OBJ_UNREF(b);
OBJ_IF_FREE(b)
{
ecore_window_reparent(e->win, 0, 0, 0);
e_icccm_release(e->win);
OBJ_FREE(b);
}
now is this:
if (e_object_get_usecount(E_OBJECT(b)) == 1)
{
ecore_window_reparent(e->win, 0, 0, 0);
e_icccm_release(e->win);
}
e_object_unref(E_OBJECT(b));
object.h and observer.h are completely commented, it shouldn't be
too hard to understand. This'll need to be documented in the manual
anyway.
* E_Objects are now used in lots of places where void* were used as
pointers to objects before, especially in the actions code. This is
obviously better, as it will generate compiler warnings when people
want to pass things to functions that expect E_Objects. This could
probably be more restrictive.
* Added typedefs for the function prototypes in E_Action_Impl. Those
fat signatures were just painful to read in the function
declarations/implementations.
* I've also tried to give parameters more useful names. Calling an
object "o" is a lot of fun when you want to grep for it.
* Included is also Graham's latest menu.c patch. Sorry for the
delay, Graham.
* I've added checks to the menu code that make sure that menus
don't pop up when they're empty (which resulted in a little useless
rectangle).
I guess that's it for now. Sorry if I broke anything, but this was
necessary imho.
SVN revision: 5605
rephorm
cpk
Carsten Haitzler