This CL adds moves the installation rules of boot images to soong. This
will eventually allow us to build devices by skipping `katiBuild` and
moving straight to `katiPackaging`
Details
1. Drop `no_full_install` from dex_bootjars singleton. This ensures that
the files installed by this singleton module does not get skipped
when generating the soong installs file (out/soong/installs-*)
2. Replace PackageFile with InstallFile. This registers the installation
rules for both make-built and soong-built images (e.g.
aosp_cf_system_x86_64)
3. Implement `AndroidMkEntries` for dex_bootjars singleton. OutputFile
needs to be non-nil so that this module does not get elided when
generating out/soong/Android-*. `OutputFile` was abritrarily set to one
of the many files installed by this singleton.
Test: no diff in
target/product/vsoc_x86_64/obj/PACKAGING/system_intermediates/file_list.txt
(top of stack)
Bug: 355700341
Bug: 355703904
Change-Id: I3531defa6bba58ef78f6d66e881502a8222fc229
ResourceProcessorBusyBox generates the R.jar binary from R.txt without
creating an intermediate R.srcjar file. Disable this behavior in kythe
builds (ctx.Config().EmitXrefRules()) to support xrefs to R imports
Test: XREF_CORPUS=$internal_corpus m nothing
Test: aninja -t query out/soong/.intermediates/frameworks/libs/systemui/animationlib/animationlib_tests/android_common/animationlib_tests.kzip
Test: verified that an R.srcjar is present in its deps
Bug: 354854007
Change-Id: I2d63c3393c5bc58103c267c4593172ce77fbc79c
Revert submission 2982300-java_api_library_non_sdk
Reason for revert: DroidMonitor: Potential culprit for b/357648959 - verifying through ABTD before revert submission. This is part of the standard investigation process, and does not mean your CL will be reverted.
Reverted changes: /q/submissionid:2982300-java_api_library_non_sdk
Change-Id: I5ef7afd9ec3e10ea99f82d02172843ad9b2cfda9
Partner worksapces contain two versions of mainline prebuilts - BigAndroid
and Go. These two prebuilts export dexpreopt'd system server artifacts
to be installed in system image. Since the install paths are same, we
run into duplicate installation rules issue unless one of them is
hidden.
This hiding was previously done by creating a dependendency between
source aosp apex to BA and Go google prebuilts. However, this
implementaion had the unfortunate side effect on the packaging name of
the Google mainline prebuilts - the name becomes the aosp apex name.
Instead of creating the dependency to aosp apex, this CL hides all
mainline apex_set(s) if it has not been flagged using
RELEASE_APEX_CONTRIBUTIONS_*. Since there are some non mainline apex
prebuilts, apex_name will be used to determine whether the prebuilt is a
mainline module.
Test: m nothing --no-skip-soong-tests
Test: In partner workspaces, downloaded the CLs in b/355682304#comment7
Test: m out/target/product/generic/obj/PACKAGING/system_intermediates/file_list.txt
verified that aosp apexes are not installed, but mainline prebuilts are
installed
Test: unset RELEASE_APEX_CONTRIBUTIONS_ADSERVICES to build from source
Test: m out/target/product/generic/obj/PACKAGING/system_intermediates/file_list.txt
verified that aosp adservices is installed, and adservices prebuilt is
**not* installed.
Bug: 355682304
Change-Id: Idacb65313553bdea5c0593976694de478034229e
The installation rules for soong built system images are generated by
soong, but the installation rules rules for make built images are still
generated by make in dex_preopt_libart.mk. There is an existing
discrepancy between the two. Make built images generates three
installation rules for
1. system/framework/<primary_arch>/$bootjar.vdex (symlink)
2. system/framework/<secondary_arch>/$bootjar.vdex (symlink)
3. system/framework/$bootjar.vdex (actual file)
Soong copies the file to (1), creates a symlink from (2) to (1) and
skips (3) altogether. This CL makes the Soong installation rules match
Make installation rules. This will eventually allow
us to build devices by skipping `katiBuild` and moving straight to
`katiPackaging`.
Test: no diff in make built installed files
target/product/vsoc_x86_64/obj/PACKAGING/system_intermediates/file_list.txt
(top of stack)
Test: debugfs out/target/product/vsoc_x86_64/system/etc/aosp_cf_system_x86_64.img
verified system/framework/boot-apache-xml.vdex exists
verified system/framework/x86/boot-apache-xml.vdex exists as a symlink
verified system/framework/x86_64/boot-apache-xml.vdex exists as a symlink
Bug: 355700341
Change-Id: I52853c07674b77a984b5a5ac5dcd69236b642b46
Resource_dirs is queried using PathsWithOptionalDefaultForModuleSrc,
which includes all the infrastructure to resolve module references,
but needs to be tagged android:"path" to be able to add the module
references as dependencies.
Test: Manually
Change-Id: Ie3f75332c9a4cc0ee4b4c93268188440ff7ce249
This change allows from-text stubs to be generated for all
java_sdk_library modules, unlike how it is currently limited to the
modules that contribute to the api surfaces (i.e. the SDK).
This change accomplish this by modifying the dependency of the
from-text stubs generation, so that the sdk_library generated from-text
stubs generating java_api_library modules no longer depend on the full
api surface stubs, but generate the stubs in the module level, and
combined later to generate the full api surface stubs.
This change also removes the java_api_library modules defined in
core-libraries/Android.bp, which are passed to generate the system
modules. Given that the from-text vs from-source toggle is done within
the java_sdk_library stubs level, these modules no longer need to exist.
Implementation details:
- Allow sdk_version to be specified in java_api_library modules. For
java_sdk_library-generated java_api_library modules, they inherit that
of the sdk_library module. Some java_sdk_library modules that do not
contribute to the api surface are allowed to set sdk_version to
something other than "none" or "core".
- Implement java_api_library to implement `SdkContext`. This allows
java_api_library to collect required deps from sdk_version (classpath,
bootclasspath, system modules), and pass the collected jars when
generating the stubs srcjar in metalava and compiling the stubs srcjar
in javac.
- Remove hardcoded list of sdk_library modules that are allowed to
genereate stubs from the api signature files, and allow from-text
stubs generation by default. Modules that are not able to generate
stubs from the api signature files are specified by setting the newly
introduced `Build_from_text_stubs` property to `false`.
Test: ENABLE_HIDDENAPI_FLAGS=true m
Bug: 327507877
Change-Id: Ia35d2f3cf9fae48fc8c4bd99a84ae18d7c0e7bee
Since only a single rule can write to a given WritablePath, it is
unecessary to hold on to the WritablePath once the rule has been
created. Keeping a WritablePath causes complications, as it
prevents using the input Path as the output when no modifications to
the input file are necessary. The normal pattern should be to create
a WritablePath using PathForModuleOut, build the rule that writes to
the WritablePath, and then store the WritablePath as a Path for use
as an input to any future rules.
WithoutRel previously only existed on OutputPath, which required
keeping the output path of the module as an OutputPath for as long
as possible in order to call WithoutRel on it at the end of the module.
Add WithoutRel to Path, make it always return a Path type, and implement
it on all the types that implement Path by using a helper in basePath.
Replace long-lived WritablePaths with Paths.
Test: all soong tests pass
Flag: EXEMPT refactor
Change-Id: I40f28075ce151e4be80d6cfc7ec173dfa46f9bbf
Note that this property is currently a no-op. This change is submitted
separately in order to prevent submitting the entire topic in git_main
to mark a main-only module with this property.
Test: m nothing
Bug: 327507877
Change-Id: I62c2e3522bf9d0c5f7955e7c9bd2e0f66458e48c
The only place basePath is used as a Path is when being returned from
PathForArbitraryOutput. Using it as a Path requires implementing the
RelativeToTop() method on it, which then allowed that method to be
inherited by SourcePath, breaking the contract on RelativeToTop that
specifies that the returned Path is of the same concrete type as the
input Path.
Make PathForArbitraryOutput return an OutputPath, and update OutputPath
to support base paths that are not out/soong. This also fixes
RelativeToTop, which was previously not working for PathForArbitraryOutput
paths.
Test: all soong tests pass
Flag: EXEMPT refactor
Change-Id: Ie8d8e2290961f35280e97137d2bd641c4d57ab87
Call t.Parallel() to make the subtests of TestClasspath run in
parallel, which drops the runtime from 12 seconds to 1.7 seconds.
Test: TestClasspath
Flag: EXEMPT refactor
Change-Id: Id2194f6b5f925b480280e9cf8da5d5767a467430
- android_ravenwood_test can now have `resource_apk: "ANDROID_APP_MODULE"`.
This will copy the APK to the test directory as "ravenwood-res.apk".
- android_ravenwood_libgroup can now have `data: ["MODULE"]`.
We use this to add framework-res.apk (as `data:["framework-res"]`)
to the ravenwood runtime path.
Bug: 339614874
Test: m nothing --no-skip-soong-tests
Test: ATEST="atest --no-bazel-mode" $ANDROID_BUILD_TOP/frameworks/base/ravenwood/run-ravenwood-tests.sh
Do it on both aosp/main and goog/main
Flag: EXEMPT: Host side change only
Change-Id: I36516fec068a6359688a1dbedcb10a49c4631441
The validation checks that run for presigned apps currently expect that
privileged apps have uncompressed dex files but if
DONT_UNCOMPRESS_PRIV_APPS_DEXS is true (UncompressPrivAppDex false) this
doesn't have to be the case so make the validation consistent with this.
Also make TestAndroidAppImport_Preprocessed verify that extra arguments
to 'check_prebuilt_presigned_apk.py' are correct for both privileged and
non-privileged apps.
Test: m nothing --no-skip-soong-tests
Test: Add presigned, privileged android_app_import module with
compressed dex to PRODUCT_PACKAGES for aosp_cf_x86_64_phone,
lunch aosp_cf_x86_64_phone-trunk_staging-userdebug &&
m DONT_UNCOMPRESS_PRIV_APPS_DEXS=true no longer fails due to validation
error.
Change-Id: I7e22cf525cd9d99d1ecb24e4e2e99c3f9de48146
In preparation for a subsequent CL convert the test to use test
fixtures.
Test: m nothing --no-skip-soong-tests
Change-Id: I12f3802419105c7ffc8e00e95a069432c6b99d53
Adds a filegroup "metalava-config-files" whose contents are passed to
Metalava using the `--config-file` option.
Bug: 354699349
Test: m checkapi
Change-Id: I1e246517c2ef678e41214c975aecee97f9faca67
java_sdk_library's "magic" will always resolve to stubs when either of
the following conditions are met:
- the module sets `api_only` property to "true": this signifies that
the module does not create an implementation library, thus the "magic"
always returns stubs
- the module sets `default_to_stubs` property set to "true": this
signifies that the reverse dependencies outside of the same apex will
always get the stubs, even when the rdep does not specify the
sdk_version.
This change utilize this information and mark the top level sdk_library
as a stubs module when any of the above conditions are met, in order to
minimize the false positives in container violation errors.
Test: Run container enforcement and observe results
Bug: 354029496
Change-Id: I04b52c5662f635ab1837eb33a39f187ae8998238
It's the same mechanism that aconfig uses, but now it's exposed directly
to Android.bp files. (I had wanted to test it out there before making
it a property just in case it didn't work well)
Bug: 352385368
Test: go test -v -run TestJarJar.*
Change-Id: I72d797d08a250be93779ee4b038a1b8fc62123b7
