Deprecate PackageParser#Package
Replaces PackageParser#Package and it's related structures with
ParsingPackage, ParsedPackage, and AndroidPackage.
This is a large scoped CL, the first step in an eventual goal
to refactor how data is handled in the package parsing and install
process. It introduces as few logic changes as necessary. Mostly
migrating to interfaces and renaming, moving parsing data calls into
3 separate interfaces that outline the intended flow for parsing.
ParsingPackage is built and used during what was PackageParser, now
replaced by ApkParseUtils and it's related classes. This is almost
exactly what was parsed from the XML/files on disk.
ParsedPackage is used when the object exits PackageParser and is
adjusted by PackageManagerService to what is considered the final
"parsed" state, adjusted from literal values, but consistent given
the same APK on disk. This should eventually be moved out of PMS
and possibly collapsed into the previous interface entirely.
AndroidPackage is the final state of the package after parsing and
adjustment has completed and no more mutations should be expected.
There are a few exceptions to this, included in AndroidPackageWrite,
which will eventually be refactored into PackageSetting or another
state class.
This marks PackageParser#Package and all the old infrastructure with
@Deprecated, as none of them are used internally. All usages were
converted, and the legacy Package is only built for un-converted tests
and @UnsupportedAppUsage methods.
There are numerous TODOs still outstanding, but addressing them
in this initial CL would introduce several logic changes. They've been
marked with the bug number and will be handled in follow ups.
This is being merged with caution thrown to the wind because
testing this on devices and in development will be the best way
to debug differences introduced by the migration. Getting it merged
as early as possible gives the most amount of time to fix regressions.
Waiting for tests of all the functionality could take literal years
before covering enough to merge this with all regressions verified.
Given a sample size of 4 heap dumps and the caveat it was taken very
early in the migration, there is a memory overhead of about 200 KB
versus the legacy implementation. This should be verified more
accurately and addressed in follow ups.
This CL also kills child/parent package support, since that's
broken already, and difficult to support with the new interface
structure.
Bug: 135203078
Test: booted an emulator, works generally as expected
Specific tests which failed / failed to build were fixed, but because
not all tests are currently passing before this change, not all were
verified.
Change-Id: I4ba050c228e6c60b8f63a9e3347b1f9a57ef794a
112 files changed