* Factor out the logic that sets up my_pack_module_relocations
from dynamic_binary.mk to pack_dyn_relocs_setup.mk.
* Use stand-alone relocation_packer only if my_pack_module_relocations
is true and my_use_clang_lld is false.
Bug: 80093890
Bug: 73768157
Test: build and boot with USE_CLANG_LLD=true
Change-Id: I7c4b5fcdce0754c57cff4acf86185cac65a26c40
Apparently empty PRIVATE_RS_TARGET_API is not the same as current,
pass 0 explicitly.
Test: forrest ub-renderscript-test
Change-Id: I8b2001b12161f07c404f78f63a80d82dffa1075f
Renderscript was using LOCAL_RENDERSCRIPT_TARGET_API := 0 to mean
current. Use current instead, and check for it when comparing
if LOCAL_RENDERSCRIPT_TARGET_API >= 21.
Fixes an issue where LOCAL_RENDERSCRIPT_TARGET_API := 0 was being
treated as < 21, causing the renderscript resources to be passed
to aapt2:
out/target/common/obj/APPS/RSTestBackward_intermediates/flat-res/zip_res.flata.contents/raw/bc32: error: resource file cannot be a directory.
Bug: 79481102
Test: m java FORCE_AAPT2=true
Change-Id: I7726d0463ce1693bc9653b4d869ced181c491036
LLVM r328258 turned on a feature called temporary dtor inlining by
default for all of C++ in clang-tidy. This feature appears to be
somewhat over-aggressive when objects are being passed by value. For
example, given:
void foo(std::unique_ptr<int> i);
void bar() {
auto x = std::make_unique<int>();
int *i = x.get();
foo(std::move(x));
*i = 99;
}
...clang-tidy will complain about `*i = 99;` being a definite
use-after-free. This is incorrect, however: `foo` could stash the
`unique_ptr` it's given in a global, or a class member, or ...
Until upstream fixes this bug, it's probably best to keep this disabled.
Bug: None
Test: Ran the analyzer across Android locally. Nothing broke; number of
complaints dropped significantly.
Change-Id: I742eedf598a72a533285d913d191bfbbf0ce1688
When USE_CLANG_LLD is true, my_use_clang_lld is true,
and *GLOBAL_LLDFLAGS will be used instead of *GLOBAL_LDFLAGS.
Bug: 73768157
Test: make checkbuild
Change-Id: I2598b72e9edde32f8e57df860571fd107c005540
* Current default is not using lld.
* When USE_CLANG_LLD or LOCAL_USE_CLANG_LLD is true or 1,
* Use *GLOBAL_LLDFLAGS instead of *GLOBAL_LDFLAGS.
GLOBAL_LLDFLAGS should call lld and with correct lld flags.
* set my_pack_module_relocations to false.
Bug: 73768157
Test: make checkbuild
Change-Id: I3e63cf8ae0865d01d2bc1f36e9304f4a5d092cb8
Normally, when building with VNDK, platform modules are not allowed to
link against vendor libraries, because the ABI of the vendor libraries
are not guaranteed to be stable and may differ across multiple vendor
images.
However, the vendor public libraries are the exceptions. Vendor public
libraries are vendor libraries that are exposed to 3rd party apps and
listed in /vendor/etc/public.libraries.txt. Since they are intended to
be exposed to public, their ABI stability is guaranteed (by definition,
though it is up to the vendor to actually guarantee it).
This change allows platform modules to link to vendor public libraries.
When the name of a vendor public libraries (whose stub is defined via
vendor_public_library in Android.bp) is listed in
LOCAL_SHARED_LIBRARIES, it is actually linked to the stub library which
is available to platform modules.
Bug: 74275385
Test: m -j
Merged-In: I063b43431196190b59332b1bcad59f267073f89f
Change-Id: I063b43431196190b59332b1bcad59f267073f89f
(cherry picked from commit c3f1356a3a)
Normally, when building with VNDK, platform modules are not allowed to
link against vendor libraries, because the ABI of the vendor libraries
are not guaranteed to be stable and may differ across multiple vendor
images.
However, the vendor public libraries are the exceptions. Vendor public
libraries are vendor libraries that are exposed to 3rd party apps and
listed in /vendor/etc/public.libraries.txt. Since they are intended to
be exposed to public, their ABI stability is guaranteed (by definition,
though it is up to the vendor to actually guarantee it).
This change allows platform modules to link to vendor public libraries.
When the name of a vendor public libraries (whose stub is defined via
vendor_public_library in Android.bp) is listed in
LOCAL_SHARED_LIBRARIES, it is actually linked to the stub library which
is available to platform modules.
Bug: 74275385
Test: m -j
Change-Id: I063b43431196190b59332b1bcad59f267073f89f
When building vendor modules with BOARD_VNDK_VERSION=current, the
API of the vendor modules will be current PLATFORM_VNDK_VERSION.
__ANDROID_API_FUTURE__ will be used as before if the version is a
CODENAME.
If BOARD_VNDK_VERSION is not "current", that means the VNDK version
of the vendor modules is BOARD_VNDK_VERSION.
Bug: 74833244
Test: Build and check boot.
Change-Id: Iccc603f1b5ffa066e74ba1357f3fbbcc13c58f76
Merged-In: Iccc603f1b5ffa066e74ba1357f3fbbcc13c58f76
(cherry picked from commit 65c680e285)
When building vendor modules with BOARD_VNDK_VERSION=current, the
API of the vendor modules will be current PLATFORM_VNDK_VERSION.
__ANDROID_API_FUTURE__ will be used as before if the version is a
CODENAME.
If BOARD_VNDK_VERSION is not "current", that means the VNDK version
of the vendor modules is BOARD_VNDK_VERSION.
Bug: 74833244
Test: Build and check boot.
Change-Id: Iccc603f1b5ffa066e74ba1357f3fbbcc13c58f76
* Use the definitions of ClangDefaultBase and ClangDefaultVersion
in build/soong/cc/config/global.go.
* scan-build is too old;
replaced by better clang-tidy's -clang-analyzer-* checks.
* WITH_STATIC_ANALYZER is still effective
to invoke ccc-analyzer or c++-analyzer.
Test: make checkbuild
Change-Id: Ifd159dd28e529d5d3c9807f5e391911365da8cdb
This build variable is unnecessary now that GCC is completely
unsupported for Android platform builds.
Bug: http://b/64032869
Test: Builds
Change-Id: I9d44ebd7129cb2bdfbb26d37922db19c6fb9efc9
* clang -fno-caret-diagnostics
suppresses warning source lines, carets, and
the stats line of "n warnings generated"
* clang-tidy -extra-arg-before=-fno-caret-diagnotics
only suppresses the "n warnings generated" line.
* Pass this flag and -quiet to clang-tidy when
WITH_TIDY is not 1 or true.
* Remove redundant quotation marks around -extra-arg-before.
Bug: 69051430
Test: normal build and build with WITH_TIDY=1
Change-Id: I9021ee315edc977c15dda6ab246a00182f31fad1
* Default builds calls clang-tidy only if enabled locally.
In this case, clang-tidy should be quiet.
* If WITH_TIDY is 1 or true, let clang-tidy emit default messages.
* Even with -quiet and all warnings are suppressed,
clang-tidy emits one line message of the number of suppressed messages.
This one-liner could be suppressed in future changes.
Bug: 69051430
Test: default build and build with WITH_TIDY=1
Change-Id: I3138ef45fc4372f904787bd1a391e8d88050f0b3
With LOCAL_PROTO_OPTIMIZE_TYPE set to one of the lite variants, the build system thinks
the proto should be compiled with lite, and will link in the line runtime libs. But if
the .proto files don't contain the directive to compile the source as lite
(option optimize_for = LITE_RUNTIME;), then the generated code will be full and the
libraries will be full, and it won't link.
Test: make
Merged-In: Ib8a135218d62fa42fa6448c49c97f7aeb2755c42
Change-Id: Ib8a135218d62fa42fa6448c49c97f7aeb2755c42
There are now two NDK sysroot timestamp files: ndk_base which depends
on the headers and stub libraries, and ndk which depends on ndk_base
and the static libraries. For building modules that need the NDK, we
only need to depend on the base sysroot.
Test: make native
Bug: None
Change-Id: I803032eda125a632d780c9cde10f47270491d0d2
The system_$(VER) can be set in LOCAL_SDK_VERSION, and the apk will use
android_system.jar at build time.
If LOCAL_SDK_VERSION is not defined and this module is installed in
vendor.img, LOCAL_SDK_VERSION is set to system_current.
Bug: 67724799
Test: 1. build && run on taimen
2. LOCAL_SDK_VERSION:=system_27 in ims.apk && build ims.apk && check
the vsdk_v27_intermediates.
Merged-In: I5b11c78b8fcd4a2f2a5e3b141527cd34dbe80018
Change-Id: I5b11c78b8fcd4a2f2a5e3b141527cd34dbe80018
(cherry picked from commit ef212cbe8d)
Use $(prefix)CLANG_SUPPORTED from Soong to control whether to use clang
for host builds or not. It's fairly general, but I only expect it to be
used to so that Windows builds can be switched between GCC and Clang
with a single switch.
Also separate native-host-cross from native-host so that it's easier to
build all possible windows modules.
Bug: 69933068
Test: m native-host-cross
Test: switch clang on and off, grep w64.*g++ out/build-aosp_arm.ninja
Change-Id: I205c5bc6739b72a539ece935381107ec1dd5bd48
This CL repoints static dependencies to their .cfi variants for CFI
enabled targets. It also disables CFI for host targets because the
version of ar intended for hosts does not have plugin support (which
CFI requires).
Bug: 67507323
Test: m -j40
Change-Id: Id11afd0c8765469858f406aace2a192afff6d042
* Skip Android.bp, prebuilt, and HEADER_LIBRARIES targets.
* If neither -Wno-error nor -Werror is used, add default -Wall -Werror.
However, if the build target is in a project
under one of ANDROID_WARNING_ALLOWED_PROJECTS, add only -Wall.
* Dump name of modules using -Wno-error or added -Wall -Werror
into file $(PRODUCT_OUT)/wall_werror.txt.
Bug: 66996870
Test: normal build
Change-Id: Ief8cf776e8613ed073a64dd271534688845d5c2c
Specify paths to instrument with coverage using space separated lists
instead of comma separated. This makes it easier to specify the lists in
a typical format in the device mk.
Test: add COVERAGE_PATHS to device-sailfish.mk, make -j60
Bug: 67998360
Change-Id: I26c67127d72bd7a153eb4a2c23eae4881a813d54
For module installed to /vendor partition, direct linking to the libs
marked as `vendor_available: false` is not allowed. The
Bug: 64730695
Test: Add vendor_available: false to libft2 and
libcompiler_rt. Add the two libs into LOCAL_SHARED_LIBRARIES of a vendor
lib (e.g. libdrm). Build fails with the link_type check error message.
Change-Id: Iaf23574ceddb0c087111e1d95997e9ddd60cdf87
libm is a default library for device builds, so default it for host
builds as well.
Also removes duplicate additions of -ldl, -lpthread, -lm and -lrt.
Test: m host
Change-Id: I6a07e12053090eb6997b79d4091c28ac9a9022de
Soong has a list of allowed host libraries. Check that list for make
modules as well.
Test: build/soong/build_test.bash (Linux&Darwin, aosp&internal)
Test: Add bad library in LOCAL_LDLIBS, ensure it errors
Change-Id: Icc48533487576998b049dc6c704c410172a91521
Instead of requiring every host module to specify -ldl -lrt -lpthread
and then break the mac build because -lrt doesn't exist, make them
implicit to match the behavior of modules built for the device.
Test: m -j checkbuild
Change-Id: I7b5b5289a41cd8f6e7dc2da59400955d867ee783
cxx_stl_setup is adding libunwind_llvm to the dependency even for
modules from soong. However, since this dependency was already handled
in soong, those modules don't need to go through it again.
Bug: 64815735
Test: BOARD_VNDK_VERSION=current mma -j (under /hardware/libhardware)
Change-Id: Id6e2547c752ec831aa56a14c0bf4db0a55d51243
Soong now adds .vendor suffix only for modules having both core and
vendor variants. Furthermore, names listed in LOCAL_SHARED_LIBRARIES
are correct (= have .vendor suffix when the dependent lib has variants).
Therefore, make does not need to force add .vendor suffix when parsing
modules from soong.
Bug: 37480243
Test: BOARD_VNDK_VERSION=current m -j <name> is successful, where <name>
is one of the vendor-only libraries in Soong. (i.e.
android.hardware.renderscript@1.0-impl)
Test: m -j does not break anything
Change-Id: Id8d0d01313c63496a10de4cd3ddb9f75180efef6
This is already blocked in Soong. Do the same for make.
Bug: 63553556
Test: BOARD_VNDK_VERSION=current m -j libSampleAuthJNI cannot be built
Change-Id: I04a89b3f9de903a6b0384f13d5a80886344c14b4
Since device_kernel_headers is marked as 'vendor:true', it is exported
to the make world as device_kernel_headers.vendor. Use the correct name
with the .vendor suffix.
Bug: none
Test: BOARD_VNDK_VERSION=current m -j does not complain about
device_kernel_headers.
Change-Id: If3eaa3c5832820c914ef427668d70fa8d8d0bf97
Device-specific kernel headers (TARGET_PROJECT_SYSTEM_INCLUDES) have
been added to the global include paths. However, since global include
paths are no longer provided when a lib is built for vendors, the
device-specific kernel headers becomes inaccessible. To solve this
problem, a pseudo header library 'device_kernel_headers' is defined
and it is added to all Android.mk defined lib/bins.
Also, in order to give this info to the soong world,
TARGET_PROJECT_SYSTEM_INCLUDES is exported as a new product config
DeviceKernelHeaders.
Bug: 62939405
Test: choosecombo to aosp_sailfish (or any other Pixel/Nexus targets)
BOARD_VNDK_VERSION=current m -j gralloc.msm8996 (or any other vendor
libs using vendor-specific kernel headers)
Change-Id: I3f0a2b893cc9352d6c13f3151e8834477d15c07b
Many vendor modules are using headers that are copied to
$(TARGET_OUT_HEADERS)/some/path via LOCAL_COPY_HEADERS. In order to let
them use the headers via #include <some/path/header.h>,
$(TARGET_OUT_HEADERS) is again added to the system include path.
Bug: 63340459
Test: BOARD_VNDK_VERSION=current m -j gives less 'cannot find header'
errors.
Change-Id: I472e74533b437653c76dc416f7f4527c0e90750d
Note that libdl is already (always) implicitly loaded
because libc.so has a dependency on it.
Also make sure that libc.so always preceding libdl.so
in dt_needed list.
Bug: http://b/62815515
Test: make
Change-Id: I94c9d676b7fa98438b452d24f6c3bbf93166c6a9
We have code that acts slightly differently when the static analyzer is
running, so that it can produce more accurate diagnostics (e.g. less
false positives). It uses __clang_analyzer__ to detect the static
analyzer.
When the static analyzer is run via clang-tidy, __clang_analyzer__
doesn't get defined.
Bug: None
Test: WITH_TIDY=1 m. clang-tidy now acts as expected in code made for
the static analyzer
Change-Id: Ib2a815c0bd67553af465b64207bb480fb52cfaf8
Soong already uses it, just not as a prebuilt module.
Test: Compare soong_to_convert.txt before/after
Change-Id: I6ffa0d967635d52b7b4f8b2b0465cd50254f22b9
libprotobuf-cpp-* libs are VNDK, However, since the '.vendor' prefix is
added before libprotobuf-cpp-* libs are added to my-*-libraries by
LOCAL_PROTOC_OPTIMIZE_TYPE, the clients are still linked against to the
platform version of the protobuf libs. As a result, link type error
occurs.
In order to avoid this, '.vendor' prefix is appended after
my-*-*libraries are all updated.
Test: normal build successfull
Test: add vendor_provided:true to libprotobuf-cpp-* and build successful
with BOARD_VNDK_VERSION=true
Change-Id: I1ed7978f03e0db56b4cbee86e286c0787f7a3afc
All the new features are turned off for now, since multiple branches and
products need to be verified before they can be turned on. So everything
should behave the same as today, except for no partition-based
warnings.
Instead of the current link type checks that happen during the build,
run as many as possible immediately after loading all the Android.mk
files. If we're allowing missing dependencies ('mm',
ALLOW_MISSING_DEPENDENCIES, tapas, etc), we'll defer the link type
checks to during the build. If we're not allowing missing dependencies,
we'll produce a better error message to the user about the missing
dependencies.
See core/main.mk for a description of the storage format.
This also remove the partition-based type checking. It hasn't worked all
that well, particularly with ASAN builds. The new VNDK checks will
handle the most pressing cases.
Test: Verify all link_type files and dependencies are the same:
grep link_type: out/build-aosp_arm64.ninja | sed -E "s/ rule[0-9]+//" | sort
Change-Id: Id643658b9d9e84f99f5db0d526aad88c1f5d3417
The build system currently uses -fPIC for position independent
libraries, and -fpie for position independent executables. However,
these levels are incompatible and conflict when an executable includes
a static lib with the -fPIC flag - such as when building with
LTO. This CL changes the flag for executables to the compatible -fPIE
instead, which is equivalent for all architectures except PowerPC.
Bug: 30227045
Test: ENABLE_CFI=true m -j40 checkbuild # builds and boots
Change-Id: Ieb20bc2528300d88ff05e8787c5840ab6f03fb38
Soong modules may be split into both /system and /vendor variants.
Ensure that Make libraries link against the correct version based on
whether LOCAL_USE_VNDK is set.
Bug: 36426473
Bug: 36079834
Test: build-aosp_arm64.ninja files are the same before/after
Test: attempt building with BOARD_VNDK_VERSION:=current
Change-Id: I229bc290373743406275f3ca16081eae04c27987
Instead of using the NDK headers and libraries, add LL-NDK specific
headers and library stubs for VNDK users. This allows us to provide an
expanded liblog interface.
Test: aosp_arm; m -j
Test: Enable BOARD_VNDK_VERSION on aosp_arm; m -j
Change-Id: I0197f44c91218c73b9567a05320c91a2baaae39b
This lets Soong pass -isystem or -I as necessary (or potentially, even
other cflags in the future).
This is not available for Android.mk use, nor exposed directly to
Android.bp users.
Test: m -j
Merged-In: Id37d4692d5fbddce467bd777903b20169f44dd6e
Change-Id: Id37d4692d5fbddce467bd777903b20169f44dd6e
This lets Soong pass -isystem or -I as necessary (or potentially, even
other cflags in the future).
This is not available for Android.mk use, nor exposed directly to
Android.bp users.
Test: m -j
Change-Id: Id37d4692d5fbddce467bd777903b20169f44dd6e
BOARD_VNDK_VERSION controls which version of the VNDK is used for the
build. We only support compiling against the current VNDK, and we don't
support installing old prebuilt VNDK libraries yet, so ensure this
variable is set to "current".
Once we support installing old prebuilt VNDK libraries, we'll also need
to disable building modules that use the VNDK.
Test: build-aosp_arm.ninja is the same before/after
Test: Ensure there are no boards that set BOARD_VNDK_VERSION
Test: Set BOARD_VNDK_VERSION := invalid
Change-Id: Ic26c9f44e356a0734f522b4538cb866d5b901d43
external/dbus and external/dbus-binding-generator was removed from
the manifest.
Bug: 31602715
Test: make checkbuild
Change-Id: Iea0277720acad8ac0fa630b8745f90fb3c3b5f00
Instead of using all_objects and translating .o to .gcno, use
cpp_objects, gen_cpp_objects, c_objects, gen_c_objects, objc_objects,
and objcpp_objects. Fixes bug with .s files breaking coverage builds.
Bug: 35843991
Test: lunch sailfish-userdebug && make -j60 NATIVE_COVERAGE=true
COVERAGE_PATHS="vendor/google_devices/marlin/proprietary/mm-camera/mm-camera2"
Change-Id: Ib2f8eef889f075ec82f79d8c3ae0fe31962b653d
Specify list of paths to exclude from coverage instrumentation.
Test: make NATIVE_COVERAGE=true COVERAGE_PATHS=hardware/interfaces
COVERAGE_EXCLUDE_PATHS=hardware/interfaces/graphics
Bug: 35769817
Change-Id: I3747fdddb381101b9ebf51909b9686e820148aad
Statically link libwinpthread.a, so that we don't have to ship a
libwinpthread dll with all of our windows exectuables.
Bug: http://b/31665213
Test: wine adb.exe
Test: wine fastboot.exe
Change-Id: I96414e980d8894f8f6a58c1c7bbd9dc5e0f5169c
We currently use static libraries without any source files to represent
header libraries, but Soong actually has cc_library_headers. So to
export those in a separate namespace from static libraries, implement
them in Make as well.
This also adds a nice pretty-warning / pretty-error macro that can be
used to print out standard warning messages pointing to the real source
file having the problem.
Test: Use a header library exported by Soong in a Make module
Change-Id: I3486539e247524cb82a20620745fc7be03014e14
Soong only supports exporting include directories under the current
directory. So bring non-local directories up as a potential problem in
soong_to_convert.txt
Test: m -j $OUT/soong_to_convert.txt, inspect
Change-Id: I7a15b92e10a1d8b8d3496c6f0529a0d0824f301e
We support these in Soong now, so stop marking them in soong_to_convert.
Test: Compare soong_to_convert.txt
Change-Id: I89befdfd8c4eaa04ab626d8be090765a1b879d8b
There are no users left, so remove all of this.
Test: lunch aosp_arm-eng; m -j native
Test: build/tools/kati_all_products.sh on aosp and internal master
Change-Id: I32f5c8b470a43dd203d7e20c192167630e4e6888
Add BOARD_VNDK_VERSION and LOCAL_USE_VNDK to specify the version of the
VNDK that will be used globally, and whether to use the VNDK on a module
basis.
If the board is using the VNDK:
* LOCAL_COPY_HEADERS may only be used by modules defining LOCAL_USE_VNDK
* LOCAL_USE_VNDK modules will compile against the NDK headers and stub
libraries, but continue to use the platform libc++.
* LOCAL_USE_VNDK modules will not have the global includes like
system/core/include, but it will use device-specific kernel headers.
This change does not attempt to enforce any linking constraints, that
will come in a later patch.
Test: out/build-aosp_arm.ninja is identical before/after
Change-Id: Icce65d4974f085093d500b5b2516983788fe2905
This make aidl-cpp write out a dep file that ninja parses correctly.
Test: Manually inspect ninja depfile
Change-Id: I4890a91eb29a6388e17b1ffac23a3dc0ffe6c212
Ninja doesn't need the phony make targets, in fact, the parser doesn't
handle them correctly and ends up duplicating the dependencies. This
shrinks a `m native` deps file from 54M -> 35M on AOSP.
Test: Compare out/build-aosp_angler.ninja before/after
Test: wrote a tool to dump the .ninja_deps, added dedup feature, files
identical after dedup.
Change-Id: Iec7a9a0739e8678c1f4db79c68e423a39b9aad4b
There are some source files in our tree that use some text processing to
find source files, and end up with source file names that look like:
/../../dir/file.c
This isn't fixed by our dotdot workarounds because the filename doesn't
start with '../', so strip the initial '/' from any source files, since
they are all relative to $(LOCAL_PATH) anyways.
Bug: 32514380
Test: Build internal source with this problem
Test: Compare build-aosp_angler.ninja before/after
Change-Id: If869419c3d3ba35d55f6e19d9332515695e47d24
Host builds were compiling without -Wl,--no-undefined because of an ASAN
issue. Pass -Wl,--no-undefined for host builds unless sanitzers are
enabled. Also fix LOCAL_ALLOW_UNDEFINED_SYMBOLS on darwin, where
disallowing undefined symbols is the default.
Test: m -j host
Test: m -j SANITIZE_HOST=address host
Bug: 32305815
Change-Id: I0344c321610c55eb8b2092b01e50e4a93a0e85df
Change COVERAGE_PATHS to access a comma-separated list of paths
instead of a space-separated list to make for easier escaping in
configuration files.
Test: tested using local bullhead build with coverage on lights and nfc
Change-Id: Iec014c8b4fbefcbcc5d865e6acba92bf6364ee3a
Fix a whitespace bug causing coverage to be enabled on all modules
whenever more than one module is specified. This is caused by
comparing whitespace to the empty string, which are not equal. Fix
by stripping the output to check for path matches.
Test: tested using local build with coverage enabled on NFC, lights
Change-Id: I61a775cfd135f94ea1e1ee9fdbed3c64c5bb1a2e
* Add -Werror if LOCAL_PATH is in the WARNING_DISALLOWED project list,
or not in the WARNING_ALLOWED project list.
Test: Build for major targets.
Change-Id: I12235ee1ca1c1837530693699e705e1955275565
The toc optimization had been disabled for host builds to ensure that
the timestamp of the final binary changed whenever its implementation
changed, in order to support rerunning host tools that were modified
during incremental builds. However, only the final install rule must be
re-run to update the timestamp, and not the link rule.
Update the shared library install dependencies to use normal
dependencies instead of order-only dependencies for host modules, and
then enable the the toc optimization for host modules. If the
implementation of a library changes it will be reinstalled, and
libraries or binaries that depend on it will also be reinstalled.
Bug: 26015464
Test: m -j; touch art/disassembler/disassembler_x86.cc; m -j, verify
out/host/linux-x86/bin/oatdump is updated
Change-Id: I0a14decc1994eb55ad269d841943aef66e320c63
$(shell) isn't particularly fast in Kati, and they have to be executed
both when reading the makefiles and determining whether the ninja file
needs to be regenerated.
Right now, the regen time is mostly hidden because we run them in
parallel. We've also configured it to ignore any commands that contain
"echo", "date", or the output directory. That happens to remove most
commands that contain side effects, so running them in parallel is fine.
But the side effects contain some important things, like the clean up
necessary when switching products. So I'm removing those filters, and
then we'll need to run the shell commands in sequence, since there will
be side-effects. That makes regen take longer though, so use pure-Make
implementations instead of $(shell) where possible.
This set of changes reduces aosp/master aosp_arm64-eng build $(shell)
usage and time by 2/3:
*kati*: func shell time: 3.135095 / 709
*kati*: func shell time: 1.067331 / 236
Bug: 30947985
Test: Manual test lines for math functions
Test: Compare build-aosp_arm64.ninja before/after
Change-Id: I4fc9d6318957992921972994f277c17918e7e1eb
This shouldn't be necessary for most makefiles, since we add
$(LOCAL_GENERATED_SOURCES) to this already included, but for prebuilts
that aren't actually prebuilts (Soong modules), we need to pass
dependencies so that the generated headers are built before being used.
Bug: 31742855
Test: Use, inspect build-*.ninja
Change-Id: I8d9f675af639d3f40780d48a016fc079a82531c2
This was printing "error:", but not actually triggering an error.
Instead of trying to write a single line bash script to handle this,
move the actual check into python. This allows us to print all of the
errors for a single module before triggering the failure.
Also updates the warning format and the warn.py script to properly parse
these warning. Many of the java:sdk -> java:platform warnings are false
positives due to the lack of LOCAL_SDK_VERSION markings on prebuilts.
Individual tags can be marked as warnings now, which lets us check for
system libraries linking against vendor libraries (which won't work on
AOSP). I'm not sure this is a completely valid check, which one reason
that it's just a warning.
Test: m all_link_types (with some missing libs commented out)
Change-Id: I333e418c9a4511b7c7e826891ae481da08fbf6f9
We want to build things in CTS (things built against the NDK) with
integer overflow checks. Some projects in the tree also make explicit
calls to the overflow checked builtins, and those projects need to be
built with the NDK (external/dng_sdk is the specific example that
brought this up).
Test: make checkbuild tests
Bug: None
Change-Id: Iab27dd0d931f723140de242049fa72f3b1ce0fb8
Dan Willemsen