For better type safety, use enum class when sending tool type.

Bug: 198472780
Test: atest libinput_tests inputflinger_tests
(cherry picked from https://googleplex-android-review.googlesource.com/q/commit:09a8fe42ba1d218c2f445e4fd5bc387e260ae067)
Merged-In: I371f08087b9513b6f75966c124de77bc12f8324e
Change-Id: I371f08087b9513b6f75966c124de77bc12f8324e

Bug: 272103317
Test: n/a doc only change
Change-Id: I8bebb31f0bd38e36139026a539030e5b61b75b4c

Using optionals rather than returning a boolean forces the caller to
check whether the property is present, enforced by the compiler, and
also allows a more succinct syntax in some cases.

Bug: 245989146
Test: atest inputflinger_test
Test: connect Apple Magic Trackpad 2, check that gesture properties are
      set correctly in dumpsys input
Change-Id: Ia0fde1f67269e6e7149f297def626e572fd7790f

a53cb97a

Change-Id: I5463f938665212362b8780e1c646a4b2bf8ad10a

When tests inject input events at a location on the screen, the
dispatched event is expected to have the same coordinates. However, on
scaled devices, this may not always be the case due to precision losses
incurred through floating point arithmetics. For example, it was
possible for an injected event with a coordinate of 1.0 to end up with
a value of 0.9997.

To combat this issue, we will round transformed axis values that are
leaving a MotionEvent to a precision of 0.001. After this CL, even if
the injection process results in precision losses, they should be
overcome by rounding, assuming injection does not require greater
precision.

This will solve the issue where input injected to an inclusive edge of
the View bounds was not getting dispatched to the View due to precision
losses.

Bug: 264978231
Bug: 260965930
Test: atest libinput_tests
Test: atest inputflinger_tests
Test: atest HoverTest (with screen size override)
Change-Id: I81062597058361a1218e6873d34b9b0d2fbfad96
Merged-In: I81062597058361a1218e6873d34b9b0d2fbfad96

When tests inject input events at a location on the screen, the
dispatched event is expected to have the same coordinates. However, on
scaled devices, this may not always be the case due to precision losses
incurred through floating point arithmetics. For example, it was
possible for an injected event with a coordinate of 1.0 to end up with
a value of 0.9997.

To combat this issue, we will round transformed axis values that are
leaving a MotionEvent to a precision of 0.001. After this CL, even if
the injection process results in precision losses, they should be
overcome by rounding, assuming injection does not require greater
precision.

This will solve the issue where input injected to an inclusive edge of
the View bounds was not getting dispatched to the View due to precision
losses.

Bug: 264978231
Bug: 260965930
Test: atest libinput_tests
Test: atest inputflinger_tests
Test: atest HoverTest (with screen size override)
Change-Id: I81062597058361a1218e6873d34b9b0d2fbfad96

Revert submission 21750284-touchpad-specific-tuning

Reason for revert: resolves flake b/272308130 according to culprit assistant https://android-build.googleplex.com/builds/culprit-assistant/run/6257e5da-1bb0-437a-8fab-f04c27d462b3 verified by 6 runs in abtd https://android-build.googleplex.com/builds/abtd/run/L75900000959044402

Reverted changes: /q/submissionid:21750284-touchpad-specific-tuning

Change-Id: I62078104a96d588bd321459e5a069dd0de5933bd

Bug: 185536303
Test: ftl_test
Change-Id: Idf0bac4711111c8cb28e8624bb146a76ec1ff38b

Specific touchpads often require small tweaks to the Gestures library's
gesture properties to work well. Examples of such tweaks include palm
thresholds, or the parameters to the stationary wiggle filter. To
support this, allow properties to be set using the 'gestureProp.' prefix
in IDC files, with the spaces in property names replaced by underscores.
For example, to set the "Pressure Calibration Offset" property to 30,
add this line to the relevant IDC file:

    gestureProp.Pressure_Calibration_Offset = 30

Only single-valued boolean, integer, or real properties can be set this
way. There's currently no use case for strings. The only use case for
arrays is for acceleration curves, which wouldn't work well with this
system anyway since they'd be overridden by the pointer speed settings.

Bug: 271251605
Test: atest inputflinger_tests
Test: connect Apple Magic Trackpad 2, check properties are set correctly
      in dumpsys input
Change-Id: I1bf97c9753e48c00d3dc3098cb676f7baebc84ce

Before this CL, the testing binary could not execute properly when
invoked directly. This is because the current working directory does not
get set.

To fix this, use the model file from /system/etc when __ANDROID__ is
defined, and refer to the current directory for all other cases.

Test: m libinput_tests && $ANDROID_HOST_OUT/nativetest64/libinput_tests/libinput_tests
Test: atest libinput_tests
Test: atest --host libinput_tests
Bug: 271134652
Change-Id: I0547ca9666eab153208a581fb60f6bef656b946a

After further discussion about the potential usage of the
MotionPredictor API, we decided that returning a List<MotionEvent> is
inconvenient for the application.

Update the native MotionPredictor to return a nullable MotionEvent
instead.

Bug: 167946763
Test: native
Change-Id: Ib1382431d894b9561d9a9e2e96c360e2b7a19b5c

Bug: None
Test: Upload
Change-Id: I1500315dce96953763a67a942204b9c4f028e9ba

This variable is left uninitialized. On some platforms, this causes the
libinput_tests to fail when sanitizers are enabled.

Bug: 251249010
Test: m libinput_tests && adb sync data && adb shell -t /data/nativetest64/libinput_tests/libinput_tests --gtest_filter="*keyCharacterMapParcelingTest*"
Change-Id: I05063efdde9a826f7001dca828b84bbec5beb0df

Devs should be able to determine if they can use an api just from
reading the api docs.

Bug: 270405876
Test: none
Change-Id: I62df5447bc547be52f6b00fedf1006c8d66a5faf

When logging evdev events, use the human-readable labels for recognized
values for the event type, the event code, and event value.

Bug: 193231132
Test: Manual, by enabling raw events logging:
    setprop log.tag.InputReaderRawEvents DEBUG
Change-Id: I8ed700a1e2d8659031514d2d8f9aef420b129dbc

Add an event verifier to check the touch streams that are being sent to
the input channel.

If a stream is inconsistent, crash.

For now, only touch streams are being verified. Disable the feature
while the bugs are being fixed.

Bug: 267082966
Test: m inputflinger_tests && $ANDROID_HOST_OUT/nativetest64/inputflinger_tests/inputflinger_tests
Merged-In: Ibb8ab70b0019789c1a17dae0077be6b23d3d9139
Change-Id: Ibb8ab70b0019789c1a17dae0077be6b23d3d9139
(cherry picked from commit f06b672b)

Add VirtualInputDevice base class and a set of subclasses for each
device type. Each virtual input device wraps a fd representing the
uinput device and a set of write...event() methods.

Most logic are moved from InputController JNI code: see ag/21294055 for
reference.

Test: atest VirtualInputTest VirtualMouseTest VirtualKeyboardTest
VirtualTouchscreenTest VirtualDpadTest

Bug: 267515782

Change-Id: Ie3a580acc890ac5af7461f012e05eb9ed3709a5f
Merged-In: Ie3a580acc890ac5af7461f012e05eb9ed3709a5f

Bug: 241001465
Test: make / silkfx demo
Change-Id: I043474aeda46a65ac93c124d854d71e1a8c082c9

Bug: 267050081
Test: atest libinput_tests
Change-Id: I09666da123a58786e8a6d47d4c29a475e92f2bbf

The buffers in the model file are used directly by TFLite, and so a
small memory saving can be achieved by backing those memory pages with
the file itself.

Bug: 267050081
Test: atest libinput_tests
Change-Id: I743a3c94477d4bb778b6e0c4b4890a44f4e19aa4

This allows us to only include the ops required to run the model and
have the linker strip the rest out, reducing memory overhead.

Bug: 267050081
Test: atest libinput_tests
Change-Id: I4055a0c8971ed4308ccfa425ab5e5ba560deb58c

Revert submission 21161765-sessionhint_api

Reason for revert: this broke the main-finalization-1 build... somehow. see b/268111957

Reverted changes: /q/submissionid:21161765-sessionhint_api

Change-Id: I2a8b6609729bba06fc9409f4583886915c82577c

Bug: 167946763
Test: atest libinput_tests
Change-Id: I7e50d38ed0c593aebc5fdc6af4b25868505d48bc

Let PointerController (MouseCursorController) know the source of
the current event, either mouse or stylus.
MouseCursorController will show the proper default pointer for
the active source, if the requested pointer type is TYPE_NOT_SPECIFIED.

Test: Manual Test
Bug: b/215436642
Change-Id: I91f702db661846fc7ad857f71656903c8aa4334a

The API to get a native attached choreographer from
ASurfaceControl.

Bug: 255838011
Test: atest GraphicsTest
Test: atest ChoreographerNativeTest (test no regression)
Change-Id: I5db99d8888fd2adbc4a8fc2ce9a7c07f4de146ee

Change NDK to use "SessionHint" enum consistently across definitions.

Bug: 266596626
Test: manual
Change-Id: I6ad095655b2dc05daf95cb5e7f506b3725915596

This model generates probabilistic motion predictions based on a
sequence of relative input movements. The input movements are converted
into polar coordinates (distance and angle) based on an axis that
follows the current path. This ensures that the orientation of the
device and of the inputs do not affect the predictions. The orientation
of the input device is also transformed to be relative to the path axis.

The test cases verifying model efficacy are consolidated into CTS.

Bug: 167946763
Test: atest libinput_tests
PiperOrigin-RevId: 492068340
Change-Id: Icd8d90bd5a7ce79c699bfdb6367a4cbd8130441a

Bug: b/266595015
Test: atest PerformanceHintManagerTest
Change-Id: I33755bb30cceb9953db293092e417a1a14288226

Before this CL, a number of checks for kl file validity were incorrect.
Some of the APIs were supposed to return an invalid value, but instead
were always returning a valid value, no matter what the input was.

Correct these values by switching to std::optional.

Bug: 266400536
Test: m libinput_tests && adb sync data && adb shell -t /data/nativetest64/libinput_tests/libinput_tests
Change-Id: I4ef45f3249dca4f4f033fb85e9fecbc2ad1f1395

It's useful sometimes to print out the events produced by the
dispatcher. In this CL:
- Switch (partially) to the C++-style prints from android-base
- Add a way to print keyevents, motionevent into a stream
- Add InputEventInjectionResult print

Also, improve the debug prints for outgoing events. When an entry is
getting dispatched, the dispatcher may modify its action, among other
variables. With this CL, this will be observable in the logs.

Bug: 211379801
Test: atest AccessibilityEndToEndTest
Change-Id: I221161af7903ae4da77733265c67a426a3e5b557

This dumps the state from the gesture converter, and a list of gesture
properties, which is useful for debugging settings.

Bug: 251196347
Test: run dumpsys input with touchpad connected, check output
Change-Id: I036d0251b06489b645b883a239ff345a98448497

Update:
Compared to the first version of this CL that got reverted, this version
also adds Flags::FOREGROUND to the windows whenever appropriate.

Before this CL, hovering window was tracked separately inside
InputDispatcher. This window was getting updated in various places.
Inconsistent motion streams, like HOVER_ENTER->DOWN->UP->HOVER_EXIT were
possible.

In this CL, we track hovering pointers inside TouchedWindow. At all
times, the currently tracked pointer must always be in the touch state.
The hovering pointer is removed when HOVER_EXIT is received.

This CL also establishes the foundation for multi-device, multi-pointer
streams, by storing hovering pointers inside TouchedWindow per-device.

Eventually, we can look into separately creating touched targets from
updating the touch state. This approach is partially used in this CL.

TouchState is used to keep track of where the hovering pointer is
currently. The 'addHoveringWindowsLocked' function returns the
equivalent of InputTargets. Eventually, we can change this to return
InputTargets.

39d37cfb

Bug: 211379801
Test: atest android.accessibilityservice.cts.AccessibilityEndToEndTest
Test: atest VirtualMouseTest
Test: m inputflinger_tests && adb sync data && adb shell -t /data/nativetest64/inputflinger_tests/inputflinger_tests

Change-Id: I0aa77bc9f680786b154312c4c936da2cf6efffa1

The InputManager#getHostUsiVersion(displayId) API can be used to tell if
a display supports the Universal Stylus Initiative protocol for styluses,
and can be used to address version compatibility issues.

We first dynamically attempt to find an InputDevice associated with the
display that supports USI. If there are none, we check the statically
add USI version for the display in config_displayUsiVersionArray.
Checking the config as a fallback means the USI InputDevice does not
have be registered/open perpetually for the getHostUsiVersion() API to
return the correct values.

Bug: 261596890
Test: manual with device
Test: atest TouchScreenTest
Change-Id: Ifabd00aa996ba88258d8112e5de3582adaf27629

Add TYPE_HANDWRITING index on native input.h to use on
inputManagerService.cpp

Test: Manual Test(hover pointer on handwriting area)
Bug: b/215436642
Change-Id: I41caec44d81c414b4acc1e3575c5acc6f99146d8

We would like to skip resampled data in VelocityTracker when the
velocity is being computed. To make this happen, we need to first
process the data per-pointer. This will then allow us to skip individual
data points.

To minimize the impact to older VT strategies that are largely untested,
the main change here is to make 'mMovements' and 'mIndex' become
per-pointer, rather than converting to a vector.

Also, use std::array instead of [] because [] cannot be used inside a
std::map easily.

There should be no functional change in this CL. The actual skipping of
resampled values will be done in a separate CL.

Bug: 167946721
Test: m libinput_tests && $ANDROID_HOST_OUT/nativetest64/libinput_tests/libinput_tests

Change-Id: I1c3c845bca0d4bb7d2c3973bfe84462139ac36f3

Need to allow resetting of key character map by setting overlay
map to null. This is required when moving from a IME setting
that has a KCM to another IME setting that has none.

Test: atest KeyboardLayoutChangeTest
Bug: 20805588
Change-Id: I62dc65970b61628486aff698a8ab8ccb31c109d6

The active pointer id is not always valid. Rather than using a magic
value for it (-1), let's use std::optional.

Some other fixes:
- convert to enum class, for better type safety / switch checking
- ident switch statements

It's necessary to ident switch statements to avoid weird clang-format
problems when changing code that follows those sections.

Bug: 167946721
Test: m libinput_tests && $ANDROID_HOST_OUT/nativetest64/libinput_tests/libinput_tests
Change-Id: I6b78301e2c1d2fba1a789fa9491f7d6ab8b37d02

Bug: 237579711
Test: UTs + build + boot

Change-Id: Ie94a626e8c92aecc0180cf71e965c905ae9f1f8d

Bug: 251196347
Test: check events received by a custom tester app, and touches shown by
      pointer location overlay
Test: atest inputflinger_tests
Change-Id: I249ca6208091e3c4291c5be68c77339bf5f69a5b

Bug: 251196347
Test: try out two-finger scroll gestures with an Apple Magic Trackpad 2
Test: atest inputflinger_tests
Change-Id: I9e693350873b90bd50605ab348821224ae1121a8