As discussed in Bug 33038203 on certain platforms there is a demand
that the power button can change the behavior depending on whether an
IME window is shown on the screen or not. The behavior requested here
can be summarized into two parts:
* Hide the IME window if it is shown [1]
* Go to the home screen if no IME window is shown
This CL implements the above request by introducing a new config mode
for config_shortPressOnPowerBehavior. Note the definition of when an
IME is shown is often tricky than people would expect. The way this
CL is implemented is to propagate IME window state from
InputMethodManagerService (IMMS) to PhoneWindowManager via
WindowManagerService regarding when the back button on the NavBar for
phones/tablets should be shown as an IME dismiss key [2].
[1]: Even with this CL the IME still is allowed to ignore the request
to hide the software keyboard. Currently there is no official
protocol to forcefully hide the software keyboard. How to deal
with such a situation is a long standing TODO task.
[2]: Internally this is controlled by the following IMMS fields:
- InputMethodManagerService#mImeWindowVis
- InputMethodManagerService#mBackDisposition
Note that those fields rely on self-report from the IME. To be
precise, the base implementation of InputMethodService is
responsible for report back its internal state to IMMS when
necessary. The important point is that, although this could
allow a malicious IME to confuse the system UI to some extent,
supporting malicious IMEs is not clearly a goal of Android.
Anyway, the definition of when an IME is shown is a kind of
hot topic in several system services recently. Hopefully we
can come up with better definition and reliable mechanism in
a future release.
Fixes: 33824860
Test: Manually verified as follows
1. Change config_shortPressOnPowerBehavior to "5"
2. Rebuilt the OS image and flash it to the device
3. Make sure that the power button works like a home button
if software keyboard is not shown.
4. Open dialer and focus in to the text field shown on top
5. Make sure that the AOSP keyboard is shown.
6. Run 'adb shell dumpsys input_method' to observe the
following line:
mImeWindowVis=Active|Visible
7. Tap the power button to make sure that the AOSP keyboard
gets dismissed.
8. Tap the power button again to make sure that it works
as if a home button.
Test: Manually tested as follows
1. Open dialer and focus in to the text field to show an IME
2. Run 'adb shell dumpsys window policy' to make sure
mDismissImeOnBackKeyPressed=true
3. Tap the back button to dismiss the IME
4. Run 'adb shell dumpsys window policy' to make sure
mDismissImeOnBackKeyPressed=false
Change-Id: I20721547c73360a70b5fc5cbe06824d577d1768a
The job queue now handles URI permissions associated with
the Intent of each job. Just (kind-of) like Service!
Also do the second pass of locking in job scheduler, getting
rid of all the async dispatching on a handler, and just executing
calls right in line with simple locking. This probably fixes
a few other race issues, and allows us to make sure that we
always finish a job correctly when dequeuing the last work (we
will always atomically dequeue and finish, so no new work can
slip in between).
And fix a little debug output in IntentFilter.
Test: ran CtsJobSchedulerTestCases, added new test for URI perms.
Change-Id: I52f700ef0cd5be3ff70050f9c0f5fe3e8a5ccac1
- As a part of transitioning into PiP, we remove the content insets, and
as a result, we also need to adjust the source hint rect for the change
in insets as well.
Bug: 37418994
Test: Enter PiP from YT, ensure no jump at the end
Change-Id: I74219b05c91d3c8b8466dd8fff4d6dd692f08725
- JobServiceEngine now takes a concrete Service instead of
generic Context in its constructor, since it really must be
associated with a real Service.
- Expand documentation of how dequeueWork() operates.
- Fix some job scheduler implementation to hopefully actually
match the docs: transfer remaining executing work to the new
job, and actually correctly transfer state from old and new
jobs if we are rescheduling due to a true return from onStopJob().
Test: bit CtsJobSchedulerTestCases:*
Change-Id: Ia66797049883eefb566264f930070afb69d469b1
- Fixing up edge case when SysUI resizing conflicts with updating the
display rotation bounds. When an interaction causes both a display
rotation and a resize from SystemUI, we should defer the resize animation
until the rotation has been propagated to SystemUI, otherwise the bounds
used will be incorrect.
Bug: 36879891
Test: android.server.cts.ActivityManagerPinnedStackTests
Change-Id: Ife1b7ab0c2f1f11f33cbc9614778ff49a28c79f6
The NotificationManager.startServiceInForeground() experiment is over,
and will not ship as API, so it's time to tidy up and get rid of it.
Bug 36130212
Test: manual
Change-Id: I834d1ce059aa464ff27f69f5e5d3625cc5e61d8a
Based on API council feedback, switch to using real UUID objects
instead of Strings. Since UUID is a general-purpose utility class
that will be passed around quite a bit, add it to Parcel and Bundle.
Define well-known namespaced UUID values for "default" and "primary
physical" storage devices, which will let us annotate a bunch of
things with @NonNull.
Define new extras for MANAGE_STORAGE intent that apps can use to
signal where and how much space they'd like the user to free up.
Test: cts-tradefed run commandAndExit cts-dev -m CtsAppSecurityHostTestCases -t android.appsecurity.cts.StorageHostTest
Bug: 37325923, 35812899, 35806020
Change-Id: I8421b126d680f69141a361c1e77223fe2bf4a325
An instance of ResourceKey may correspond to different instances
of Resources at different points in time. This can lead to old
instances of Display returned from cache even if resources have
changed.
We need to ensure 1:1 correspondence between Resources and
ResourceKey before using this kind of cache. Disabling it for now
to fix CTS.
Change-Id: Ib53550c4f2d969c06a570ab6051529269b04b38d
Fixes: 37328072
Fixes: 37305567
Bug: 33430498
Test: android.server.cts.ActivityManagerAppConfigurationTests
Test: android.app.cts.DisplayTest#testRotation
Adds a method isTheFinalCountDown that allows to correctly
determine whether it is the final countdown.
Test: None
Change-Id: I786ae3455479bac25ccf25efba1c3dce18185117
