Prepare for using Config when adding singletons by passing
Config to NewTestContext and NewContext instead of to ctx.Register.
This will enable a followup change to store SingletonMakeVarsProviders
registered on the Context in the Config, which is necessary to run
multiple tests in parallel without data races.
Test: all soong tests
Change-Id: Id229629a4e42ff4487d317241673837726c075fc
Bazel overlay is an experimental feature. This renames the feature to 'queryview' to better describe its purpose, and also move away from the already overloaded 'overlay' term in Android.
Test: m queryview && bazel query --package_path=out/soong/queryview //...
Change-Id: I8b5068c7db46cb61a03a8e87af9c7c9077ebeff9
A recent change to Blueprint [1] added the feature to infer the type of
list elements to more than just "list of strings", but this also changed
the text to "list of string" (without s). This CL fixes the bazel
overlay, which uses "list of strings".
Test: m bazel_overlay && bazel query //... --package_path=out/soong/bazel_overlay
[1]: 764a771917%5E%21/#F1
Change-Id: I7522d15b1ce751807e1881ce1e94f068d226b316
There is no need for multiproduct_kati to run the tests for
every product, they don't vary by product config. --skip-soong-tests
can also be used for local development to run soong_build even if
the tests don't pass.
Bug: 156428456
Test: m --skip-soong-tests nothing
Change-Id: I9c00e3d1b6e51d17bb290339c3f124d4d1c9e69f
We need to capture the user build command to understand what additional
arguments is passed in to Soong. This data is useful to know which
targets are being executed and have statistics on top build targets
are specified.
Bug: b/169453712
Test: ran "m blueprint_tools", "m nothing", "lunch 1" and checked
in soong_metrics proto.
Change-Id: I7ab94dcf037211f1abffe6afd012b03ffbf99756
This keeps the top level directory clean, and leverages the
new "build/bazel" repo location / package path.
Test: m bazel_overlay && cd out/soong/bazel_overlay && bazel cquery //...
Bug: 171180724
Fixes: 171180724
Signed-off-by: Jingwen Chen <jingwen@google.com>
Change-Id: I151b8f3d2177167711e5974e5eabad8d6e910276
This reverts commit 323dc60712.
Reason for revert: Possible cause of test instability
Bug: 170513220
Test: soong tests
Change-Id: Iee168e9fbb4210569e6cffcc23e60d111403abb8
With this change, bazel_module is a specifiable property on
genrule module definitions. With bazel-enabled mode, soong_build will
defer to Bazel for information on these modules.
source build/soong/bazelenv.sh to enter bazel-enabled mode.
Test: Manually verified on bionic/libc genrules using aosp_cf_x86_phone-userdebug
Change-Id: I3619848186d50be7273a5eba31c79989b981d408
Prints the following warning messsage:
WARNING:
Setting OUT_DIR to a path other than out may result in slow RBE builds.
See http://go/android_rbe_out_dir for a workaround.
Bug: b/169676232
Change-Id: Id1ad06c4dc672c24373642e0b624833eb6a0dbcf
Putting t.Parallel() in each test makes them run in parallel.
Additional t.Parallel() could be added to each subtest, although
that requires making a local copy of the loop variable for
table driven tests.
Test: m checkbuild
Change-Id: I5d9869ead441093f4d7c5757f2447385333a95a4
surface module properties as attributes.
This CL maps int, bool, string and string_list module props from Soong
modules into their respective Bazel targets.
With this CL, you can now query modules based on their properties. For
example:
$ bazel query 'attr(nocrt, 1, //...)'
$ bazel query 'attr(apex_available, //apex_available:platform, //...)'
$ bazel query //art/dalvikvm:dalvikvm--linux_glibc_x86_64 --output=build | grep compile_multilib
Test: m bazel_overlay && cd out/soong/bazel_overlay && bazel cquery 'attr(apex_available, com.android.runtime, //...)'
Test: soong_build tests
Fixes: 162720644
Fixes: 164320355
Change-Id: Iea8e594b952feccac3281f36dd6bdee8e7d62c3a
This patchset changes bazel_overlay to generate soong_module as a macro,
instead of a rule, and generate module properties in the BUILD files as
kwargs to the soong_module macro.
Here's a sample of the new BUILD files with module properties:
bionic/libdl/BUILD.bazel:
https://paste.googleplex.com/6484466996346880?raw
art/build/apex/BUILD.bazel:
https://paste.googleplex.com/5461276001042432?raw
bionic/apex/BUILD.bazel:
https://paste.googleplex.com/4932795173437440?raw
soong_module is now a macro that conditionally expands to underlying
soong_<module type> rules with statically defined attributes. In this
CL, we are starting with a hardcoded filegroup rule definition to
demonstrate the conditional rule loading within the soong_module macro.
If the module_type matches an existing Bazel rule, soong_module forwards
the entire **kwargs into the rule, which Bazel typechecks.
Non-filegroup module types will be expanded into generic_soong_module,
but with the kwargs dropped.
This approach allows us to:
1) Programmtically generate soong_<module type> rules for all module
types available in Soong, together with the statically defined attribute
types in `attrs`.
2) Incrementally migrate and test individual module types from
generic_soong_module to their module rule shims.
3) Swap out the module rule shims to the actual Bazel rules (e.g
cc_library, java_library) and perform attribute manipulation in Starlark
itself.
Example of querying against the 'srcs' attribute in soong_filegroup:
```
$ bazel cquery 'kind(soong_filegroup, //...)' | wc -l
590
$ bazel cquery --output=build 'attr(srcs, "linker.cpp",
kind(soong_filegroup, //bionic/...))'
INFO: Analyzed 3907 targets (0 packages loaded, 0 targets configured).
INFO: Found 3907 targets...
/usr/local/google/home/jingwen/aosp/out/soong/bazel_overlay/bionic/linker/BUILD.bazel:4144:13
soong_filegroup(
name = "linker_sources",
generator_name = "linker_sources",
generator_function = "soong_module",
generator_location = "bionic/linker/BUILD.bazel:4144:13",
srcs = ["dlfcn.cpp", "linker.cpp", "linker_block_allocator.cpp",
"linker_dlwarning.cpp", "linker_cfi.cpp", "linker_config.cpp",
"linker_debug.cpp", "linker_gdb_support.cpp", "linker_globals.cpp",
"linker_libc_support.c", "linker_libcxx_support.cpp",
"linker_namespaces.cpp", "linker_logger.cpp",
"linker_mapped_file_fragment.cpp", "linker_phdr.cpp",
"linker_relocate.cpp", "linker_sdk_versions.cpp", "linker_soinfo.cpp",
"linker_tls.cpp", "linker_utils.cpp", "rt.cpp"],
deps = [],
)
/usr/local/google/home/jingwen/aosp/out/soong/bazel_overlay/soong_module.bzl:32:23
in <toplevel>
```
This CL is known to be lacking the following features, and will be looked at in follow up CLs:
1) Pretty printing reflect.Interface properties, like arch, multilib and
dists.
2) Generating module Bazel rule shims for all module types, instead of
hardcoding them like `soong_filegroup`.
Bug: 162720644
Test: bazel_overlay_test.go (soong build test)
Test: m bazel_overlay && cd out/soong/bazel_overlay && bazel cquery //...
Signed-off-by: Jingwen Chen <jingwen@google.com>
Change-Id: Ic1e448887eb540ed15a55bc4090cf75a4d832d41
The Bazel overlay is a directory at out/soong/bazel_overlay that
replicates the layout of the AOSP Soong module tree, but as a Bazel
workspace. Each Soong module variant is represented as a BUILD target
created with the `soong_module` rule.
To create this overlay, run `m bazel_overlay`.
A `soong_module` target can depend on other `soong_module` targets.
These dependencies replicate each module's `directDeps` in the Blueprint
graph, just before `PrepareBuildActions`.
This enables users to use bazel query as a way to introspect the Soong
module graph. For example,
- Direct reverse dependencies of //bionic/libc:generated_android_ids in
//bionic/libc/...:
$ bazel query 'rdeps(//bionic/libc/...,
//bionic/libc:generated_android_ids, 1)'
//bionic/libc:libc_bionic_ndk--android_recovery_arm_armv7-a-neon_static
//bionic/libc:libc_bionic_ndk--android_ramdisk_arm_armv7-a-neon_static
//bionic/libc:libc_bionic_ndk--android_arm_armv7-a-neon_static_com.android.runtime
//bionic/libc:libc_bionic_ndk--android_arm_armv7-a-neon_static
//bionic/libc:generated_android_ids
- Why does com.android.runtime depend on lzma?
$ bazel query
'somepath(//bionic/apex:com.android.runtime--android_common_com.android.runtime_image,
//external/lzma/...)'
//bionic/apex:com.android.runtime--android_common_com.android.runtime_image
//bionic/libc/malloc_debug:libc_malloc_debug--android_arm_armv7-a-neon_shared_com.android.runtime
//system/core/libunwindstack:libunwindstack--android_arm_armv7-a-neon_shared_com.android.runtime
//external/lzma/C:liblzma--android_arm_armv7-a-neon_shared_com.android.runtime
- What does the dep graph of //bionic/libc:crtbegin_so look like?
$ bazel query
'deps(//bionic/libc:crtbegin_so--android_arm_armv7-a-neon)'
--output=graph > graph.in && dot -Tpng < graph.in > graph.png
https://photos.app.goo.gl/DfsdoFRNsRjGwTmy8
Test: croot && m bazel_overlay && cd out/soong/bazel_overlay && bazel
query //... && bazel query 'rdeps(//bionic/libc/...,
//bionic/libc:generated_android_ids, 1)'
Signed-off-by: Jingwen Chen <jingwen@google.com>
Change-Id: I3bf40309bfb2d963bb8a688706385a57ee304c37#
Currently, the build timestamp saved to the build_date_timestamp
field in the MetricsBase metrics protobuf usually comes from a text
file named build_date.txt. It may not be accurate for metrics purposes.
Instead, use the build start time from Soong main UI for a better
timestamp.
Bug: b/140638454
Test: Ran and checked the build_date_timestamp by running the printproto
command.
Change-Id: I4d893fbbf9830ba21911e56ae13bc5272be47ae6
From aosp/1329396, the RBE metrics protobuf file is part of the
metrics uploading process. The RBE metrics protobuf file is
generated by running the bootstrap shutdown command. A new function
named DumpRBEMetrics was written in order to generate the RBE metrics
protobuf file before sending to the uploading process.
Bug: b/140638454
Test: * Unit test cases
* Ran RBE build on my local host and verified the
metrics protobuf file is created.
* Ran non-RBE build after RBE build and verified that
the previous metrics protobuf file was deleted.
Change-Id: I4b8068905cb67c4b8c2d94793917b98974fed707
Soong and Make have no ways to figure out what splits will be outputted
from a given android_app_set, so it's impossible for them to provide
full PACKAGES.$(LOCAL_MODULE).CERTIFICATE entries, which are required to
build a final apkcerts.txt. This change makes extract_apks produce
apkcerts.txt files for each input modules instead. The Make-side
counterpart of this change merges all local apkcerts.txt into a final
one.
Fixes: 160119159
Test: main_test.go
Test: m apkcerts-list
Merged-In: I321e80fd636a955213761f56a3ac64bfe7f7f7c0
Change-Id: I321e80fd636a955213761f56a3ac64bfe7f7f7c0
Specifically, make Android.bp.list and soong.variables explicit inputs
While this is not a comprehensive list of all inputs
of this action (as the action depends on all blueprint
files in the source tree), this is closer to the truth.
This is a rollforward CL, which was originally rolled
back, as path validation failed for when OUT_DIR was
an absolute path. Validation has now been relaxed.
Test: Manually verified ninja output, checkbuild approved validation
for aosp-crosshatch, and manually ran frameworks/rs/build_rs.py with
patch fix
Change-Id: I4eb0d517f57336dd54eaa4bd31f46df9e93e6da2
While this is not a comprehensive list of all inputs of this action (as the action
depends on all blueprint files in the source tree), this closer to the truth.
Test: Manually verified ninja output, checkbuild approved validation for aosp-crosshatch
Change-Id: I77a0f612afb025232bee7d3ac31257b808991829
The metrics uploader was currently running on foreground where it
would copy the metrics files in a separate directory and then forked
into the background for the upload process. As a result, running the
lunch command would take a second longer to run since each metrics
uploader run had an average of half a second.
Bug: 140638454
Test: * Wrote and updated unit test cases.
* Set ANDROID_ENABLE_METRICS_UPLOAD to point to the latest
metrics_uploader bash script. Executed the "lunch 1" command
and measured the running time. Executed "m nothing" command
and checked that the metrics were uploaded.
* Ran "lunch 1" and "m nothing" with
ANDROID_ENABLE_METRICS_UPLOAD=""
* Removed oauth from metrics_uploader and ran "m nothing" and
"lunch 1". The oauth Message appeared only to "m nothing"
Change-Id: I13c61e666c8f44613dee291a704cef6a27335188
Soong now supports the ability to upload metrics to another location
by setting the ANDROID_ENABLE_METRICS_UPLOAD to an uploader that
accepts the upload.proto proto buffer message. When the environment
variable is set, a set of build metrics files (soong_metrics,
rbe_metrics.pb and build_error) is uploaded.
Bug: 140638454
Test: * Wrote unit test cases
* Setup the uploader, built a succcessful and failed aosp_arm-eng
target and monitor the uploading of the metrics.
Change-Id: I76a65739c557dc90345e098ca03119a950ece2d2
For an end user hitting these errors, the most likely reason is that
they have mixed up linker and crtbegin_dynamic.o from different builds
in their out tree.
Test: Build a linux_bionic executable in a tree exhibiting b/157549171.
Test: m nothing
Bug: 157549171
Change-Id: I64bee0848ed61226da6654caf3bafb51cf46d923
apex_available/min_sdk_version are required to be available for mainline
modules(APK/APEX).
- java_import/java_library_static
- android_library/android_library_import
Bug: 156996905
Test: prebuilts/sdk/update_prebuilts.sh
Change-Id: I201ebabaa533c5ee19e394ab30ca2fa0e3d7ce0e
"dlv attach":
- Spams the terminal
- Requires sysctl -w kernel.yama.ptrace_scope=0
- Apparently, does not allow the debugger to inspect variables
Test: Manual.
Change-Id: I625369effaf5abda2b5d884c8ce5bde7247774eb
apex_set takes an .apks file that contains a set of prebuilt apexes with
different configurations. It uses extract_apks to select and install the
best matching one for the current target.
Bug: 153456259
Test: apex_test.go
Test: com.android.media.apks
Change-Id: I1da8bbcf1611b7c580a0cb225856cbd7029cc0a7
This was a temporary migration tool, and has been broken ever since the
ninja environment started being restricted. Remove it, since it has
lasted two releases.
Test: TEMPORARY_DISABLE_PATH_RESTRICTIONS=true m
Change-Id: I2e3709525598030f5c24d12fa966e531dfed33ae
Golang's FileInfo() implementation recognizes FAT filesystem directories
correctly, but doesn't actually add executable bits, which essentially
makes them inaccessible. Use 0777 when we know we're dealing with
directories.
(This is a cherry-pick change.)
Fixes: 153207404
Test: manually tested with an aar file with FAT permission bits.
Change-Id: Ie9f5320bc44d24b2f9d54d4036bb1747456655bb
Merged-In: Ie9f5320bc44d24b2f9d54d4036bb1747456655bb
Soong UI generates a proto file named build_progress.pb in $(OUT_DIR)
output directory that contains build action numbers (how many are executing,
finished and total) during the course of a build. This is for external
systems that invokes the Platform Build Systems and would like to know
the completion status.
Bug: b/150401146
Test: Wrote a bash script that continuously read the
build_progress.pb file and computed the build completed percentage
while building the aosp_arm-eng target. Compared the percentage between
the Soong output console and the one reported by the bash script.
Change-Id: I7c7347bc8e41958093892d8e2731c4f4169937dd
Collect the number of modules and variants and some basic statistics
on memory usage inside soong_build by writing out a proto that is
read back in by soong_ui.
Test: examine soong.log
Change-Id: I6926876377a4f6229cf41fdbf166ae03c885ea55
This change automates one of the best practices for
custom tools. The best practice says "Prefer to list
input files on the command line, otherwise we may not
know to re-run your command when a new input file is
added." [1]
Normally you'd reference these inputs with $(in) or
one of the forms of the $(location*) substituions on
the command line for the custom tool. However, if the
custom tool does not accept the list of inputs on the
command line, the build system could fail to re-run
the custom tool if the list changes.
This change adds a hash of the list of input names to
the sbox command that wraps the custom tool. If the
list of inputs change, the hash will change, and
therefore the sbox command will change, causing ninja
to re-run the custom tool. The hash is visible to
(but ignored by) the sbox command, and hidden from
your custom tool.
[1] https://android.googlesource.com/platform/build/soong/+/refs/heads/master/docs/best_practices.md#custom-build-tools
Test: TestGenruleHashInputs
Bug: 149397658
Change-Id: I18b547ea3c4296ee15bd6150a4778a8f376d80b7
Create a highmem pool based on the total RAM and the number of CPUs,
with an override via the NINJA_HIGHMEM_NUM_JOBS variable. Put
metalava into the highmem pool.
Ninja does not support nested pools, and when goma or RBE is enabled
the maximum ninja parallelism is set very high with local jobs in a
local pool. When both the local pool and highmem pool are enabled,
the total number of local jobs will be as high as the sum of the sizes
of the two pools. Keep the highmem pool limited to 1/16th of the
local pool when remote builds are enabled to try to minimize the
effect while still limiting highmem jobs.
Fixes: 142644983
Test: m nothing, examine pools
Test: m USE_GOMA=true nothing, examine pools
Change-Id: Id79f11f44948992960ac34ecf831dacbe21bd332
Test: touch empty.zip; merge_zips out.zip empty.zip
(output should contain `empty.zip: zip: not a valid zip file`)
Change-Id: If707876af2301d7ad7d065c004cffb0c2491c7df
Apparently we have some instances of Soong taking 16GB of RAM. While
that's a problem to solve on it's own, make multiproduct_kati's
auto-parallelism detection smarter by taking into account the total RAM
on the machine instead of just the number of CPUs.
Bug: 146925549
Test: check soong.log for autodetected parallelism values
Change-Id: Ice34c2cf73ea4f8f235521cbefc8654a79a04eef
This is so that we can split the build_test build on large branches into
multiple builds, each testing a subset of the products.
Test: m blueprint_tools (runs the go unit tests)
Test: build/soong/build_test.bash --only-config
vs
Test: build/soong/build_test.bash --only-config --shard-count=2
Test: build/soong/build_test.bash --only-config --shard-count=2 --shard=2
Change-Id: I40ccc1aa477bc0ffa74ff564d155068509be18f0
soong_javac_wrapper is silencing a useless warning:
warning: [options] bootstrap class path not set in conjunction with -source 1.9
but recent versions of javac have started also printing:
1 warning
Read the warning count, subtract the number of silenced warnings, and
reprint it if the non-silenced warning count is nonzero.
Fixes: 144118634
Test: javac_wrapper_test.go
Change-Id: Ie1d0a978188ab7b1c41027f718a1274608628123
Force soong_ui in dumpvars mode to use a dumb terminal so that it
doesn't print an empty status table. Also move the logs to files
that are prefixed with "dumpvars-" so that they don't rotate the
real build logs.
Fixes: 139307849
Test: lunch
Change-Id: If92421af07731fedc9ca42670f309e95922a2224
sbox will generate a random directory for the output root, and most
tools will encode that directory name in the output target of the
depfile.
So embed the library from dep_fixer into sbox so that it can rewrite the
output filename to a static (reproducible) value. Ninja doesn't care
what that value is, so it's just "outputfile".
Also fix up rule_builder to actually tell sbox about the depfile.
Test: mmma system/iorap; check the contents of:
out/soong/.intermediates/system/iorap/libiorap-binder/android_arm_armv7-a-neon_core_static/gen/aidl/system/iorap/binder/com/google/android/startop/iorap/IIorap.cpp.d
Change-Id: I3640a2e8b0c034f143a35e398a8418a6d621b265
Android build generates 180K+ zip files that are to be merged and the
result passed to the indexing pipeline to generate the source code
cross-reference. Their names cannot be passed on the command line,
hence the need for this change. That required extensive changes to
the code as it no longer can be assumed that all the input files can
be kept open.
Bug: 121267023
Test: internal + treehugger + combine index files
Change-Id: I0a829f96ff7187ad967fb6b8cede387501ec57cc
mm is now doing the same thing as mma. Now print a deprecation message
for other tools that may be setting ONE_SHOT_MAKEFILE directly (or users
that are still using a very old envsetup.sh).
Test: mm # does not print
Test: ONE_SHOT_MAKEFILE=bionic m
Change-Id: I98221657024d824a61c0d721d5c7dd1a2ea2fb1e
1. Update pom2bp to support "host_support" device module.
2. Allow other module type in support library
(The intention is to support the room annotation processor)
Test: m
Bug: 137918883
Change-Id: I9a8fbcff903cc395d5dc38430774202ce389d6bc
Merged-In: I9a8fbcff903cc395d5dc38430774202ce389d6bc
(cherry picked from commit bd7c2f9a1d)
This is the first part of deprecating the mmm command. mmm will simply
be redirected to mmma by building the list of modules with their
dependencies.
Bug: b/135187558
Test: Ran mmm external/libusb:libusb.
Change-Id: I50cf0d6a07197556ab6130e612561c6d5a3fb31e
This is the first part of deprecating the mm command. mm will simply
be redirected to mma by building the modules and their dependencies.
Bug: b/135187558
Test: Ran mm in external/protobuf directory.
Change-Id: Idad40f89e3768aeb459172f632a44fd7b0a201bc
Allow running Soong in a headless delve debugger by passing
SOONG_DELVE=<listen addr> in the environment.
Bug: 80165685
Test: SOONG_DELVE=:1234 m nothing
Change-Id: Icfc893c8a8354a9bbc99112d9c83259cb41906d1
Keep a running map of the critical path to each edge as it
finishes. At the end of the build, find the edge with the
longest critical path and print the path to out/soong.log.
Test: critical_path_test.go
Change-Id: Ie01d26b068b768156b166bff00cdc3273e8124ca
Soong_ui produces a build_error raw file to $(OUT_DIR) where
the file contains a list of build action errors. Each build action
error represents an error raised by a command. The build_error file
is populated if there was a build error.
Bug: b/132969697
Test: Ran m for successful build. Introduced a broken build change,
ran m and verified using printproto that build_error was
generated successfully.
Change-Id: I690ca1778b4e56f144a3173ba1d16d8494c32c15
There was one case that did not work:
1) Create dir in root source tree
2) Run "m libc"
The build would fail as it was using the mma logic. A separate
one named BUILD_MODULES was added to allow building specific
modules in any directories using the "m" command.
Bug: b/130049705
Test: unit test cases, ran "m libc" in a tmp directory inside
of the source tree.
Change-Id: I8d23e685a673a311001ee8edd89bd73b662392dd
ctx.Writer is recommended stream to use for standard io.
Fixes: b/135487952
Test: Lunch and executed ./out/soong_ui --dumpvar-mode and
--dumpvars-mode to check the usage is output to the console.
Change-Id: I38351ebdb05f49f92d6b4bfb0423092299d1c5ea
The --build-mode flag is a new flag in soong_ui that accepts a build
action (m, mm, mmm, mma, mmma), the directory where the build action
is occuring and an optional list of build arguments and invokes the
build option. This is to deprecate the build actions behavior in
envsetup.sh and start using the one defined in soong_ui.
Bug: b/130049705
Test: Ran lunch and ran all the build action commands in soong_ui.
Ran a bunch of invalid commands such as no Android.mk file in
a specific directory, invalid target name, etc...
Change-Id: I88e9e59f47f8c5cdc872fb83a0f3f0a0919885b6
The metrics.proto needs to be imported to Google3 in order to unmarshal
the build metrics data from bigstore. Cleaned up the enum names to use
the Camel naming convention and renamed the build_metrics to
soong_metrics to be more specific to soong and to allow quering
soong_metrics files from bigstore.
Bug: b/135280521
Test: lunch andf m -j. Checked if soong_metrics was generated correctly
by using printproto command.
Change-Id: I998c8d05db592e94a653d6ca32250b80df3c9b21
Write log output through StatusOutput so that the status implementation
can synchronize it with its own output.
Test: status_test.go
Change-Id: I917bdeeea4759a12b6b4aa6d6d86ee18a2771723
terminal.Writer is now just a wrapper around stdio.Stdout() without
any useful functionality. Replace it with stdio.Stdout() as an
io.Writer.
Test: ui/terminal/status_test.go
Change-Id: I5bc5476afdca950b505642f0135a3af9d37fbe24
Instead of extracting the aar, and passing the res directory to another rule
(along with a dependency to a different known file), don't keep the
extracted res directory, and pass the aar directly to the resource
compliation rule, which will extract just the res directory.
I need this for my RBE experiments, where non-listed output files won't
exist in other rules.
Test: m
Change-Id: I99074381052cbcebb6a402484abae9ab2e40284f
