Failures in strip.sh are still occurring with darwinStripPool set to
10, try 5.
Bug: 132822437
Test: none
Change-Id: I2df247c89b6d8f6e741d4e17057b06ff9ec72b40
New strip option named keep_symbols_and_debug_frame, that will keep the
symbols and the .debug_frame. This is meant for use by libc.so only on
arm32. Other libraries might want to use it to keep better unwinding
information on device.
Bug: 132992102
Test: Built libc.so with this option and verified that it contains
Test: the .debug_frame section.
Change-Id: I823a28199dec8316e8b26fe31ff9f17e6b11d406
strip.sh can use a file descriptor per .o file when run on .a files,
which can hit the system file descriptor limit on darwin. This
causes failures when manay variants of libgcc_stripped are built
simultaneously. Put all strip rules on darwin into a pool that
limits them to 10 concurrent processes, which will limit the file
descriptor usage to ~7500.
Fixes: 132822437
Test: no mention of darwinStripPool in out/soong/build.ninja on linux
Test: m libgcc_stripped on darwin
Change-Id: I3d4fbbd8d44d2e9059a79df113ab95336ec2c658
We use libgcc as fallback for symbols not present in libclang_rt
builtins, however we didn't know what exact symbols were being used,
some may not be intended to fallback.
Create libgcc_stripped, which only contains unwind symbols from libgcc.
Bug: 29275768
Test: bionic-unit-tests
Change-Id: I5b349fa6138e51663bf3b67109b880b4356da8e8
This reverts commit 61166dc047.
One difference from the earlier change is that import libraries are now
using the '.lib' extension instead of '.a' to prevent clash with
AdbWinApi.a.
Bug: http://b/110800681
The following flags that the binutils linkers support are not
available in lld for Windows:
-soname
--no-undefined
-rpath
Windows also uses "import libraries", which are stub libraries used only
for linking. The binutils linkers accepted a DLL and treated them as an
import library. But lld issues the following error:
lld-link: error: ...DLL: bad file type. Did you specify a DLL instead
of an import library?
To resolve this, pass '-out-implib=libFoo.lib' to lld when linking
libFoo.dll to get lld to generate an import library. Add libFoo.lib as
an implicit output to the 'ld' build rule.
Rewrite the shared libraries when building a library/binary to use the
import library instead of the DLL. As a side-effect, this also uses the
newly-created AdbWinApi.lib that's alongside
development/host/windows/prebuilt/usb/AdbWinApi.dll
Test: Run Windows tests (go/android-llvm-windows-testing) and check
absence of regressions. Also check that the following commands pass:
$ adb.exe devices
$ fastboot.exe devices
Change-Id: I34e07d345e0207086ac8e8ea12525d8c322b20fd
So that <module>/gen/yacc/... is (re)created by a single rule, previous
files are removed, and location.hh is in the build graph when it is
produced.
Test: treehugger
Change-Id: I2f6e47ea07f315e10ae1cb8ad50697e7123d0285
This commit add `--llndk` to the command line option for
`create_reference_dumps.py` as well.
Bug: 124620774
Bug: 130324828
Test: lunch aosp_arm64-userdebug && make
Change-Id: I117864209b8cf8c934fd3aa1086b7691c1c449e3
Looks like the reason for it existing has been fixed. It should probably
just be removed.
Bug: None
Test: WITH_TIDY=1 m
Change-Id: I770b2fec4ac44f265ff31731c9c0bd4da14d5b0f
Add a proto.plugin property to allow specifying a custom protoc
plugin to generate the code.
Fixes: 70706119
Test: m am StreamingProtoTest
Change-Id: I1ecdd346284b42bbcc8297019d98d2cd564eb94c
Using blueprint.Rule for protoc commands was causing code duplication
because there was no good way to run the same protoc for cc, java and
python but then run custom source packaging steps for java and python.
Move most of the code into a common function that returns a
RuleBuilder, and then let java and python add their own commands at
the end of the rule.
Bug: 70706119
Test: All Soong tests
Test: m checkbuild
Change-Id: Ic692136775d273bcc4f4de99620ab4878667c83a
clang -Wl,--out-implib doesn't update its output file if it hasn't
changed, always restat the outputs.
Fixes: 129553377
Test: m checkbuild && m checkbuild
Change-Id: I2cb19e1b8280fd69c8a9b2143f55760ec05630f8
Bug: http://b/110800681
The following flags that the binutils linkers support are not available
in lld for Windows:
-soname
--no-undefined
-rpath
Windows also uses "import libraries", which are stub libraries used only
for linking. The binutils linkers accepted a DLL and treated them as an
import library. But lld issues the following error:
lld-link: error: ...DLL: bad file type. Did you specify a DLL instead
of an import library?
To resolve this, pass '-out-implib=libFoo.a' to lld when linking
libFoo.dll. Add libFoo.a as an implicit output to the 'ld' build rule.
Rewrite the shared libraries for a library/binary to use the import
library instead of the DLL. As a side-effect, this also (correctly)
uses the AdbWinApi.a that's alongside
development/host/windows/prebuilt/usb/AdbWinApi.dll
Test: Run Windows tests (go/android-llvm-windows-testing) and check
absence of regressions.
Change-Id: I15a178589aa6882caa6e7e38650cc6ef48109764
This commands adds `--llndk` to `create_reference_dump.py` command if
the module with incompatible ABI changes is an LLNDK or a NDK lib. This
prevents people from uploading a CL that simply removes ABI dumps (the
result even when the uploader runs the command).
Bug: 124620774
Test: Make an incompatible ABI changes in libaaudio and see the flag.
Change-Id: I9f07f879193bd4d73f5924baff298d28d6fe1506
... except for Darwin Mach-O, as it is not supported by llvm-strip.
Test: m checkbuild
Bug: 119221035
Change-Id: I021637b6dd3530bff1f563f2ec7c2168e1083b7e
This commit adds a header_abi_checker section so that the library owner
can have a fine-grained control over the ABIs that must be checked.
For example, a library "libexample" may have following configurations:
cc_library {
name: "libexample",
header_abi_checker: {
symbol_file: "libexample.map.txt",
exclude_symbol_versions: ["LIBEXAMPLE_PRIVATE"],
exclude_symbol_tags: ["platform", "apex"],
},
}
Bug: 122845490
Test: Add header_abi_checker to libc to filter out LIBC_PRIVATE
Change-Id: I60cfea868f815afe6213c242ed0ca818161d55c6
This commit fixes a typo related to abidiffs. The generated abidiffs
should be copied to `$$DIST_DIR/abidiffs` instead of
`$$DIST_DIR/abidiff`.
(See also. https://android-review.googlesource.com/796458/)
Bug: 123491909
Test: lunch aosp_x86-userdebug && make dist
Change-Id: I89c13580344a27cac5ea0d00497f5bba3227cf61
.s files (unlike .S files) aren't run through the preprocessor, so -M*
doesn't actually write out a depfile.
Since our ninja is now going to be verifying that the depfile is created
(https://android-review.googlesource.com/861510), don't specify a
depfile for .s files.
Bug: 121058584
Test: apply https://android-review.googlesource.com/861510
Test: cd external/libavc; mma
Change-Id: I1697aa020c63639317c8f4771147026601ae72fc
llvm-ar takes a --plugin argument but it is ignored for compatibility,
so passing it has no effect.
Change-Id: I4fc51d226d66cf2a43462d3d4ccc12e6e5ebb226
We're only using it to distribute files in case of failure, which isn't
well supported currently, but can be handled for now by using the
DIST_DIR environment variable during the command execution.
This was at least one cause that we'd be re-running Soong during every
build server build, as the DIST_DIR values are unique.
Test: m dist
Change-Id: Ibd5e6b6c46695350de80b745bfb6a6aa685033a0
This commit switches the include path of header-abi-dumper clang headers
to `prebuilts/clang-tools/${os}-x86/clang-headers`. This decouples the
cross git repositories dependencies.
Bug: 111579848
Test: development/vndk/tools/header-checker/utils/create_reference_dumps.py
Change-Id: I53083298ca7be39b3fbaffb123146c12c189e7c3
Clang is always used now, so we can remove all the GCC checks. Removing
GCC-specific configuration will happen in the next CL.
Test: m
Change-Id: I4835ecf6062159315d0dfb07b098e60bff033a8a
Instead, hardcode the ~dozen paths into build/soong/Android.bp, which
will unblock removing more GCC support.
Bug: 114286031
Test: m
Change-Id: I2508432e00b1469141f01e667f3c6a2fe30cd805
* PATH_TO_CLANG_TIDY_SHELL is ${config.ClangTidyShellPath},
wich is "build/soong/scripts/clang-tidy.sh"
Bug: 110538415
Test: build with WITH_TIDY=1
Change-Id: I27d7be095c998583fa9ad4d705aa1332bd018140
* C++ source files could be compiled with "-fwhole-program-vtables"
in cppflags. We need to filter out those flto dependent flags
in flags.CppFlags and use flags.toolingCppFlags instead of
flags.cppFlags.
Test: build with WITH_TIDY=1
Change-Id: Ic2b0f99b95a5f0422f879226e6f6060cab71456a
* Some flags are accepted by clang but not clang-tidy.
They could cause the diagnostic-unused-command-line-argument warning.
Flag -flto messed up the -I flags.
* Add clang-diagnostic-unused-command-line-argument to
default clang-tidy checks.
* Move CLANG_TIDY_UNKNOWN_CFLAGS to build/make/core/clang/tidy.mk.
Bug: 111850071
Bug: 111885396
Test: build with WITH_TIDY=1
Change-Id: Iabeeb27715acf83ef6aafe3e77206b9a01a0d889
* Pass LLVM_{OBJCOPY,STRIP} to makefile rules.
* strip.sh is used only in soong-only mode.
It will use llvm-strip and and llvm-objcopy
when --use-llvm-strip is given.
Keep flags of strip.sh in alphabetic order.
* Tested build/soong/scripts/build-ndk-prebuilts.sh
by adding "UseClangLld": true, to soong.variables file.
Bug: 80093681
Test: build/soong/scripts/build-ndk-prebuilts.sh
Change-Id: I612267304eea434c7a33cc086b27b577d5f64094
-check-all-apis currently, produces noise for some libraries
(eg:libstagefright_soft_vpxenc), so remove this till a good solution
arrives.
Bug: 79928919
Test: create abi reference for libstagefright_soft_vpxenc; m -j
libstagefright_soft_vpxenc.vendor; no warnings about struct LAYER_CONTEXT
layout changing.
Change-Id: I8f30700c1b619c72504ca9aef69075f6e334ba5f
This change is to let people to just copy & paste the update command
Test: break the ABI of libbinder intentionally and make.
The message includes absolute path to the create_reference_dump.py tool.
Change-Id: If7703d64be1f51ea4b054aa1e815276e1d6dbc14
When we're doing ABI diffing, we run something like:
(run-abi-diff) || (echo && exit 1)
When `dist` is added, we attempt to copy out the result:
(run-abi-diff) || (echo) && (copy && exit 1)
But this always fails, since it will always run exit 1. We really want
the parenthesis around the entire second section. This change switches
it to:
(run-abi-diff) || (echo && (copy) && exit 1)
This matches the behavior before Iae25374fe937a0cbe8a8ddf9e23c3bc1f62bbb2a
Test: treehugger on P
Change-Id: Ic39da4d4630ee950b811ec7854d55d65c52f8661
Test: create reference dump for libjpeg; add exported function to libjpeg;
m -j libjpeg.vendor, build fails with helpful message.
Test: create reference dump for libjpeg; add exported function to libjpeg;
m -j libjpeg.vendor dist DIST_DIR=dist, build fails with helpful message.
Change-Id: Iae25374fe937a0cbe8a8ddf9e23c3bc1f62bbb2a
* Suppress more noisy new warnings at global level.
* Add -no-pie to partial link .o files, with -r.
* Revert workaround of b/72706604, no need of
-Wl,-plugin-opt,-emulated-tls
* Filter out clang 7.0 unknown flag "-Wno-extended-offsetof"
Bug: 72706604
Bug: 72412006
Test: make checkbuild
Change-Id: I7ff45465c4bd771991f42b40f68dc35586045656
Instead of open-coding the logic of whether there is one, or where to
find it.
Test: diff out/soong/build.ninja without dist
Test: diff out/soong/build.ninja with dist specified
Change-Id: Ia3f1ef335e2d6e2175343338d04867d778a50300
So that the Path and similar functions can be used directly, without
manually adding something like configErrorWrapper (it just uses it all
the time now).
Test: out/soong/build.ninja is identical
Change-Id: I8cb524b09a84d0b8357d7c3501c71270af411e17
Instead of whichever xz happens to be on the PATH.
Bug: 36130900
Test: prebuilts/build-tools/build-prebuilts.sh
Change-Id: Ic9b66062ca7947f0eca9d839c0864da492d4ef71
This is needed so that we can embed the paths of lsdump files in a file,
in order to make lookup faster, while creating reference dumps.
Test: mm -j64 in external/libjpeg-turbo for aosp_arm64_ab;
out/soong/make_vars-aosp_arm64_ab.mk contains
SOONG_LSDUMP_PATHS := <list of lsdump paths>
Change-Id: I04608429e1add307cc1ee79d2f0c348fb041613c
Historically, we've always passed '-I .' as the first argument to
protoc, essentially treating all proto file package names as their full
path in the android source tree. This would make sense in a monorepo
world, but it makes less sense when we're pulling in external projects
with established package names.
So keep the same default (for now), but allow individual builds to opt
into using local paths as the default names with
'canonical_path_from_root: false'. A cleanup effort and/or large scale
change in the future could change the default to false.
As part of this, run protoc once per input proto file, since the flags
may need to change per-file. We'll also need this in order to specify
--dependency_out in the future.
Bug: 70704330
Test: aosp/master build-aosp_arm.ninja is identical
Test: aosp/master soong/build.ninja has expected changes
Test: m
Test: Build protobuf test
Change-Id: I9d6de9fd630326bbcced1c62a4a7e9546429b0ce
LLVM-AR does not allow passing --plugin options more than once. The
--plugin ARFLAGS that lto want to add, may already exist if sanitizer is
also turned on.
Fixed this by adding a new bool Flags.ArGoldPlugin. Set this variable to
true whenever LLVM gold plugin is needed for ArFlags. In function
TransformObjToStaticLib(), add this option to arFlags using global value
${config.LLVMGoldPlugin} if the bool value is true.
Bug: http://b/73160350
Test: build the image with make and succeeded.
Change-Id: I62785829b0a4b663225926e4aed98defc1b6da2c
(cherry picked from commit 4917049f6e)
Bug: http://b/72343691
Change https://android-review.googlesource.com/c/572758, in addition to
dependencies from the compiler flags, also marked all exported
dependencies as implicit. This can cause lots of unnecessary
recompiles. This change moves exported dependencies back as order-only
dependencies.
Test: 1. mma in art after changing profile_compilation_info.h triggers
only a limited number of recompiles.
2. verify that changes to PGO profile files trigger recompiles.
Change-Id: Icb0f4cd2b6da0add3b6e5206661e6aa7a577602f
This commit adds `extends: "name"` property and provides basic support
to VNDK extensions. This is the simplest example:
```
cc_library {
name: "libvndk",
vendor_available: true,
vndk {
enabled: true,
},
}
cc_library {
name: "libvndk_ext",
vendor: true,
vndk: {
enabled: true,
extends: "libvndk",
},
}
```
A vndk extension library must extend an existing vndk library which has
`vendor_available: true`. These two libraries must have the same
`support_system_process` property.
VNDK-ext libraries are installed to `/vendor/lib[64]/vndk` and
VNDK-SP-ext libraries are installed to `/vendor/lib[64]/vndk-sp` by
default.
If there is a matching abi-dumps in `prebuilts/abi-dumps`,
`header-abi-diff` will be invoked to check for ABI breakages.
Bug: 38340960
Test: lunch aosp_walleye-userdebug && make -j8 # runs unit tests
Test: lunch aosp_arm-userdebug && make -j8 # build a target w/o VNDK
Test: Create a lsdump for a vndk lib, add an exported API to vndk lib,
and build fails as expected.
Test: Create a lsdump for a vndk lib, create an vndk extension lib with
extra API, and build succeeds as expected.
Test: Create libutils_ext, add an extra function to libutils_ext, and
call it from a HIDL service.
Change-Id: Iba90e08848ee99814405457f047321e6b52b2df0
While diffing abi, for libraries which do not export headers, use
-advice-only.
Test: Create reference dump for libdl; mm -j64; header-abi-diff gets
invoked with '-advice-only'.
Test: Create reference dump for libjpeg; mm -j64; header-abi-diff gets
invoked without '-advice-only'.
Bug: 71768219
Change-Id: I678346081f15088388f223f1276d0b0c42d2a8c2
Version scripts were earlier used as a symbol map, primarily for llndk
libraries. Since they do have stub libraries which contain symbols, we
shall use them instead, to internally form symbol maps in
header-abi-linker.
Test: mm -j64 in bionic/libdl, header-abi-linker gets invoked with
-so <so-file> rather than -v <version-script>
Change-Id: Ifb67dc34457a997f37cc9f71ca16ad068e9b44c1
This properly sets -format=gnu for all non-Darwin targets, because
llvm-ar is cross-platform (but defaults to the host environment).
Bug: http://b/71618641
Test: m checkbuild
Change-Id: Ic5bce7e237ea0cadfa1f96ea151d41115d8c56b8
