The comment on the interface of GetDirectDep states if there are
multiple deps with different tags, the first will be returned; however,
the current behavior is to panic if there are multiple deps. The
behavior now:
* a single dep, return the module and tag
* a single module with different tags: return module with first tag
* multiple modules: panic
* no module: return nil, nil
Bug: 186488405
Test: ~/aosp/build/bazel/ci/bp2build.sh
Test: m nothing
Change-Id: Id1e7315e7874b4a683ad7357ed2793822315821f
This allows us to remove libasync_safe from the mixed builds denylist.
Test: mixed_libc.sh CI script
Change-Id: Ibafd231284864078bf30340f919d39e5098843ce
Previously, buildBootImage had to be called for the default boot image
config before the art boot image config as it created some rules which
were specific to default boot image. The creation of these rules was
encapsulated in Config().Once(..) to ensure that they were only ever
created once.
This change extracts that functionality out of buildBootImage to the
calling method which means that they are only called once and so no
longer need protecting against being called again.
Bug: 177892522
Test: m droid
Change-Id: I464477de1a08df15e577873a9accf7db2bc088d1
Previously, the buildBootImage function maintained a list of
missingDeps that was returned from findAndCopyBootJars and was passed
to various other methods that created ninja rules that depended on the
files created by findAndCopyBootJars. The purpose of that list was
basically to record that findAndCopyBootJars could not find some files
so that the subsequent ninja rules could be marked as being in error.
This changes avoids having to explicitly propagate the missing deps to
each rule by instead having findAndCopyBootJars generate an error rule
instead of a Cp rule for any missing file. That means that any attempt
to build any rule that depends on the missing file will fail when the
missing file is created.
Bug: 177892522
Test: Add an unknown module to PRODUCT_BOOT_JARS
lunch art_module_arm64
- Run each of the following targets and made sure it failed with a
message indicating that the unknown module was missing.
m SOONG_ALLOW_MISSING_DEPENDENCIES=true out/soong/module_arm64/dex_bootjars/boot.prof
m SOONG_ALLOW_MISSING_DEPENDENCIES=true out/soong/generic_arm64/dex_bootjars/boot.bprof
m SOONG_ALLOW_MISSING_DEPENDENCIES=true out/soong/generic_arm64/dex_bootjars/updatable-bcp-packages.txt
Change-Id: Ib7196c128419fbffdfeb61c1c5d47a3b08d1e132
The code that has been removed triggers in many cases when it should
not, e.g. an unknown module on the bootclasspath. The code was
originally added to try and ensure that if the attempt to add useful
error messages failed it would help debug the issue. However, it ended
up just creating more failures.
If this does become an issue then this will be revisited.
Bug: 177892522
Test: Add an unknown module to PRODUCT_BOOT_JARS
m SOONG_ALLOW_MISSING_DEPENDENCIES=true
Change-Id: I08dbd4b29d301f3f1f78a0fb396687062b47deb8
Take two: this time, put libasync_safe on the mixed build denylist
because it seems to break things.
Test: Presubmits + testing the offending action using a Ninja
invocation. "m droid dist" also running.
Change-Id: I7912b6670980776d981d874dab157b0268693ab2
OS-specific exclude_srcs and srcs will be done in a follow-up, due to
complexities from merging multiple select statements together.
Test: TH
Bug: 186153868
Change-Id: I01b881e9a5a7cd41b4a507f8be7e9e65eab37bdc
In a non-top level Android.bp file, exclude_srcs was not working at all
due to a bug in expandSrcsForBazel. GlobFiles was expanding a glob
relative to root, but the expandedExcludes list was relative to the
module dir, causing the glob function to not consider the
expandedExcludes list at all.
Add tests to demonstrate that this is working now.
Test: TH
Bug: 186388919
Change-Id: Ice8254231d085b39126e91b823a09ec328ee0ae0
Not needed anymore for bp2build-incremental since https://android-review.googlesource.com/q/topic:no-include-check.
Not needed for mixed builds either, since cc compile actions aren't sandboxed.
Fixes: 186488830
Test: treehugger and go tests
Change-Id: Ib5d4908dcce6bf910a653c457bb251d726e717d4
Further restrictions on a restricting build configuration. Before, we
restricted certain paths from being included. Now, we restrict certain
paths from including.
This adds system/libfmq for now, but I'm planning to add more in the
future, as they are cleaned up.
Bug: 35624006
Test: m nothing
Change-Id: I913f190f93ab63059ee0372d342daa8f5584806b
Remove llndk_library in favor of cc_library with llndk.symbol_file.
Bug: 170784825
Test: m checkbuild
Test: TestLlndkLibrary
Change-Id: I43580976589a7a2a176d7442be53fa043c0c8324
Set llndk.llndk_headers: true to allow a cc_library_headers
module to be used by a cc_library module with llndk.symbol_file set.
Bug: 170784825
Test: TestLlndkHeaders
Change-Id: Ib65a4b70717dc9a54ae30f2991485bb1bb9b8409
Header libraries don't compile anything, so they don't need the
default stl and libclang_rt dependencies. Trimming the dependencies
simplifies using header libraries for LLNDK libraries, which need
extra variants that don't exist for the default dependencies.
Test: m checkbuild
Change-Id: Ib9864c3cc7048b2cc52acf304830777b65a51877
Vendor public libraries are checked for product modules, but never
reached because the VNDK check occurs first, and wouldn't work anyways
because vendor_public_library did not have the product_available
property. Reorder the rewrite checks and add VendorProperties, and
add a test.
Bug: 178231622
Test: vendor_public_library_test.go
Change-Id: Idbd2802fbd134d22c30600762cb0465633703506
This reverts commit cba7b0c43c.
Reason for revert: mixed_droid failing: libbacktrace fails due to libasync_safe, linkerconfig (and others) due to duplicate symbols
Change-Id: Ib6dd6bbb6f76048a7abd0970d457cf75921beae6
Roman Kiryanov