Revert submission 2651299-bazel_sandwich
Reason for revert: b/293883239, checking if this breaks the build.
Reverted changes: /q/submissionid:2651299-bazel_sandwich
Change-Id: I5f1d4bb13d21e6599f5c353dcaba2375f5ec234d
The "bazel sandwich" is a mechanism for bazel to depend on make/soong
outputs. The name comes from the fact that bazel is now at the top
and bottom of the build graph. This is intended to allow us to work
on converting the partition builds to bazel while not all of the
dependencies of the partition have been converted.
It works by adding the bazel_sandwich_import_file rule, which emits a
dangling symlink that starts with bazel_sandwich:, and includes
information that the aquery handler in soong reads. The aquery handler
rewrites the symlink so that it points to a file generated by
make/soong, and adds a ninja dependency from the symlink to the file
it's targeting.
This allows us to depend on make-built files from bazel, but notably
it doesn't allow us to depend on analysis-time information from make.
This shouldn't be a problem for the partitions, but limits the use of
the bazel sandwich to similar, less complicated types of builds.
go/roboleaf-bazel-sandwich
Bug: 265127181
Test: m bazel_sandwich
Change-Id: Ic41bae7be0b55f251d04a6a95f846c50ce897adc
This mnemonic has '' args, which would become literal empty when
combined by strings.Join. To prevent this, add it to the
special-handling function.
Bug: 284483729
Test: N/A
Change-Id: Ibd3c3ebc55284882e7336487319070b9710076d3
This is prep work to add additional special handling for genrule
generated headers as there will be similar partitioning for those
headers.
Test: go test soong tests
Change-Id: Ib63e7e4f7554b2b7b7bc78b2825b20c05403216a
The use case for this is for building rules_go's root builder which runs
into issues when built in a directory that contains a symlink to
prebuilts/go
The implementation will involve two changes of working dir
- `sbox` to change the working directory to
__SBOX_SANDBOX_DIR__
- the generated manifest will change the working
directory to mixed build execution root relative to that
Implemenation details
1. Create a unique intermediate path by hashing the outputs of a buildAction.
"out/bazel/output/execroot/__main__/" was deliberately not chosen as
the outpuDir for the sandbox because ruleBuilder would wipe it.
`sbox` will generate the files in __SBOX_SANDBOX_DIR__ and then place
the files in this intermediate directory.
2. After the files have been generated in (1), copy them to
out/bazel/output/execroot/__main__/...
3. For bazel depsets that are inputs of an action, copy the direct
artifacts into the sandbox instead of the phony target
4. Make sandboxing an opt-in. Currently we will only use it for
`GoToolchainBinaryBuild`
In the current implementation, (3) will increase the size of the ninja
file. With sboxing turned on for only GoToolchainBinaryBuild, this will
increase the size of the ninja file by around 1.3% on aosp's cf
Test: m com.android.neuralnetworks (will build soong_zip from source
using rules_go)
Test: OUT_DIR=out.other m com.android.neuralnetworks
Bug: 289102849
Change-Id: I7addda9af583ba0ff306e50c1dfa16ed16c29799
These are essentially FileWrite actions, like SourceSymlinkManifest.
However, currently aquery appears to give non-deterministic results
for the file content. Emit an empty file for now to avoid the
non-determinism.
Fixes: 288292933
Bug: 287539324
Test: Presubmits
Change-Id: Ieaf8aff0d58d2d69087e4e34dc99074a015e700c
GoCompilePkg action's aquery results contains ''. e.g (the interesting
ones)
```
/bin/bulder ... -embedroot '' -gcflags '' -asmflags ''
```
strings.Join would cause it to appear as a whitespace and cause issues.
(in ninja without this CL)
```
/bin/builder ... -embedroot -gcflags -asmflags
```
Convert the empty literal to '' before strings.Join
To limit inadvertent side-effects, this limits it to GoCompilePkg mnemonic
Test: TH
Change-Id: I67161c194dabac6f857ff49b85d4b2471970a9b2
prebuilt_file Bazel targets do not have any build actions per se, but
return a PrebuiltFileInfo provider that determines installation rules.
To integrate this info into mixed builds, add a `GetPrebuiltFileInfo` method to
the API.
Bug: 280094612
Test: unit tests for prebuilt_etc in the next CL of this stack
Change-Id: I79da6b75b6a6cbd30869860d3620aeda731ea36a
This allows USE_BAZEL_VERSION to be set for m builds, which will use
Bazelisk for any Bazel invocations during those builds.
This should be used only for manual debugging, typically to either test
new Bazel features, verify Bazel compatibility with Android, or culprit
find new Bazel breakages.
Test: Manually run builds with USE_BAZEL_VERSION, toggled off and on to
ensure the build was rerun. Tested with a broken commit, a working
commit, and 'last_green' special term
Change-Id: I8b475dca5c8d4bd849ee3724a8c3aca9b631bcb8
host_ldlibs are used during linking. Set these in `linkopts` for now.
Note that this CL does not do the `CheckBadHostLdLibs` validation of
Soong. There are some different ways to do this, and these are being
discussed in b/216626461. It is likely that we will need to create a new
property `host_ldlibs` to do the validation. But for now, re-use
`linkopts`.
Bug: 216626461
Test: bp2build unit tests
Change-Id: Id4c77e4460fb1fb003fa58ea27bab5b50ea8cefe
* changes:
Special case platform variant of bootstrap libs
Select stub/impl per apex variant
For test apexes, base_apex_name is the api domain
Broaden the granularity of config_setting from apex_name to api_domain
Print default val if all vals in axis match default val
For converting the art plugins to pure soong, it would be useful to
have a property that's qualified on both a soong config variable and
the OS. Soong had very little-known support for this by saying your
soong config variable changes the "target.android.cflags" property,
and we didn't supporting bp2building that. Add the bp2build support.
This cl also refactors product variable and soong variable bp2building
so that they're separate from each other, which I think makes the code
easier to understand.
Test: go test
Change-Id: Ic74dc75da8103fa2523da95c3560c9ce3c5e5672
Create a select statement for every api_domain a library could be
included in. The stub/impl
selection heuristics per apex domain are
1. If dep has stubs and the api domain appears in dep's apex_available, use
impl
2. If dep has stubs and the api domain does not appear in dep's
apex_available, use stubs
(Category 3: If dep does not have stubs and the apex does not appear in
dep's apex_available, then a separate apex_available validation in Bazel
will emit an error).
Platform variants have been special-cased for now to use equality of
apex_available for stub/impl selection
Test: go test ./bp2build
Bug: 272378496
Change-Id: Ibd29efd763c8863c7e6d2a9af0da30bbde07175d
The use case for this is for creating a select statement for stub/impl
selection. Since Bazel propagates apex_name from the top-level apex,
the source apex and test apex builds a specific library in two different
configurations. We need select statements for both these two
configurations, but this metadata might not always exist in Android.bp
since test apexes are not necessary to be listed in Android.bp files.
To overcome this, the select statements will be created per api domain
instead. This CL uses a naming convention to infer the api domain of
apexes.
Test: go test ./bp2build
Change-Id: Iad2b45f736bc98a24d2ce1cf2f69aad67973092d
These are created by bp2build in /build/bazel/rules/apex. Eventually
these config_settings should likely be colocated with the source apex
definition.
Another alternative was to make Bazel's apex a macro that generates a
config_setting. I did not pursue this further for now since it requires the
apex_available of every allowlisted cc_library to also be allowlisted.
This might not always be true (e.g. com.android.runtime)
Test: go test ./bp2build
Change-Id: Ibbb14b0d9c1491b3c79b7634a18d9d35b03922c1
The use case for this is creating config_setting(s) specific to an apex
variant and selecting stub/impl in that config_setting. We likely need
only a handful of such config_setting(s), but determining that list
requires iterating the build graph.
Test: go test ./bp2build
Change-Id: I9aa552e3d0bcf67513023c3a7d4bbf8fae464ee4
- Creation of a bazel command is independent of test/real implementation
- Use a custom struct instead of cmd.Exec
- Move mock bazel runner to the test
Bug: 270989498
Test: m nothing
Test: USE_PERSISTENT_BAZEL=0 m nothing
Test: Treehugger
Change-Id: Ieec35dad5e21aac644d5b8dc79a92397d42db861
This reverts commit 1db4348734.
Changes from original:
- extracted function to add the current version to stub
versions for bp2build.
- added libc++ to mixed builds denylist
Reason for revert: re-uploading with fix in topic
Change-Id: Ifa0ed456bf8cb4a7f861d6826263adfedb4fdd9c
This reverts commit ca438e6b72.
Reason for revert: ci post-submit failures in aosp-master-bazel of targets mixed-droid-clean and mixed-droid-incremental:
FAILED: ninja: 'out/target/product/generic_arm64/obj/SHARED_LIBRARIES/libc_intermediates/libc.so.toc', needed by 'out/target/product/generic_arm64/obj/EXECUTABLES/updater_intermediates/LINKED/updater', missing and no known rule to make it
12:21:27 ninja failed with: exit status 1
Change-Id: I081b499d23f2568cdf6227c4e3b0278164086b69
Bug: 270408757
Test: go test
Test: m NeuralNetworksTest_shared_partial --bazel-mode-staging
&& verify that Ninja commands link libneuralnetworks via stubs
Change-Id: I19bf5a829cea7fd00c0f82511ad2bddbc40fae4f
Now that the json module is available in the cquery starlark execution
enviornment.
Bug: 245315355
Test: m nothing
Change-Id: Ie7c314e2f4e2ef4257bd84abc1add01ddb668f0b
This CL adds a few things:
1) Populate the filesInfo struct with cquery'd information from an
apex's ApexMkInfo provider. This filesInfo is then used in
apex/androidmk.go to generate Make modules (soong_cc_rust_prebuilt.mk),
which are then used in packaging to generate zip files of symbols in $PRODUCT_OUT.
2) Make a list of dicts of primitives JSON-encodable.
3) Tests.
Bug: 271423316
Bug: 271423062
Test: presubmits
Change-Id: Iaa34f51044de310510e580d9cf1fe60bbef801c1
This feature is toggled on with USE_PERSISTENT_BAZEL, which is off by
default. Those that opt-in will have a bazel server running between
builds (with a 3hr default TTL) which will greatly improve analysis on
subsequent builds. (As Bazel maintains a cache of analysis results).
Bug: 266983462
Test: Manual `m nothing` runs (timing with and without the feature)
Test: New integration test
Test: Presubmits
Change-Id: I3af4948baa0c490e9b87c48ffdbe9f67732586c7
Creation of build statements is largely parallelizable because each
action is independent apart from updates/reads to
depsetHashToArtifactPathsCache. Locally resulted in build statements
taking ~.45 seconds on staging mode to ~.02 seconds
Test: CI
Change-Id: Iab00c8394a9eab17353f71230885ff0870e17f24
This prevents unnecessary copies of data from the query by using a
pointer to the proto or its contents
Test: go test soong tests
Test: m nothing
Change-Id: I5962b27fe31f7103d2af6cef159f11747cff3ed9
Previously in mixed builds, only the tidy files for the boundary module
were built, whereas all of its transitive dependencies' tidy files were
not being built. Instead we should export the list of transitive tidy
files for a module so that we can run clang-tidy for the boundary module
as well as its dependencies.
Bug: 195029134
Test: WITH_TIDY=1 DISABLE_ARTIFACT_PATH_REQUIREMENTS=true mss tidy-packages-modules-NeuralNetworks --bazel-mode-dev
Change-Id: I463646d2ae1fc4aa075a54c264e1c34571c3fd5c
This fixes two main sources of nondeterminism:
1. Fix a bug in the ConfigurationAxis comparator (which caused
ConfigurationAxis sorting to be nondeterministic)
2. Process C++ dependencies using the sorted ConfigurationAxis order. In
theory, the order in which dependencies are processed shouldn't
matter (as they should end up in different select stanzas). However,
in the case of InApex stubs, this is not the case; we now ensure
that lists are concatenated in a predictable order.
Added bonus: Some cleanup with SortConfigurationAxes which
makes use of go generics (this made it easier to debug this issue).
Will follow-up with regression tests.
Test: Manually verified that build.ninja checksum and BUILD.bazel checksums do not change after running `m nothing` 6 times in AOSP (with comment-only Android.bp changes in between each run)
Change-Id: I81168e45bdbbcd61ea95ff665cf6c4bc180aa4e0
Instead of using a sentinel file that was created under bazel_workspace
prune them out. The sentinel file was being created under bazel_workspace,
which would be deleted and thus rendered useless from the perspective of
not retriggering a build.
Test: `touch Android.bp` then `NINJA_ARGS="-d explain" m --bazel-mode-staging adbd_test`
Bug: 265155778
Change-Id: If5d8f0ea7f3b8828fda0646faafd1a621f0cb27c
This piggybacks onto the ApexInfo cquery handler, so we're issuing a
single bazel query call that reads two providers in the starlark expr.
Also rename requiredDeps to makeModulesToInstall to differentiate it from
APEX's required/provided libs in the apex manifest.
Test: unit test
Test: mkdiff
Fixes: 263123189
Change-Id: Ib7e43f1586f29864eee8627dba3631bfaff27afa
The adbd_test androidmk definition is missing some cc-related variables
for cc_test in mixed builds. These variables should be populated from
information from Bazel.
Bug: 265758350
Change-Id: I59d017e2eb2f139188ba3383c457cc0055372b61
