We really should be re-running Kati when a file in here changes, since
it may actually affect the result. The general problem we've run into is
when there are two kati output files that are clean according to the
current rules, but if they actually ran we'd force an installclean.
This also had the side-effect of not running shell commands that touched
$OUT_DIR, like writing files using $(shell echo ... >out/...), now those
may be run.
This will cause Kati to re-read the makefiles more often -- twice when a
CleanSpec.mk changes, twice when switching products on the same device.
Basically every time we read the old state, do something about it, and
write out the new state, we need to rerun the next time so we aren't
continuously cleaning up what changed.
Bug: 30947985
Test: Manual, switching between configurations; enable --regen_debug
Change-Id: I888f9f8295f318f8214b2c79525cf94f457a5a8b
So that an external tool can read the configuration using
dump-many-vars, and write out the soong.variables file while only
loading config.mk.
Also remove dumpvar-% from PARSE_TIME_MAKE_GOALS, since it's only used
with CALLED_FROM_SETUP / config.mk. This provides an easy way to test
this change as well.
Test: get_build_var NINJA
Test: get_build_var ANDROID_GOALS
Test: get_build_var NINJA_GOALS
Test: get_build_var KATI_GOALS
Test: MAKECMDGOALS="droid sdk dumpvar-ANDROID_GOALS" get_build_var ANDROID_GOALS
Test: MAKECMDGOALS="droid sdk dumpvar-NINJA_GOALS" get_build_var NINJA_GOALS
Test: MAKECMDGOALS="droid sdk dumpvar-KATI_GOALS" get_build_var KATI_GOALS
Test: MAKECMDGOALS="out/build-a.ninja dumpvar-ANDROID_GOALS" get_build_var ANDROID_GOALS
Test: get_build_var SOONG_MAKEVARS_MK
Test: get_build_var SOONG_VARIABLES
Test: CALLED_FROM_SETUP=true BUILD_SYSTEM=build/core make -f build/core/config.mk out/soong/soong.variables showcommands
Test: Compare soong.variables from above, and before/after with `m`
Test: Compare out before/after with `m doesnotexist`
Test: `m dist doesnotexist`
Change-Id: I1bc291aa165297c930fe600067edad6ce4979210
Bug: 31788743
Test: $ make cts dist
# check contents of generated package and API coverage
# report
$ make checkbuild
Merged-In: Ieec069bafbcc6df5a5b3b06befb4a3a5678a06fd
Change-Id: Ieec069bafbcc6df5a5b3b06befb4a3a5678a06fd
(cherry picked from commit f8f0880b3b8fb6a80ed766068d5f1714625b8960)
bash doesn't kill children when it receives SIGTERM. Since makeparallel
is the last command run by the shell, run it with exec to replace the
shell with makeparallel so make can send the signal directly to
makeparallel.
Bug: 31907490
Test: m -j & killall make; pgrep -a makeparallel
Change-Id: If8aeb51ec224234da5f05eae3382ce77f48a57e2
We never want Kati to see MAKEFLAGS, as forcefully overriding variables
is terrible. The variables in MAKEFLAGS are still available in the
environment, so if part of the build wants input from the user, it
should be explicitly checking for an environment variable or using ?=
This commit shouldn't actually change any behavior -- the version of
Kati checked into our tree doesn't actually read MAKEFLAGS (the latest
version does, which is why this change is necessary). We're also already
clearing MAKEFLAGS during makeparallel, so the USE_GOMA case already
doesn't pass through MAKEFLAGS.
Test: Manual $(warnings $(origin MYTEST) $(MYTEST)) with old and new Kati
Change-Id: I5e4ffa66b3539a539cc7a26da24645b11445c469
Pass the USE_GOMA flag to soong, and export the CC_WRAPPER environment
variable after adding gomacc to it.
Bug: 31142427
Change-Id: I6de698a40817c0615b03fa17bd27075465d79ffd
Currently, the only way to properly disable goma is to unset USE_GOMA.
If you set USE_GOMA=false, we won't turn on goma, but we'll pass -j500
to ninja.
Test: USE_GOMA={x,false} m -j48 showcommands (inspect)
Change-Id: I6b327d08ebadbe614a3bf7646fd597054bbfaaa2
This was printing all of the Soong modules in addition to the modules
defined in the ONE_SHOT_MAKEFILE. So replicate what we did with
MODULES-IN-* with GET-INSTALL-PATH-IN-* and update all the users.
Bug: 30191725
Change-Id: Ib1e4bf00f64d7a3e8cd91f44eaf1ff82fdfbf162
AUX is a new class, similar to TARGET
While TARGET defines toolchain for Application Processors
AUX is defining toolchains for arbitrary utility cores (DSPs, GPUs,
MCUs, etc). This allows building of non-android sources as part
of Android tree and avoid using prebuilts if source code is avaliable
Bug: 29635686
Change-Id: Ie755ea054b16c3e86369f5fb2ba6eb0b384af77f
Signed-off-by: Alexey Polyudov <apolyudov@google.com>
Kati only supports a single use of --no_ignore_dirty, so we were
ignoring the SOONG_ANDROID_MK file, and only detecting changes to
SOONG_MAKEVARS_MK. Fix this by using a pattern that should apply to both
of those makefiles.
Change-Id: I46390f9887f95f1db0efe4e93339667f35ebc67c
This is to ensure that Make and Soong agree on configuration variables
like TARGET_GLOBAL_CFLAGS. Only a few variables are actually checked
currently until we make more actually the same.
Bug: 23566674
Change-Id: Ibede43d933ede4d470e182f9490ea6ec4ef52fbc
am: 667a5fe33b
* commit '667a5fe33b8a9bb3e812a4923758e466926338cb':
Only pass NINJA_ARGS to the main ninja compile
Change-Id: Ie713aa9cd7d0d9d10dabb8b999a8ebd2e029588c
This way commands like "-t query <target>" will work against the main
build, instead of failing to find the target during the initial soong
build and stopping.
Change-Id: I9b6bd6ac431b95e12db69ec7f272669b9d292761
My previous attempt at properly handling symlinked OUT_DIRs only worked
if the symlink was to a different directory, not one in the same
directory.
This time, make sure that both make and soong use the same
representation of the output directory by passing BUILDDIR to
bootstrap.bash. Soong has been updated to pick whether to use a relative
or absolute path back to the source tree depending on what makes sense.
If the BUILDDIR or the path back to the source tree change, re-run the
bootstrap.
Also, move the $OUT_DIR/Android.mk and $OUT_DIR/CleanSpec.mk generation
to before Kati runs. In the case that $TOP/out was symlinked to
$TOP/out.angler:
1) Soong generates out.angler/soong/Android.mk
2) Kati's find generator produces in-memory tree of the filesystem
3) $(shell ) creates out.angler/Android.mk
4) The emulated findleaves Android.mk search finds
out.angler/soong/Android.mk since the in-memory tree is never updated.
This doesn't happen in the normal case, because we pass --prune=$OUT_DIR
to findleaves.
Change-Id: Ib0fdae2e80f75ddcf33a3c8c5ea0978f5308b437
(cherry picked from commit cc60f01357)
[ccross: resolved conflicts in nyc-dev]
am: 4aca571
* commit '4aca5715523d1b8a85abde00d7ccc2989c90a96b':
Remove -d keepdepfile from ninja command line DO NOT MERGE
Change-Id: Ie978a4113e3098b11eb263d326ba31ec7b2275fa
Also specify "-d keepdepfile" to ninja so .P files won't be
deleted by ninja.
-include for .s files are removed because GCC doesn't
generate .d files for .s files.
Bug: 26839129
Bug: 27886012
Change-Id: If00e93c7a33449ec314a5cdba438475a32979f4e
(cherry picked from commit 4037c4225a)
My previous attempt at properly handling symlinked OUT_DIRs only worked
if the symlink was to a different directory, not one in the same
directory.
This time, make sure that both make and soong use the same
representation of the output directory by passing BUILDDIR to
bootstrap.bash. Soong has been updated to pick whether to use a relative
or absolute path back to the source tree depending on what makes sense.
If the BUILDDIR or the path back to the source tree change, re-run the
bootstrap.
Also, move the $OUT_DIR/Android.mk and $OUT_DIR/CleanSpec.mk generation
to before Kati runs. In the case that $TOP/out was symlinked to
$TOP/out.angler:
1) Soong generates out.angler/soong/Android.mk
2) Kati's find generator produces in-memory tree of the filesystem
3) $(shell ) creates out.angler/Android.mk
4) The emulated findleaves Android.mk search finds
out.angler/soong/Android.mk since the in-memory tree is never updated.
This doesn't happen in the normal case, because we pass --prune=$OUT_DIR
to findleaves.
Change-Id: Ib0fdae2e80f75ddcf33a3c8c5ea0978f5308b437
Without switching on USE_SOONG, this lets us verify that all builds have
blueprint/soong/go, and can generate a build.ninja file without
producing errors.
Change-Id: I79451ce24c8b5b6745e41d951153265441c33e8a
Currently, we're building against the native host version of libc++ and
other headers. Instead, use prebuilts of ckati/makeparallel built with
Soong against the included libc++ and sysroot. Building ckati /
makeparallel doesn't work well in unbundled branches that don't have all
of the libc++ dependencies to do successful host builds.
KATI is default variable defined when running under Kati, don't use it
for other purposes. Use CKATI instead.
Bug: 27533820
Change-Id: I82d66f5e34e3341a3cee4c5c83cd1d9a54d057ff
We've been including the system libc++ headers even if we're building
against our version of libc++. Stop doing that, and only add the headers
to our path if we're using the system libraries.
If nothing is specified, on recent OSX versions, libc++ is the default
c++ library instead of libstdc++. We've been explicitly including the
libc++ headers on all versions, but that breaks old versions. Force us
over to libc++, since the system libstdc++ does not support C++11, and
libc++ is still supported on our oldest version (10.8).
Change-Id: I1fccee8da0f425e10ccc9d3247ed40664eb6ada0
Also specify "-d keepdepfile" to ninja so .P files won't be
deleted by ninja.
-include for .s files are removed because GCC doesn't
generate .d files for .s files.
Bug: 26839129
Change-Id: If00e93c7a33449ec314a5cdba438475a32979f4e
Places whitelisted brillo tests and the whitelist itself
in a zip when run.
BUG: 27385399
Change-Id: I93c2ea8cc521292a6de811bb47bc87a727edd21f
TEST: manual make dist brillo_tests, confirmed desired files were in zip.
Dan Willemsen