Most of the fields in the bootImageConfig/Variant structs are assigned
inside a Once func so are guaranteed to be only set once. However, some
are assigned outside. This change adds comprehensive tests for those
structs and verifies that the constant fields are preserved and the
mutated fields have the correct value.
The check for the constant fields is added in a new TestBootImageConfig
test.
The check for the mutated fields is added into
TestSnapshotWithBootclasspathFragment_ImageName as that test checks an
art bootclasspath_fragment in the following configurations:
* source on its own
* prebuilt on its own
* source and prebuilt with source preferred
* source and prebuilt with prebuilt
It reveals a couple of interesting facts:
* All the *installs fields are set to the same value irrespective of
whether the source or prebuilt is preferred. The information is
constructed solely from information already within the
bootImageConfig/Variant and so can be moved within Once.
* The licenseMetadataFile is incorrect when prebuilt is preferred.
That is due to both the source and prebuilt modules setting it and
the source module always wins as the source module depends on the
prebuilt so always runs its GenerateAndroidBuildActions after it.
Those issues will be cleaned up in following changes.
Bug: 245956352
Test: m nothing
Change-Id: If917cfbcb3b1c842a8682d51cc1ee1fed1c51add
Sometimes it is necessary for some functionality to be disabled while
running in unit tests, e.g. functionality that requires external
information such as error prone tools and configuration. Sometimes it
is necessary for some functionality to be enabled while running in
unit tests, e.g. functionality that makes state available for
testing but which is not necessary at runtime.
Previously, that was done by checking to see if TestProductVariables
was nil. This change adds a method to abstract that.
Bug: 245956352
Test: m nothing
Change-Id: I7845b79328e7180623161a9bf897568089da4e4f
Previously, CommonOs was not added to Config.Targets for the test which
caused AndroidMk to crash when trying to determine if a target needed
the 2ND_ prefix added to some make variables.
Bug: 245956352
Test: m nothing
Change-Id: Ia815537ee5ce8e83df9a01c57250aa888453d138
* changes:
Remove deviceInstalls from bootImageVariant
Remove profilePathOnHost from bootImageConfig
Document bootImageConfig and bootImageVariant structs
aosp_riscv64 isn't ready for build_test yet, it still requires
ALLOW_MISSING_DEPENDENCIES=true to build.
Bug: 250918230
Change-Id: I9151fa84bf6ad0fa67dd06d3268311ae716b5185
Test: none
Bug: b/239044089
Test: run build/bazel/ci/incremental_build --bazel-mode com.android.adbd and see the additional metrics in summary.csv
Change-Id: I2c3b9cead2ee43c700a9ad52e669a64aab1499ce
* changes:
Move fuzzer's CollectAllSharedDependencies into GenerateAndroidBuildActions
Support AllowMissingDependencies in prebuilt_apex modules
Support AllowMissingDependencies for apex dependencies
Add AllowMissingDependencies support for prebuilt_etc module with no src property
Make OutputFileForModule work for AllowMissingDependencies
Fix panics when target arch is riscv64
Previously, a java module with sdk_version: "system_server_current",
would only be able to access the system server or public API of a
java_sdk_library. This change allows it to access the system server,
module lib, system and public APIs in that order.
The apiScope structs define the characteristics of each of the
different API scopes used as required by the java_sdk_library. They are
organized into a hierarchy which is used for two different purposes.
The first purpose is to define an extension hierachy. If scope X
extends Y then X provides a superset of all API elements (classes,
fields, methods, etc) provided by Y. That is reflected in the fact that
the .txt file for X would be a delta on the .txt file for Y. So, system
extends public and so system_current.txt only contains additional API
elements to add to current.txt.
The second purpose is when a java_sdk_library/import is asked to
provide a specific API scope. e.g. a library that has:
sdk_version: "module_current"
will ask each of the SDK libraries it depends upon for a module-lib
API. However, not all of them will provide an API for that scope. In
that case it will find the closest suitable API scope.
Previously, it did that by traversing up the API extension until it
found an API scope that it did provide and return that. As
system_server_current extended the public API that meant that a library
which has:
sdk_version: "system_server_current"
would provide a system server API if available, and if not fall
straight back to public. That meant that the library could not access
system or module-lib APIs even though it is running in the system
server which should be able to access all APIs.
One way to fix this would have been to just have system server API
scope extend module-lib but that would have had a number of nasty
side effects:
* It would have created a significant overhead as every module that
provides a system server API would also have to provide a module-lib
and system API, along with their corresponding .txt files.
* Each existing java_sdk_library that provided a system server API
would need those .txt files created.
* Generating sdk snapshots for older releases would have been more
complicated.
* It would have confused developers.
All of that would be unnecessary because the system server API scope is
intended to be provided by libraries that are used solely by the system
server so there is no point in them providing anything other than a
system server API.
So, instead a separate access hierarchy was added which is the same as
the extension hierarchy for all existing scopes except for the
system server scope, which instead of just being able to access the
public API will be able to access the module-lib scope, which can in
turn access system and it can in turn access public.
That achieves what we want which is a library that is loaded into the
system server to be able to access all API scopes.
Bug: 204176972
Test: m nothing
Change-Id: I854df63fcaeba32afbc1eb0d1a501238022673d0
The use of this field to return information from buildBootImageVariant
up the call stack to one of the callers resulted in data races being
detected. This change simply passes the deviceInstalls up the call
stack.
Bug: 245956352
Test: m nothing
go test -race ./sdk/... -run TestSnapshotWithBootclasspathFragment_ImageName -test.count 100
# Run the previous command without this change and sometimes it
# shows the data race around deviceInstalls. When run with this
# change it reports no data races.
Change-Id: I3c73920dcb17a6c89a63c6a9c3a0bb049a98a690
The use of this field to return information from bootImageProfileRule
up the call stack to one of the users resulted in data races being
detected. This change simply returns the profile path back up the call
stack.
Bug: 245956352
Test: m nothing
go test -race ./sdk/... -run TestSnapshotWithBootclasspathFragment_ImageName -test.count 100
# Run the previous command without this change and sometimes it
# shows the data race around profilePathOnHost. With this change
# that data race is not reported. Although there is still another
# data race.
Change-Id: I03b09e514cc94f2a6c9d5117d3b2f130cc2e4f5b
Add warnings to the structs to explain how they are supposed to be used
and deprecate fields that are used incorrectly.
Bug: 245956352
Test: m nothing
Change-Id: I090698287b96fd37102b88beb5d7252977bddc54
Apexes for new architectures have to increase their minSdkVersion
to the minimum supported version for the architecture.
Bug: 250918230
Test: lunch aosp_riscv64-userdebug && m ALLOW_MISSING_DEPENDENCIES=true nothing
Change-Id: I49220cbec628f1508709741dc56b62aaac7786d9
Make rust and cc fuzzers collect their shared libraries once in
GenerateAndroidBuildActions and store it for later use by the
packaging singleton. Also use android.OutputFileForModule to get
the paths. Together this will fix fuzzers that depend on architecture
specific prebuilt shared libraries that are missing a prebuilt for an
architecture when building with AllowMissingDependencies.
Bug: 250918230
Test: lunch aosp_riscv64-userdebug && m ALLOW_MISSING_DEPENDENCIES=true nothing
Change-Id: I154a6f3a767c883e9fe7067003615db73ee78e2d
Currently, tagging a symbol with #apex is not required when the symbol
is in a non-NDK library. However, this is considered dangerous because
such a symbol will automatically be promoted to NDK APIs when the
library is promoted to an NDK library. When that happens, the native API
council won't be able to notice the promotion because promoting a
non-NDK library into an NDK library doesn't require an update of the
map.txt file, but Android.bp only.
To prevent that, we should mandate #apex (or #systemapi if the api is
provided by the system, not apex) tag for Mainline APIs regardless of
whether the library the API belongs to is an NDK library or not.
Bug: 184712170
Test: m
Change-Id: Iad589b606bc6ed06ef1237f4c9f0c22518b0db75
By default, ndkstubgen does not omit NDK symbols as long as they
satisfy the API version and the CPU architecture requirements. This
change adds a new flag --no-ndk to ndkstubgen which is used to
override the default behavior. If the flag is set, NDK symbols are
omitted leaving only the annotated symbols (e.g. #apex, #systemapi,
etc.).
This will be used for stub libraries that are not part of NDK. So far,
symbols in such libraries haven't needed to be annotated as #apex, and
that has caused a confusion that those symbols belong to NDK. The
follow-up change will ensure that those symbols are always annoated as
either #apex or #systemapi so that their roles are clearly visible.
Bug: 184712170
Test: atest test_ndkstubgen
Test: atest test_symbolfile
Change-Id: Ic8d2c7d0b32bdef79f7563621035e60f406e4131
* This is a temporary work around until the RBE input processor
can handle those flags.
* It's okay to remove some cflags in clang-tidy command
if they do not affect tidy check results.
Bug: 248371171
Test: enable RBE; RBE_CLANG_TIDY_EXEC_STRATEGY=remote; make tidy-soong_subset
Change-Id: If499bd7e1f0f89ccc3b7f7df7d67cfc64dc67028
Arch-specfic prebuilt_apex modules may be missing prebuilts for new
architectures. Mark the module as missing dependencies and allow
build analysis to continue when AllowMissingDependencies is set.
Bug: 250918230
Test: lunch aosp_riscv64-userdebug && m ALLOW_MISSING_DEPENDENCIES=true nothing
Change-Id: I79f0c20a0c9443fe443f9ed9bc846aa649f6b6a6
Use android.OutputFileForModule instead of cc.Module.OutputFile,
which will already handle the AllowMissingDependencies case.
Bug: 250918230
Test: lunch aosp_riscv64-userdebug && m ALLOW_MISSING_DEPENDENCIES=true nothing
Change-Id: I605ae9afe06ac450adec1d2a856e529e45ed5cb5
Arch-specific prebuilt_etc modules may be missing source files for new
architectures. Allow build analysis to continue when there is no
source file when AllowMissingDependencies is set.
Bug: 250918230
Test: lunch aosp_riscv64-userdebug && m ALLOW_MISSING_DEPENDENCIES=true nothing
Test: TestPrebuiltEtcAllowMissingDependencies
Change-Id: I647c7305339e3ed80283be5e59e6f4ef15ae2384
Report missing output files as missing dependencies when
AllowMissingDependencies is enabled. This will fix
AllowMissingDependencies builds where a dependency module is
present but missing support for a specific architecture.
Bug: 250918230
Test: lunch aosp_riscv64-userdebug && m ALLOW_MISSING_DEPENDENCIES=true nothing
Test: TestOutputFileForModule
Change-Id: I95e96e3e7cb3df7bf678ed628b45baf659addbad
Fix panics in api_level.go and apex.go when using riscv64 as the target
arch.
Bug: 250918230
Test: lunch aosp_riscv64-userdebug && m ALLOW_MISSING_DEPENDENCIES=true nothing
Change-Id: I85c7685f3d14fa2dc7ffbcdea7f490feca304ef7
The module types in scope are
1. ndk_library
2. ndk_headers, versioned_ndk_headers
3. api_domain
These modules will no longer generate Bazel targets in bp2build
workspace. A new soong mode `api_bp2build` will be used to generate API
specific targets in a separate Bazel workspace at out/soong/api_bp2build
Test: go test ./bp2build
Change-Id: I01d06b6d8b8364c0d56d6d3f07f8f8bf21db31d7
The generated Bazel workspace will only contain api specific targets.
This is feasible since these targets do not have any cross dependencies
with the targets in the bp2build workspace
The advantages of a new mode are
1. Does not pollute bp2build workspace with api targets
2. Does not block api targets with the current allowlist conversion
mechansims in bp2build
(In the future we might want to combine these two workspaces)
A Soong module type will generate a Bazel target if it implements
ApiProvider interface
Test: m apigen
Test: m nothing
Change-Id: I69c57ca6539f932e0ad554ce84a87fb7936fdba0
Overrides properties will now be captured in json flag files, which will
be copied to installed vendor snapshot modules.
Bug: 216567575
Test: soong test && manual install
Change-Id: Ife5e84b126e798fba7802b9cff000c9197756cb9
Context
- Android-<target>.mk currently does not contain information about partition for its dependent unembedded jni libraries, but only lists the name of the unembedded jni libraries.
- If an android_app module depends on an unembedded jni library that is located in a different partition, make cannot find the library.
Implementation
- Create a string field partition in jniLib struct.
- Add variable "LOCAL_SOONG_JNI_LIBS_PARTITION_<target>", an array of mappings of the name of the jni library to its partition.
Bug: 154162945
Test: m
Change-Id: I6b8e1272ff59dc70e3dd6ce8c6c8e4686dad76df
Add barebones riscv64-linux-android support. This should be enough
to add riscv64-specific entries to Android.bp files, but can't
actually compile anything until there are riscv64 toolchains.
Test: arch_test.go
Change-Id: I0dcc7e797d9352dd38243be908a7f19004ff3db1
The certificate module is handled the same in Soong between android apps
and apexes, so share the bp2build code as well.
There are a few changes in this CL:
- If override_apex.certificate is unset, the generated apex also unsets
it. This prevents the generated apex from using the base apex's
certificate, which is most likely incorrect (e.g. google variant using
the cert for the aosp variant). Instead, rely on the default
certificate handling in the macro.
- If the certificate prop is a string, then it gets generated into
certificate_name in order to disambiguate. This behavior is identical
to android_app.
Test: added various unit tests.
Bug: 249089160
Fixes: 249089160
Change-Id: I99e18964ff546429a985d0f64dc21e2c69d35d9d
