They're deprecated, but still needed by most users, as the replacement
is not available until API 29. Restore the docs and explain the bugs.

Bug: None
Test: treehugger
Change-Id: I4227ce3f0ee57a73a13e394cd48373accf4a58a3

Add the soon-to-be-required removal message for ALooper_pollAll, and
expand the docs for both ALooper_pollAll and ALooper_pollOnce to make
the reason for removal clear, and to explain the fix callers should
apply.

Bug: https://github.com/android/ndk/issues/2014
Test: treehugger
Change-Id: I020a280346d1ff9f5fcbec4c153dee8319cdf539

Bug: None
Test: treehugger
Change-Id: I3a52883fba0e31f05e50cd51193bbb306418b18a

By declaring the ABI of `enum` types, the `enum` type can now be used in
place of a raw and untyped integer in function parameters and structure
fields, self-documenting what the possible values are.

Test: treehugger
Change-Id: I6b9cf8b36b16e23c0e992f8155452bd194e6e1ad
Signed-off-by: Marijn Suijten <marijns95@gmail.com>

Bug: https://github.com/android/ndk/issues/1920
Test: treehugger
Change-Id: Ib8296e7b9c86809d16fbde0787d016095a189387

Because of the existence of extra spaces, the owner of keycodes.h is
not configured as expected. We should delete this extra spaces.

Bug: none
Test: After merging, check whether the owner of keycodes.h includes
all users under INPUT_OWNER

Change-Id: I634c49f53a4633bdee3cdefbf0debae4dff90a47

This is far from the only C incompatibility in this file, but it's the
one that's most likely to not break builds, so I'm starting small.

This is a source compat break, but not an ABI break.

Bug: https://github.com/android/ndk/issues/1920
Test: treehugger
Change-Id: I5a32762fa61b1399fb354397479603858184ea05

The looper.h header introduced __REMOVED_IN, which requires the
sys/cdefs.h header to provide definitions for the macro. Add them here
and update files that include looper.h to not provide their own
definitions of __REMOVED_IN. Further, update comments to clarify that
the additional defines are in place to satisfy builds that use glibc or
Windows/MacOS libcs.

Bug: 14257980
Change-Id: Ia524b7e180c5137e84c205a77dc219b9e9e41608

We don't yet have a way to determine if a device can actually report a
usage code. Adding a flag to mark usage code as fallback only mapping.

Bug: 297094448
Bug: 317838112
Test: atest inputflinger_tests
Change-Id: I137113d0dc9a1c8abf523b96942486be176d0240
Merged-In: I137113d0dc9a1c8abf523b96942486be176d0240

ftl::Function<F, N> is a container for function object, and can mostly
be used in place of std::function<F>.

Unlike std::function<F>, a ftl::Function<F, N>:

  * Uses a static amount of memory (controlled by N), and never any
    dynamic allocation.
  * Satisfies the std::is_trivially_copyable<> trait.
  * Satisfies the std::is_trivially_destructible<> trait.

However to satisfy those constraints, the contained function object must
also satisfy those constraints, meaning certain types (like
std::unique_ptr's) cannot be part of the contained function object type.

The size of a ftl::Function<F, N> is guaranteed to be:

    sizeof(std::intptr_t) * (N + 2)

If not specified, N defaults to zero, which is big enough to store a lambda
that captures a single pointer (such as "this" for forwarding to a
member function.

By comparison, sizeof(std::function) == sizeof(std::intptr_t) * 4, at
least with on x86-64 with clang 15.

Compile time checks are performed that the constraints are all satisfied,
and that the value of N is large enough to contain the desired function
object type.

ftl::make_function is a helper function to construct a ftl::Function,
and will deduce the template type arguments. In addition to constructing
a ftl::Function for a function object, ftl::make_function has overloads
for creating a ftl::Function which will invoke a member function or a free
(non-member) function.

ftl::no_op is a helper value to construct a ftl::Function<F, N> that
does nothing, except default construct a return value, if one is needed.

A unit test is also included to demonstrate and verify the
implementation, including asserting that function objects which don't
meet the requirements cannot be used. The test also asserts some
non-obvious corner cases for handling argument and return value
conversions to match how std::function behaves.

Bug: 279581095
Test: atest ftl_test
Change-Id: I268facb106a248d0766e931595291036bc606fb7

This makes it more convenient to dump windows from other places during
debugging. For example, this could also be called from setInputWindows
to inspect all of the window flags that might be changing.

Bug: 312714754
Test: adb shell dumpsys input
Change-Id: I2ace23bec1231f3a318e1b6a19f146b588e077b6

DisplayViewport.h:133:29: warning: format specifies type 'int' but the argument has type 'ui::Rotation' [-Wformat]

Test: presubmit
Change-Id: Ie656a3da47f2155b106014dff967d5b2d05998d9

The MetricsManager needs to receive UP/CANCEL events to trigger
atom reporting. I must have moved these lines around during the
refactor and overlooked this mistake.

This change also modifies MotionPredictor and MetricsManager to
hold a "ReportAtomFunction" to facilitate testing.

Test: `statsd_testdrive 718` shows atoms reported with `adb shell setenforce 0`.
Test: `atest frameworks/native/libs/input/tests/MotionPredictor_test.cpp -c` passes.
Test: `atest frameworks/native/libs/input/tests/MotionPredictorMetricsManager_test.cpp -c` passes.

Bug: 311066949

Change-Id: Icbb709bbb7cf548512e0d9aa062783d554b857e3

Previously we introduced the reportActualWorkDuration API without
specifying the work duration for each components, this patch introduces
a separate API that allows clients to send work duration with each
component to allow fine grained scheduling strategy.

Bug: b/284324521
Test: atest PerformanceHintNativeTest
Test: atest PerformanceHintManagerTest
Change-Id: Ie68657dece0d7fef5843df798472adb9108f534f

Also, reuse dupFd method instead of calling ::dup directly.

Test: m libinput
Test: m libinputdispatcher
Bug: 302723053
Change-Id: Ie6e4f1ea9cfc00b4c2be44dcc8922a17de129cfb

Revert submission 24997733-adpf-gpu-plumbing

Reason for revert: b/311102554

Reverted changes: /q/submissionid:24997733-adpf-gpu-plumbing

Change-Id: Ifd3973cc97b7be525c5e936ab6764b727de1804d

Test: mma
Bug: 302723053
Change-Id: I52f14cadb027b3f854946d5315dce3d23aa21b19

Previously we introduced the reportActualWorkDuration API without
specifying the work duration for each components, this patch introduces
a separate API that allows clients to send work duration with each
component to allow fine grained scheduling strategy.

Bug: b/284324521
Test: atest PerformanceHintNativeTest
Test: atest PerformanceHintManagerTest
Change-Id: Ia2e66ae173255acee3f05fa99177659604976aa1

Bug: b/288119641
Test: atest NativeThermalTest
Change-Id: I2ba820da74b290a25ff1edf8d278b9200dfaf950

Initial version of a fuzzer for InputDispatcher.

The goal is to reproduce crashes that are triggered by the fatal logs,
mainly around:
1) mismatching downtime / eventTimes
2) unexpected hover events

Currently, the fuzzer runs without hitting those targets.
However, it hits an ODR due to rect so currently it has to run without
checking for ODRs.

It also currently hits an out of memory issue after a short run, finding
a problem in
AStatsManager_setPullAtomCallback packages/modules/StatsD/lib/libstatspull/stats_pull_atom_callback.cpp:397:46

Bug: 281806933
Test: FUZZER=inputflinger_input_dispatcher_fuzzer; m $FUZZER && ASAN_OPTIONS=detect_odr_violation=0 $ANDROID_HOST_OUT/fuzz/x86_64/$FUZZER/$FUZZER
Test: atest inputflinger_benchmarks
Change-Id: I465ea11520fc9cc21886646c0ecf20dc529b2698

Bug: 210460522
Bug: 309154937
Test: atest inputflinger_tests
Change-Id: If07a2d1dd0043e4959ed20b086fa5afef26bd751

To fully support palm rejection, we should not produce any velocity for
pointers that have been marked as palm.

This can be achieved by erasing the corresponding pointer state in
VelocityTracker for ACTION_POINTER_UP with FLAG_CANCELED events, or for
all pointers in ACTION_CANCEL events.

The pointers that have not been canceled will still produce the correct
velocity.

Bug: 236772648
Test: TEST=libinput_tests; m $TEST && $ANDROID_HOST_OUT/nativetest64/$TEST/$TEST
Change-Id: I67ecae6b4bff2eb63f9cff416841031dea3ed680

When KeyCharacterMap is read from parcel, there's no need to force the
caller to store the object in a shared pointer. We can return a
unique_ptr first, and let the caller decide on how exactly that lifetime
should be managed.

Bug: 274058082
Test: presubmit
Change-Id: I8c5ec1e32a9304f6ad186bc0279f4c7bcbab77d8

This function is now getting moved to Input.h.

We are also making its behaviour deterministic: if at least one pointer
is stylus, then the event is a stylus event (unless the source wasn't
set to "stylus").

This way, an event with 1 finger and 1 stylus pointers will have the
same consistent behaviour, regardless of the order that the pointers are
stored inside the event.

Bug: 211379801
Test: TEST=inputflinger_tests; m $TEST && $ANDROID_HOST_OUT/nativetest64/$TEST/$TEST
Change-Id: Ie164e32b7c0e9cf21b3819b01a03ac2885666191

Bug: https://github.com/android/ndk/issues/1933
Test: treehugger
Change-Id: Ifbb4d08722aa953f3e86d770e8d6c9017ab9d680

We don't yet have a way to determine if a device can actually report a
usage code. Adding a flag to mark usage code as fallback only mapping.

Bug: 297094448
Test: atest inputflinger_tests
Change-Id: I137113d0dc9a1c8abf523b96942486be176d0240

Check for the source inside InputVerifier. Sources like MOUSE_RELATIVE
could send ACTION_MOVE events without a prior DOWN event. Verifying such
streams is tricky, so let's simply skip such events for now.

Also in this CL, add some verifications to the number of pointers inside
the event.

Bug: 211379801
Test: enable event verification and run native tests
Test: atest inputflinger_tests libinput_tests
Change-Id: I3703ba57af7ede77712b91b7429ac46c0624a616

Change-Id: I3a0c6b46197debe5985d386b56bb1178e24d85f9

Bug: 300701739
Test: presubmit
Change-Id: I235275b5d299efea779851434c392ca4e07ca4fd

Current wrong formatting:
https://developer.android.com/ndk/reference/group/bitmap#androidbitmap_compress

Bug: https://github.com/android/ndk/issues/1933
Test: treehugger
Change-Id: I7174c728fbe589e1d80be82a5f0df56e054c9609

These words are meant to be in the code _font_, not in their own
separate code _blocks_. See
https://developer.android.com/ndk/reference/group/thermal#athermal_getthermalheadroom
for the current wrong formatting.

Bug: https://github.com/android/ndk/issues/1933
Test: treehugger
Change-Id: I2a30439bd58dd1435c4efb525a90e06631fbe6cf

Also fix an 'a'/'an' typo.

Bug: https://github.com/android/ndk/issues/1933
Test: treehugger
Change-Id: Ifd505772dce79ebd68d928c67775030e2fd98d3f

Adding a timer provider allows the Gestures library to perform some
asynchronous tasks, including ones which improve tap-to-click detection.

Bug: 297192727
Test: set the
      persist.device_config.aconfig_flags.input.com.android.input.flags.enable_gestures_library_timer_provider
      sysprop to true, restart, then make fast tap-to-click gestures
      (where the finger is contacting for fewer than 3 frames) on the
      touchpad, and check they're reported immediately
Change-Id: Ib9b8dacc71c88b6fd47bdd747f90ef6a44b37cc4

In the change with ID Ibee1e7dc5021296bdb5871dec59d8d4978fcf0c9,
we introduced lambdas at the tracing sites so that we could
conditionally execute the code to generate the tracing string only when
tracing was enabled. However, it seems like the introduction of the
lambdas, especially ones that capture everything by reference, have
added a performance burden, causing ~5% increase in latency in the input
pipeline.

In this CL, we remove the lambdas from the tracing sites, and instead
rely on the expression evaluation guarantees of the ternary to ensure
that the expression to generate the trace message will only be evaluated
when the condition is true.

Bug: 297462790
Test: Will evaluate perf metric after submitting
Change-Id: I69b37551c4b23256c64544e0ddf4b5b39a9403fb

VelocityTracker may use different strategies, depending on the system
flags. By default, the behaviour is like this:
AXIS_X -> lsq2
AXIS_Y -> lsq2
AXIS_SCROLL -> impulse

However, if we provide a specific strategy to VT, then things would
change. The new map would look like this:
AXIS_X -> provided strategy
AXIS_Y -> provided strategy
AXIS_SCROLL -> provided strategy

This works fine if the user specifies "impulse" as the desired strategy,
because impulse supports all of the axes.
However, lsq2 only works in non-differential mode.

The combination of "AXIS_SCROLL -> lsq2" is not allowed.

To fix this, we only allow the specified strategy to affect
non-differential axes.

This is fine, because currently, impulse is the only strategy that can
work in the differential mode.

To reproduce the issue:
1. Run `atest VelocityTrackerTest`. Test should pass, because the
   default strategy is being used.
2. Connect the device to internet
3. Reboot the device
4. Run `atest VelocityTrackerTest` again. Test should fail because the
   device would pick up an "lsq2" strategy value and then try to run the
   axis_scroll tests with it.

Bug: 295290374
Test: atest VelocityTrackerTest
Change-Id: I702a2a3e58db3ce2e0ff0c33122839a527eebab2
Merged-In: I702a2a3e58db3ce2e0ff0c33122839a527eebab2
(cherry picked from commit 8a2e589e)

Previously, we traced using ATRACE_NAME inside an
if condition after checking if atrace was enabled. This was so that we
could avoid the addtional overhead of formatting the trace string when
there's no ongoing trace.

Since ATRACE_NAME is expected to be scoped trace of the call, doing it
inside an if condition means the scope of the trace ends immediately
once we leave the inner scope of the conditional. This essentially makes
all of our scoped traces point traces.

We get around this by adding a new helper macro ATRACE_NAME_IF that
takes a lambda that formats the message. This way, we can avoid the
formatting overhead and ensure the entire scope of the method is traced.

Bug: 210460522
Test: manual: take a trace with Perfetto
Change-Id: Ibee1e7dc5021296bdb5871dec59d8d4978fcf0c9

Fills out the implementation and tests for MotionPredictorMetricsManager.

(main cherry pick: ag/24451072)

Test: atest frameworks/native/libs/input/tests/MotionPredictorMetricsManager_test.cpp
Test: Manual testing on-device, computed metric values seem reasonable.

Bug: 268245099

Merged-In: Iec18415de9c3070f2b285c5c82f5a5e0ceaaf471
Change-Id: I0def0dae626adbbe33e1ed90c08b2cae867cde01

Fills out the implementation and tests for MotionPredictorMetricsManager.

(Cherry pick of ag/23861881 from udc-qpr-dev)

Test: atest frameworks/native/libs/input/tests/MotionPredictorMetricsManager_test.cpp
Test: Manual testing on-device, computed metric values seem reasonable.

Bug: 268245099
Change-Id: Iec18415de9c3070f2b285c5c82f5a5e0ceaaf471

To ensure Java is in sync with native, move these consts to
IInputConstants.aidl.

Test: none
Bug: 295290374
Change-Id: I616f4e969e5b7ec72263db62f225f85786f5e923

VelocityTracker may use different strategies, depending on the system
flags. By default, the behaviour is like this:
AXIS_X -> lsq2
AXIS_Y -> lsq2
AXIS_SCROLL -> impulse

However, if we provide a specific strategy to VT, then things would
change. The new map would look like this:
AXIS_X -> provided strategy
AXIS_Y -> provided strategy
AXIS_SCROLL -> provided strategy

This works fine if the user specifies "impulse" as the desired strategy,
because impulse supports all of the axes.
However, lsq2 only works in non-differential mode.

The combination of "AXIS_SCROLL -> lsq2" is not allowed.

To fix this, we only allow the specified strategy to affect
non-differential axes.

This is fine, because currently, impulse is the only strategy that can
work in the differential mode.

To reproduce the issue:
1. Run `atest VelocityTrackerTest`. Test should pass, because the
   default strategy is being used.
2. Connect the device to internet
3. Reboot the device
4. Run `atest VelocityTrackerTest` again. Test should fail because the
   device would pick up an "lsq2" strategy value and then try to run the
   axis_scroll tests with it.

Bug: 295290374
Test: atest VelocityTrackerTest
Change-Id: I702a2a3e58db3ce2e0ff0c33122839a527eebab2