A recent change is causing some apps to request their
SurfaceView size before requesting visibility. We won't
apply this because of the View.GONE check in WMS. Once the app
becomes visible, it thinks the surface size hasn't changed
locally, so it appends PRESERVE_GEOMETRY to let the render thread
and reposition child control the layout size. Now we will never
set a requested width/height though and end up with 0. The line
to not update requested size while preferring geometry was
required when we were using SCALING_MODE_NO_SCALE_CROP, as
making the size smaller would apply immediately (with
the surface size acting as a reduced crop until we got a new
buffer), but now that we are using SCALING_MODE_FREEZE there
are no problems.
Bug: 27914300
Change-Id: Ic0997ab3950f3eac360260e9939b6f2ee6667453
performReceive() is called with the activity manager lock held.
Don't call out to other system services (such as user state
callbacks) with the lock held. Post a runnable to the handler
instead.
Bug: 27589217
Change-Id: I5c2f8debe1ae365682b27af65e0bf02bcce50d4a
Note:
DevicePolicyManagerService is changed to inject ContentObserver notifier
Test: all test cases in DevicePolicyManagerTest pass
BUG: 25710621
Change-Id: I347cec71769d0e9dd6a334d7d6339d5ce6a3fa6a
Add content to Settings and SettingsLib to add a checkbox for PBAP
client profile, and remove the box for PBAP server profile when the
client profile is enabled.
Bug: 27642222
Change-Id: I125559904a8e017a01ae90b1f6425129a87cbee3
Fix error in audio focus dispatching log: the message value (here
MSSG_FOCUS_CHANGE, which is 0) was logged instead of the focus
value being sent to the audio focus listener.
Bug 25790720
Change-Id: I534ce8d1059c09a52e56c51acd60405ef3c4cda1
It turns out that BaseInputConnection has still depended on a private
API named BaseInputConnection#reportFinish(), which was introduced
4 years ago to work around a UI freeze due to an unbalanced batch edit
count [1]. Note that such an unbalanced batch edit count cannot always
be avoidable. It can easily occur in the following situations.
- The current IME crashed during batch edit.
- The user changed the View focus during batch edit.
- The current IME called IMM#switchToNextInputMethod() during batch
edit.
The remaining problem is that #reportFinish() is still an internal API
and only subclasses of BaseInputConnection can implement it, and IMM
calls it when and only when the current InputConnection is
BaseInputConnection or its subclass. InputConnectionWrapper and any
other InputConnection implementations will never receive such a callback
to clean up InputConnection#{begin, end}BatchEdit(), which is considered
to be a major contributor to UI freeze.
To address the above issue, we unhide BaseInputConnection#reportFinish()
as InputConnection#closeConnection() so that application developers can
receive an appropriate callback to clean up internal state including
unfinished batch edit.
[1] I5525d776916f0c42d5e6d4a4282aed590d7f0e9a
9d69ecbf61a4a142c3f4cbb9d5659faa6f85e832
Bug: 24688781
Bug: 25332806
Change-Id: I234309c5880c9fe0b299b8bd0f8862796d4dda0d
The VirtualRefBasePtr is used to track the usage of native objects from
the java side by reference counting. Implementing the class delegate to
make sure we dispose the objects correctly.
Change-Id: I3dd4717944b0dbe79f30e49b3083bf65c6e5276d
We still need to be able to compile all the Java source code in the tree
using javac, though the compiled code never ends up running on device.
Bug: 25958912
Change-Id: Id76663cdf768db579517515d7b7f51519b515fa8
Bug: 26912651
By setting debug.hwui.filter_test_overhead to true, hwui's
janktracker will attempt to filter out overhead caused
by the event injection that automated testing uses
Change-Id: I75c8dc5e7798e06e3009baf396108507c7240eec
