Instead of passing the entire contents of
$(COMPATIBILITY.$(suite).FILES) to eval, which may keep that string
around, delay the evaluation of that, and the new files until inside the
eval.
This saves ~2.8GB: 7.4GB -> 4.6GB of ckati max resident memory for a
relatively small internal build. It also saves ~10% of the makefile
loading time (81 -> 73 seconds).
Test: build-aosp_arm.ninja is identical
Change-Id: If45a4796f1bbf6d67dff388ea877a6115a4e06f4
This is a partial revert of I43b645658f468c23a5b9ebcfcd9d4516537db540
On at least a generic_x86 build internally:
art/build/Android.gtest.mk:121: error: overriding commands for target `Uncompressed', previously defined at art/build/Android.gtest.mk:101
Bug: 77611511
Test: none
Change-Id: I78ca65e6f0c81f09e7da848eda797b3a8f97a521
Many boards have warnings like this, saying that we defined a build
rule, but later something else came in and overrode it with something
else:
art/build/Android.gtest.mk:677: warning: overriding commands for target `test-art-target-gtest-cmdline_parser_test'
art/build/Android.gtest.mk:674: warning: ignoring old commands for target `test-art-target-gtest-cmdline_parser_test'
Beyond the obvious problem of replacing the rule with something else,
target-specific variables can be combined as well, leading to some very
strange problems.
Since so many boards still have problems like this, but we don't
currently have any global problems, add a flag so that we can mark
boards as not broken. This should prevent regressions while we clean up
the individual problems.
Once the non-broken devices number significantly more than the broken
devices, we'll switch this default. And once they're all cleaned up this
variable will become obsolete, and these warnings will always be errors.
Bug: 77611511
Test: lunch aosp_arm-eng; m nothing
Test: lunch aosp_marlin-eng; m nothing
Test: build_test on all downstream branches
Change-Id: I43b645658f468c23a5b9ebcfcd9d4516537db540
Our binary was rather old, and for a variety of reasons we haven't kept
it updated. We've been running into a handful of reliability issues that
would have been fixed with an update, and a few reproducibility /
correctness issues that may or may not be fixed with newer versions.
For local no-change full rebuilds, ccache can still save ~35% of the
build time (but adds a few minutes to initially populate the cache). But
most local uses should be using incremental builds anyways, not clean
rebuilds. Or you're doing builds of different configurations, which
wouldn't be cache hits either, and would make your cache even larger.
At a large scale, we haven't seen a significant performance difference
between having ccache on or off. This may be different if you've got
very good build locality, or a very large cache -- but if you've got
good build locality, it's reasonable to do incremental builds (not for
release builds, and while running `m installclean` in between builds).
So for our cases, we'd prefer the stability and correctness of not using
ccache, but if you still want to use ccache, continue setting USE_CCACHE
and also set CCACHE_EXEC to the path of your ccache executable.
Bug: 32748498
Bug: 72408185
Test: performance testing of USE_CCACHE=false vs true locally
Test: turned off ccache for a collection of targets
Test: CCACHE_EXEC=/usr/bin/ccache USE_CCACHE=true m
Change-Id: I7117fe3107bd98521051ae343038a38f7e855502
LOCAL_VINTF_FRAGMENTS/vintf_fragments are
used to specify what manifest fragments should be installed
by a target.
Test: fragments get installed to the right location
Test: broken fragment gets detected
Test: boot device and verify service is working and manifest is updated
Test: verify OTA package contains fragments
Bug: 66917623
Change-Id: I21abe65a31b8c3d255c8ccd80e102ff3acb23105
Start deprecating add_lunch_combo, preferring a list of common choices
in each AndroidProducts.mk file.
This list will be validated so that we don't have typos:
https://android-review.git.corp.google.com/c/device/generic/car/+/619533
Or targets that no longer exist.
Bug: 77599627
Test: lunch
Test: lunch 34
Test: lunch aosp_x86_64-eng
Test: lunch aosp_<tab>
Change-Id: Ie0ddaa94cbd6cee26584f56f1706a8ec1333f87e
The install filter is speed-profile in order to enable the use of profiles
from the dex metadata files. Note that if a profile is not provided or if
it is empty speed-profile is equivalent to (quicken + empty app image).
Test: build & install an app
Bug: b/30934496
Change-Id: I895b5e2a1d8ff478f56b0f08cb589789d0743b54
This CL adds a list of components to enable CFI for on Arm64 product
configs.
This list is not directly used anywhere, but product configs can
inherit this to easily enable CFI for all tested components.
Bug: 63926619
Test: Inherit the product config; m -j60; verify that CFI is enabled
Change-Id: I56b6dfd5495639974a19623bc7aea9efeb9acff6
The function along with LOCAL_TEST_DATA are very helpful, but it took me
some extra minutes to figure out the issue with the arguments.
Test: N/A
Change-Id: I2c2878fca99ab77ea881d86635c27c34f9ecf855
This is a list of modules which should be installed by default, but only on
builds which are set up for Address Sanitizer (via SANITIZE_TARGET=address).
Additionally, add sanitizer-status to PRODUCT_PACKAGES_DEBUG so that all
userdebug builds will have the binary. Currently, the module has the
"debug" tag but it may go away in the future.
Change-Id: I0bea4faf9c2a65380292471437e51ef8324b5af3
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
This CL changes PRODUCT_CFI_INCLUDE_PATHS to be included in all
product configs by default. To maintain the status quo, the sanitizer
logic has been modified to only respect this product config for Arm64
devices (where this was previously enabled).
Bug: 63926619
Test: m -j60 # the device still has CFI enabled thanks to the default
opt-in
Change-Id: I22788d92be881d3290568488f5458c85e02ee8c7
If there are no symbols to keep (i.e. all function symbols are also in
the dynsym table), then the $@.keep_symbols file has size 0.
When objcopy parses a --keep-symbols file, it has a special case for
0-sized files where it silently fails (exits with status 1, no error
message). On the other hand, objcopy is happy with a file containing no
symbols, as long as there is some whitespace or a comment. Avoid the
special case by appending a newline to keep_symbols.
Bug: b/62815515
Bug: b/77242617
Test: manual
Change-Id: I348593164fe78ece5c85db3d28c7daccce3a52fb
AUX modules are not necessarily using the same toolchain
as the TARGET modules. they don't have to depend on
this library.
Change-Id: Ib50cf0eb26c257ae3eb69a43aa1a12c41d5d39b0
Signed-off-by: Alexey Polyudov <apolyudov@google.com>
This variable is equal to the sepolicy version for tip-of-tree development.
Bug: 74622750
Test: ./build/soong/soong_ui.bash --dumpvars-mode -vars PLATFORM_SEPOLICY_VERSION
is unchanged.
Change-Id: I2a62f7399fe245ec0c681e9a789da6b1b83dd9d8
