Bug 35928527
During optimized transactions, a fragment may be removed without
being created. That leaves the state of the fragment in INITIALIZING
and previously, that state wasn't ever saved. This CL allows a
fragment that is being removed to be brought up to the CREATED state
so that it can be saved during saveAllState().
Test: manual and Ie7207cc647312d38b377405bc5ec8721db757d2e
Change-Id: I649f1931745be43087ec3578e9195624e80821dc
This CL cleans up APIs around font variation settings.
- Remove FontConfig and FontManager public API.
- Remove FontManagerService from system service.
- Extract inner class FontConfig.Axis as top-level class FontVariationAxis.
This is used by Typeface.Builder public API to create new Typeface.
- Introduce and expose FontVariationAxis utility functions from/to string.
- Throws if the invalid font variation settings is passed.
Test: android.text.cts.FontVariationAxisTest passes
Test: android.graphics.cts.TypefaceTest passes
Test: android.graphics.cts.PaintTest passes
Change-Id: I9ccafe7a53935960566243e2856e166878ca59ae
If the snapshot starting window has different dimensions than the
snapshots we have taken, we do the following:
- Create a child Surface that has exactly the dimensions of the
snapshot.
- We fill the parent surface with the app background color, as well
as all screen background decorations (status bar background,
navigation bar background).
- We also clip of the status bar/navigation bar background in some
cases, as it looks ugly if it's not behind the system bars.
- Furthermore, we inherit all layout flags on the window and all
layout relevant SystemUI flags on the window such that it's very
likely that the size will match, and the system bars are drawn
correctly.
- In order to make the transition from the snapshot to the real
window a bit more predictable/less messy, we enforce a minimum
duration the snapshot is visible, which is slightly more than our
app transitions.
Test: TaskSnapshotSurfaceTest
Test: Open app, go home, go landscape, open app again
Test: Go to multi-window, open app from recents with a snapshot
taken in fullscreen.
Fixes: 36703868
Change-Id: Ia2d4add6971a18ab7aa2942d2b644d6e87a402af
(Finally) introduce a new ServiceConnection callback to
tell you when the binding has died. This allows you to robustly
have a weak service monitoring, and also is an easy way to find
out about breakages due to app updates etc.
Also clean up some debug output.
Test: moved to own suite and ran them.
Change-Id: I526cc00816c384fa9eb1312b92406f38085cbff9
- This CL has two main changes:
1) It modifies the activity multi-window and picture-in-picture mode
changed callbacks to provide the configuration of the activity with
the mode applied.
2) It modifies the order in which the multi-window and picture-in-picture
mode callbacks are made, to ensure that when going in and out of
picture-in-picture: first PiP, then MW, and then the config change.
- Previously, the ordering of the two callbacks was inconsistent. When
calling moveActivityToPinnedStack(), we reparent the task into the pinned
stack (triggering the picture-in-picture mode change), followed by the
resize animation (causes configuration changes). Inversely, when we
expand the task to fullscreen (and not just remove it), we run the
animation first, which resizes the task to the final size (causes
configuration changes) then reparent after the animation completes
(triggering the picture-in-picture mode change).
In this CL, we ensure that for both the transition in and out of PiP, we
defer to the bounds animation to trigger the PiP mode change. Normal
calls to reparent or adding a new task are unchanged. When the PiP
mode change is called from the animation, it provides the final target
bounds which we use to calculate the target configuration of the activity
for the callback. If the bounds animation is interrupted, an update will
also be scheduled if we change the fullscreen state we are animating to.
To work around the issue where we are scheduling MW/PiP mode changes in
both the animation and the configuration change, we also now keep track
of each state internally in the ActivityRecord.
Bug: 36099777
Test: android.server.cts.ActivityManagerPinnedStackTests
Test: #testConfigurationChangeOrderDuringTransition
Change-Id: I03513bc3a4d4a72c250983f22f079ce9d7a2cb40
Signed-off-by: Winson Chung <winsonc@google.com>
Yum!
Also needed to have a Context.revokeUriPermission() variant that is sane,
so reasonable CTS tests can be written.
Test: new ClipDataJobTest added.
Change-Id: Ia3135ea788a6e32c971bae7dab3a844d0ef4139c
When in VR mode, launch all activities into the virtual display ID as
provided by the Compatibility display. This includes two cases -
- New activity launches
- Existing activity in the background.
Testing Done: Tested with PlanarVirtualDisplay app and Settings,
Calculator and GestureApp with different intent flags.
Bug: 36071574
Bug: 36071445
Test: android.server.cts.ActivityManagerDisplayTests
Test: #testVrActivityLaunch
Test: #testVrActivityReLaunch
Change-Id: Ic590a7cbd6f9b339dc83b22a8ffb1252219ef22e
Signed-off-by: Karthik Ravi Shankar <karthikrs@google.com>
The ViewStructure typically represents a View, but it it can also be a virtual
view; in particular, WebView uses virtual views to represent HTML elements.
Although most of the properties of the HTML element maps to properties of
Android Views, some properties (such as 'name' and 'id' on <INPUT> fields)
don't, and those are crucial for autofilling web pages.
Rather than trying to artificially map these properties, it's better to create
a generic representation, for the following reasons:
1. Web standards move in a different velocity than Android APIs
2. Android APIs cannot be changed easily. Deprecated APIs continue to work,
and new added APIs don't work in older versions
3. The data used for autofill is opaque to the Framework - it's only relevant
to the node producers (like WebView) and consumers (Autofill services).
Also removed the setIdEntry() that was used for the same purpose.
Fixes: 36696757
Bug: 36718508
Test: VirtualContainerActivityTest with new checks pass
Change-Id: Ia626bd1f640b0b5861e81a5915504b95029874c9
Bug 36702993
Instead of throwing an exception during commitAllowStateLoss()
or commitNowAllowingStateLoss(), silently drop the transaction
when the FragmentManager is detached.
Test: I7093457c2b4936b95ab086dbe947571f0b525f63
Change-Id: I77641b69f2573fc895a27ec9da63c2d2ab6d7a4e
Rather than require an a-priori Notification be supplied in order to
start a service directly into the foreground state, we adopt a two-stage
compound operation for undertaking ongoing service work even from a
background execution state. Context#startForegroundService() is not
subject to background restrictions, with the requirement that the
service formally enter the foreground state via startForeground() within
5 seconds. If the service does not do so, it is stopped by the OS and
the app is blamed with a service ANR.
We also introduce a new flavor of PendingIntent that starts a service
into this two-stage "promises to call startForeground()" sequence, so
that deferred and second-party launches can take advantage of it.
Bug 36130212
Test: CTS
Change-Id: I96d6b23fcfc27d8fa606827b7d48a093611b2345
(cherry picked from commit 79047c62b5)
