Many test authors are starting to use `CheckFlagsRule` across suites
like CTS, so instead of recommending that they use `null` when on
Ravenwood, offer them a clear behavior that is either "all-on" or
"all-off".

Eventually we'll explore providing default flag values on Ravenwood,
but let's get something cleaner available now.

Bug: 318841620
Test: ./frameworks/base/ravenwood/run-ravenwood-tests.sh
Change-Id: Id234a375426fad38f64d3a999339bb720eae84bf

Ravenwood test authors should only need to call setServicesRequired()
for the services they directly interact with, but often those
services will have indirect dependencies on other OS internals, and
those dependencies will evolve over time.

To handle this, we add the concept of dependencies to SystemService,
where a service declares the other services required to run.  (They
can choose to omit services from their dependencies where they
gracefully handle a missing service at runtime.)

Bug: 325506297
Test: atest RavenwoodServicesTest
Change-Id: I42187b3494fef499d619ad882891e832b0fd6bca

One of our eventual goals with Ravenwood is to support usage of
system services from test code.  Robolectric takes the approach of
publishing "shadows" which are effectively fakes of the Manager
objects visible to app processes, and it unfortunately doesn't offer
a mechanism to run "real" services code.

In contrast, Ravenwood aims to support API owners progressively
offering their system services either via a "fake" approach, or by
using various levels of the "real" code that would run on a device.

This change wires up the foundational support and uses the simple
`SerialManager` example to demonstrate using the same "real" code
on both Ravenwood and devices.  It also demonstrates the `Internal`
pattern being used to customize behavior for tests.

To offer as hermetic as a test environment as possible, we start
new instances of each requested service for each test.  Requiring
developers to be explicit about the services they need will help
keep overhead low, especially for tests that don't need services.

Bug: 325506297
Test: atest RavenwoodServicesTest
Change-Id: Ie22436b38f2176f91dfce746b899ebab7752bbb8

Enable android.util.Log's native code

Test: ./ravenwood/run-ravenwood-tests.sh
Bug: 318393625

Change-Id: I0ec607586d71b1cc34b71db67249e110edc304df

- Also make sure framework-minus-apex is prioritized
- Also some build rule cleanup

Bug: 315031371
Test: ./ravenwood/run-ravenwood-tests.sh

Change-Id: I2d0eb822517abfc17f351810a53176ad9904f908

We had temporarily been using the binary XML implementation in all
situations, but this change brings in the existing `kxml2-android`
library to offer reading/writing of human-readable XML.  We also
use the upstream SAX parser to offer an event-based parser.

In all cases we carefully keep the existing XML libraries in place
untouched on physical devices to avoid regressions.

Bug: 324417456
Test: atest CtsUtilTestCasesRavenwood
Test: atest FrameworksCoreTestsRavenwood
Test: atest FrameworksUtilTestsRavenwood
Change-Id: I2effc7fca5c586968658073f9652d13cf50a2c66

Common test utilities like `TestableLooper` rely on being able to
drive `Looper` interactions via `TestLooperManager` which can only
be obtained via `Instrumentation`.

This change integrates with the common `InstrumentationRegistry`
library which is how most tests obtain their `Instrumentation`
instance.  We play an odd dance with `java_host_for_device` and
`java_device_for_host` to get a version of `androidx.test.monitor`
that we can link against our runtime.  (The build system otherwise
gets excited that we're trying to compile an Android-linked library
into what appears to be a Java-linked library, but we know that we're
offering the Android runtime.)

Fix subtle bug in `MessageQueue_host` where a requested wake might
race with the `for (;;)` loop in `MessageQueue.next()`; if a wake is
pending, it's a valid reason to return from `nativePollOnce()`.

Add timeout logic to `RavenwoodRuleImpl` which can be enabled to
dump all our stack traces just before Tradefed times out, to aid in
debugging deadlocks; disabled by default.

Bug: 319647875
Test: atest SystemUiRoboTests
Test: atest SystemUiRavenTests
Test: atest FrameworksCoreTestsRavenwood:TestLooperManagerTest
Change-Id: I13605d32df0c3e6758f541d5e75a6bdaf44368db

Bug: 292141694
Test: atest CtsUtilTestCasesRavenwood
Change-Id: I4068c227b4d42b34d2a7e1e5da2032a870b3cfc2

Some CTS tests desire to link against `test_current` to prove to
themselves that they're not touching hidden APIs.  This means if they
also want to use our `ravenwood-junit` rules, that dependency also
needs to link against `test_current`.

Since we're reaching down into plenty of hidden details when we're
actually running on a Ravenwood runtime, we split into a `stub` and
an `impl` library, so that we give a cleanly compiling library to
to clients, which is then replaced under the Ravenwood runtime.

Bug: 292141694
Test: atest-dev CtsTextTestCasesRavenwood CtsTextTestCases
Change-Id: I96dbc9643324a7853a992e97b51a88f70eb721f3

We have the XML interfaces from libcore, but the implementations
there are tied closely to things like StringPool.  We thankfully
have our own human-readable XML serializer that we could use, but
there's unfortunately not a good pull parser.

To get us unstuck for the moment, pivot the internals to
unconditionally use the binary XML format, which at least we
know is thoroughly tested, and any data written through this path
can then also be parsed.

This is enough to support PersistableBundle, and we'll circle back
to get ourselves a human-readable XML implementation in the future.

Bug: 292141694
Test: atest-dev CtsOsTestCasesRavenwood CtsOsTestCases
Change-Id: I875a3a2b0e95e52861afe187e2a5e9f1e740d8d5

The Ravenwood environment doesn't support IPC, but much of the code
that developers intend to test (such as AIDL stubs) still relies on
baseline same-process behavior to be intact.

This change adds UID and PID support to Process and Binder, but we
carefully communicate that it's only available when the test author
has configured it via a RavenwoodRule.  This ensures that tests
don't accidentally rely on unexpected defaults, as some tests will
want AID_SYSTEM, and others want a normal app UID.

Also bring along SystemClock, PatternMatcher, and ParcelUuid
along with relevant tests.

Finally, now that we have an IgnoreUnderRavenwood annotation, we
can pivot back our default failure to be a RuntimeException instead
of an AssumptionViolatedException.

Bug: 292141694
Test: atest-dev CtsOsTestCasesRavenwood CtsOsTestCases
Change-Id: I5f54c3179b2d305b9ab9144c43fd063c6b756e44

Also add some presubmit tests as a smoke test.

Bug: 292141694
Test: run-all-tests.sh
Test: atest --no-bazel-mode CtsUtilTestCasesRavenwood

Change-Id: I30569313b764dfa30db9c2c5627050da4a6a1d67

We have baseline implementations of the Parcel and Bundle, so we're
ready to support them under the Ravenwood environment.  Brings along
the relevant CTS to prove that it works consistently on both devices
and Ravenwood.  Make a few test-driven bug fixes to Parcel code.

Currently does not support FDs or Binders.

Add new `@IgnoreUnderRavenwood` annotation that is respected by a
new `RavenwoodRule`, letting the same test code be shared to run
under both devices and Ravenwood.  Currently packaged under our own
area of the tree, but named to match layout of `platform_testing`
for future migration.

Add "ThrowClass" policy.

Bug: 292141694
Test: atest-dev CtsOsTestCasesRavenwood CtsOsTestCases:ParcelTest CtsOsTestCases:BundleTest
Change-Id: I88d19afff47fbea6557e922a10e8555ee0f70ba3

Soong uses the directory structure to guess the java package, and
"annotations-src" is not a package considered for APIs. Remove the
prefix such that soong sees the correct package android.*.

Hide the Ravenwood classes for now.

Test: m checkapi
Change-Id: I430009deb7df2beac1736c4a68ff6fa1a3be572a

- Add "framework-minus-apex.ravenwood".
- Update the behavior of: --default-throw:
  - Now this will "keep" the subject classes and fields.
- When a 'throw' directive is specified on a static initializer,
  it's now handled as 'keep', because otherwise the class
  wouldn't be usable.

Test: ./scripts/run-all-tests.sh
Test: m framework-minus-apex.ravenwood
Bug: 292141694
Change-Id: If4d650dc175a5a78cf8844dfed2217c19f2d5790