The GLib main loop allows destroying sources while still in the dispatcher
function, so whether the Clutter threading lock is being held or not is
actually unpredictable by the library. This might be fixed in GLib, so we
just do what GTK+ does for the time being and wait.
See bug: http://bugzilla.gnome.org/show_bug.cgi?id=459555
When we are in the the event translation function sometimes we need to
synthesise events: the double and triple click events are synthetic events
placed on the queue after a sequence of events has been received, for
instance.
Until now, the events were placed on the queue after the translation from
the native events was successful. This led to a loss of ordering because
we put the synthesised event on the queue before the last event that
triggered it.
This patch puts the events on the queue before translating them, with a
"pending" flag set; if the translation sequence is completed then the flag
is removed - otherwise the event is removed from the queue altogether. The
queue manipulation functions have been modified to ignore the "pending"
flag when looking for events.
This patch also adds a private structure overlayed on the ClutterEvent
struct so that we can extend the events with private data without exposing
it in the public API.
* clutter/Makefile.am:
* clutter/clutter-enum-types.h.in:
* clutter/clutter-enum-types.c.in: Use template files for
glib-mkenums; this makes the Makefile template a bit more
clean and the enum types generation more customisable (e.g.
when GLib 2.14 has been released, we can use g_once_init_enter()
and g_once_init_leave() to have thread-safe GType functions
for the enum types as well).
Emit the ClutterTimeline::new-frame signal only if we arrive there with an
active timeout source. Also, make sure to take a reference on the timeout
when emitting signals.
If the timeline is advancing of zero frames, bump it up to one frame. (bug
471, Rob Bradford)
If the user called clutter_timeline_pause() while in the ::new-frame handler
the timeout is removed and the timeline is reset, while the correct behaviour
should be to simply return. (#371, Johan Bilien)
If we skipped the last frames and we ended up over the frame number, emit
the ::new-frame signal with the last frame before emitting the ::completed
signal.
Remove the call to clutter_timeline_stop() when reaching the end of the
timeline: it was used just to remove the timeout source and it emitted the
::paused signal, which doesn't make any sense. Also rewind the timeline
after emitting the ::completed signal, so that calling get_current_frame()
in the ::completed signal handlers still work.
The ownership of the ClutterAlpha object is transferred to the ClutterBehaviour
used by the effect; we just need to unref the behaviour to make the alpha
object disappear.
The clutter_timeout_pool_insert_sorted() function caused an inversion of
the timeout sources in the pool; this led to a wrong behaviour in the
execution of the timeout functions. See bug 471.
This patch drops clutter_timeout_pool_insert_sorted() in favour of the
standard g_list_insert_sorted(), which produces identical behaviours with
and without the pool.
A new test, written by Rob Bradford, has been added to the regression test
suite in order to identify sorting issues with the timeout pools.
Apply a patch from Neil Roberts (bug 468) that makes the get_center() method
of ClutterBehaviourRotate return the correct tuple of coordinates.
Also, fix some coding style issues and make sure to freeze and thaw the
GObject notification queue when emitting the notify signal on multiple
properties in one method.
Some of the API landed in trunk has been backported to the stable branch
after testing. This patch updates the Since: tag in the API reference to
reflect the time when the API was first added to a release.
When computing the increasing smoothstep alpha value use the right type
for the results, to avoid overflowing the [0, 0xffff] range of the alpha
value.
Also use the right format when printing out the opacity and alpha values
in the opacity behaviour.
Language bindings will have problems with raise() and lower() bein in both
ClutterContainer and ClutterActor; hence, this patch renames
clutter_container_raise() and clutter_container_lower() to
clutter_container_raise_child() and clutter_container_lower_child(),
respectively.
* clutter/clutter-actor.c:
* clutter/clutter-actor.h:
* clutter/clutter-event.c:
* clutter/clutter-event.h:
* clutter/clutter-main.c:
* clutter/clutter-main.h:
* clutter/clutter-private.h:
* clutter/clutter-stage.c:
* clutter/clutter-stage.h:
* clutter/clutter-types.h:
Initial implementation of actors emmitting event signals (423);
- Actors set_reactive() to receive mouse events.
(call clutter_enable_motion_events() for per action motion events)
- clutter_stage_set_key_focus () to direct key events.
- Events bubble up to parents (ending at stage)
(original source identified by clutter_event_get_source())
TODO:
- enter/leave notifys for actors.
- stage specific events - fullscreen
- grabs
* tests/test-events.c:
Extend a little to use new API
* clutter/cogl/gl/cogl.c:
* clutter/glx/clutter-backend-glx.c:
Move get_proc_address into cogl and out of backend.
(shaders will need it)
* clutter/clutter-group.c: (clutter_group_real_lower):
Fix a minor compile warning.
* TODO:
Sync up.
Every direct reference to ClutterGroup has been removed, in favour of
ClutterContainer - which is the official way of creating container
actors as of 0.4.
Since ClutterActor directly calls ClutterGroup methods we need ClutterContainer
to provide them, so that every container actor behaves the same way.
This patch makes ::raise, ::lower and ::sort_depth_order virtual functions
of the ClutterContainer interface, implemented by ClutterGroup.
The documentation has been added and deprecation warnings have been added
as well.
ClutterStage now uses the newly added clutter_container_find_child_by_id()
function to find the right child in the scenegraph, using the picking id.
This fixes the test-boxes test case.
Finding a child by its ID is a method that should be used by every container
and not only by ClutterGroup. This is needed to actually fix the picking in
ClutterStage.
* clutter/clutter-timeout-pool.c: Fix removing and adding timeouts
to the timeout pool during a dispatch of a timeout source already
inside the pool. (#456, based on a patch by Neil Roberts)
(clutter_timeout_dispatch), (clutter_timeout_pool_dispatch): Hold
the main Clutter lock in the pool dispatch function, instead of
the per-timeout dispatch; this guarantees that the ref+unref of
the single timeouts are done under the main lock.
Ported from 0.4 branch.
* clutter/clutter-texture.c:
Fix typo in clutter_texture_get_pixbuf (#458, #442?, Neil Roberts).
Use take_object when getting pixbuf prop to avoid ref leak
The threads_enter() and threads_leave() functions are no-ops because Clutter
is not thread-aware nor thread-safe. Leaving them in is a source of confusion
so we just remove them.
Update the NEWS and the README files with the release notes.
