Merge commit '1f4421048206ca1745ddd6366588759c9c5c38f9'
* commit '1f4421048206ca1745ddd6366588759c9c5c38f9':
Improve orientation sensing when waking up while flat.
Use the layout xml attribute splitMotionEvents="true" or the ViewGroup
method setMotionEventSplittingEnabled(true) to enable motion event
splitting. Rules for splitting are as follows:
* Splitting is enabled per ViewGroup. When splitting is enabled any
MotionEvent dispatched to that ViewGroup can potentially be split
into several and dispatched to children independently.
* Each pointer is assigned a target child view when the ACTION_DOWN or
ACTION_POINTER_DOWN event is received. That will be the pointer's
target until it goes up, the target returns false from onTouchEvent,
or the MotionEvents are intercepted.
* Multiple pointers may be assigned to the same target. All pointer
data sent to a target are bundled into a single MotionEvent. Child
views do not need to be aware that splitting has occurred.
Change-Id: I993f838e2f6b455da9812f4742a016dfcd1c4cc9
This change makes the window manager pass the currently displayed
orientation to the WindowOrientationListener each time to calls to get
the currently sensed orientation. Now, when the
WindowOrientationListener doesn't know for sure what the orientation
of the device should be (for example, if the sensed orientation is
flat on a table), then it trusts the already-displayed orientation,
rather than trusting whatever sensors reading it most recently
received.
The main use case this addresses is:
* user uses device in portrait
* user turns off device, puts it in pocket
* while in pocket, screen gets turned on, activating orientaiton
sensor, which senses landscape orientation
* screen timed out
* user takes device out, places it on a table (or looks down at it,
holding it flat), and unlocks it
Previously, when the device was unlocked, the
WindowOrientationListener would return landscape orientation based on
its previous sensor readings. Now, it will override those previous
readings because it knows they never actually took effect and the
device is presently flat.
This change also slightly modifies the way we handle flat orientation,
now completely ignoring data when the tilt distrust is nonzero (even
if the current sensor reading shows non-tilted).
Change-Id: Ia4950a41827f8c53a80487d0c2e83b556df781b9
This enforces the new policy if menu items are created
programmatically instead of inflated from menu xml.
This reverts commit fbb72fdbe5.
Change-Id: I102e93b80a3248634e98e0d7dc3321bf1d8b263b
The MenuInflater returned by ActionModes will set the default
showAsAction value to "ifRoom" if not otherwise specified.
Change-Id: I86045dc4e877a97a3464a1a5f44d8d1e2120b086
Merge commit 'f66f8a58cc4412bd258e1d5b11cac17a617bdc58'
* commit 'f66f8a58cc4412bd258e1d5b11cac17a617bdc58':
Add support for the PointerLocation overlay.
This change involves adding a new method to IWindowManager,
monitorInput() that returns an InputChannel to receive a copy of all
input that is dispatched to applications. The caller must have
the READ_INPUT_STATE permission to make this request (similar to
other window manager methods such as getKeycodeState).
Change-Id: Icd14d810174a5b2928671ef16de73af88302aea0
Merge commit '8279c754d176d5c2aa23f6f7ad16fcea03278788'
* commit '8279c754d176d5c2aa23f6f7ad16fcea03278788':
Making sure that the prefix is handled properly.
Merge commit 'd3cf0544cc7eec8ace3dc3b7fc17b7dac69e0aab'
* commit 'd3cf0544cc7eec8ace3dc3b7fc17b7dac69e0aab':
Optimize VelocityTracker to run in linear time.
Merge commit '75493be9c0b4dd24fc5cddce6ee993ce5f0b2887'
* commit '75493be9c0b4dd24fc5cddce6ee993ce5f0b2887':
Make sure profiling is done only for views that are actually measured, laid out and drawn.
Uses a linked list for efficient pointer addition and removal.
When possible, makes use of the fact that pointer ids are usually in
sorted order to avoid quadratic time lookups when adding new data.
Fixed an incorrect assumption that the pointer count would always change
when old pointers were removed.
Also fixed a bug in InputQueue FinishedCallback recycling.
Change-Id: Ie048d3bb022d39cf4185e2fe43923a861d94c4f3
This change also removes the use of SoftReferences for View's
drawing cache.
A bitmap now creates a PhantomReference enqueued in a reference
queue provided by the new Finalizers class. This queue is polled
from a thread started after forking zygote. That thread is in charge
of clearing the references after GC runs and of calling reclaim()
on them. The reclaim() method is now how finalizers are run.
Note that a PhantomReference cannot be kept in the instance it
refers to, which is why they are kept in a separate List.
Change-Id: If3c1a5e9dc23fa49e34857860d730f5cf5ad5926
The makefile variable USE_OPENGL_RENDERER must be set to true to compile
libhwui and the related code in the JNI layer.
This change also removes obsolete APIs from Canvas that must not be used
and would be confusing if left in. These APIs were remnants of our first
attempt at an OpenGL renderer for the view hierarchy and had not been
taken out before Android 1.0 was released.
Change-Id: I2475ff1307212bab26c926724f3c508681c7dae1
Since orientation sensing has been an issue for numerous users, I
decided a spend a little time experimenting with some possible
improvements. I've settled on a couple major changes:
* Perform all lowpass filtering in spherical coordinates, not
cartesian. Since the rotations are what we're really concerned
with, this makes more sense and gives more consistent results.
* Introduce a system of tracking "distrust" in the current data, based
on external acceleration and on tilt. The basic idea is after a
signal of unreliable data -- repeated acceleration or
nearly-horizontal tilt -- we wait for things to "stabilize" for some
number of ticks before we start trusting the data again. This is an
extension of the basic lowpass filtering. One simple example is
after the phone is picked up off a table, we ignore the first few
readings. Another example is while the phone is under external
acceleration for a while (i.e. in a car mount on a rough road), if a
single "good" reading comes in, we distrust it, under the assumption
that it was probably just a lucky reading (i.e. the magnitude
happened to be close to that of gravity by chance).
These changes have allowed me to relax other constraints, such as the
filtering time constants, the maximum deviation from gravity, and the
max tilt before we start distrusting data.
The net effect is that orientation changes happen more quickly and can
happen under a wider variety of conditions, but false changes due to
tilt and acceleration are still avoided well. I think the improvement
is subtle, but it's the best I've come up with in my limited time.
I've also included some refactoring and additonal comments to try and
further clarify the (somewhat twisted) logic.
Change-Id: I34c7297bd2061fae8317ffefd32a85c7538a3efb
Rendering is implementing by rasterizing the paths into A8 textures.
This cna be extremely inefficient if the path changes often.
Change-Id: I609343f304ae38e0d319359403ee73b9b5b3c93a
