Not all drivers that support the "VK_EXT_global_priority" extension
support "VK_QUEUE_GLOBAL_PRIORITY_REALTIME_EXT".  Instead of assuming
REALTIME by the existence of the extension, properly query for the
highest priority, using HIGH if available.

Bug: 281820575
Test: Manual testing
Change-Id: I846e829b775c7df90a44a0c97c0007cc6b44b438

It has been reported to us that our TEST_MAPPING file JSON is not
valid, which causes it to execute less tests on presubmit than
expected.

Bug: 288639518
Test: will check the presubmit logs to confirm
Change-Id: Ifa0ebea26db8f487d28821c1cbdf004e37f0d461

Previously the per-app opt in only works for ANGLE apk, this patch
refactors the logic to allow per-app opt in to work for system ANGLE.
This patch checks whether system ANGLE is supported and if it presents,
uses it if the ANGLE apk setup fails.

With this patch, the per-app opt in will function this way:

1) When full ANGLE apk is installed
Regardless of whether system ANGLE is present, the "angle" option will
use ANGLE apk, the "native" option will switch to native GLES drivers
specified by ro.hardware.egl, the "default" option will use whatever the
default graphics driver is loaded.

2) When full ANGLE apk is not installed and system ANGLE is present
The "angle" option will use system ANGLE, the "native" option will
switch to native GLES drivers specified by ro.hardware.egl, the
"default" option will use whatever the default graphics driver is
loaded.

3) When full ANGLE apk is not installed and system ANGLE doesn't exist
The per-app option will not function at all.

To Properly load ANGLE, the namespace of system ANGLE should be
inherited from sphal namespace so that search paths and libraries
linkage are properly constructed.

Minor: clean up unused and unnecessary variables.

Bug: b/283858001
Test: test with camera
Test: atest CtsAngleIntegrationHostTestCases
Change-Id: I2c9d4e1747796d2b0ae58c4456b781b9940fbb26

We are doing per-device pointer tracking now, so this CL helps remove
the assumptions around having a single active device.

For now, the behaviour is preserved. In the future, we will not need to
check the currently active device at the TouchState level.

Bug: 211379801
Test: m inputflinger_tests && $ANDROID_HOST_OUT/nativetest64/inputflinger_tests/inputflinger_tests
Change-Id: Ie13e9c93eefc04c2955de3b8a06c9c2da28c9d72

Previously, input dispatcher assumed that a single input device is
active. However, we are trying to add multi-device support.

In this CL, store the pointers inside touched windows on a per-device
basis. This CL should not cause any behaviour changes. We expect there
to be a single device active at a time still.

Bug: 211379801
Bug: 281806933
Test: m inputflinger_tests && $ANDROID_HOST_OUT/nativetest64/inputflinger_tests/inputflinger_tests
Change-Id: I73fc4adcf306bb49e153f3daeed68753e0632c86
Signed-off-by: Siarhei Vishniakou <svv@google.com>

Test: atest libinput_tests
Fixes: 266747937
Change-Id: Ic5ec548d2edc8bad5e8b88aaf8511cd297a89275

Merge "Fix the bug in image compression that failed to specify correct usage, type, flags" into udc-qpr-dev

Revert submission 23491348-enable-metrics-collector

Reason for revert: b/287676652

Reverted changes: /q/submissionid:23491348-enable-metrics-collector

Change-Id: I8b34f4d0afb9cbaed46ce9e45cbabf7a7f1b7770

This reverts commit 2db205d8.

Reason for revert: b/285086044, b/285401355

Change-Id: I0f642f0ead5a70daec649095855ec0209d2c0682

We don't need this, and likely no one needs this lib anymore.

Bug: 271455682
Test: m checkinput
Change-Id: I472ade3a212b42c082e7b877f7fd80ad1d6ed968

The previous implementation of the vulkan swapchain incorrectly implemented
vkGetPhysicalDeviceSurfaceFormats2KHR. As a workaround at least one
driver vendor resorted to a code workaround in their driver to correct
the bad call in flight.

The failure was during a call to vkGetDeviceImageFormatProperties2 where
the swapchain was failing to include the correct image usage and image
type, along with a failure to include the correct image compression
flags.

This CL amends the mistake to add these missing components to the vulkan
vkGetDeviceImageFormatProperties2 call.

Any driver vendors that have not yet finalized their image compression
implementations will benefit from not needing to add an android specific
workaround for this bug. The CL as written does not require the driver
workaround that was already written to be revised. The driver will work
correctly without source changes. When the driver is updated the
workaround can cleanly be removed without side-effects.

Bug: 288100893
Test: dEQP-VK.api.image_compression_control.swapchain.android
Change-Id: Id10e5f3e42ae4e16b2cf4874a0ff4e0e2f21cd2d

Merge "Force HALs to explicitly enable legacy method for clearing buffer caches" into udc-dev am: def2c211 am: 900a24cb

Original change: https://googleplex-android-review.googlesource.com/c/platform/frameworks/native/+/23715006



Change-Id: I3a1422bdbd090b0a3803bcc4cf2b1779d934bc72
Signed-off-by: Automerger Merge Worker <android-build-automerger-merge-worker@system.gserviceaccount.com>

Merge "Force HALs to explicitly enable legacy method for clearing buffer caches" into udc-dev am: def2c211

Original change: https://googleplex-android-review.googlesource.com/c/platform/frameworks/native/+/23715006



Change-Id: I0aa11e113daddcba91b7fb771391b4092b81f9fc
Signed-off-by: Automerger Merge Worker <android-build-automerger-merge-worker@system.gserviceaccount.com>

Modify the snapshot update logic so its easier to maintain. Instead of
walking the tree and updating all the snapshots, this cl introduces a
preliminary step of walking though all the requested changes and
updating the affected snapshots by merging the changes to the snapshot.
This is followed by walking down the tree and updating properties that
are dependent on the parent snapshots.

If the changes are confined to buffer updates, then the fast path avoids
walking down the tree.

Bug: 238781169
Test: presubmit

Change-Id: Ic9aa66c376bf7ea80e38b321dd08b8d0f69559a9

Prior to this CL, the dispatcher death tests used to run in the "fast"
style. That made their execution faster, but also resulted in occasional
hangups of the testing method.

The test would fail because it would get stuck forever.

The failure was reproducible every ~100 runs on device. Failure could
not be reproduced on host, though.

In this CL, we switch to the "threadsafe" approach for running this
test. It makes the test slower, but in practice it's still fast enough,
taking less than 100ms per execution.

At the same time, I could not reproduce a failure in a similar death
test, the UntrustedSpy_AbortsDispatcher one. That test ran successfully
10000 times on the device, with about 15 ms per execution.

So we are only adding this parameter to the untrusted overlay one for
now.

Bug: 282837934
Test: m inputflinger_tests && adb sync data && adb shell -t data/nativetest64/inputflinger_tests/inputflinger_tests --gtest_filter="*UntrustedOverlay_AbortsDispatcher*" --gtest_repeat=1000
Change-Id: I747e17a3e191d910c02ad71241d3fad9a9599158

Bug: 282110579, 238781169
Fixes: 277265947
Test: presubmit

Change-Id: I2066f606337005eb049ad721ff80f34c9989b8c1

Original change: https://googleplex-android-review.googlesource.com/c/platform/frameworks/native/+/23755377



Change-Id: Id0302d602aaa816dbc9f84a5c3a6e3dc67534569
Signed-off-by: Automerger Merge Worker <android-build-automerger-merge-worker@system.gserviceaccount.com>

Bug: 286428012
Test: presubmit
Change-Id: I6a3572ff10b259feae74863c0203bb5f085f233f
Merged-In: I6a3572ff10b259feae74863c0203bb5f085f233f

This is needed to further understand the dispatching decisions.

In one of the bugs, the dispatcher sent an event to two windows and
ended up with two foreground windows at the same time. That shouldn't
occur.

This logging in needed to figure out how the dispatcher got there.

Bug: 287860440
Test: none
Change-Id: I1021ab19ee450361974a61d39c45fd5f134b100b

As in HWUI's CanvasContext::draw, check whether the context has been
abandoned at the beginning of a frame. Previously, RE never checked for
this, though it may have been responsible for other spurious failures,
such as "Unable to generate SkSurface".

When RE fails to generate an SkSurface/SkImage, query the backend to
determine which method to call to get more information about the
texture. This allows us to get meaningful information when running
SkiaVk.

Bug: 286058063
Test: manual - build and boot with gl and vk
Test: manual - change fatal logs to ALOGD always
Change-Id: I532f9073698a920619ffd665d05d8e63163366d8

Before this CL, the policy state was asserted too early. The dispatcher
might not have finished processing the commands yet.

To fix this, force the tests to wait until dispatcher is idle by
providing a dispatcher reference to the assertion function.

Bug: 282837934
Test: atest inputflinger_tests
Change-Id: Ie427071f774d2bfb4b9ebfef94c7971656a69509

Before this CL, NotifyMotionArgs stored PointerCoords and
PointerProperties in a fixed-size array. Upon creation of a new object,
some of that data typically remained uninitialized.

At the same time, the copy assignment operator was defaulted, which
meant that the uninitialized data was getting accessed in order to copy
the object.

The sanitizers identify this as a problem and crash.

To fix this, store these objects inside vectors.

Bug: 271455682
Test: atest inputflinger_tests
Change-Id: I9dba29f75df59a21f8ed7fd0f46fd1f6d45f2eef