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
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
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
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
Dumping files in the *_intermediates/src causes incremental build
problems, because we don't know when to delete them if they are
no longer generated. Switch to outputting a srcjar instead, and
pass it in to the turbine and javac compiles.
Bug: 73885582
Test: m checkbuild
Change-Id: Ibef28507522339d930c09426d0c98372eb864cb6
Merged-In: Ibef28507522339d930c09426d0c98372eb864cb6
(cherry picked from commit 2d1cddd567)
Replace RenderScript.stamp with a srcjar, with a resource zip as
an implicit output. This makes the renderscript rules self contained,
which will make it easier to move them before aapt.
Bug: 73885582
Test: m checkbuild
Change-Id: I12d33e52019aebac6ea33271939228e4690a4173
Merged-In: I12d33e52019aebac6ea33271939228e4690a4173
(cherry picked from commit e00e2fae52)
Use zipsync instead of extract_srcjars.sh.
Also fixes an issue where unzipping a resource zip would
ignore anything that wasn't a *.java file.
Bug: 73885582
Test: m checkbuild
Change-Id: I1c25f65afa196614879071c8655866febebac185
Merged-In: I1c25f65afa196614879071c8655866febebac185
(cherry picked from commit 3f99cd5933)
Some javac and turbine compiles have no source files but sources
from srcjars. Continue with compilation if the java source list
is empty but the srcjar source list is not.
Bug: 73885582
Test: m checkbuild
Change-Id: Ica412de2dd1b1bfc12c01c3eafca884220d42771
Merged-In: Ica412de2dd1b1bfc12c01c3eafca884220d42771
(cherry picked from commit 0c0adb993c)
Allow aapt2 rules to take zipped resource files similar to the
way the javac rules take srcjars. For now, unzip the files using
extract_srcjars.sh, and then use aapt2 compile --dir. Eventually
it could be done with aapt2 compile --zip.
Bug: 73885582
Bug: 74574557
Test: m checkbuild
Change-Id: I97ea3ff305e7a9ef2a908b4cf0bca903af62c6b1
Merged-In: I97ea3ff305e7a9ef2a908b4cf0bca903af62c6b1
(cherry picked from commit bc0c494d3a)
Allow java modules to take jars containing .java files as an input.
For turbine, pass them directly to turbine. For javac, extract them
to a directory and pass a list of the files to javac.
Bug: 73885582
Test: m checkbuild
Change-Id: Ib4351a5b4b165a55e5fb02ee2680466db368983a
Merged-In: Ib4351a5b4b165a55e5fb02ee2680466db368983a
(cherry picked from commit 11e2d55c0f)
Sharding is not used in make any more, the large java modules
(libcore and framework) have moved to Soong. Keeping sharding
support complicates java compiling, so remove it.
Bug: 73885582
Test: m checkbuild
Change-Id: I2932f9e96bfc8f2ebf82095b16d6c0b3c51613f0
Merged-In: I2932f9e96bfc8f2ebf82095b16d6c0b3c51613f0
(cherry picked from commit 5ef6e79c21)
Dumping files in the *_intermediates/src causes incremental build
problems, because we don't know when to delete them if they are
no longer generated. Switch to outputting a srcjar instead, and
pass it in to the turbine and javac compiles.
Bug: 73885582
Test: m checkbuild
Change-Id: Ibef28507522339d930c09426d0c98372eb864cb6
Merged-In: Ibef28507522339d930c09426d0c98372eb864cb6
(cherry picked from commit 2d1cddd567)
Replace RenderScript.stamp with a srcjar, with a resource zip as
an implicit output. This makes the renderscript rules self contained,
which will make it easier to move them before aapt.
Bug: 73885582
Test: m checkbuild
Change-Id: I12d33e52019aebac6ea33271939228e4690a4173
Merged-In: I12d33e52019aebac6ea33271939228e4690a4173
(cherry picked from commit e00e2fae52)
Use zipsync instead of extract_srcjars.sh.
Also fixes an issue where unzipping a resource zip would
ignore anything that wasn't a *.java file.
Bug: 73885582
Test: m checkbuild
Change-Id: I1c25f65afa196614879071c8655866febebac185
Some javac and turbine compiles have no source files but sources
from srcjars. Continue with compilation if the java source list
is empty but the srcjar source list is not.
Bug: 73885582
Test: m checkbuild
Change-Id: Ica412de2dd1b1bfc12c01c3eafca884220d42771
The latest version of turbine requires a -- after the javac
options, and uses multiple --bootclasspath arguments instead
of a javac-style colon separated list.
Test: m javac-check
Bug: 74339924
Change-Id: I3f694ac375eab61954d4fe63c1166a18314bf75f
Allow aapt2 rules to take zipped resource files similar to the
way the javac rules take srcjars. For now, unzip the files using
extract_srcjars.sh, and then use aapt2 compile --dir. Eventually
it could be done with aapt2 compile --zip.
Bug: 73885582
Bug: 74574557
Test: m checkbuild
Change-Id: I97ea3ff305e7a9ef2a908b4cf0bca903af62c6b1
Allow java modules to take jars containing .java files as an input.
For turbine, pass them directly to turbine. For javac, extract them
to a directory and pass a list of the files to javac.
Bug: 73885582
Test: m checkbuild
Change-Id: Ib4351a5b4b165a55e5fb02ee2680466db368983a
Sharding is not used in make any more, the large java modules
(libcore and framework) have moved to Soong. Keeping sharding
support complicates java compiling, so remove it.
Bug: 73885582
Test: m checkbuild
Change-Id: I2932f9e96bfc8f2ebf82095b16d6c0b3c51613f0
Silence warnings when merging header jars that were missed by
Ifb67dbf7734b5f53941d110f578c4fe642e36005.
Bug: 69316739
Test: m checkbuild
Change-Id: I5cfea62e46d99f9b315a589d8915bf62307220aa
Instead of relying on whatever version of xmllint is on the host system,
build and use the version in external/libxml2.
Test: diff build_aosp-marlin.ninja, expected changes.
Test: m $(xmllint targets in build_aosp-marlin.ninja)
Test: introduce xml error, build fails
Change-Id: I39579f06db3777e3b5c8dda7c7541c25a35887b2
So they can be used by art/test/Android.run-test.mk.
Also remove the line wrapping, which was inserting extra characters
at the beginning of the result, and add a sanity check.
Bug: 26275726
Test: no change to out/build-hikey960-test-art.ninja
Change-Id: I35b54c258a1dc76bb1a8a758ec4a08915a2b226d
This was ok for apk because they are in different parent directory.
Test: marlin built
Bug: 63920015
Bug: 30972906
Change-Id: I5ed00824fda9e7e0a99c825402753c21a06ba6c5
This reverts commit 4a2926a44e.
Reason for revert: The original CL caused two issues:
1) Increased app startup because `profman` failed to verify boot
jars and created an empty profile for the boot image. All methods
in the framework therefore ran interpreted/JITed. New flag
"--skip-apk-verification" has been added to `profman` to fix
the issue in CL Iad614e47feada293218c11cff985a8012ca49a0f.
2) Caused a crash in Mockito CTS tests because jvmti did not
remove hiddenapi access flags from boot jars before using them.
This has been fixed with CL Ibc9a96a6b541c06844f276db009ac29514f7a3bb.
Bug: 64382372
Bug: 72550707
Test: make test-art-host
Change-Id: Ie8c58714455662f042f56b2977b01b75c5ec7d2f
Kati has a `--warn` option that adds some extra Kati-specific warnings
-- like using undefined make functions, undefined user functions, likely
bad variable lookups, etc. Some of these are bugs, others are fine doing
nothing. This fixes up all of them in the core build system.
It also complains about $(eval) usage in a recipe not being recommended.
Those aren't handled as part of this change.
Bug: 72661763
Test: build-aosp_arm.ninja is identical before and after
Change-Id: I8e00af142a7745236d3ad4efc9e91ec3ce71a511
This reverts commit b7f00ed2af.
Bug: 72550707
Bug: 64382372
Test: atest CtsInlineMockingTestCases
Reason for revert: Causing invalid dex files to be passed to jvmti
agents. This causes CtsInlineMockingTestCases and
other tests to fail.
Change-Id: I50129e94791b9e2bef919fe398b4250ae461195b
When building with OpenJDK 9's javac with -source 1.9 -target 1.9,
a runtime image (with system modules) is used via the --system
command line flag, instead of --bootclasspath.
The runtime image only contains the default libraries (libcore)
but is missing other libraries that might be on the bootclasspath
for a particular build target.
This CL fixes compilation for this case by adding the missing
jars onto the classpath instead.
This already used to work but was broken by CL [1]. I attempted
to conditionally revert some of that CL's changes to java_common.mk
for the case of language level 1.9, but couldn't get it to work.
Therefore this CL follows a different approach.
[1] http://r.android.com/519552
Bug: 38177569
Test: Treehugger
Test: Ran "EXPERIMENTAL_USE_OPENJDK9=true make checkbuild docs"
and spot-checked some javac invocations to confirm that the
system modules libs were not included in the bootclasspath,
but other libs were.
Change-Id: I48fd11aac9b310bfa58dee0f9cfb3ef33f10bca8
Restrictions on usage of private APIs require encoding new information
into the dex files of the boot class path. ART now contains a new build
tool called `hiddenapi` which takes three lists of class member
signatures (blacklist, light and dark greylist), finds the class members
in their respective dex files and modifies their access flags in place.
This patch invokes the `hiddenapi` tool on all JARs in
PRODUCT_BOOT_JARS. For Java libraries built with Makefiles the tool is
invoked after the dexer directly on DEX files. For Soong-built
libraries, the build system has to unzip the JAR produced by Soong,
apply `hiddenapi` and rezip again. This is due to the fact that the
PRODUCT_BOOT_JARS variable is not available to Soong.
Bug: 64382372
Test: m
Change-Id: I6ce897d204459c8b6f46ed49e0909ff76c08a9ed
This reverts commit f22aca1e47.
Switching to use -systemjars instead of -libraryjars means that it is no longer
necessary to filter out the junit and android.test classes from jars to which
Proguard is being applied.
Bug: 30188076
Bug: 69156675
Test: make checkbuild
Change-Id: I36e14d28e01517927b458d69ade8ccb2c79a3340
- Move logic to uncompress dexs in an APK in dex_preopt_odex_install.mk
and definitions.mk.
- Explicitly mark nostripping cases where dexpreopt will not embed the dex
file in the APK.
bug: 63920015
Test: m
Test: verify priv-apps dexs are uncompressed and unstripped
Test: Verify a non priv-app APK with uncompressed dex doesn't get stripped.
Change-Id: I624a03e3d965cebc0cae43fd6f7a6260178e6b8a
Specifying this option causes the make target's sources
and classpath dependencies to be patched into the specified
module via --patch-module (if the build toolchain implements
the module system, ie. when targeting 1.9). This allows tests
to compile that have classes in packages under test.
Currently, only java.base is on the module path (everything
else is on the classpath), and only when using -target 1.9
(ie. EXPERIMENTAL_USE_OPENJDK9=true).
Bug: 71699916
Test: Treehugger
Test: Checked that I was able to compile some ojluni & icu
test sources that had classes in (libcore) packages
under test.
Change-Id: I044fb9d904754ed281f574ab3dd52235cf0d7fec
Also use zip2zip to extract classes into another jar
to avoid hitting parameter limit in the shell invocations.
Bug: 69329508
Test: m checkbuild tests
Change-Id: Ic047aa699519b6b3fd100cb5395de945f96e4d0b
Some test targets include dex files as resource to be loaded
at runtime. This is an issue since full D8 also take dex files
as input.
This CL extracts only the class files and feed that to D8 instead
of the whole jar.
Test: USE_D8_DESUGAR=true m checkbuild tests
Bug: 69329508
Change-Id: I52d80ec01933573f18eb497472a80b6a649c9dae
Add an option to use D8's desugar instead of the standalone version.
USE_D8_DESUGAR=true m would trigger DCHECK in dex2oat unless
https://android-review.googlesource.com/c/platform/art/+/562595 is
patched in as well.
Bug: 69329508
Test: m && USE_D8_DESUGAR=false m
Change-Id: I601dada44c361e45bbc0e2227ae1544a7d9efa90
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)
Jack supported '?' and '*' wildcards anywhere in a filter. Tighten
the restrictions to only support '*' at the end of a filter, and
to disallow '?'. This will allow jacoco support to be built on
top of the existing zip2zip tools.
Add checks that all filters meet the new requirements, and run
the checks outside the LOCAL_EMMA_INSTRUMENT==true check so that
mistakes can be caught without having to run a build with
coverage enabled.
Bug: 69629238
Test: m EMMA_INSTRUMENT=true EMMA_INSTRUMENT_STATIC=true
Test: m in internal master
Test: https://android-build.googleplex.com/builds/view-workplan?viewType=Table&workplanId=L67200000122054710&nodeType=Trybot
Change-Id: Ib78bd3c8685fbc6bdcdb517df874186efd1cff33
This CL topic breaks the single flag value EXPERIMENTAL_USE_OPENJDK9
into two flags USE_OPENJDK9 and TARGET_OPENJDK9 which are computed
independently in config.mk and config.go; this means that later
build logic no longer depends on / duplicates the semantics of an
unset EXPERIMENTAL_USE_OPENJDK9, because that later build logic
can use USE_OPENJDK9 and TARGET_OPENJDK9 instead. Thus, it becomes
slightly easier to change this default in future CLs (touching
only config.mk and config.go).
It also makes it more straightforward to convert .mk files
to .bp.
This CL also introduces a new allowed value "false" for
EXPERIMENTAL_USE_OPENJDK9. This fourth possible value currently
has the same semantics as a default/unset
EXPERIMENTAL_USE_OPENJDK9, but allows people to explicitly
switch back to the old semantics when the default changes.
Test: make showcommands core-oj (in environments with
EXPERIMENTAL_USE_OPENJDK9 set to "", "1.8", and "true").
Bug: 38177295
Change-Id: Iaade9610b237722e5d175143808b8653b1e98996
Instead of building R8 from external/r8 use the prebuild jar in
prebuilds/r8.
Test: m -j USE_R8=true
Change-Id: I9680ae430cdbc2b3beb910873c8eed27354f6af1
Proguard doesn't support v53 class files. For now, this
CL simply replaces the Proguard step with a plain 'cp',
which is essentially a no-op.
To avoid apps exceeding the dex limit in this mode, this
CL also enables --multi-dex for those builds.
Test: Treehugger
Bug: 67673860
Change-Id: I17dc5d827360d606a0cac12e324f97e8dedbb630
This reverts commit 3d3e1cf260.
Rationale: part of a group of commits that left aosp_x86_64 not
building. (See https://android-build.googleplex.com/builds/
submitted/4426589/aosp_x86_64-eng/latest/logs/build_error.log)
Bug: 30227045
Test: builds
Change-Id: Ie22590abe3d1cdccb8d141baf6480d49dedf8789
This CL makes the following changes:
(a) It disables diagnostics for CFI which requires the runtime ubsan
library (which isn't included in static executables).
(b) It applies the ar flags for CFI correctly for nested .a
archives.
(c) Applies the version script to export CFI shadow for non-static
binaries
(d) Doesn't apply cross-dso CFI for static executables
Bug: 30227045
Test: Static executables build correctly and do not complain about
missing symbols from the ubsan runtime library.
Test: Nested .a files correctly use the gold plugin.
Change-Id: Id8fe3c13f6b76565aafbf1266e95f50d1447a790
Desugar default methods if requested by LOCAL_MIN_SDK_VERSION < 26.
Also remove --desugar_try_with_resources_if_needed=false, it is no
longer necessary if we always proguard against frameworks.jar.
Fixes d8 issues when it sees unexpected default methods that dx
was passing through.
This reapplies I81796b452d5a70a951e175727953d4295ca32c0c after
If30943efcaea44b2db1b38a6c1d558113324757f.
Test: m checkbuild
Change-Id: I3bafbb2d63de2a6b50bc8c7ff40259b54d24f128
When LOCAL_SDK_VERSION is not set, frameworks.jar is put
in the classpath for javac. If it is also in the classpath
(as opposed to the bootclasspath) for desugar then desugar
tries to rewrite classes that implement interfaces with
default methods to contain calls to the generated companion
classes for those interfaces. frameworks.jar will not
contain the companion classes, which causes proguard to
fail.
Move frameworks.jar to the bootclasspath, which matches more
closely with builds against the SDK stubs jars.
Also remove the static libraries from the classpath, they have
already been merged into the input jar.
Test: m checkbuild
Change-Id: If30943efcaea44b2db1b38a6c1d558113324757f
Desugar default methods if requested by LOCAL_MIN_SDK_VERSION < 26.
Also remove --desugar_try_with_resources_if_needed=false, it is no
longer necessary if we always proguard against frameworks.jar.
Fixes d8 issues when it sees unexpected default methods that dx
was passing through.
Test: m checkbuild
Change-Id: I81796b452d5a70a951e175727953d4295ca32c0c
With this change building with USE_R8=true will run R8 instead of Proguard + dx
for all targets that have LOCAL_PROGUARD_ENABLED set. R8 (through the
compatproguard wrapper) will use the same options as Proguard for now.
Additional Proguard configurations for R8 can be added using LOCAL_R8_FLAG_FILES
on individual targets. This makes it possible to make configuration updates
without touching the Proguard + dx configuration
Test: m -j
Test: m -j USE_D8=true USE_R8=true
Test: m -j USE_D8=true USE_R8=false
Test: m -j USE_D8=true
Change-Id: I78e8a0fd17cc5cefedccffe95b93c94a60a727f1
When building the recovery image on marlin (m bootimage):
----- Making recovery image ------
Copying baseline ramdisk...
Modifying ramdisk contents...
/bin/bash: line 0: [: 0x02000000: integer expression expected
----- Made recovery image: out/target/product/marlin/boot.img --------
Because if -gt doesn't like hexadecimal numbers. Fix the issue by doing
a dummy arithmetic to get integer $$maxsize.
Test: `m bootimage` with aosp_marlin-userdebug. Also change the
BOARD_BOOTIMAGE_PARTITION_SIZE closer to the actual size to
trigger the different size check paths.
Change-Id: I75c7727664d7255b9c18f57ae38076ff90b8a957
Soong has support for building system modules. Use the directories
produced by Soong with --system to replace -bootclasspath arguments
when using javac -target 1.9.
Since soong can't generate current SDK stubs yet, and no existing
SDK stubs need -target, only use -target 1.9 for modules that are
not compiling against the SDK. That means in practice the only
system modules that will be used for now is the default one,
core-system-modules.
Bug: 63986449
Test: m -j EXPERIMENTAL_USE_OPENJDK9=true makes some progress
Change-Id: I350ef50aedf36fdd72458c23d4fe8a2edf1a9a02
JAVAC_WRAPPER seems not be used when SOONG_JAVAC_WRAPPER is used.
Since JAVAC_WRAPPER just wraps actual javac call, the result must
be the same.
Bug: 67723445
Change-Id: I944f9534587c00ffd1bc79806c534432467ba4a5
This wasn't initializing the STATS.MODULE_TYPE.* variables, leading to
106kB of stack usage in ckati for sdk_x86_64-eng. With this fixed, the
next highest stack usage is 35kB.
Bug: 36182021
Test: Run $(flavor) over all the variables
Test: Check stack usage with custom ckati patches
Test: build-sdk_x86_64.ninja is identical before/after this change
Change-Id: Id1d8ffb8415ad6613d0093d31be5472d42d300fd
Combining $(call dexpreopt-remove-classes.dex) into a
$(call copy-one-file) rule is hacky and error prone. Make a helper
dexpreopt-copy-jar function that takes LOCAL_DEX_PREOPT and copies
the jar, removing the dex files if necessary.
Test: only description message changes to out/build-${TARGET_PRODUCT}.ninja
Change-Id: Id2fa2d18571ca01bc81302c912b8de25f3dde7a1
Those dependencies are jars/apks on which we want to
uncompress their dexs.
bug: 30972906
bug: 63920015
Test: sailfish build
Change-Id: Ic96ffe9dbe39abc1c28e7de134892d689207c9ca
When checking *.img size with the partition size, the build system
reserves additional bits for spare bits and spare bad blocks.
However, for emmc and ufs, the space bits and blocks are entirely
managed by the underlying controller and thus not visible from outside
of the controller. In fact the check routine was made for legacy MTD
storages where raw flash blocks are directly exposed.
This makes the size checking a little bit conservative in modern devices.
Builds were failed even though the *.img can actually fit into the
partition. To handle this problem, the additional size is no longer
reserved when checking *.img size with the partition size.
This change also removes following build flags that are meaningful
only for devices having MTD storages:
BOARD_NAND_PAGE_SIZE
BOARD_NAND_SPARE_SIZE
Further use of them breaks the build
Bug: 35790399
Bug: 66399382
Test: build
Merged-In: I954bf261441b53844e75d05788866f1692a2ad43
Change-Id: I954bf261441b53844e75d05788866f1692a2ad43
When checking *.img size with the partition size, the build system
reserves additional bits for spare bits and spare bad blocks.
However, for emmc and ufs, the space bits and blocks are entirely
managed by the underlying controller and thus not visible from outside
of the controller. In fact the check routine was made for legacy MTD
storages where raw flash blocks are directly exposed.
This makes the size checking a little bit conservative in modern devices.
Builds were failed even though the *.img can actually fit into the
partition. To handle this problem, the additional size is no longer
reserved when checking *.img size with the partition size.
This change also removes following build flags that are meaningful
only for devices having MTD storages:
BOARD_NAND_PAGE_SIZE
BOARD_NAND_SPARE_SIZE
Further use of them breaks the build
Bug: 35790399
Bug: 66399382
Test: build
Merged-In: I954bf261441b53844e75d05788866f1692a2ad43
Change-Id: I954bf261441b53844e75d05788866f1692a2ad43
If a resource was moved / removed without any other timestamp updates,
ninja would not re-run aapt. To fix this, add a list of resource files
into the same command as aapt (echo ... >/dev/null). That way if the
list changes, ninja will recognize the command line changing and rerun
aapt.
Bug: 67315436
Test: mmma cts/tests/autofillservice
Test: mv cts/tests/autofillservice/res/layout/welcome_activity.xml \
cts/tests/autofillservice/res/layout/welcome_activity_test.xml
Test: mmma cts/tests/autofillservice
Change-Id: Ia93ac4ff48910fed74a1f1539ce4e1f5a9db023f
Remove all support for running jack.
Bug: 65302138
Test: m -j checkbuild
Change-Id: I1ef87c88af3f2689f8b0dcf1a01c958b16159631
Merged-In: I1ef87c88af3f2689f8b0dcf1a01c958b16159631
(cherry picked from commit 5db5d31d73)
Remove all support for running jack.
Bug: 65302138
Test: m -j checkbuild
Change-Id: I1ef87c88af3f2689f8b0dcf1a01c958b16159631
Merged-In: I1ef87c88af3f2689f8b0dcf1a01c958b16159631
(cherry picked from commit 5db5d31d73)
Make R.stamp actually part of the same rule that creates R.java.
Otherwise if the generated sources directory is removed, we don't
regenerate R.java / Manifest.java.
Also move the package-export.apk and proguard options file into the same
rule.
Test: m telephony-common; rm out/target/common/obj/*/*_intermediates/src; m telephony-common
Change-Id: I71f6fcaf20593ca23271d52d15537aaa9ef4cc77
Follow Jack's behavior for now and pass LOCAL_MIN_SDK_VERSION to
dx. Don't pass it to desugar for now, desugar with low
--min_sdk_version numbers enables features we haven't verified
yet, and existing apps won't be using the new language features
that require extra desugaring.
Test: examine classes.dex from an app with LOCAL_MIN_SDK_VERSION
Change-Id: Ic7d679b1b11fba98a33418a4f79d4762a91f1f9c
m -j TURBINE_ENABLED=false will build without turbine. This is
primarily useful for A/B comparisions of turbine builds, but will
also be used to temporarily disable turbine for unbundled builds
until we have turbine prebuilts.
Bug: 64308460
Test: m -j TURBINE_ENABLED=false java
Test: m -j java
Change-Id: Ie48746c8bfc60c361be9634cb1805ca1c09aa1fe
Simplify the java-lib-* type functions, three layers of abstraction
is a bit much for a single line function, and isn't reducing enough
duplication to be worth the complexity.
Add app-lib-* functions and use them.
Test: m -j checkbuild
Change-Id: Ia47ebc95e4af2073396233416f14fdee7d856844
The bootclasspath libraries have direct dependencies now, so
remove them from LOCAL_JAVA_LIBRARIES so they don't also end
up in the classpath.
Test: m -j checkbuild
Test: m -j ANDROID_COMPILE_WITH_JACK=true java
Test: manually inspection of build-${TARGET_PRODUCT}.ninja
only has expected changes
Test: no changes to system.img
Change-Id: Icbc8987468f237aa351708b33c3abe5ea3d1816e
The libraries in the bootclasspath were getting their dependencies
because they were also in the default libraries list. Make the
dependencies explicit in preparation for rearranging the default
libraries.
Test: m -j checkbuild
Change-Id: I3ea004714a31cd648ec2ef57ed3099bce70bc6b9
- ndk/r10 is removed, we need to use SOONG ndk stubs instead.
- Use MIN_SUPPORTED_SDK_VERSION instead of hard coded version 9 for
32 bit compilation.
Bug: 38342163
Test: mm
Test: make -j40 FORCE_BUILD_RS_COMPAT=true RSTest_Compat
Change-Id: I9549446cb7f8faee8842b06f096d035535dc7f3d
Mac java build has flaky error for merge_zips. The error happend is
because both turbine.jar.premerged and static_java_libs are missing.
We allow one of them can be missing, but we don't allow both of them are
missing. So re-org the merge_zips logic to see if the input of Turbine is
empty or it is due to Turbine internal.
Bug: b/66441248
Test: m clean && m -j32
Change-Id: I3c4b15004fd67fef2b83eabc36841bc8c8ff6e4c
By doing this, the change of private_manifest file won't trigger
Javac to re-run during re-build.
Bug: b/65455145
Test: m clean && m -j32
Change-Id: Ib90d56daf0be5f31b51e07b7714a933bc5cf4496
1. Use merge_zips to merge jars without unzipping/rezipping static Java
libraries. And seperate merging process as a independent Make target so
that Javac can only depend on Turbine.
2. Remove duplicate vars: "full_java_lib_deps" in droiddoc.mk
3. Also use classes-header.jar for link_apk & link_instr
Bug: b/65455145
Test: m clean && m -j32
Change-Id: I7b0307664a6b86075c983284fce099bd3946318c
bug: 30972906
bug: 63920015
Test: doing zipinfo on the produced apks after build.
Test: fugu build
This reverts commit f69441b76b.
Change-Id: I6b029f2044385f3e3fa332091a9244cec79d2474
Broke Fugu build. Will work on a workaround for it.
bug: 30972906
bug: 63920015
This reverts commit 09dfe6739a.
Change-Id: I5a8b8de4ec98ae44808762ca1ba974ce89bebb05
To ensure the verified boot work does not make ART run
dex code uncompressed in memory, make the build system
generate uncompressed dex files in APKs.
The alignment of STORE files in an APK is already handled.
bug: 30972906
bug: 63920015
Test: doing zipinfo on the produced apks after build.
Change-Id: I0ccdeae5d674b2a409a1343c4a14699c7bf0defb
The problem is that when I added .KATI_RESTAT for javalib.jar, I didn't
change the function: create-empty-package to generate javalib.jar.tmp.
The problem causes the build breakage on MacOS/Jack.
Bug: b/64308460
Test: download <git-wear-master> branch locally, and then do:
m -j out/target/product/angelfish/package-stats.txt
Change-Id: I9a5d74caaf164b44676e456050b4f8a5b3bdded0
This reverts commit 33dcd0c5eb.
Reason for revert: <stub.jar.toc needs classes.jar.toc which is no longer existed after Turbine was introduced. So use classes.jar to represent stub.jar.toc>
Bug: b/64308460
Test: m clean && time m -j32 ANDROID_COMPILE_WITH_JACK=true
Change-Id: If618993f1e0c789d93cc15664212603f994fd460
This reverts commit 0739155c86.
Reason for revert: <sdk_google_aw_x86-sdk (linux) build 4310016 in git_master is broken>
Change-Id: I525d48557bd0425917e61a85f596663ccc63c887
1. Bundle static_java_header_libs during turbine-classes.jar if
necessary. This can be improved if we have the ability to merge jars on
the fly.
2. Copy sdk/stubs-classes.jar to sdk/stubs-classes-header.jar, and Use
.KATI_RESTAT to cut un-necessary targets rebuilt.
3. Copy prebuilt-classes.jar to prebuilt-classes-header.jar since
Turbine doesn't apply to prebuilt.
4. Run jarjar after Turbine compilation, otherwise downstream targets
won't find corresponding symbols.
5. Change classes.jar in -classpath used in Desugar to
classes-header.jar.
6. Change legacy-libs in Proguard from classes.jar to
classes-header.jar.
7. Add .KATI_RESTAT for turbine-classes.jar, and remove IJAR.
8.boot.art is re-generated during the incremental rebuild since
javalib.jar got re-generated. Then boot.art re-triggered lots of
downstream targets to be rebuilt. So we add .KATI_RESTAT for javalib.jar
to prevent downstream to be rebuilt. The ziptime & commit-change-to-toc
operation for javalib.jar(small) didn't bring too much overhead when we build
from clean state.
Next step: specify static_java_header_libs in -classpath, and merge jars
at the end.
touch frameworks/base/core/java/com/google/android/util/Procedure.java &&
time m
time: 1m58s (794 -> 49 targets)
m clean && time m java
time: 9m34s
Bug: b/64308460
Test: m clean && m checkbuild
Change-Id: Iefcc234405b9f461b6882c06bba15e21fa783d28
Move the java source list generation step into its own rule. This
has a couple of advantages. It consolidates the source list
generation so that it only has to run once for javac, jack, and
jack-check. It also massively reduces the length of the javac
command line, so that error messages are significantly shorter,
and allows easily rerunning the failing build command because
the file list is still on disk.
The primary disadvantage is that javac error messages no longer
include the list of files passed to javac, but the list is
available earlier in the build long when the file list was written,
and is still available on disk.
Test: m -j javac-check
Change-Id: I9730b352b33a060e08221b61c11c617d23320d67
jar -C <dir> . produces a jar containing files in filesystem order,
which can vary between builds. Manually find and sort the list of
files, and convert them into a list of -C <dir> <file> pairs.
Fixes: 64634025
Test: m -j checkbuild
Test: m -j out/target/product/sailfish/system/framework/ext.jar, check
that entries are sorted
Test: m -j out/target/product/generic_arm64/system/framework/ext.jar on mac
Change-Id: I7dced6acbe621a60cd49daf17872941485602732
Commit 2bdbb6e935 added
command line flags to the unzip command in unzip-jar-files
to skip the path 'module-info.class', if present.
This has led to log spam because unzip warns when an
excluded file is not present in the zip/jar file, and
that warning cannot be suppressed via command line flags.
Therefore, this CL modifies the unzip-jar-files macro
to rm -f the module-info.class file after the fact, if
it was created, which does not lead to log spam.
Note that module-info.class will be filtered out of
.jar dependencies (such as ASM 6.0) regardless of
whether one is building with an OpenJDK 9+ toolchain.
Test: Treehugger.
Bug: 64719206
Change-Id: Ic6be806a50557b4ba13fc18da91a8af12d14586c
This reverts commit 9cc3c76abd which
added command-line parameters to the invocation of SignApk to permit
it to access sun.** classes which are not accessible by default in
Java 9. This hack is no longer needed because SignApk no longer needs
this access because of tools/apksig commit
b40d3e4821bc9cb094b0ff13153340e3a0da1b3f which switched APK JAR
signature generation logic away from directly using sun.** classes.
Test: find out -name CtsPkgInstallTinyApp* | xargs rm -Rf
mmma -j74 cts/hostsidetests/appsecurity/test-apps/tinyapp
Test: find out -name CtsPkgInstallTinyApp* | xargs rm -Rf
EXPERIMENTAL_USE_OPENJDK9=target1.8 \
OVERRIDE_ANDROID_JAVA_HOME=<path to Java 9 SDK> \
PATH=${OVERRIDE_ANDROID_JAVA_HOME}/bin:${PATH} \
mmma -j74 cts/hostsidetests/appsecurity/test-apps/tinyapp
Bug: 37137869
Change-Id: I389c366f8a5bed56c496293bc871458adbedb0e8
- Prebuilts with LOCAL_MODULE_CLASS == APPS and BUILD_PACKAGE that
define LOCAL_COMPRESSED_MODULE := true will be installed compressed
in the system partition.
- The compression algorithm is simple gzip for now.
In addition :
- Compressed modules are not dex-preopted.
- JNI shared libraries are always embedded inside compressed modules,
they are never installed as separate files on the FS.
Test: Manual
Bug: 63802184
Change-Id: Id12b1ff0b1d68791ec1178783f7e78910e122a33
Test: runs "m -j checkbuild tests cts", and then --bootclasspath_entry
shows up in desugar command.
Change-Id: Ibcc8a3f51438723151c57f79e5a1184350efbd30
During the Android build process, multiple .jar files are
unzipped into the same directory. If the .jar is an OpenJDK 9
modular jar (with a module-info.class in the root directory),
the last module-info.class extracted will overwrite any earlier
module-info.class files extracted, and will cause all extracted
class files to be considered part of that module.
Therefore, this would break compilation under OpenJDK 9 (with
-target 1.9 -source 1.9). This CL fixes this by omitting the
module-info.class file (if present) when extracting the .jar.
This essentially turns any modular jar into a regular jar,
replacing the module with corresponding classes on the classpath.
This is sufficient for now because Android does not currently
support module dependencies.
Test: Treehugger
Bug: 38177569
Change-Id: Ia184e64d2f24b8ca79aeab1c00bd5da0386530bf
Desugaring try-with-resources is not necessary for platform builds,
and triggers some problems for apps built as part of the platform
that target SDK versions before 19. Disable all try-with-resource
desugaring.
Bug: 63180735
Bug: 63900665
Bug: 63901645
Test: m -j ANDROID_COMPILE_WITH_JACK=false checkbuild
Change-Id: I98b827aa1e80b43e6eb7b58254c23c7a4f7dc52d
The extension directory defaults to lib/ext and does not
exist by default. Setting it to the empty string de facto
disables this obsolete feature.
AOSP is moving to a hermetic toolchain so this argument
will stop working soon. Further, OpenJDK 9 javac no longer
supports this command line argument when compiling for
-source 1.9 -target 1.9.
This command line argument has been around since the
earliest versions of Android, but is now obsolete.
This CL drops it.
Bug: 63746471
Test: Treehugger
Change-Id: Ia0214c1b192e3ffda10772d777557a81ce346c03
Use $(JAVA) to invoke dx.jar instead of the dx wrapper script.
Also allow the dx command line or the dx jar to be specified on
the command line. Allows better dependencies on dx.jar, and
doesn't rely on having java on the path.
Bug: 62956999
Test: m -j checkbuild
Change-Id: I5f265132690ad28b3139954fea44eb0c4a43393a
Replace java -> $(JAVA), etc. to prepare for moving java binaries
out of the path and using full paths to them.
Test: builds
Bug: 62956999
Change-Id: I5dfcd0d4e855ecde69cd127169e9545e59723d12
Apparently the default bash echo doesn't support \e even though the man
page says it does. So use \033 instead.
Test: m -j nothing on mac, get colors
Test: m -j nothing on linux, still get colors
Change-Id: I608fa87e3c28bf8f99264d39f9b250008a81235a
Created math.mk and moved all math function definitions there.
Then, included this file in config.mk. This allows the functions to be
used by whatever includes config.mk (envsetup and build).
Test: manually setting combinations of PRODUCT_SHIPPING_API_LEVEL
and PRODUCT_FULL_TREBLE_OVERRIDE and looking at values of
PRODUCT_FULL_TREBLE.
Bug: 62229856
Change-Id: Icdab4214a1e65ae202411613dbcb9c9ea5f43b09
Bug: 38342163
- obj/lib is deprecated and TARGET_OUT_INTERMEDIATE_LIBRARIES will be
removed. We need to use per-module intermediates directory instead.
- prebuilts/ndk/current folder is gone. For libm and libc, we just
need ndk/r10 as stub libs for linking.
Test: mm and made sure this does not affect on device target.
Test: make -j50 FORCE_BUILD_RS_COMPAT=true RSTest_Compat works as
expected.
Change-Id: I8fbf5c10322707849a23c6b0dacc28b028db451e
signapk relies on internal APIs sun.security.{pkcs,x509},
for example in com.android.apksig.internal.apk.v1.V1SchemeSigner.
This breaks at signapk runtime under OpenJDK 9 because those
packages are not exported by the java.base module.
This CL unbreaks signapk by allowing it to access these internal
packages. In the long term, signapk should migrate away from these
internal APIs (bug 37137869).
Test: make ANDROID_COMPILE_WITH_JACK=false checkbuild tests \
&& make checkbuild tests
(with OpenJDK 8u45 toolchain on the PATH)
Test: make EXPERIMENTAL_USE_OPENJDK9=true \
ANDROID_COMPILE_WITH_JACK=false checkbuild
(with jdk 9-ea+170 toolchain on the PATH)
Bug: 37137869
Bug: 38177295
Change-Id: I64cab83e6eb7b135cf2ad7b523736cb409aaae02
Desugar reflects over internal APIs at runtime, using the
java.lang.invoke.MethodHandles.Lookup API.
On OpenJDK 9 toolchains, such reflection is only allowed
to packages to which the java.lang.invoke module is opened.
This CL adds an override to open the module to all unnamed
modules (i.e., to Desugar) when running Desugar.
Test: make checkbuild
(with OpenJDK 8u45 toolchain on the PATH)
Test: make EXPERIMENTAL_USE_OPENJDK9=true checkbuild
(with OpenJDK 9-ea toolchain on the PATH)
Bug: 38177295
Change-Id: I2cf74a96ea17366dd50b8d92af8e41e812247ef7
This CL resubmits a fixed version of commit
e2a8da2683.
This CL applies the following changes:
- explicitly set the path where desugar will dump temporary class files.
This ensures that the system property is set before it is read during
InnerClassLambdaMetafactory.<clinit>. Before this CL, the system
property was set by Desugar.createAndRegisterLambdaDumpDirectory(),
which may run too late.
- explicitly specify -source 1.8 for droiddoc's javadoc run.
Previously, the command used the language version of the build
toolchain, which might fail due the backward-incompatible restrictions
imposed by the proposed module system.
Some Android build targets use LOCAL_JAVA_LANGUAGE_VERSION 1.7,
but droiddoc combines sources from multiple build targets and there
where no backwards incompatible changes in 1.8, so it should be fine
to use 1.8 for code from either language level.
Bug: 38318052
Bug: 38225656
Bug: 38177295
Test: make clean && make ANDROID_COMPILE_WITH_JACK=false checkbuild tests \
&& make checkbuild tests
(using OpenJDK 8 toolchain)
Change-Id: I2fffe6668747f48be44e34c67332af9b8a996d2a
- explicitly set the path where desugar will dump temporary class files.
This ensures that the system property is set before it is read during
InnerClassLambdaMetafactory.<clinit>. Before this CL, the system
property was set by Desugar.createAndRegisterLambdaDumpDirectory(),
which may run too late.
- explicitly specify -source 1.8 for droiddoc's javadoc run.
Previously, the command used the language version of the build
toolchain, which might fail due the backward-incompatible restrictions
imposed by the proposed module system.
Some Android build targets use LOCAL_JAVA_LANGUAGE_VERSION 1.7,
but droiddoc combines sources from multiple build targets and there
where no backwards incompatible changes in 1.8, so it should be fine
to use 1.8 for code from either language level.
Bug: 38318052
Bug: 38225656
Bug: 38177295
Test: make clean && make checkbuild docs
(using OpenJDK 8 toolchain)
Change-Id: Ie5fe0502dfe2f99fc58b1a723b7c18d5896a2646
We shouldn't give non-clean paths to tools -- if a/b/../file was
specified, we can simplify that path to a/file, and not need to create
a/b just to make the path name work.
The testcases come from golang's filepath.Clean tests, this should be
compatible with that implementation.
Bug: 37716307
Test: TEST_MAKE_clean_path=true m -j blueprint_tools
Change-Id: I290a02b0a1e4a7c2b9255bca3c881589b521c402
Allow working around Jack bugs by using javac to compile *.java
to class files, and then pass through Jack for desugaring,
proguarding, and dexing.
Also make ANDROID_COMPILE_WITH_JACK=false override LOCAL_JACK_ENABLED.
Bug: 37483961
Test: m -j checkbuild
Test: Add LOCAL_JACK_ENABLED := javac_frontend to a target
Change-Id: I2189055aac9deccc36b8eab8e949796dddadf7a1
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
Add soong_javac_wrapper to all javac commands. soong_javac_wrapper
colorizes javac output, and hides noisy messages.
Previously attempted in I7e8fcd2e4e1ed3ff530a8ccfe931ceb7e411e0ad
and I449a61ed2c95b3e1afd51a31a11baab15d97e2f2.
Test: m -j ANDROID_COMPILE_WITH_JACK=false
Change-Id: I9d7a20d766fadab323a4cf95e5734741b3f578bb
Prebuilts may already be stripped. Instead of requiring every stripped
prebuilt to be annotated with LOCAL_STRIP_MODULE := false, just make
the default (mini-debug-info) pass through already-stripped modules
without failing.
Bug: 36793128
Test: m -j checkbuild
Change-Id: Iafc4614ceb91bbbb2e5ff82caefd18753f8961b1
The codebase is growing more uses of annotation processors,
through a mixture of LOCAL_ADDITIONAL_DEPENDENCIES and
LOCAL_JACK_FLAGS. Add LOCAL_ANNOTATION_PROCESSORS and handle
the rest in the build system.
For now we also need LOCAL_ANNOTATION_PROCESSOR_CLASSES to
work around bad behavior in grok (b/25860419).
Test: m -j java
Change-Id: I465b021e65d25ed3ea517333a82ec4ebd63fc038
(cherry picked from commit cde14987ba)
The codebase is growing more uses of annotation processors,
through a mixture of LOCAL_ADDITIONAL_DEPENDENCIES and
LOCAL_JACK_FLAGS. Add LOCAL_ANNOTATION_PROCESSORS and handle
the rest in the build system.
For now we also need LOCAL_ANNOTATION_PROCESSOR_CLASSES to
work around bad behavior in grok (b/25860419).
Test: m -j java
Change-Id: I465b021e65d25ed3ea517333a82ec4ebd63fc038
Jack can silently handle using a java library as a static java
library by merging in the java library dex file. This causes
problems when switching to javac, because dex doesn't support
dex merging with multidex enabled?
Make the output files consistent between java libraries and static
java libraries. Java libraries will now produce:
classes-pre-proguard.jar: the classes before proguard processing
classes.jar: the final jar file containing classes before dexing
javalib.jar: a jar containing classes.dex
Static java libraries will eventually only produce
classes-pre-proguard.jar and classes.jar. All inter-library
linking is done with classes.jar, so a java library can be
used as a static java library.
There are too many dependencies outside the build system that
expect javalib.jar to exist for static and host java libraries,
so for now continue to build a javalib.jar that is a copy of
classes.jar.
Test: m -j ANDROID_COMPILE_WITH_JACK=false java
Test: m -j java
Bug: 36901093
Change-Id: I6730e2d3ec38004874265b2a690442dec57b33f4
This CL addresses the problem that native tests build for multiple
architectures and also build for host.
First it restructures each testcase folder to the following layout:
<testcase>/<testcase>.config
<testcase>/<arch 1>/testcase
<testcase>/<arch 2>/testcase
Secondly it now uses the new soong-zip prefix capability to
separate out host and target testcases in the device-tests &
general-tests zip files.
Bug:36692141
Test: `make device-tests general-tests cts -j` &
ensured all built successfully.
Change-Id: I4e8b084ee56fa8940914911649ff35507d87eb27
This CL is to generate every static RRO package for its target package
automatically at build-time.
BOARD_ENFORCE_RRO build variable is added to specify whether enforcing
RRO is required or not.
BOARD_ENFORCE_RRO_EXEMPT_SOURCES build variable is added to specify
the module list of which item should be exempt from enforcing RRO.
Test: tested on bullhead and sailfish
Bug: 34097942
Change-Id: I455b2ce34e66c57a540c299b5e378b7c4e78d5b8
(cherry picked from commit 3070610b72)
This reverts commit 440079b84b.
Adding the pipe interferes with error codes, which results
in an empty classes-full-debug.jar output file being left
after the failure. The next build sees the output file with
an up-to-date timestamp and doesn't rerun the failing rule.
Bug: 36666657
Change-Id: I6658edb766d8ba3120f88e3d8d6eda6ea6c691da
DEFAULT_APP_TARGET_SDK is PLATFORM_VERSION_CODENAME (for example 'P')
for non-release builds, and PLATFORM_SDK_VERSION (for example 25) for
release builds. Use it for dx and desugar, and translate
PLATFORM_VERSION_CODENAME to 10000 to enable features in the current
development version.
Bug: 36087246
Bug: 36118520
Test: m -j ANDROID_COMPILE_WITH_JACK=false
Change-Id: I3d6bbc36a8869e4ced6bb850bc32f37b4cba3f85
Dan Willemsen