The lists of public/private APIs used for non-SDK API restriction
enforcement used to be generated by Doclava. In Q we're switching
to resolving android.jar (and other SDK stub JARs) against the boot
class path directly to avoid lowering Doclava's high-level view of
the framework and issues related to it.
This patch adds a new build rule which invokes `hiddenapi` on all
boot class path dex files simultaneously. The tool generates two
text files - one with public and one with private APIs.
Bug: 79409988
Test: m out/target/common/obj/PACKAGING/hiddenapi-private-list.txt
Change-Id: I826d8ac513a41a1d6d5e29a8300073158cff5acf
Merged-In: I826d8ac513a41a1d6d5e29a8300073158cff5acf
(cherry picked from commit a39db48511)
Recent CL added a flag to an unzip command which does not restore
the timestamps of extracted files. This is necessary because the
extracted files are the output of the rule and thus the timestamp
must be newer than the timestamp of the inputs.
Unfortunately, the sdk_mac build has an old version of UnZip (5.52
vs 6.0) which does not support the flag. Replace it with find|touch.
Test: m
Change-Id: Ib1f218af4547a1d604b46e68b6ef50d45cb91ee9
Java modules built with make and Soong currently invoke `hiddenapi`
with their own rules. Refactor the definitions so that Soong uses
make's definition. This will be useful for a subsequent CL which
needs to invoke `hiddenapi` on all boot dex files simultaneously.
Bug: 79409988
Test: m
Change-Id: I85105e650605d37c3ee8200d7f0d40c297756fcb
The `hiddenapi` build tool now expects the first argument to be
a command name. Change its build rule accordingly.
Bug: 79409988
Test: make
Change-Id: I7e34331c487f490d1de6a10c4139768e9774f03e
Bug: http://b/62424007
This breaks 64-bit Windows executables. Probably due to the Clang
migration, the entry-point need not be set even for 32-bit executables.
Test: 32-bit and 64-bit aapt2_tests.exe and ziparchive-tests.exe can
start. (There are some failed tests, though).
Change-Id: I1dea9137c79c5f26b493640cfae0856c1a7ea301
We only support relocation packing inside lld now, remove calls to
the standalone relocation packer. Any shared library built with
LOCAL_USE_CLANG_LLD := false will no longer have relocation packing.
This will also prevent relocation packing on prebuilt shared
libraries.
Also enables relocation packing on Darwin since lld will work on
Darwin.
Bug: 110715614
Test: m checkbuild
Change-Id: Ib4d4bc86f164d05dd78455c8e7605d4b4f5a2a77
This reverts commit 054f210fd5.
Reason for revert: Revert it again since BUG of 109736180 already be fixed.
Local verified for different build targets of
walleye_coverage-userdebug, marlin_coverage-userdebug
and make sure they can build pass.
BUG: 109736180
Test: lunch walleye_coverage-userdebug; make adbd_test
BUG: 73082610
Test: cd packages/apps/DocumentsUI/tests/; mma
Modify AndroidTest.xml to a not well-formed xml; mma
Then, it should build fail.
atest -v DocumentsUITests
Change-Id: Iadfcb28a4e3f685a66697de21f57ca5bdb8b7c9e
Relanding with change to not run on the darwin build since host init
verifier is not possible there.
Bug: 36970783
Bug: 110477913
Test: builds pass without failures
Test: builds fail with failures in init files used via LOCAL_INIT_RC,
PRODUCT_COPY_FILES, or TARGET_PREBUILT
Change-Id: Iae93705a9320159ed97328e615eaa3bc150e6442
A java module exports SDK library names that it is using directly or
indirectly via its dependencies. Manifest fixer uses the SDK lib names
to automatically add <uses-library> tags for the apk.
The SDK library names are exported via exported-sdk-libs file in make.
From Soong, they are exported via LOCAL_EXPORT_SDK_LIBRARIES flag.
Bug: 77575606
Test: m -j
Change-Id: I4e7a9bdc5bf845af85168abf55f8063900bacc72
Sdk addon has been allowed for libraries built with Android.mk. But some
of the libraries included in the Sdk addon should be built with
java_sdk_library. module-stubs-files function is modified for supporting
java_sdk_library modules.
Bug:77577799
Test: make -j
Change-Id: If289e5aa5b5016901ff9b69da8fc7f7d8138a456
Since this is a directory separator, it causes a good deal of
strangeness in the build whenever we include a module name in a path.
It becomes particularly problematic if used together with ".."
Test: build_test on downstream branches
Change-Id: I344eca0db3346cd6ffabff767c34159c85ebc051
Bug: 36970783
Test: builds pass without failures
Test: builds fail with failures in init files used via LOCAL_INIT_RC,
PRODUCT_COPY_FILES, or TARGET_PREBUILT
Change-Id: Ia98793776b365733ca8f0174dc228326a6edf041
* PRIVATE_OBJCOPY is still set to objcopy and used
when llvm-objcopy cannot replace objcopy.
Bug: 80093681
Test: build/boot with USE_CLANG_LLD=true
Change-Id: I4bd8a53f74864b49e007f0c5f1b0f4b16e121135
