We would like to make relayoutWindow a non-blocking call,
and of course why not, who likes blocking. However it functions
as a critical path of BLASTSync. To understand why examine the
Guarantee described in BLASTSync.md.
In order to implement this guarantee we need to know
“Which frame has the client finished drawing” when it calls
finishDrawing. The current answer to this question is
“The frame reflecting the state observed in the last call
to relayoutWindow”. The comments on mPending and mCurrentDrawHandlers
also have a lot more context on how this works currently. Since
relayoutWindow has a critical section, it also ensures that changes
to syncable state and preparation of sync will be observed atomically
(since they are both observed over relayoutWindow, which is always
called before any frame drawing updated syncable state).
We design a new protocol described in BLASTSync.md, which uses a seqId
to track which call to finishDrawing reflects which syncable state.
We implement this seqId based system. Unfortunately the WindowManager
doesn’t quite conform to the requirements specified by the protocol,
and a follow up CL ensures that syncable state changes will
always be sent with the seqId.
This CL simply adds the protocol specification and makes some required
interface changes. It can be seen to be a no-op.
Bug: 161810301
Bug: 175861051
Bug: 175861127
Bug: 200285149
Change-Id: If2dea07121fe7de5d2524c5b63678dddf955d4b7
The frame number at the server side is not used anymore since BLAST.
Bug: 161810301
Test: presubmit
Change-Id: I01c6fb0750799aa94766f2582ce6ac031c9426cf
This is a step to move the layout logic to the client side. We won't
obtain the surface size from the server then.
This CL also moves the logic about adjusting the LayoutParams of
wallpaper to the client side.
Bug: 161810301
Test: presubmit
Change-Id: I3a81e174035c67a285cab449c0701ee2fe6f6a24
Since both of splitting tasks in split screen go to overview together
now, update to snapshot all pending animations when rotating while
showing overview to ensure both of splitting task snapshots are taken
with proper bounds before rotation.
Bug: 200813008
Test: atest RecentsAnimationControllerTest
Test: enter overview after activated split screen, observed task
thumbnails showing with correct bounds after roation.
Change-Id: Id1101bf8be67593f85a43c1d4afae769ffc737d3
This reverts commit 9ba705609f019eafe1ca16fe5c02d50006416f9b.
Reason for revert: The metrics become unstable
Bug: 129527548
Change-Id: If1b9368eb55b124a8d66f06c82de193e4ddb9318
So if there are messages posted to other threads in relayout, they
won't affect the next round to execute relayout. And the continuous
invocation isn't common in the real case.
Bug: 129527548
Test: atest RelayoutPerfTest
Change-Id: I0081fc7a80271ee3dd97c19a6ae7e3bc13ae8a8b
InsetsState contains much more information than visibilities, such as
display frame, display cutout, rounded corners, privacy indicator
bounds, and frames of of insets sources. The control target only needs
to send the requested visibilities to WMS, so it can be too heavy to use
InsetsState.
This CL introduces an new class, InsetsVisibilities, which only contains
which type has which visibility. So it uses less memory, and it is more
efficient on copying and checking the equality.
Fix: 194186241
Test: atest InsetsVisibilitiesTest WindowAddRemovePerfTest
InsetsControllerTest RegisterStatusBarResultTest
CommandQueueTest LightsOutNotifControllerTest
ActivityRecordTests DisplayContentTests
DisplayPolicyLayoutTests InsetsPolicyTest
InsetsSourceProviderTest InsetsStateControllerTest
WindowFrameTests WindowManagerServiceTests WindowStateTests
Change-Id: I86c1b26b4383bfa3b924726d580e5706e13ba735
Otherwise android.perftests.utils.Stats will reject the data
and make the test fail.
Also correct a typo.
Bug: 182962701
Test: atest InternalWindowOperationPerfTest
Change-Id: I1d097f0de9a632b02d204509729f1671327285e4
This CL moves computeWindowBounds to the client side. The client can use
it to compute the frame hint on its own.
This can be a step to make client compute its window frame locally.
Bug: 161810301
BUg: 175858823
Test: atest WindowAddRemovePerfTest ActivityRecordTests
DisplayPolicyLayoutTests TaskSnapshotSurfaceTest
Change-Id: Ia5af1919b8e0e973646a63d1a4c3bf7ea7e2d1f6
Since the display cutout is a type of insets and the display cutout can
be obtained from WindowInsets, it makes sense that InsetsState has the
display cutout instance. In this way, we can send the display cutout to
client via W#insetsChanged instead of W#resized.
This can be a step to remove the class of ClientWindowFrames, and can
also be a step to make client compute its window frame locally.
Fix: 175858810
Bug: 161810301
Test: atest WindowAddRemovePerfTest ImeInsetsSourceConsumerTest
InsetsControllerTest InsetsStateTest ViewRootImplTest
WindowInsetsControllerTests ActivityRecordTests
DisplayPolicyLayoutTests LaunchParamsControllerTests
TaskSnapshotSurfaceTest WindowMetricsActivityTests
WindowMetricsWindowContextTests WindowMetricsTest
WindowFrameTests WindowStateTests WmDisplayCutoutTest
Change-Id: I9a930b1d2f7df3cea2b29629b767a4a5f31bca17
Bug: 174932174
Test: I solemnly swear I tested this conflict resolution.
Exempt-From-Owner-Approval: refactoring with team leads buy-in
Change-Id: I9262a08ffc1ccede8e519d0eed90ed2bfcf0232c
As general background, OWNERS files expedite code reviews by helping
code authors quickly find relevant reviewers, and they also ensure
that stakeholders are involved in code changes in their areas.
Some teams under frameworks/base/ have been using OWNERS files
successfully for many years, and we're ready to expand them to cover
more areas. Here's the historical coverage statistics for the last
two years of changes before these new OWNERS changes land:
-- 56% of changes are fully covered by OWNERS
-- 17% of changes are partially covered by OWNERS
-- 25% of changes have no OWNERS coverage
Working closely with team leads, we've now identified clear OWNERS on
a per-package basis, and we're using "include" directives whenever
possible to to simplify future maintenance. With this extensive
effort, we've now improved our coverage as follows:
-- 98% of changes are fully covered by OWNERS
-- 1% of changes are partially covered by OWNERS
-- 1% of changes have no OWNERS coverage
This specific change is automatically generated by a script that
identifies relevant "include" directives.
Bug: 174932174
Test: manual
Exempt-From-Owner-Approval: refactoring with team leads buy-in
Merged-In: I3480ddf2fe7ba3dfb922b459d4da01fa17a2c813
Change-Id: I3480ddf2fe7ba3dfb922b459d4da01fa17a2c813
Add WindowPerfTestBase and WindowPerfRunPreconditionBase
into apct-perftests-utils. So window manager and input
method manager can share the same functions.
Bug: 174292015
Test: WmPerfTests ImePerfTests
Change-Id: Ie2818536d6611d1ba5f4b6cd725cd2d4a95e1cac
The server won't dispatch the legacy content insets, stable insets, and
visible insets to the client. The insets would be computed from the
insets state by the client.
This CL also fixes the insets scaling issue in compatible mode.
Bug: 149813814
Fix: 169940916
Fix: 165412978
Test: atest WindowAddRemovePerfTest InsetsAnimationControlImplTest
ActivityRecordTests DisplayPolicyLayoutTests
InsetsPolicyTest InsetsStateControllerTest
Change-Id: I83570973f587a6abf887752494f750b46e3e484d
The new logic sends the requested visibility to server even when the
client doesn't have a control. This makes the requested visibility
reliable at the server side.
This CL also changes getRequestedState to getRequestedVisibility in
WindowState, because the frames in getRequestedState are not reliable.
Fix: 161247175
Fix: 166819574
Test: atest DisplayPolicyLayoutTests DisplayPolicyTests InsetsPolicyTest
InsetsSourceProviderTest InsetsStateControllerTest
WindowStateTests InsetsAnimationControlImplTest
InsetsControllerTest WindowAddRemovePerfTest
ActivityRecordTests
Change-Id: I7d32dacaa1302bab61386f5bd69aaa7dde953ebe
If Blast was enabled, WindowStateAnimator would create a container layer
and then a blast layer as a child. This caused some issues with
reparentChildren and relative Z, so instead of trying to continously fix
those issues, just use one layer. If blast is enabled, it will create a
single SurfaceControl with type blast. Otherwise, it will create a
BufferQueueLayer.
This means that we have to be careful with setting any properties on
the layer since the blast adapter could also be modifying the layer.
This is in the right direction since we will move all transactions for
the layer to the client, eventually.
Test: YT and Chrome with and without Blast
Change-Id: I43efc37663696ba8e54d341a03531f0048ee6a14
This CL is aimed to refine the AM/ATM API surface. The other
stack naming internally spread in AM and ATM will be covered
in later CLs.
Also Consolidate #removeStack() with #removeTask().
Bug: 157876448
Test: existing tests
Change-Id: I1254f972dab3bbe66a736a0402eebce9d0fb1496
Accpet the event time from RecentsAnimation#startRecentsActivity.
Also remove unused method of recents animation in activity manager.
Bug: 169221287
Test: Enable statsd log: "adb shell cmd stats print-logs"
Touch gesture navigation bar.
adb logcat | grep statsd | grep "(48)"
The line may contain 0x100000->4[I] 0x110000->20[I]
that means 4=by recents and 20=latency 20ms.
Change-Id: Ia188a6a3b0dab9edaa0052df3b18508f8cc895d3
This CL removes
- mSeq
- System UI flags used to communicate between WMS and System UI
- redundant AIDL methods
- redundant fields and methods
- redundant tests
- PolicyControl
This CL also
- refines the format in DisplayPolicy#dump
- sends a boolean to InputManager to indicate if System UI is in a low
profile mode instead of sending the legacy system UI visibility
Bug: 149813814
Fix: 169105126
Test: presubmit
Test: dumpsys window displays
Test: See if the layout of ImmersiveModeConfirmation is as expected
Change-Id: I8c8df509355bebc9b560af57d5458614557bcd2f
The frame won't be changed if there is no IWindow#resized or
IWindowSession#relayout. So it can be retrieved from these methods
directly instead of another binder transaction.
And because some parameters are usually used together for layout,
the parameters are consolidated into a new ClientWindowFrames.
That reduces changing the interface in the future if the frame
related information needs to be changed.
Also refine the resize handling in ViewRootImpl to make it easier
to read.
There should be no behavior change by this modification.
Bug: 161781274
Test: WmTests, DialogFrameTests
Change-Id: I9f711ad2023442046fa8582944320b98e7c4ecfa
- Add a readme file to describe how to lock CPU/GPU frequencies
on local test.
- Ensure gesture navigation is enabled when testing.
- Full compile the test package.
- Remove all activities before testing.
Bug: 161782101
Test: atest WmPerfTests
Change-Id: Iebcfcbc07e3fe7d3b1839b8c2e04041d7338c48e
The plugged-in type may be BATTERY_PLUGGED_AC in test environment.
And the precondition of the device may have set stay-on with any
type, then it is enough to keep device interactive. Previously
the usb type doesn't match the current state that causes device to
sleep if the screen off timeout has occurred.
Bug: 131727899
Test: atest WmPerfTests
Change-Id: Ia76a0f55e967a2d52fc4b1057b19a8af921260f3
Since commit c4e90ac, exported is required if the declared
activity has an intent filter.
Fixes: 158589759
Test: atest WmPerfTests
Change-Id: Iea4a901d684e821636d27c2bb564842a328f1d9b
