This reverts commit 6c7fb437.

Reason for revert: Master is broken; not clear if/how this CL is to blame but since it's not needed immediately backing out right now in case its involved

Change-Id: I2048e28e515c42bf36d4386acc48b060ad9afb8e

Test: Unit test included
Bug: 120429729
Change-Id: I1c42787e9571fb845188d0444af39e509474721e

This reverts commit 9da4edac.

Reason for revert: b/126374865

Change-Id: I55d05da395dcba2562db61f0f0a873105eef6302

Test: make update-api and make checkapi
Change-Id: I034f99483c53d86922f735143f6a16ef41d46093

Allow @RequiresPermission to use the special marker string
"carrier privileges" to indicate that the app having carrier
services will also satisfy the requirement.

Bug: 72967236
Test: Unit tests included
Change-Id: I1ae8ef8d70b20b24a2f907cd80a50011fa66625c

Fixes: 124811433
Test: in androidx-master-dev, ./gradlew :media2-widget:generateApi
Change-Id: I9e97d8c52b489b5b8a8f5b1ec49b9425c91125d6

This is intended for doclava and other tools to do their
own rendering of the doc content to describe a deprecation
level.

Also renames the @since tag to @apiSince to make it less
ambiguous and to not conflict with existing @since tags
that are present in various docs.

Also starts emitting @apiSince into package.html docs,
such that there is a concept of a package API level; this
is the lowest API level for any class in that package.

Also makes sure we don't emit @apiSince for SystemApi
docs, since we don't have accurate historical information
for SystemApi and TestApi sources.

Also starts writing @apiSince tags even when the since tag
is 1 (e.g. for the APIs added from the beginning). This
was omitted for optimization purposes earlier but is added
back to make the doc generation task easier.
I also added some optimizations for these two new tags since
they can be merged more quickly with some special handling
since they never appear in existing sources and can always
be listed last in the docs.

Finally, various fixes to the API lookup data base
generation; this makes the generated database more
closely mirror what's in android.jar, and, importantly
for doc generation, properly tracks in-development APIs,
such that generated docs at the moment shows up as "Q"
instead of "28".

Test: Unit test updated
Change-Id: If25a8075dc1bb2ace184d1b4d6f19717fae2bc83

Test: Unit test updated
Change-Id: I7f0b859b5f70649c611706a065f5a6411345976d

This CL adds the following flag:
--subtract-api <api file>  Subtracts the API in the given signature or jar
                           file from the current API being emitted via
                           --api, --stubs, --doc-stubs, etc. Note that the
                           subtraction only applies to classes; it does not
                           subtract members.

This lets you limit the emitted API signature files, stub files, etc,
to be just the delta relative to a subset of the API, or an older
version of the API.

Bug: 119021844: Remove duplicate classes from org.apache.http.legacy stubs library
Test: Unit test included
Change-Id: I28a9f0f49ac42626a62f5ca4faa6c35e38554228

This CL adds a new flag to tell metalava that it's invoked while
*only* checking compatibility. In that case it will ignore various
other flags being passed in to it (such as --stubs, which it's
currently unconditionally passed but is not useful and is costly
during compatibility checks.)  It also renames the --update-api
flag (meant for similar purpose during make update-api, where
we don't need to generate stubs) to be called --only-update-api
to make its purpose clearer.

Test: Unit test included
Change-Id: I3f105746c5f76356b8dca0f1709ee2a4413ff938

This CL updates the handling of Kotlin-style null syntax (used
in format v3). It now does a much more accurate job formatting
types from Java syntax (and Java type use annotation syntax)
into Kotlin-style nullness syntax. This cannot be done with
simple things like replacing @Nullable with ? since the place
for the annotations in the type expression is not the same as
where the nullness suffix goes.

This CL also adds --format=latest as an alias for --format=v3
and --format=recommended as an alias for --format=v2.
(The alias values will change over time as the formats and
recommendations evolve.)

Bug: 123415058
Test: Many new unit tests
Change-Id: I6622041f0966b99157db4fe60d534a55e658bd10

More recent versions of the parsing environment
requires some javadoc extension points to be
registered.

Test: make sdk
Change-Id: Ifee7ff006b6d11bbb58ec772e722ffef9f6f6f33

This should fix
123538809: Broken at //contentads/gfp/video/liveads/testing/playback:appletv4_invalid_test_build_test

Change-Id: I52ad8f0afd0c2c47c30a16e47dada923cbd61f25
Fixes: 123538809
Test: Unit tests updated

The severity changed; update the unit tests.

Test: This is it
Change-Id: I21fcb3442ba7cce39db3dbaf384fd41a3792b76c

Change-Id: I2d34f9c0275ff1bc70de28087d3a4b0086780b28
Test: None

When reading in annotations from Java stub files, attempt to merge
in type annotations as well.

Bug: 123223339
Test: Unit test included
Change-Id: I289b0a92d8a3741ce6aabdbb5c83b3b3cbcf32d1

Bug: 122452571
Test: make sdk before aosp/I25f6203725391dc5404cb8b6049479c9c3da1699 is
applied

Change-Id: Ie499130c037885618bac851a665b4dee6cc0cefe

Ensure that we clean up the application environment properly when
exiting. This doesn't matter when metalava is invoked as part of the
build since it's invoked and then exited, but when running unit
tests this is important.

Test: Existing unit tests
Change-Id: I83ad663dd649d234cddd0afb6d3cf8f67b1d846f

Test: Unit test included
Change-Id: I17c4e63b2feef7ff0b0ebd97719fdfad03a03543

We had some accidental nullability annotations on primitive
types in API 28, such as on Boolean.getBoolean. These were
subsequently removed, but compatibility checks flag these
as illegal removals. Ignore primitives for these purposes.

Test: Unit test included
Change-Id: Iac0eaab888a764285d98ee11d4ac1d650ca68e53

Until Android P, the SDK did not contain nullness annotations,
so in Android P all annotations were marked as @RecentlyNullable
or @RecentlyNonNull.

This CL fixes things such that as of master/Q, when a nullness
annotation on an API is not new, the annotation is instead mapped
to a non-@Migrate annotation as far as the Kotlin compiler is
concerned: android.annotation.{NonNull, Nullable}.

Also makes some internal annotation processing more efficient.

Test: Unit tests included
Change-Id: Ia2144b5142cb0ce5e5303a5c72fb9c7c6e3a69d7

When comparing type signatures,	spaces should not be significant,
eg. Map<X,Y> ==	Map<X, Y>. This	was already mostly the case, except
for when comparing types from one signature file with types from
another, as is the case when doing compatibility checks with show
annotations (such as the system API).

Test: Unit tests included
Change-Id: Icec1a679cfb9245905581bc0fe9074745f835038

Normally baselines are deleted if they are empty. This
CL adds a flag to let you leave baselines in place even
if they are empty.

--delete-empty-baselines  Whether to delete baseline files if they are
                          updated and there is nothing to include.

Also tweak baselines to not end with two blank lines.

Test: make update-api
Change-Id: I7c24d6ac531ad32dfd0e45e2ebd3f2d4c7591338

Gradle 5.x requires a newer version of the shadow jar plugin.

Test: ./gradlew shadowJar
Bug: N/A
Change-Id: I1b2603be123a4c1b36fbcb08fefbdcc0cdfc7650

Bug: 123242905
Test: m -j metalava && java -jar ../../out/host/linux-x86/framework/metalava.jar --version
Change-Id: I926d1abe4da973cd86290e69dcc0df2b4d84554b

This adds support for Kotlin 1.3, which also has the side
effect of switching packages for coroutine signatures.
This required a number of changes; for one thing, in
various places the PSI annotations would say @Nullable
for things like default parameters whereas the UAnnotations
correctly were marked as @NonNull, so metalava had to
start looking at UAnnotations instead.

Test: Unit tests updated
Change-Id: I9a7537a215ce3f8d065d6183bc5f09ebdf3bf77a

Tweak the handling of baselines: Make sure the baseline message
reflects the correct path; add a flag to allow passing the
build on baseline update errors such that the source tree can
be updated in one go, etc.

Also switch to kotlin UTF_8 charset constant.

Test: Existing
Change-Id: I6d9fd003eb50e027b45a6bf7765e4f5270e369af

This CL fixes CTS tests that depend on string equality between
the JDiff type descriptions and their own internal pretty-printer
based on reflection lookup of the API; the pretty printer always
emits a space after a comma in type lists whereas metalava's
signature file to JDiff output converter would preserve whatever
was in the signature file, which as of v2 omitted spaces.

This CL forces the output to always have spaces when converting
to JDiff output.

Test: Unit test included (plus of course the failing CTS test)
Change-Id: Icbcb3067423cb5fa259e18e028704c27c48953e4
Fixes: 123140708