In the review of ag/26174212, an unnecessary string conversion was
identified; this CL corrects that, plus two other instances in
the same function.

Bug: 314732730
Test: Presubmit
(cherry picked from https://googleplex-android-review.googlesource.com/q/commit:60e4fdfa0897b94c003ba3df3b39a16e10c886d1

)
Merged-In: Id2b620088e12266f041be864271493ffe7e1a3b1
Change-Id: Id2b620088e12266f041be864271493ffe7e1a3b1
Signed-off-by: Edward Liaw <edliaw@google.com>

Sony gamepads need a uniq attribute of a certain size in order to
obtain the MAC address, which makes our CTS test fail. This CL
adds the ability to define this attribute in our test definitions.

Bug: 314732730
Test: Compiles properly, will be tested together with the test
(cherry picked from https://googleplex-android-review.googlesource.com/q/commit:93a4c0fbdc14d99cbd382fd50be83b7058ca6b56

)
Merged-In: I83203e8eb876b27d0e6e60adfbf6de3f4a6e9872
Change-Id: I83203e8eb876b27d0e6e60adfbf6de3f4a6e9872
Signed-off-by: Edward Liaw <edliaw@google.com>

* This brings the childNafCheck method in line with the version in
  androidx.test.uiautomator.

Bug: 315023780
Test: manual
Change-Id: Ie5ba2340bedabc31f346c6e717353c7a5db0e90e

Use the size of limitSurfaceSize to set the window

Bug: 339763174
Test: 1. set the property:
	    [ro.surface_flinger.max_graphics_height]: [1080]
    	[ro.surface_flinger.max_graphics_width]: [1920]
    2. Repeatedly Turn on/off while connecting to 4K TV and check the boot animation
Change-Id: I378aab715d75b8ddb47068cb0e81503d83339021

Bug: 29055299
Test: manual
Change-Id: I4f821acb298f8d74d17afd0db7ba4eb7729620d2

The previous changes to support specifying timestamps for evemu
recordings inadvertently made injections from JSON-style recordings have
the timestamps at which the injection was scheduled, not when it
happened, causing many events to be injected with very close-together
timestamps. This broke drawing in the tests for some inking libraries,
such as Keep's.

Bug: 330844071
Test: $ atest 'PlatformScenarioTests:android.platform.test.scenario.sysui.stylus.StylusInkingTest#writeText_appearsInShowcaseApp'
      (with the @Ignore in StylusInkingTest.kt removed)
Test: replay the recording from b/330844071#comment30, check the curve
      is drawn correctly
Test: replay an evemu recording and check that the timestamps shown by
      `getevent -lt` are still correct
Change-Id: I5342a79d4f9b53875b9918443bf1f5fd16cd205e

See parent cl: aosp/3028633

Test: atest -p cmds/locksettings:postsubmit
Test: atest -p services/core/java/com/android/server/locksettings:postsubmit
Test: atest -p services/devicepolicy:postsubmit
Change-Id: I6d0d0320e0bf8927f9b8b27b471ae58ace56124a

This suggested change is automatically generated based on group
memberships and affiliations.
If this change is unnecessary or in error, vote CR -1 and the bot
will abandon it. Vote CR +1/2 to approve this change.

See the owner's recent activity for context:
https://android-review.googlesource.com/q/kwekua@google.com

To report an issue, file a bug in the Infra>Codereview component.

Change-Id: Iafe15ba5276e97d8ffc036d4b5bc71967df25329

The recent change to how event injections are scheduled causes problems
when a client waits a considerable time after a device registration
before sending a lot of events separated by delays. The long wait means
that many of the events get scheduled in the past, so they're all
injected at once, filling the kernel's buffers and causing events to be
dropped.

The updateTimeBase command allows the client to tell us that the long
delay is intentional, and that the following injections should be
scheduled relative to the current time, rather than the previous
injections or the registration.

[0]: change Ieaa4f2f06d5e0b7d13abc3afa474948996db7c0a

Bug: 330844071
Test: check for SYN_DROPPED events in the output of:
      $ adb shell getevent -lt
      while running the inking test:
      $ atest 'PlatformScenarioTests:android.platform.test.scenario.sysui.stylus.StylusInkingTest#writeText_appearsInShowcaseApp'
Change-Id: I31c596251e27149da16270a23f4f57e7bb8e3460

This change adds BMM logging in restore at install flow right from the point when PackageManager triggers restore.
The change also adds more detailed logging in Full Restore flow, around various failure points. This'll help with quantify different failures that happen in Full Restore flow, some of which are masked by the complexity in Framework code.

Bug: 331749778
Test: Ensure that the change builds (m -j frameworks/base)
	Ensure CTS and GTS B&R tests pass
Change-Id: Ieff6df433398ffff71c7eeda73174f9f1d509b58

Section 1002 (/proc/version) is primitive data (TYPE_STRING)

Bug: 277334359
Bug: 328677623
Test: adb shell incident -p AUTOMATIC 1002 | protoc --decode_raw
Change-Id: I2f9728d542becd3690522b3a113a4760c7d324d2

when -a is given, it will append a postfix to the FILENAME, such that
FILENAME_0.png, FILENAME_1.png. this won't break the existing behavior
as it doesn't mutate the name if it's taking a single display.
when both -a and -d are given, it will ignore -d and capture all the
displays.
also updated the usage doc to clarify it captures the default display
when the id is not given.

Test cases:
- no argument > prints out in the cmd window
- only -p > prints out in the cmd window as png format
- -h > correctly shows the usage
- -d with and without .png > saves as a file
- -a with and without .png > saves as files
- -d and -a > ignores -d and shows the same results as single -a

Bug: 321278149
Test: adb shell screencap with various flags
Flag: NA
Change-Id: Iecfeec1a1edbc95d7e8931ac3b22ac7f0706c3e7

In the review of ag/26174212, an unnecessary string conversion was
identified; this CL corrects that, plus two other instances in
the same function.

Bug: 314732730
Test: Presubmit
Change-Id: Id2b620088e12266f041be864271493ffe7e1a3b1

Sony gamepads need a uniq attribute of a certain size in order to
obtain the MAC address, which makes our CTS test fail. This CL
adds the ability to define this attribute in our test definitions.

Bug: 314732730
Test: Compiles properly, will be tested together with the test
definition change in the same topic

Change-Id: I83203e8eb876b27d0e6e60adfbf6de3f4a6e9872

The most common operation when getting a new Resources object
is checking if all apks and overlays are still up to date to
reuse the cached object. It makes sense to optimize it by
excluding the file by path lookups and instead keeping an open
FD to the file in the cache

+ Make IsFabricatedOverlay() more efficient via a name check
  and string_view where possible

Bug: 282215580
Test: build + boot + presubmit
Change-Id: Ib1ab20cba359c2195a72dd2e10096883d95b4453

With the changes to sys.powerctl this variable is no longer set on a
power event - the power action is simply carried out. (This is to avoid
possible DoS of these power functions.) However, this means that this
property can no longer be read to see if a power event is happening.
Since there is no longer a way of telling if a remote exception is real
or not, best not to print out an error.

Note that this bug occurs on some devices, not all. I was able to
reproduce it reliably on a Pixel 7a, but not on a Pixel 7 or Cuttlefish.

Bug: 318323013
Test: adb shell svc power reboot

Change-Id: Icbc1ec1b7882140d4d742b056ea2adaafae67ee9

Remove shell commands from the commands.Telecom binary and into
TelecomServiceImpl.

For backwards compatibility, still allow Telecom to process the
commands by retrofitting in TelecomShellCommand.

Bug: 327310548
Test: atest CtsTelecomTestCases CtsTelecomCujTestCases
Change-Id: I44fe946046504b804204e1935b02a3047328bf4e

bug: b/326318293

Test: atest FrameworksInProcessTests CrossUserPackageVisibilityTests SdkSandboxInprocessTests
Change-Id: I774f9b81f1507ccbff6fd9c55a47d48911113089

(This CL is unchanged. Original description below, with additional Test:
line)

If we want to inject events at predictable times, then we shouldn't take
into account the current time when scheduling those injections, as that
could change depending on the timings of the process sending commands to
uinput. Instead, assume that the process sending commands intends time
offsets between injected events to be as close as possible to those
specified in the delay commands.

Bug: 310958309
Test: atest CtsInputTestCases
Test: atest android.view.cts.input.InputDeviceKeyLayoutMapTest \
            android.view.cts.input.InputDeviceSensorManagerTest \
            --rerun-until-failure=10
Change-Id: Ieaa4f2f06d5e0b7d13abc3afa474948996db7c0a

The reland fixes an issue where the time at registration would be used
for timestamp values forever unless `delay` commands were introduced.

(Original description, with additional Test line:)

Last year support for specifying timestamps for uinput events was added
to the Linux Kernel [0]. This allows us to give precise timestamps for
events being played back from an evemu recording, despite the actual
times at which we inject them being imprecise due to how threads and
the kernel are scheduled.

[0]: https://lore.kernel.org/lkml/20230427000152.1407471-1-biswarupp@google.com/

Bug: 310958309
Test: with the patch added to the device's kernel, play back pointing
      stick and touchscreen recordings. Compare `getevent -lt` output
      with the timestamps in the recordings to check that the offsets
      between timestamps match (e.g. using a spreadsheet).
Test: atest UinputTests
Test: atest android.view.cts.input.InputDeviceKeyLayoutMapTest \
            android.view.cts.input.InputDeviceSensorManagerTest \
            --rerun-until-failure=10
Change-Id: Icf77e8adba075ab070806311f31c6a5e1bd98bc9

(This CL is unchanged. Original description below, with additional Test:
line)

evemu recordings use microseconds for their time intervals, but we can
only schedule handler calls in Device at millisecond precision. So far
we've converted the microseconds into milliseconds in EvemuParser, which
means that the precision losses compound over time (since each delay
will be slightly shorter than it should be, and the next delay will
start from that slightly earlier time, etc.). Keeping the delay
durations in a more precise unit up until the very last moment means
that we'll only get the precision loss once for each event. Since it's
somewhat uncommon to use microseconds elsewhere in Android code, and we
get the system time in nanoseconds, we may as well use nanoseconds
rather than microseconds.

Bug: 310958309
Test: play an evemu recording through uinput
Test: atest UinputTests
Test: atest android.view.cts.input.InputDeviceKeyLayoutMapTest \
            android.view.cts.input.InputDeviceSensorManagerTest \
	    --rerun-until-failure=10
Change-Id: Ibb968487ed114a4c464ac7061af4cda188e92498

Revert submission 26154399

Reason for revert: b/324707605

Reverted changes: /q/submissionid:26154399

Change-Id: Ie1e20baefa998f0b69fd56ffa780ea3d3cd038db

Revert submission 26154399

Reason for revert: b/324707605

Reverted changes: /q/submissionid:26154399

Change-Id: I2a9bc83e94a28691a1e3c6e3aaa8bb6d77ab17a5

Revert submission 26154399

Reason for revert: b/324707605

Reverted changes: /q/submissionid:26154399

Change-Id: Ib4ab9a80e74ad5bfec005dd2744458a60bdedde1

Seems I missed this TODO when I added version ID support in commit
d9505841.

Bug: 302297266
Test: atest UinputTests
Change-Id: If7071266be531fd82d8301c26df405476f614cb2

If we want to inject events at predictable times, then we shouldn't take
into account the current time when scheduling those injections, as that
could change depending on the timings of the process sending commands to
uinput. Instead, assume that the process sending commands intends time
offsets between injected events to be as close as possible to those
specified in the delay commands.

Bug: 310958309
Test: atest CtsInputTestCases
Change-Id: I3ceab49bd6f492244e53d3263bf10802353853c2

Last year support for specifying timestamps for uinput events was added
to the Linux Kernel [0]. This allows us to give precise timestamps for
events being played back from an evemu recording, despite the actual
times at which we inject them being imprecise due to how threads and
the kernel are scheduled.

[0]: https://lore.kernel.org/lkml/20230427000152.1407471-1-biswarupp@google.com/

Bug: 310958309
Test: with the patch added to the device's kernel, play back pointing
      stick and touchscreen recordings. Compare `getevent -lt` output
      with the timestamps in the recordings to check that the offsets
      between timestamps match (e.g. using a spreadsheet).
Test: atest UinputTests
Change-Id: If486cdb7218918aca64e6561f9fc2f30acce736a

evemu recordings use microseconds for their time intervals, but we can
only schedule handler calls in Device at millisecond precision. So far
we've converted the microseconds into milliseconds in EvemuParser, which
means that the precision losses compound over time (since each delay
will be slightly shorter than it should be, and the next delay will
start from that slightly earlier time, etc.). Keeping the delay
durations in a more precise unit up until the very last moment means
that we'll only get the precision loss once for each event. Since it's
somewhat uncommon to use microseconds elsewhere in Android code, and we
get the system time in nanoseconds, we may as well use nanoseconds
rather than microseconds.

Bug: 310958309
Test: play an evemu recording through uinput
Test: atest UinputTests
Change-Id: I68d54c05214ae778167adfd0d3fc9a345454c336

SomeArgs objects are held in a pool, which gets depleted if they're not
recycled after use, lowering performance.

Bug: 245989146
Test: play back an evemu recording
Change-Id: I0d5109f654bd61eb2b3ef68a2ce7a3ab07371ded

The most common operation when build idmaps is to find the
resource ID by name. This is the least optimized operation
in the current resources data structures, as they only
expect ID -> value lookups

This change adds an optional flag that creates a name->ID
hash map inside ResStringPool, and use it when loading
APKs in idmap2d

Bug: 282215580
Test: build + boot + perf record

Change-Id: I82f4d684cb48e2dcddcd677b882b11497c1c13b1

Bug: 282215580
Test: build + boot + perf record

Change-Id: I5c76515aa3dad405a2938d8d0e87048cc52cf61b

Bug: 302297266
Test: check rendering and links in Gitiles
Change-Id: I47e9c1efd5656c3dca20672e7da4fe4417b28803

This adds an api that allows creation of nine patch frros.
We process them the same as aapt2 does so that
when they are used at runtime they work correctly.

API-Coverage-Bug: 314168567
Test: manual and automatic
Bug: 296324826
Change-Id: I40da020189e9ec914fbea0c17f181209347d83de

Bug: 302297266
Test: atest UinputTests
Test: play a recording with version ID set, check `getevent -li` output
      while it's running
Change-Id: Idd3b6166a3912c26720dddbf4ad72b5a07a604a5

Bug: 302297266
Test: run with various broken recordings, check error messages are
      helpful and have correct line numbers
Change-Id: Iad0e788815c9d146babe67568a6ceef27b68007f

Bug: 302297266
Test: atest UinputTests
Change-Id: I7b4c6fa87f6ba50147f3ddeb9b8e4ceaa238ffc1

evemu [0] is a system used by the wider Linux community to record
sequences of evdev events and descriptions of the device that created
them. Together with the evemu-record implementation added to
frameworks/native, implementing support for evemu recordings in uinput
gives us a system for event recording and replay that's compatible with
other Linux systems.

Since the format looks quite different from the existing JSON-style one,
we can automatically detect which type of data is being passed in,
instead of having to change the command-line interface.

As part of the implementation, the Event.Bus enum is replaced with plain
integers. This allows the tool to support new bus IDs that are added to
Linux's input.h without code changes, at least for evemu files.

[0]: https://gitlab.freedesktop.org/libevdev/evemu

Bug: 302297266
Test: replay recordings made using Android and FreeDesktop
      evemu-record implementations
Change-Id: Ie2f969da24db9aa04037335d5b697cdc0db0b3ca

Although the binary looks for a vendor provided shared library,
it is an otherwise vendor agnostic tool. Moving it off the
vendor partition also has the added benefit of it not requiring
root to be run.

Bug: b/273907878
Test: manual
Change-Id: I81726ffb7d96acef69b7b77c864be254be9a74c5

Parsing strings into enum values should be done by the parser, not the
builder, and this means that the evemu parser will be able to pass
properly typed values to setCommand rather than "magic" strings.

Bug: 302297266
Test: m uinput
Change-Id: Icac5764a66025412c87c326280d1895f3da78cac

Previously the tool just carried on, with event injections being
silently dropped, which isn't very obvious behaviour.

Bug: 302297266
Test: cause device creation to fail somehow, check the tool exits
Change-Id: I94e931729339bff0f6fc7ab5b591f062043a1d7a