The last frame of an animation stays stuck on the screen for a couple of frames.
Specifically, the "Quick Contact" animation that animates the picture
closed (fades/scales it away) animates all the way to the end... then hangs there
briefly before being taken down.
The problem is a rendering bug where we correctly detect that a DisplayList
has nothing to draw (since the last frame is completely transparent, alpha==0),
but incorrectly ignore the fact that we cleared the transparent-background
window prior to not-drawing that DisplayList. When we detect that there's
nothing to draw, we don't bother swapping buffers. So even though we drew
the right thing (clearing the buffer), we didn't actually post the buffer to the
screen.
This change factors in both the clear and the draw to decide when to swap buffers.
Issue #8564865 Quick contact close animation jank redux
Change-Id: Ib922cff88a94f025b62f7461c1a29e96fe454838
The new implementation more accurately tracks the velocity
of flings and takes care to avoid obvious discontinuities.
The main goal is for a fling to appear to be a linear
extension of the movement already in progress. The minimum
fling velocity is set to ensure that flings appear to be
fairly smooth despite being discretized.
Use sequences of repeated key events instead of individual
down/up events to represent continuous motions in one
direction which can be helpful for stopping flings at boundaries
such as when flinging the cursor position within a text view.
Compute the movement thresholds based on the physical
size of the touch pad, if known. If not known, we assume a
nominal size.
Support stopping flings with a tap just like we do for
normal touch events elsewhere in the framework.
Moved the detection of ASSIST swipes into the InputReader
where it belongs. These swipes must be detected globally
to ensure consistent behavior across the all applications.
Added a custom protocol in EventHub to enable input device
drivers to override the timestamp of the following events
in a packet. This change enables input device drivers
that have a better idea of when an input event was actually
generated to pass this information to the input system.
Particularly useful with uinput.
Bug: 8583760
Change-Id: I8ef4e827804786d549cfaa00793a2b9dd0fda465
The goal is to better encapsulate this code to make it easier
to maintain and to facilitate some upcoming changes.
Some of the variables have been renamed but the logic is unchanged.
Bug: 8583760
Change-Id: I45501f7dabebcb938e42c386291d615d088a4c8c
The trackball to dpad synthesis was using the MotionEvent's precision
field to determine a threshold for movement but the calculations
involved did not actually make sense for any value of precision
less than 2.0. This worked fine before since the InputReader
hardcodes the trackball's precision to 6.
Injected trackball events may have a different precision which can
result in the thresholds being set inappropriately. For example,
it was not possible to move focus by one unit at a time when
the precision was set to 1.0.
The old code was probably using precision as a way to set a
threshold based on the trackball moving by some minimum number
of physical ticks, in this case 2. But the code will work just
as well if we set an absolute threshold based on distance
traveled given that the input system is already expected to
normalize the trackball movements before delivering them to the
application.
So stop using precision.
Bug: 8473020
Change-Id: I3c6f7fb1b507f8cf5608b47550e7345fea3352fa
Redesigned how ViewRootImpl delivers input events to views,
the IME and to native activities to fix several issues.
The prior change to make IME input event delegation use
InputChannels failed to take into account that InputMethodManager
is a singleton attached to the main looper whereas UI may be
attached to any looper. Consequently interactions with the
InputChannel might occur on the wrong thread. Fixed this
problem by checking the current thread and posting input
events or callbacks to the correct looper when necessary.
NativeActivity has also been broken for a while because the
default event handling logic for joysticks and touch navigation
was unable to dispatch events back into the native activity.
In particular, this meant that DPad synthesis from touch navigation
would not work in any native activity. The plan is to fix
this problem by passing all events through ViewRootImpl as usual
then forwarding them to native activity as needed. This should
greatly simplify IME pre-dispatch and system key handling
and make everything more robust overall.
Fixed issues related to when input events are synthesized.
In particular, added a more robust mechanism to ensure that
synthetic events are canceled appropriately when we discover
that events are no longer being resynthesized (because the
application or IME is handling or dropping them).
The new design is structured as a pipeline with a chain of
responsibility consisting of InputStage objects. Each InputStage
is responsible for some part of handling each input event
such as dispatching to the view hierarchy or to the IME.
As a stage processes an input event, it has the option of
finishing the event, forwarding the event to the next stage
or handling the event asynchronously. Some queueing logic
takes care to ensure that events are forwarded downstream in
the correct order even if they are handled out of order
by a given stage.
Cleaned up the InputMethodManager singleton initialization logic
to make it clearer that it must be attached to the main looper.
We don't actually need to pass this looper around.
Deleted the LatencyTimer class since no one uses it and we have
better ways of measuring latency these days using systrace.
Added a hidden helper to Looper to determine whether the current
thread is the indicated Looper thread.
Note: NativeActivity's IME dispatch is broken by this patch.
This will be fixed later in another patch.
Bug: 8473020
Change-Id: Iac2a1277545195a7a0137bbbdf04514c29165c60
SurfaceView and TextureView do not currently support overlays
correctly; the docs now reflect this constraint.
Change-Id: I79183c02b51ae4cd14638198d0668b2c2e3e22e1
Adding views to views (possible with the new Overlay API) is weird.
This change moves the view-management facilities of Overlay to a subclass
that is specific to the overlay returned from ViewGroup.getOverlay().
So now you can add drawables to all view overlays, but only add/remove
views to/from the overlay returned from ViewGroup.getOverlay().
Also, the previous approach of using an interface for Overlay was
changed to classes for both ViewOverlay and ViewGroupOverlay.
Finally, this change makes not handling touch correctly the proper,
and documented, behavior of overlay views. There are various tricky issues
to sort out with input in overlays (including click handling as well as focus)
and we don't want developers starting to use overlays as some kind of general
container hierarchy, so we're purposely constraining overlays to have visual-only
behavior.
Issue #8459085 Overlay needs to handle touch correctly
Change-Id: I207b8dbf528f87c92369d270d8b0a6556826d207
InputChannels are normally duplicated when sent to a remote process
over Binder but this does not happen if the recipient is running within
the system server process. This causes problems for KeyGuard because the
InputMethodManagerService may accidentally dispose the channel
that KeyGuard is using.
Fixed the lifecycle of InputChannels that are managed by the IME
framework. We now return a duplicate of the channel to the application
and then take care to dispose of the duplicate when necessary.
In particular, InputBindResult disposes its InputChannel automatically
when returned through Binder (using PARCELABLE_WRITE_RETURN_VALUE).
Bug: 8493879
Change-Id: I08ec3d13268c76f3b56706b4523508bcefa3be79
Remove remnants from times long gone. We don't need to redraw the
union of the previous frame's dirty region and the new dirty
region.
Change-Id: I9fb96f99a6a72c2233f9ca563cf6432a42b2b65b
RectEvaluator is useful when animating object bounds.
The other change is a hidden API that allows temporary suspension
of layout, useful for animations which need to animate view bounds
without conflicting with layout passes that might happen in the middle
of the animation.
Change-Id: I3dc08cb6ec455dfa3409e825506b218d3ea63d7a
This fixed an issue where an InputEventSender might outlive
its usefulness and continue to be used well after it should
have been disposed or recreated.
Also improves the queue management somewhat.
Bug: 8493879
Change-Id: I7e0b6a3c43cbe72f8762991f5d36560feebd214b
If the receiver or sender was not properly disposed, then
the underlying input channel might be leaked because the
native peer was holding a strong reference to the object.
Switched to using a weak reference.
Also updated Binder to use a new helper created for this purpose.
Change-Id: I19680bf96d0548777bff02aa1d91874d1e8e41da
Relaunching a running activity will sometimes cause animations in that
activity to run, causing jank due to too much happening (window animations
on that same window running at the same time).
An earlier fix pauses application rendering while window animations are running,
but only the first time the activity comes up. If the window is animated
after that (such as is the case with re-launching it while it is running
in the background), rendering, and therefore animations, happen as usual,
causing the jank.
The fix is to simply broaden the scope of when application rendering is
paused to include anytime the window is animating.
Issue #8472972 Make app animations play nice with window animations
Change-Id: I1dd8da039fcb4d1faf6c0811bae97ef2847deb84
bug:8501661
Makes the performance issues and interaction with layer type/paint
more clear.
Additionally, corrects change from 47ab7d6612
to still allow displayList alpha to override layer paint
Change-Id: Ic94d75865700820489370461cd8ac9f9077a8d90
When a tablet rotates, FUL must be stopped and restarted in a new
position. 90 degree rotations cause a configuration change, causing
FUL to be automatically reconstructed in the new location. However,
a 180 degree rotation is not a configuration change, so FUL was not
restarting. A 180 degree rotation happens more often than one might
think. If you set the tablet down and later picked it up in the
opposite orientation, FUL would not work prior to this fix.
This change adds a rotation watcher to KeyguardFaceUnlockView. It
watches for 180 degree rotations and stops and restarts FUL
accordingly.
The rotation watcher callback must be unregistered when
KeyguardFaceUnlockView is recreated (as during 90 degree rotation
changes), otherwise the number of rotation watcher callbacks will keep
growing and they will never go away. This is a problem not just
because there are many callbacks hanging around, but also because the
old callbacks end up trying to access biometric unlock views that no
longer exist, resulting in crashes. So, a simple function was added
to the window manager to unregister a rotation watcher.
Change-Id: Ie1ef20a9a22b8f4e39918987dff2b8ad444fcfd1
When the pipelining optimization was added we never updated
handleImeFinishedEvent to look through the whole queue of items, so it was only
looking at the head.
Bug: 8498214
Change-Id: I79c62392a93b47e3e1eab3f4fe54a5c999dfb566
- use private flag bit field for ViewGroup.MarginLayoutParams
- make ViewDebug support "byte" type for FlagMapping
Change-Id: Iff7c31544e3ce2d29919c9424425f2bf87042b8b
- follow up to the fix for bug #8480245 ViewGroup layout margins can be wrong in RTL mode
- deal with "RTL compatibility mode": if left/right margins are not defined and if we
haev some start/end ones then use the start/end ones.
Change-Id: I98fe3276de2bd14f60a1c423a47569a68046f7be
The input method manager service now supplies an input channel for
communication while creating an IME session on behalf of the
application.
This change significanly reduces the overhead of IME event dispatch
by using a standard input channel to send input events rather than
using binder. This results in fewer thread context switches
and fewer object allocations.
What's more, the IME may perform additional batching of the motion
events that it receives which may help it catch up if it is
getting behind while processing them.
Bug: 7984576
Bug: 8473020
Change-Id: Ibe26311edd0060cdcae80194f1753482e635786f
Bug #8378964
This change defers drawing into layers until after the renderer for FBO0
is ready to draw. At that point, all the precaching is done which means
all glyphs can be uploaded at once in the font caches.
Change-Id: Ie1f7a7ff30f76f06fb3dbc72c7d05e66207d1ecb
