Specifies the set of environment variables available to actions with target configuration. Variables can be either specified by `name`, in which case the value will be taken from the invocation environment, by the `name=value` pair which sets the value independent of the invocation environment, or by `=name`, which unsets the variable of that name. This option can be used multiple times; for options given for the same variable, the latest wins, options for different variables accumulate. Note that unless `--incompatible_repo_env_ignores_action_env` is true, all `name=value` pairs will be available to repository rules.

If discarding the analysis cache due to a change in the build system, setting this option to false will cause bazel to exit, rather than continuing with the build. This option has no effect when `--discard_analysis_cache` is also set.

If true, an analysis failure of a rule target results in the target's propagation of an instance of `AnalysisFailureInfo` containing the error description, instead of resulting in a build failure.

If enabled, issue only a warning instead of an error for loads of experimental .bzls.

If set, allow at least one action to run even if the resource is not enough or unavailable.

Specified the module versions in the form of `{module1}@{version1},module2@{version2}` that will be allowed in the resolved dependency graph even if they are declared yanked in the registry where they come from (if they are not coming from a [`NonRegistryOverride`]). Otherwise, yanked versions will cause the resolution to fail. You can also define allowed yanked versions with the `BZLMOD_ALLOW_YANKED_VERSIONS` environment variable. You can disable this check by using the keyword `all` (not recommended). [`NonRegistryOverride`]: https://github. com/bazelbuild/bazel/blob/master/src/main/java/com/google/devtools/build/lib/bazel/bzlmod/NonRegistryOverride. java

Allowed values for the `--cpu` flag.

Filters spawn strategies by the execution platform without affecting order. For example: ``` common --spawn_strategy=remote,sandboxed,worker,local common --strategy=Genrule=local common --allowed_strategies_by_exec_platform=@platforms//host:host=local, sandboxed,worker common --allowed_strategies_by_exec_platform=//:linux_amd64=remote ``` With the above options; - Actions configured for the host platform will be given `remote,sandboxed, worker`. - Actions configured for the `//:linux_amd64` platform will be given `remote`. - Actions configured for the `//:linux_amd64` platform with mnemonic `Genrule` will be given no strategies and fail to spawn.

Sets the maximum number of transitive dependencies through a rule attribute with a `for_analysis_testing` configuration transition. Exceeding this limit will result in a rule error.

The Android target compiler.

Generate AndroidX-compatible data-binding files. This is only used with databinding v2. This flag is a no-op.

Use android databinding v2 with 3.4.0 argument. This flag is a no-op.

Determines whether C++ deps of Android rules will be linked dynamically when a cc_binary does not explicitly create a shared library. 'default' means bazel will choose whether to link dynamically. 'fully' means all libraries will be linked dynamically. 'off' means that all libraries will be linked in mostly static mode.

Selects the manifest merger to use for android_binary rules. Flag to help the transition to the Android manifest merger from the legacy merger.

Sets the order of manifests passed to the manifest merger for Android binaries. ALPHABETICAL means manifests are sorted by path relative to the execroot. ALPHABETICAL_BY_CONFIGURATION means manifests are sorted by paths relative to the configuration directory within the output directory. DEPENDENCY means manifests are ordered with each library's manifest coming before the manifests of its dependencies.

Sets the platforms that android_binary targets use. If multiple platforms are specified, then the binary is a fat APKs, which contains native binaries for each specified target platform.

Enables resource shrinking for android_binary APKs that use ProGuard.

Whether to announce rc options.

Implementation to use to sign APKs

Whether to generate debug symbol(.dSYM) file(s).

Comma-separated list of aspects to be applied to top-level targets. In the list, ifaspect `some_aspect` specifies required aspect providers via`required_aspect_providers`, `some_aspect` will run afterevery aspect that was mentioned before it in the aspects list whose advertisedproviders satisfy `some_aspect` required aspect providers. Moreover,`some_aspect` will run after all its required aspects specified by`requires` attribute.`some_aspect` will then have access to the values of those aspects'providers.`{bzl-file-label}%{aspect_name}`, for example `//tools:my_def.bzl%my_aspect`, where`my_aspect` is a top-level value from a file `tools/my_def.bzl`.

Specifies the values of the command-line aspects parameters. Each parameter value is specified via `<param_name>=<param_value>`, for example `my_param=my_val` where `my_param` is a parameter of some aspect in `--aspects` list or required by an aspect in the list. This option can be used multiple times. However, it is not allowed to assign values to the same parameter more than once.

When printing the location part of messages, attempt to use a path relative to the workspace directory or one of the directories specified by -- package_path.

If --output_filter is not specified, then the value for this option is used create a filter automatically. Allowed values are 'none' (filter nothing / show everything), 'all' (filter everything / show nothing), 'packages' (include output from rules in packages mentioned on the Blaze command line), and 'subpackages' (like 'packages', but also include subpackages). For the 'packages' and 'subpackages' values //java/foo and //javatests/foo are treated as one package)'.

Maximum number of open files allowed during BEP artifact upload.

Specifies the build event service (BES) backend endpoint in the form `[SCHEME://]HOST[:PORT]`. The default is to disable BES uploads. Supported schemes are `grpc` and `grpcs` (grpc with TLS enabled). If no scheme is provided, Bazel assumes `grpcs`.

Sets the field `check_preceding_lifecycle_events_present` on `PublishBuildToolEventStreamRequest` which tells BES to check whether it previously received `InvocationAttemptStarted` and `BuildEnqueued` events matching the current tool event.

Specify a header in `NAME=VALUE` form that will be included in BES requests. Multiple headers can be passed by specifying the flag multiple times. Multiple values for the same name will be converted to a comma-separated list.

Specifies the instance name under which the BES will persist uploaded BEP. Defaults to null.

Specifies a list of notification keywords to be added the default set of keywords published to BES (`command_name={command_name}`, `protocol_name=BEP`). Defaults to none.

Specifies whether to publish BES lifecycle events. (defaults to 'true').

Specifies how long bazel should wait for the BES/BEP upload to complete while OOMing. This flag ensures termination when the JVM is severely GC thrashing and cannot make progress on any user thread.

Specifies the maximal size of stdout or stderr to be buffered in BEP, before it is reported as a progress event. Individual writes are still reported in a single event, even if larger than the specified value up to `-- bes_outerr_chunk_size`.

Specifies the maximal size of stdout or stderr to be sent to BEP in a single message.

Connect to the Build Event Service through a proxy. Currently this flag can only beused to configure a Unix domain socket (`unix:/path/to/socket`).

Specifies the base URL where a user can view the information streamed to the BES backend. Bazel will output the URL appended by the invocation id to the terminal.

Specifies a list of notification keywords to be included directly, without the `user_keyword=` prefix included for keywords supplied via `--bes_keywords`. Intended for Build service operators that set `-- bes_lifecycle_events=false` and include keywords when calling `PublishLifecycleEvent`. Build service operators using this flag should prevent users from overriding the flag value.

Specifies how long bazel should wait for the BES/BEP upload to complete after the build and tests have finished. A valid timeout is a natural number followed by a unit: Days (d), hours (h), minutes (m), seconds (s), and milliseconds (ms). The default value is `0` which means that there is no timeout.

Specifies whether the Build Event Service upload should block the build completion or should end the invocation immediately and finish the upload in the background. - `wait_for_upload_complete`: blocks at the end of the current invocation until all events (including lifecycle events if applicable) are uploaded and acknowledged by the backend. - `nowait_for_upload_complete`: blocks at the beginning of the next invocation until all events (including lifecycle events if applicable) are uploaded and acknowledged by the backend. - `fully_async`: blocks at the beginning of the next invocation until all events are uploaded but does not wait for acknowledgements. Events may be lost in case of (transient) failures and backends may report streams as incomplete in this mode. There is no guarantee that `FinishInvocationAttempt` or `FinishBuild` lifecycle events are sent.

If true dex2oat action failures will cause the build to break instead of executing dex2oat during test runtime.

Execute the build; this is the usual behaviour. Specifying `--nobuild` causes the build to stop before executing the build actions, returning zero if the package loading and analysis phases completed successfully; this mode is useful for testing those phases.

If non-empty, write a varint delimited binary representation of representation of the build event protocol to that file. This option implies `--bes_upload_mode=wait_for_upload_complete`.

Convert paths in the binary file representation of the build event protocol to more globally valid URIs whenever possible; if disabled, the `file://` uri scheme will always be used

Specifies whether the Build Event Service upload for `-- build_event_binary_file` should block the build completion or should end the invocation immediately and finish the upload in the background. Either `wait_for_upload_complete` (default), `nowait_for_upload_complete`, or `fully_async`.

If non-empty, write a JSON serialisation of the build event protocol to that file. This option implies `--bes_upload_mode=wait_for_upload_complete`.

Convert paths in the json file representation of the build event protocol to more globally valid URIs whenever possible; if disabled, the `file://` uri scheme will always be used

Specifies whether the Build Event Service upload for `-- build_event_json_file` should block the build completion or should end the invocation immediately and finish the upload in the background. Either `wait_for_upload_complete` (default), `nowait_for_upload_complete`, or `fully_async`.

The maximum number of entries for a single `named_set_of_files` event; values smaller than 2 are ignored and no event splitting is performed. This is intended for limiting the maximum event size in the build event protocol, although it does not directly control event size. The total event size is a function of the structure of the set as well as the file and uri lengths, which may in turn depend on the hash function.

Whether all actions should be published.

If non-empty, write a textual representation of the build event protocol to that file

Convert paths in the text file representation of the build event protocol to more globally valid URIs whenever possible; if disabled, the `file://` uri scheme will always be used

Specifies whether the Build Event Service upload for `-- build_event_text_file` should block the build completion or should end the invocation immediately and finish the upload in the background. Either `wait_for_upload_complete` (default), `nowait_for_upload_complete`, or `fully_async`.

The maximum number of times Bazel should retry uploading a build event.

Forces test targets tagged 'manual' to be built. 'manual' tests are excluded from processing. This option forces them to be built (but not executed).

Custom key-value string pairs to supply in a build event.

Build python executable zip; on on Windows, off on other platforms

If true, build runfiles symlink forests for all targets. If false, write them only when required by a local action, test or run command.

If true, write runfiles manifests for all targets. If false, omit them. Local tests will fail to run when false.

Specifies a comma-separated list of tags. Each tag can be optionally preceded with '-' to specify excluded tags. Only those targets will be built that contain at least one included tag and do not contain any excluded tags. This option does not affect the set of tests executed with the 'test' command; those are be governed by the test filtering options, for example '--test_tag_filters'

If enabled, when building C++ tests statically and with fission the .dwp file for the test binary will be automatically built as well.

If specified, only *_test and test_suite rules will be built and other targets specified on the command line will be ignored. By default everything that was requested will be built.

If greater than 0, configures Bazel to cache file digests in memory based on their metadata instead of recomputing the digests from disk every time they are needed. Setting this to 0 ensures correctness because not all file changes can be noted from file metadata. When not 0, the number indicates the size of the cache as the number of file digests to be cached.

If set to `auto`, Bazel reruns a test if and only if:1. Bazel detects changes in the test or its dependencies,2. The test is marked as `external`,3. Multiple test runs were requested with `--runs_per_test`, or4. The test previously failed.If set to `yes`, Bazel caches all test results except for tests marked as`external`. If set to `no`, Bazel does not cache any test results.

Output the canonical policy, after expansion and filtering. To keep the output clean, the canonicalized command arguments will NOT be shown when this option is set to true. Note that the command specified by -- for_command affects the filtered policy, and if none is specified, the default command is 'build'.

Comma-separated list of architectures for which to build Apple Catalyst binaries.

Whether to generate and analyze .d files.

Whether to narrow inputs to C/C++ compilation by parsing #include lines from input files. This can improve performance and incrementality by decreasing the size of compilation input trees. However, it can also break builds because the include scanner does not fully implement C preprocessor semantics. In particular, it does not understand dynamic #include directives and ignores preprocessor conditional logic. Use at your own risk. Any issues relating to this flag that are filed will be closed. At Google without this flag your build will most likely fail.

Specifies a suffix to be added to the configuration directory.

Sets the suffixes of header files that a cc_proto_library creates.

Sets the suffixes of source files that a cc_proto_library creates.

Check bazel version compatibility of Bazel modules. Valid values are `error` to escalate it to a resolution failure, `off` to disable the check, or `warning` to print a warning when mismatch detected.

If disabled, .bzl load visibility errors are demoted to warnings.

Check if the direct `bazel_dep` dependencies declared in the root module are the same versions you get in the resolved dependency graph. Valid values are `off` to disable the check, `warning` to print a warning when mismatch detected or `error` to escalate it to a resolution failure.

Don't run tests, just check if they are up-to-date. If all tests results are up-to-date, the testing completes successfully. If any test needs to be built or executed, an error is reported and the testing fails. This option implies --check_up_to_date behavior. Using this option will also add: --check_up_to_date

Don't perform the build, just check if it is up-to-date. If all targets are up-to-date, the build completes successfully. If any step needs to be executed an error is reported and the build fails.

If disabled, visibility errors in target dependencies are demoted to warnings.

If specified, Bazel will instrument code (using offline instrumentation where possible) and will collect coverage information during tests. Only targets that match `--instrumentation_filter` will be affected. Usually this option should not be specified directly - `bazel coverage` command should be used instead.

Use terminal controls to colorize output.

Specifies desired cumulative coverage report type. At this point only LCOV is supported.

Specify the mode the binary will be built in. Values: `fastbuild`, `dbg`, `opt`.

Compile a single dependency of the argument files. This is useful for syntax checking source files in IDEs, for example, by rebuilding a single target that depends on the source file to detect errors as early as possible in the edit/build/test cycle. This argument affects the way all non-flag arguments are interpreted; instead of being targets to build they are source filenames. For each source filename an arbitrary target that depends on it will be built.

The C++ compiler to use for compiling the target.

Selects additional config sections from the rc files; for every <command>, it also pulls in the options from <command>:<config> if such a section exists; if this section doesn't exist in any .rc file, Blaze fails with an error. The config sections and flag combinations they are equivalent to are located in the tools/*.blazerc config files.

Additional option to pass to gcc when compiling C source files.

Additional options to pass to gcc.

Location of the binary that is used to postprocess raw coverage reports. This must be a binary target. Defaults to `@bazel_tools//tools/test:lcov_merger`.

Location of the binary that is used to generate coverage reports. This must be a binary target. Defaults to `@bazel_tools//tools/test: coverage_report_generator`.

Location of support files that are required on the inputs of every test action that collects code coverage. Defaults to `//tools/test:coverage_support`.

Deprecated: this flag is not used internally by Blaze although there are legacy platform mappings to allow for backwards compatibility. Do not use this flag, instead use `--platforms` with an appropriate platform definition.

Configures a credential helper conforming to the [Credential Helper Specification][ch-spec]to use for retrieving authorization credentials for repositoryfetching, remote caching and execution, and the build event service.Credentials supplied by a helper take precedence over credentials supplied by`--google_default_credentials`, `--google_credentials`, a `.netrc` file, or the auth parameter to `repository_ctx.download()` and`repository_ctx.download_and_extract()`.May be specified multiple times to set up multiple helpers.See [Configuring Bazel's Credential Helper - Engflow Blog][ch-example] for instructions.[ch-spec]: https://github.com/EngFlow/credential-helper-spec[ch-example]: https://blog.engflow.com/2023/10/09/configuring-bazels-credential-helper/

How long to cache credentials for if the credential helper doesn't return an expiration time. Changing the value of this flag clears the cache.

Configures the timeout for a credential helper.Credential helpers failing to respond within this timeout will fail the invocation.

Use CSFDO profile information to optimize compilation. Specify the absolute path name of the zip file containing the profile file, a raw or an indexed LLVM profile file.

Generate binaries with context sensitive FDO instrumentation. With Clang/LLVM compiler, it also accepts the directory name under which the raw profile file(s) will be dumped at runtime. Using this option will also add: --copt=-Wno-error

The cs_fdo_profile representing the context sensitive profile to be used for optimization.

Use terminal cursor controls to minimize scrolling output.

Specifies a custom malloc implementation. This setting overrides malloc attributes in build rules.

Additional option to pass to gcc when compiling C++ source files.

Override the default resources amount for tests. The expected format is`{resource}={value}`. If a single positive number is specified as `{value}`it will override the default resources for all test sizes. If 4comma-separated numbers are specified, they will override the resourceamount for respectively the `small`, `medium`, `large`, `enormous` test sizes.Values can also be `HOST_RAM`/`HOST_CPU`, optionally followedby `[-|*]{float}` (eg. `memory=HOST_RAM*.1,HOST_RAM*.2,HOST_RAM*.3,HOST_RAM*.4`).The default test resources specified by this flag are overridden by explicitresources specified in tags.

Each `--define` option specifies an assignment for a build variable. In case of multiple values for a variable, the last one wins.

A comma-separated list of names of packages which the build system will consider non-existent, even if they are visible somewhere on the package path.Use this option when deleting a subpackage 'x/y' of an existing package 'x'. For example, after deleting x/y/BUILD in your client, the build system may complain if it encounters a label '//x:y/z' if that is still provided by another package_path entry. Specifying --deleted_packages x/y avoids this problem.

Whether to desugar Java 8 bytecode before dexing.

Whether to include supported Java 8 libraries in apps for legacy devices.

If set, and compilation mode is not 'opt', objc apps will include debug entitlements when signing.

Discard the analysis cache immediately after the analysis phase completes. Reduces memory usage by ~10%, but makes further incremental builds slower.

A path to a directory where Bazel can read and write actions and action outputs. If the directory does not exist, it will be created.

Additional places to search for archives before accessing the network to download them.

Specify a file to configure the remote downloader with. This file consists of lines, each of which starts with a directive (`allow`, `block` or `rewrite`) followed by either a host name (for `allow` and `block`) or two patterns, one to match against, and one to use as a substitute URL, with back-references starting from `$1`. It is possible for multiple `rewrite` directives for the same URL to be given, and in this case multiple URLs will be returned.

How many milliseconds should local execution be delayed, if remote execution was faster during a build at least once?

The local strategies, in order, to use for the given mnemonic - the first applicable strategy is used. For example, `worker,sandboxed` runs actions that support persistent workers using the worker strategy, and all others using the sandboxed strategy. If no mnemonic is given, the list of strategies is used as the fallback for all mnemonics. The default fallback list is `worker,sandboxed`, or`worker,sandboxed,standalone` if `experimental_local_lockfree_output` is set. Takes [mnemonic=]local_strategy [,local_strategy,...]

Determines whether C++ binaries will be linked dynamically. 'default' means Bazel will choose whether to link dynamically. 'fully' means all libraries will be linked dynamically. 'off' means that all libraries will be linked in mostly static mode.

The remote strategies, in order, to use for the given mnemonic - the first applicable strategy is used. If no mnemonic is given, the list of strategies is used as the fallback for all mnemonics. The default fallback list is `remote`, so this flag usually does not need to be set explicitly. Takes [mnemonic=]remote_strategy[,remote_strategy,...]

Embed source control revision or release label in binary

If true, Bazel picks up host-OS-specific config lines from bazelrc files. For example, if the host OS is Linux and you run bazel build, Bazel picks up lines starting with build:linux. Supported OS identifiers are linux, macos, windows, freebsd, and openbsd. Enabling this flag is equivalent to using --config=linux on Linux, --config=windows on Windows, etc.

If set, any use of absolute paths for propeller optimize will raise an error.

If set, any use of absolute paths for FDO will raise an error.

Enable runfiles symlink tree; By default, it's off on Windows, on on other platforms.

Checks the environments each target is compatible with and reports errors if any target has dependencies that don't support the same environments

Comma-separated list of aspects to be applied to exec-configured targets, regardless of whether or not they are top-level targets. This is an experimental feature and is subject to change.

Log the executed spawns into this file as length-delimited SpawnExec protos, according to src/main/protobuf/spawn.proto. Prefer --execution_log_compact_file, which is significantly smaller and cheaper to produce. Related flags: --execution_log_compact_file (compact format; mutually exclusive), --execution_log_json_file (text JSON format; mutually exclusive), --execution_log_sort (whether to sort the execution log), --subcommands (for displaying subcommands in terminal output).

Log the executed spawns into this file as length-delimited ExecLogEntry protos, according to src/main/protobuf/spawn.proto. The entire file is zstd compressed. Related flags: --execution_log_binary_file (binary protobuf format; mutually exclusive), --execution_log_json_file (text JSON format; mutually exclusive), --subcommands (for displaying subcommands in terminal output).

Log the executed spawns into this file as newline-delimited JSON representations of SpawnExec protos, according to src/main/protobuf/spawn.proto. Prefer --execution_log_compact_file, which is significantly smaller and cheaper to produce. Related flags: --execution_log_compact_file (compact format; mutually exclusive), --execution_log_binary_file (binary protobuf format; mutually exclusive), --execution_log_sort (whether to sort the execution log), --subcommands (for displaying subcommands in terminal output).

Whether to sort the execution log, making it easier to compare logs across invocations. Set to false to avoid potentially significant CPU and memory usage at the end of the invocation, at the cost of producing the log in nondeterministic execution order. Only applies to the binary and JSON formats; the compact format is never sorted.

Expand test_suite targets into their constituent tests before analysis. When this flag is turned on (the default), negative target patterns will apply to the tests belonging to the test suite, otherwise they will not. Turning off this flag is useful when top-level aspects are applied at command line: then they can analyze test_suite targets.

How long the server must remain idle before a garbage collection of the action cache is attempted. Ineffectual unless -- experimental_action_cache_gc_max_age is nonzero.

If set to a nonzero value, the action cache will be periodically garbage collected to remove entries older than this age. Garbage collection occurs in the background once the server has become idle, as determined by the -- experimental_action_cache_gc_idle_delay and -- experimental_action_cache_gc_threshold flags.

The percentage of stale action cache entries required for garbage collection to be triggered. Ineffectual unless -- experimental_action_cache_gc_max_age is nonzero.

Deprecated in favor of aspects. Use `action_listener` to attach an `extra_action` to existing build actions.

Active directories for Skyfocus and remote analysis caching. Specify as comma-separated workspace root-relative paths. This is a stateful flag. Defining one persists it for subsequent invocations, until it is redefined with a new set.

Compress Java resources in APKs

Use android databinding v2. This flag is a no-op.

Enables resource shrinking for android_binary APKs that use ProGuard.

use rex tool to rewrite dex files

Use dex2oat in parallel to possibly speed up android_test.

If set to true, Bazel is allowed to run action in a virtual thread. The number of actions in flight is still capped with `--jobs`.

The number of maximum concurrent actions to run with async execution. If the value is less than `--jobs`, it is clamped to `--jobs`.

Whether to publish `TargetSummary` events.

If true, expand Filesets in the BEP when presenting output files.

Specify how an output group's files will be represented in `TargetComplete`/`AspectComplete` BEP events. Values are an assignment of an output group name to one of `NAMED_SET_OF_FILES_ONLY`, `INLINE_ONLY`, or `BOTH`. The default value is `NAMED_SET_OF_FILES_ONLY`. If an output group is repeated, the final value to appear is used. The default value sets the mode for coverage artifacts to BOTH: `--experimental_build_event_output_group_mode=baseline.lcov=both`

Initial, minimum delay for exponential backoff retries when BEP upload fails. (exponent: 1.6)

Selects how to upload artifacts referenced in the build event protocol. In Bazel the valid options include `local` and `remote`. The default value is `local`.

If enabled, adds a `visibility()` function that .bzl files may call during top-level evaluation to set their visibility for the purpose of load() statements.

If `on_failed` or `on_passed`, then Blaze will cancel concurrently running tests on the first successful run with that result. This is only useful in combination with `--runs_per_test_detects_flakes`.

If set to true, rule attributes and Starlark API methods needed for the rule cc_shared_library will be available

Whether to double-check correct desugaring at Android binary level.

Specifies the strategy for the circuit breaker to use. Available strategies are "failure". On invalid value for the option the behavior same as the option is not set.

If specified, Bazel will also generate collect coverage information for generated files.

If enabled, the profiler collects the system's overall load average.

If enabled, the profiler collects the Linux PSI data.

If enabled, the profiler collects CPU and memory usage estimation for local actions.

If enabled, the profiler collects SkyFunction counts in the Skyframe graph over time for key function types, like configured targets and action executions. May have a performance hit as this visits the ENTIRE Skyframe graph at every profiling time unit. Do not use this flag with performance- critical measurements.

If enabled, the profiler collects the system's network usage.

If enabled, the profiler collects worker's aggregated resource data.

Records a Java Flight Recorder profile for the duration of the command. One of the supported profiling event types (cpu, wall, alloc or lock) must be given as an argument. The profile is written to a file named after the event type under the output base directory. The syntax and semantics of this flag might change in the future to support additional profile types or output formats; use at your own risk.

This flag controls how the convenience symlinks (the symlinks that appear in the workspace after the build) will be managed. Possible values: - `normal` (default): Each kind of convenience symlink will be created or deleted, as determined by the build. - `clean`: All symlinks will be unconditionally deleted. - `ignore`: Symlinks will not be created or cleaned up. - `log_only`: Generate log messages as if `normal` were passed, but don't actually perform any filesystem operations (useful for tools). Note that only symlinks whose names are generated by the current value of `--symlink_prefix` can be affected; if the prefix changes, any pre-existing symlinks will be left alone.

This flag controls whether or not we will post the build event `ConvenienceSymlinksIdentified` to the Build Event Protocol. If the value is true, the BEP will have an entry for `convenienceSymlinksIdentified`, listing all of the convenience symlinks created in your workspace. If false, then the `convenienceSymlinksIdentified` entry in the BEP will be empty.

Enables the experimental local execution scheduling based on CPU load, not estimation of actions one by one. Experimental scheduling have showed the large benefit on a large local builds on a powerful machines with the large number of cores. Reccommended to use with --local_resources=cpu=HOST_CPUS

If set to true, the auto-generated //external package will not be available anymore. Bazel will still be unable to parse the file 'external/BUILD', but globs reaching into external/ from the unnamed package will work.

How long the server must remain idle before a garbage collection of the disk cache occurs. To specify the garbage collection policy, set --experimental_disk_cache_gc_max_size and/or --experimental_disk_cache_gc_max_age.

If set to a positive value, the disk cache will be periodically garbage collected to remove entries older than this age. If set in conjunction with --experimental_disk_cache_gc_max_size, both criteria are applied. Garbage collection occurrs in the background once the server has become idle, as determined by the --experimental_disk_cache_gc_idle_delay flag.

If set to a positive value, the disk cache will be periodically garbage collected to stay under this size. If set in conjunction with --experimental_disk_cache_gc_max_age, both criteria are applied. Garbage collection occurrs in the background once the server has become idle, as determined by the --experimental_disk_cache_gc_idle_delay flag.

Specify a Docker image name (e.g. "ubuntu:latest") that should be used to execute a sandboxed action when using the docker strategy and the action itself doesn't already have a container-image attribute in its remote_execution_properties in the platform description. The value of this flag is passed verbatim to 'docker run', so it supports the same syntax and mechanisms as Docker itself.

If enabled, Bazel will pass the --privileged flag to 'docker run' when running actions. This might be required by your build, but it might also result in reduced hermeticity.

If enabled, injects the uid and gid of the current user into the Docker image before using it. This is required if your build / tests depend on the user having a name and home directory inside the container. This is on by default, but you can disable it in case the automatic image customization feature doesn't work in your case or you know that you don't need it.

If enabled, Bazel will print more verbose messages about the Docker sandbox strategy.

If set to true, attr.label(materializer=), attr (for_dependency_resolution=), attr.dormant_label(), attr. dormant_label_list() and rule(for_dependency_resolution=) are allowed.

When set, targets that are build "for tool" are not subject to dynamic execution. Such targets are extremely unlikely to be built incrementally and thus not worth spending local cycles on.

Takes a list of OS signal numbers. If a local branch of dynamic execution gets killed with any of these signals, the remote branch will be allowed to finish instead. For persistent workers, this only affects signals that kill the worker process.

Controls how much load from dynamic execution to put on the local machine. This flag adjusts how many actions in dynamic execution we will schedule concurrently. It is based on the number of CPUs Blaze thinks is available, which can be controlled with the --local_resources=cpu= flag. If this flag is 0, all actions are scheduled locally immediately. If > 0, the amount of actions scheduled locally is limited by the number of CPUs available. If < 1, the load factor is used to reduce the number of locally scheduled actions when the number of actions waiting to schedule is high. This lessens the load on the local machine in the clean build case, where the local machine does not contribute much.

If >0, the time a dynamically run action must run remote-only before we prioritize its local execution to avoid remote timeouts. This may hide some problems on the remote execution system. Do not turn this on without monitoring of remote execution issues.

If set to true, enables the APIs required to support the Android Starlark migration.

Enable Docker-based sandboxing. This option has no effect if Docker is not installed.

If set to true, enables the `macro()` construct for defining symbolic macros.

If set to true, .scl files may be used in load() statements.

If true, enable the use of --experimental_active_directories to reduce Bazel's memory footprint for incremental builds. This feature is known as Skyfocus.

If true, enable the set data type and set() constructor in Starlark.

Whether to enable thread dumps. If true, Bazel will dump the state of all threads (including virtual threads) to a file every -- experimental_thread_dump_interval, or after action execution being inactive for --experimental_thread_dump_action_execution_inactivity_duration. The dumps will be written to the <output_base>/server/thread_dumps/ directory.

If true, enable package()s to set the transitive_visibility attribute to restrict which packages may depend on them.

Deprecated in favor of aspects. Filters set of targets to schedule extra_actions for.

Deprecated in favor of aspects. Only schedules extra_actions for top level targets.

If true, then Bazel fetches the entire coverage data directory for each test during a coverage run.

Filter the ProGuard ProgramJar to remove any classes also present in the LibraryJar.

Strategies to handle potential incorrectness from changes beyond the frontier (i.e. outside the active directories)

If true, Bazel will print instructions for fixing Skycache violations

If true, coverage for clang will generate an LCOV report.

If set to true, exposes a number of experimental pieces of Starlark build API pertaining to Google legacy code.

Whether Starlark options set to their default values are included in the output.

If set, add a "requires-xcode:{version}" execution requirement to every Xcode action. If the Xcode version has a hyphenated label, also add a "requires-xcode-label:{version_label}" execution requirement.

If enabled, C++ .d files will be passed through in memory directly from the remote build nodes instead of being written to disk.

If enabled, the dependency (.jdeps) files generated from Java compilations will be passed through in memory directly from the remote build nodes instead of being written to disk.

If set to true, the contents of stashed sandboxes for reuse_sandbox_directories will be tracked in memory. This reduces the amount of I/O needed during reuse. Depending on the build this flag may improve wall time. Depending on the build as well this flag may use a significant amount of additional memory.

How long an install base must go unused before it's eligible for garbage collection. If nonzero, the server will attempt to garbage collect other install bases when idle.

If true, enables the <code>isolate</code> parameter in the <a href="https: //bazel.build/rules/lib/globals/module#use_extension" ><code>use_extension</code></a> function.

Enables reduced classpaths for Java compilations.

If materializing param files, do so with direct writes to disk.

Uses these strings as objc fastbuild compiler options.

If true, use libunwind for stack unwinding, and compile with -fomit-frame- pointer and -fasynchronous-unwind-tables.

When enabled, enforce that a java_binary rule can't contain more than one version of the same class file on the classpath. This enforcement can break the build, or can just result in warnings.

Which model to use for where in the output tree rules write their outputs, particularly for multi-platform / multi-configuration builds. This is highly experimental. See [GH-6526] for details. Starlark actions can opt into path mapping by adding the key `supports-path-mapping` to the `execution_requirements` dict. [GH-6526]: https://github.com/bazelbuild/bazel/issues/6526

Each entry should be of the form `label=value` where label refers to a platform and values is the desired shortname to override the platform's CPU name in `$(TARGET_CPU)` make variable and output path. Only used when `--experimental_platform_in_output_dir`, `-- incompatible_target_cpu_from_platform` or `--incompatible_bep_cpu_from_platform` is true. Has highest naming priority.

Enable persistent aar extractor by using workers.

If true, a shortname for the target platform is used in the output directory name instead of the CPU. The exact scheme is experimental and subject to change: 1. First, in the rare case the `--platforms` option does not have exactly one value, a hash of the platforms option is used. 2. Next, if any shortname for the current platform was registered by `--experimental_override_name_platform_in_output_dir`, then that shortname is used. 3. Then, if `--experimental_use_platforms_in_output_dir_legacy_heuristic` is set, use a shortname based off the current platform Label. 4. Finally, a hash of the platform option is used as a last resort.

If set to true, enables a number of platform-related Starlark APIs useful for debugging.

If true, use the most recent Xcode that is available both locally and remotely. If false, or if there are no mutual available versions, use the local Xcode version selected via xcode-select.

Specifies additional profile tasks to be included in the profile.

Includes the extra "out" attribute in action events that contains the exec path to the action's primary output.

Includes target configuration hash in action events' JSON profile data.

Includes target label in action events' JSON profile data.

Run extra actions for alternative Java api versions in a proto_library.

py_binary targets include their label even when stamping is disabled.

Controls the output of BEP ActionSummary and BuildGraphMetrics, limiting the number of mnemonics in ActionData and number of entries reported in BuildGraphMetrics.AspectCount/RuleClassCount. By default the number of types is limited to the top 20, by number of executed actions for ActionData, and instances for RuleClass and Asepcts. Setting this option will write statistics for all mnemonics, rule classes and aspects.

Controls the output of BEP BuildGraphMetrics, including expensiveto compute skyframe metrics about Skykeys, RuleClasses and Aspects.With this flag set to false BuildGraphMetrics.rule_count and aspectfields will not be populated in the BEP.

Whether to make source manifest actions remotable

The minimum blob size required to compress/decompress with zstd. Ineffectual unless --remote_cache_compression is set.

The maximum number of attempts to retry if the build encountered a transient remote cache error that would otherwise fail the build. Applies for example when artifacts are evicted from the remote cache, or in certain cache failure conditions. A new invocation id will be generated for each attempt.

If set to true, Bazel will extend the lease for outputs of remote actions during the build by sending `FindMissingBlobs` calls periodically to remote cache. The frequency is based on the value of `--experimental_remote_cache_ttl`.

The guaranteed minimal TTL of blobs in the remote cache after their digests are recently referenced e.g. by an ActionResult or FindMissingBlobs. Bazel does several optimizations based on the blobs' TTL e.g. doesn't repeatedly call GetActionResult in an incremental build. The value should be set slightly less than the real TTL since there is a gap between when the server returns the digests and when Bazel receives them.

A path to a directory where the corrupted outputs will be captured to.

If set to true, discard in-memory copies of the input root's Merkle tree and associated input mappings during calls to GetActionResult() and Execute(). This reduces memory usage significantly, but does require Bazel to recompute them upon remote cache misses and retries.

A Remote Asset API endpoint URI, to be used as a remote download proxy. The supported schemas are grpc, grpcs (grpc with TLS enabled) and unix (local UNIX sockets). If no schema is provided Bazel will default to grpcs. See: https://github.com/bazelbuild/remote-apis/blob/master/build/bazel/remote/asset/v1/remote_asset.proto

Whether to fall back to the local downloader if remote downloader fails.

Whether to propagate credentials from netrc and credential helper to the remote downloader server. The server implementation needs to support the new `http_header_url:<url-index>:<header-key>` qualifier where the `<url-index>` is a 0-based position of the URL inside the FetchBlobRequest's `uris` field. The URL-specific headers should take precedence over the global headers.

Sets the allowed number of failure rate in percentage for a specific time window after which it stops calling to the remote cache/executor. By default the value is 10. Setting this to 0 means no limitation.

The interval in which the failure rate of the remote requests are computed. On zero or negative value the failure duration is computed the whole duration of the execution.Following units can be used: Days (d), hours (h), minutes (m), seconds (s), and milliseconds (ms). If the unit is omitted, the value is interpreted as seconds.

If set to true, Bazel will mark inputs as tool inputs for the remote executor. This can be used to implement remote persistent workers.

HOST or HOST:PORT of a remote output service endpoint. The supported schemas are grpc, grpcs (grpc with TLS enabled) and unix (local UNIX sockets). If no schema is provided Bazel will default to grpcs. Specify grpc:// or unix: schema to disable TLS.

The path under which the contents of output directories managed by the --experimental_remote_output_service are placed. The actual output directory used by a build will be a descendant of this path and determined by the output service.

If set to true, enforce that all actions that can run remotely are cached, or else fail the build. This is useful to troubleshoot non-determinism issues as it allows checking whether actions that should be cached are actually cached without spuriously injecting new results into the cache.

Enables remote cache key scrubbing with the supplied configuration file, which must be a protocol buffer in text format (see src/main/protobuf/remote_scrubbing.proto).This feature is intended to facilitate sharing a remote/disk cache between actions executing on different platforms but targeting the same platform. It should be used with extreme care, as improper settings may cause accidental sharing of cache entries and result in incorrect builds.Scrubbing does not affect how an action is executed, only how its remote/disk cache key is computed for the purpose of retrieving or storing an action result. Scrubbed actions are incompatible with remote execution, and will always be executed locally instead.Modifying the scrubbing configuration does not invalidate outputs present in the local filesystem or internal caches; a clean build is required to reexecute affected actions.In order to successfully use this feature, you likely want to set a custom --host_platform together with --experimental_platform_in_output_dir (to normalize output prefixes).

If set to true, repository_rule gains some remote execution capabilities.

If set, the repository cache will hardlink the file in case of a cache hit, rather than copying. This is intended to save disk space.

If true enables the repository_ctx `load_wasm` and `execute_wasm` methods.

The maximum number of attempts to retry a download error. If set to 0, retries are disabled.

When enabled, `--trim_test_configuration` will not trim the test configuration for rules marked testonly=1. This is meant to reduce action conflict issues when non- test rules depend on `cc_test` rules. No effect if `--trim_test_configuration` is false.

Enable experimental rule extension API and subrule APIs

Whether to validate java_* sources.

Whether to include the command-line residue in run build events which could contain the residue. By default, the residue is not included in run command build events that could contain the residue.

If 0, sandboxes are deleted as soon as actions finish, blocking action completion. If greater than 0, sandboxes are deleted asynchronously in the background without blocking action completion. Asynchronous deletion uses a single thread while a command is running, but ramps up to as many threads as the value of this flag once the server becomes idle. Set to `auto` to use as many threads as the number of CPUs. A server shutdown blocks on any pending asynchronous deletions.

If true, actions whose mnemonic matches the input regex will have their resources request enforced as limits, overriding the value of -- experimental_sandbox_limits, if the resource type supports it. For example a test that declares cpu:3 and resources:memory:10, will run with at most 3 cpus and 10 megabytes of memory.

If > 0, each Linux sandbox will be limited to the given amount for the specified resource. Requires --incompatible_use_new_cgroup_implementation and overrides --experimental_sandbox_memory_limit_mb. Requires cgroups v1 or v2 and permissions for the users to the cgroups dir.

If > 0, each Linux sandbox will be limited to the given amount of memory (in MB). Requires cgroups v1 or v2 and permissions for the users to the cgroups dir.

No-op

Save the state of enabled and requested feautres as an output of compilation.

Scale all timeouts in Starlark repository rules by this factor. In this way, external repositories can be made working on machines that are slower than the rule author expected, without changing the source code

If enabled, could shrink worker pool if worker memory pressure is high. This flag works only when flag experimental_total_worker_memory_limit_mb is enabled.

If set to true, non-main repositories are planted as symlinks to the main repository in the execution root. That is, all repositories are direct children of the $output_base/execution_root directory. This has the side effect of freeing up $output_base/execution_root/__main__/external for the real top-level 'external' directory.

If enabled, the register_toolchain function may not include target patterns which may refer to more than one package.

For debugging Skyfocus. Dump the focused SkyKeys (roots, leafs, focused deps, focused rdeps).

For debugging Skyfocus. If enabled, trigger manual GC before/after focusing to report heap sizes reductions. This will increase the Skyfocus latency.

Enable dynamic execution by running actions locally and remotely in parallel. Bazel spawns each action locally and remotely and picks the one that completes first. If an action supports workers, the local action will be run in the persistent worker mode. To enable dynamic execution for an individual action mnemonic, use the `--internal_spawn_scheduler` and `--strategy=<mnemonic>=dynamic` flags instead. Expands to: --internal_spawn_scheduler --spawn_strategy=dynamic

If true, then Bazel will run coverage postprocessing for test in a new spawn.

Enables type checking in files and functions that contain type annotations or related syntax. (When this flag is disabled, Bazel is more forgiving of invalid types in type annotations.)

Enables type annotations and related syntax in .bzl files. Locations of files where these are allowed is further restricted by `-- experimental_starlark_types_allowed_paths`. Type syntax is never permitted in .scl files regardless of this flag.

List of canonical Label prefixes under which Starlark type annotations are allowed.

Stream log file uploads directly to the remote storage rather than writing them to disk.

If this option is enabled, filesets will treat all output artifacts as regular files. They will not traverse directories or be sensitive to symlinks.

If true, checks that a Java target explicitly declares all directly used targets as dependencies.

If true, repository rules and module extensions will only inherit `PATH`, `PATHEXT` (on Windows), and environment variables explicitly specified by `-- repo_env`. Note that unless `--incompatible_repo_env_ignores_action_env` is true, `--action_env=NAME=VALUE` will also be included.

Dump the threads when action execution being inactive for this duration. If zero, no thread dumps are written for action execution being inactive.

How often to dump the threads periodically. If zero, no thread dumps are written periodically.

If this limit is greater than zero idle workers might be killed if the total memory usage of all workers exceed the limit.

The maximum size of the stdout / stderr files that will be printed to the console. -1 implies no limit.

Whether to narrow inputs to C/C++ compilation by parsing #include lines from input files. This can improve performance and incrementality by decreasing the size of compilation input trees. However, it can also break builds because the include scanner does not fully implement C preprocessor semantics. In particular, it does not understand dynamic #include directives and ignores preprocessor conditional logic. Use at your own risk. Any issues relating to this flag that are filed will be closed. At Google without this flag your build will most likely fail.

If set to true, do not mount root, only mount whats provided with sandbox_add_mount_pair. Input files will be hardlinked to the sandbox instead of symlinked to from the sandbox. If action input files are located on a filesystem different from the sandbox, then the input files will be copied instead.

If specified, Bazel will generate llvm-cov coverage map information rather than gcov when collect_code_coverage is enabled.

Please only use this flag as part of a suggested migration or testing strategy. Note that the heuristic has known deficiencies and it is suggested to migrate to relying on just `--experimental_override_name_platform_in_output_dir`.

Whether to run validation actions using aspect (for parallelism with tests).

Use Windows sandbox to run actions. If "yes", the binary provided by -- experimental_windows_sandbox_path must be valid and correspond to a supported version of sandboxfs. If "auto", the binary may be missing or not compatible.

Path to the Windows sandbox binary to use when -- experimental_use_windows_sandbox is true. If a bare name, use the first binary of that name found in the PATH.

If true, experimental Windows support for --watchfs is enabled. Otherwise --watchfsis a non-op on Windows. Make sure to also enable --watchfs.

If non-empty, only allow using persistent workers with the given worker key mnemonic.

If enabled, Bazel may send cancellation requests to workers that support them.

If this limit is greater than zero, workers might be killed if the memory usage of the worker exceeds the limit. If not used together with dynamic execution and `--experimental_dynamic_ignore_local_signals=9`, this may crash your build.

The interval between collecting worker metrics and possibly attempting evictions. Cannot effectively be less than 1s for performance reasons.

If enabled, multiplex workers with a 'supports-multiplex-sandboxing' execution requirement will run in a sandboxed environment, using a separate sandbox directory per work request. Multiplex workers with the execution requirement are always sandboxed when running under the dynamic execution strategy, irrespective of this flag.

If enabled, workers are run in a hardened sandbox, if the implementation allows it. If hardening is enabled then tmp directories are distinct for different workers.

A worker key mnemonic for which the contents of the sandbox directory are tracked in memory. This may improve build performance at the cost of additional memory usage. Only affects sandboxed workers. May be specified multiple times for different mnemonics.

If enabled, actions arguments for workers that do not follow the worker specification will cause an error. Worker arguments must have exactly one @flagfile argument as the last of its list of arguments.

Log certain Workspace Rules events into this file as delimited WorkspaceEvent protos.

Causes the build system to explain each executed step of the build. The explanation is written to the specified log file.

Explicitly specify a dependency to JUnit or Hamcrest in a java_test instead of accidentally obtaining from the TestRunner's deps. Only works for bazel right now.

The platforms that are available as execution platforms to run actions. Platforms can be specified by exact target, or as a target pattern. These platforms will be considered before those declared in the `WORKSPACE` file by `register_execution_platforms()`. This option may only be set once; later instances will override earlier flag settings.

The toolchain rules to be considered during toolchain resolution. Toolchains can be specified by exact target, or as a target pattern. These toolchains will be considered before those declared in the `WORKSPACE` file by `register_toolchains()`.

No-op flag. Will be removed in a future release.

Generate binaries with FDO instrumentation. With Clang/LLVM compiler, it also accepts the directory name under which the raw profile file(s) will be dumped at runtime. Using this option will also add: --copt=-Wno-error

Use FDO profile information to optimize compilation. Specify the name of a zip file containing a .gcda file tree, an afdo file containing an auto profile, or an LLVM profile file. This flag also accepts files specified as labels (e.g. `//foo/bar:file.afdo` - you may need to add an `exports_files` directive to the corresponding package) and labels pointing to `fdo_profile` targets. This flag will be superseded by the `fdo_profile` rule.

Use cache prefetch hints.

The fdo_profile representing the profile to be used for optimization.

The given features will be enabled or disabled by default for targets built in the target configuration. Specifying `-{feature}` will disable the feature. Negative features always override positive ones. See also `--host_features`.

Allows the command to fetch external dependencies. If set to false, the command will utilize any cached version of the dependency, and if none exists, the command will result in failure.

Specifies which compilation modes use fission for C++ compilations and links. May be any combination of {'fastbuild', 'dbg', 'opt'} or the special values 'yes' to enable all modes and 'no' to disable all modes.

Sets a shorthand name for a Starlark flag. It takes a single key-value pair in the form `{key}={value}` as an argument.

Each test will be retried up to the specified number of times in case of any test failure. Tests that required more than one attempt to pass are marked as 'FLAKY' in the test summary. Normally the value specified is just an integer or the string 'default'. If an integer, then all tests will be run up to N times. If 'default', then only a single test attempt will be made for regular tests and three for tests marked explicitly as flaky by their rule (flaky=1 attribute). Alternate syntax: regex_filter@flaky_test_attempts. Where flaky_test_attempts is as above and regex_filter stands for a list of include and exclude regular expression patterns (Also see --runs_per_test). Example: --flaky_test_attempts=//foo/. *,-//foo/bar/.*@3 deflakes all tests in //foo/ except those under foo/bar three times. This option can be passed multiple times. The most recently passed argument that matches takes precedence. If nothing matches, behavior is as if 'default' above.

The command for which the options should be canonicalized.

If enabled, all C++ compilations produce position-independent code ("- fPIC"), links prefer PIC pre-built libraries over non-PIC libraries, and links produce position-independent executables ("-pie").

At any point after an invocation has been running for at least one minute, if Blaze has spent at least this percentage of the invocation's wall time doing full GCs, Blaze will give up and fail with an OOM. A value of 100 effectively means to never give up for this reason.

If set to a value in [0, 100] and this is a command that takes top-level targets (e.g. build but not query) and there are multiple such top-level targets, overrides --gc_churning_threshold. Useful to configure more aggressive OOMing behavior (i.e. a lower value than -- gc_churning_threshold) when they are multiple top-level targets so that the invoker of Bazel can split and retry while still having less aggressive behavior when there is a single top-level target.

Limits which, if reached, cause GcThrashingDetector to crash Bazel with an OOM. Each limit is specified as <period>:<count> where period is a duration and count is a positive integer. If more than --gc_thrashing_threshold percent of tenured space (old gen heap) remains occupied after <count> consecutive full GCs within <period>, an OOM is triggered. Multiple limits can be specified separated by commas.

The percent of tenured space occupied (0-100) above which GcThrashingDetector considers memory pressure events against its limits (-- gc_thrashing_limits). If set to 100, GcThrashingDetector is disabled.

If enabled, Bazel profiles the build and writes a JSON-format profile into a file in the output base. View profile by loading into chrome://tracing. By default Bazel writes the profile for all build-like commands and query.

Specify how to execute genrules. This flag will be phased out. Instead, use --spawn_strategy=<value> to control all actions or -- strategy=Genrule=<value> to control genrules only.

A comma-separated list of Google Cloud authentication scopes.

Specifies the file to get authentication credentials from. See https://cloud.google.com/docs/authentication for details.

Whether to use 'Google Application Default Credentials' for authentication.See [Authentication methods at Google - Google Cloud][gc-auth-methods] for details.Disabled by default.[gc-auth-methods]: https://cloud.google.com/docs/authentication

Configures keep-alive pings for outgoing gRPC connections. If this is set, then Bazelsends pings after this much time of no read operations on the connection, butonly if there is at least one pending gRPC call. The value 0 disables the keep-alives.

Configures a keep-alive timeout for outgoing gRPC connections. If keep-alive pings areenabled with `--grpc_keepalive_time`, then Bazel times out a connection if it doesnot receive a ping reply after this much time. Times are treated as secondgranularity; it is an error to set a value less than one second. If keep-alivepings are disabled, then this setting is ignored.

A label to a checked-in libc library. The default value is selected by the crosstool toolchain, and you almost never need to override it.

Set this to 'full' to enable checking the ctime of all input files of an action before uploading it to a remote cache. There may be cases where the Linux kernel delays writing of files, which could cause false positives. The default is 'lite', which only checks source files in the main repository. Setting this to 'off' disables all checks. This is not recommended, as the cache may be polluted when a source file is changed while an action that takes it as an input is executing.

Whether to manually output a heap dump if an OOM is thrown (including manual OOMs due to reaching --gc_thrashing_limits). The dump will be written to <output_base>/<invocation_id>.heapdump.hprof. This option effectively replaces -XX:+HeapDumpOnOutOfMemoryError, which has no effect for manual OOMs.

If true, Blaze will remove FileState and DirectoryListingState nodes after related File and DirectoryListing node is done to save memory. We expect that it is less likely that these nodes will be needed again. If so, the program will re-evaluate them.

Specifies the set of environment variables available to actions with execution configurations. Variables can be either specified by `name`, in which case the value will be taken from the invocation environment, by the `name=value` pair which sets the value independent of the invocation environment, or by `=name`, which unsets the variable of that name. This option can be used multiple times; for options given for the same variable, the latest wins, options for different variables accumulate.

Specify the mode the tools used during the build will be built in. Values: `fastbuild`, `dbg`, `opt`.

No-op flag. Will be removed in a future release.

Additional option to pass to the C compiler when compiling C (but not C++) source files in the exec configurations.

Additional options to pass to the C compiler for tools built in the exec configurations.

The host CPU.

Additional options to pass to C++ compiler for tools built in the exec configurations.

The given features will be enabled or disabled by default for targets built in the exec configuration. Specifying `-{feature}` will disable the feature. Negative features always override positive ones.

If specified, this setting overrides the libc top-level directory (-- grte_top) for the exec configuration.

The Java launcher used by tools that are executed during a build.

Additional options to pass to javac when building tools that are executed during a build.

Additional options to pass to the Java VM when building tools that are executed during the build. These options will get added to the VM startup options of each java_binary target.

Additional option to pass to linker when linking tools in the exec configurations.

Minimum compatible macOS version for host targets. If unspecified, uses 'macos_sdk_version'.

Additional options to selectively pass to the C/C++ compiler when compiling certain files in the exec configurations. This option can be passed multiple times. Syntax: regex_filter@option_1,option_2,...,option_n. Where regex_filter stands for a list of include and exclude regular expression patterns (Also see --instrumentation_filter). option_1 to option_n stand for arbitrary command line options. If an option contains a comma it has to be quoted with a backslash. Options can contain @. Only the first @ is used to split the string. Example: --host_per_file_copt=//foo/.*\.cc,-//foo/bar\. cc@-O0 adds the -O0 command line option to the gcc command line of all cc files in //foo/ except bar.cc.

The label of a platform rule that describes the host system.

The maximum number of attempts for http downloads.

The maximum timeout for http download retries. With a value of 0, no timeout maximum is defined.

The maximum number parallel http downloads.

Scale all timeouts related to http downloads by the given factor

If true, Bazel ignores `bazel_dep` and `use_extension` declared as `dev_dependency` in the `MODULE.bazel` of the root module. Note that, those dev dependencies are always ignored in the `MODULE.bazel` if it's not the root module regardless of the value of this flag.

If set to true, tags will be propagated from a target to the actions' execution requirements; otherwise tags are not propagated. See https: //github.com/bazelbuild/bazel/issues/8830 for details.

Check the validity of elements added to depsets, in all constructors. Elements must be immutable, but historically the depset(direct=...) constructor forgot to check. Use tuples instead of lists in depset elements. See https://github.com/bazelbuild/bazel/issues/10313 for details.

If true, native rules add `DefaultInfo.files` of data dependencies to their runfiles, which matches the recommended behavior for Starlark rules ([runfiles features to avoid]). [runfiles features to avoid]: https://bazel. build/extending/rules#runfiles_features_to_avoid

When enabled, an exec groups is automatically created for each toolchain used by a rule. For this to work rule needs to specify `toolchain` parameter on its actions. For more information, see [GH-17134]. [GH-17134]: https://github.com/bazelbuild/bazel/issues/17134

A comma-separated list of rules (or other symbols) that were previously part of Bazel and which are now to be retrieved from their respective external repositories. This flag is intended to be used to facilitate migration of rules out of Bazel. See also https://github. com/bazelbuild/bazel/issues/23043. A symbol that is autoloaded within a file behaves as if its built-into- Bazel definition were replaced by its canonical new definition in an external repository. For a BUILD file, this essentially means implicitly adding a load() statement. For a .bzl file, it's either a load() statement or a change to a field of the `native` object, depending on whether the autoloaded symbol is a rule. Bazel maintains a hardcoded list of all symbols that may be autoloaded; only those symbols may appear in this flag. For each symbol, Bazel knows the new definition location in an external repository, as well as a set of special-cased repositories that must not autoload it to avoid creating cycles. A list item of "+foo" in this flag causes symbol foo to be autoloaded, except in foo's exempt repositories, within which the Bazel-defined version of foo is still available. A list item of "foo" triggers autoloading as above, but the Bazel-defined version of foo is not made available to the excluded repositories. This ensures that foo's external repository does not depend on the old Bazel implementation of foo A list item of "-foo" does not trigger any autoloading, but makes the Bazel- defined version of foo inaccessible throughout the workspace. This is used to validate that the workspace is ready for foo's definition to be deleted from Bazel. If a symbol is not named in this flag then it continues to work as normal -- no autoloading is done, nor is the Bazel-defined version suppressed. For configuration see https://github. com/bazelbuild/bazel/blob/master/src/main/java/com/google/devtools/build/lib/packages/AutoloadSymbols. java As a shortcut also whole repository may be used, for example +@rules_python will autoload all Python rules.

If enabled, `bazel test --run_under=//:runner` builds `//:runner` in the exec configuration. If disabled, it builds `//:runner` in the target configuration. Bazel executes tests on exec machines, so the former is more correct. This doesn't affect `bazel run`, which always builds `--run_under=//foo` in the target configuration.

Whether to emit a strip action as part of objc linking.

If true, Bazel will fail a sharded test if the test runner does not indicate that it supports sharding by touching the file at the path in `TEST_SHARD_STATUS_FILE`. If false, a test runner that does not support sharding will lead to all tests running in each shard.

If enabled, check testonly for prerequisite targets that are output files by looking up the testonly of the generating rule. This matches visibility checking.

If enabled, the `{binary}.repo_mapping` file emits a module extension's repo mapping only once instead of once for each repo generated by the extension that contributes runfiles.

If incompatible_enforce_config_setting_visibility=false, this is a noop. Else, if this flag is false, any config_setting without an explicit visibility attribute is //visibility:public. If this flag is true, config_setting follows the same visibility logic as all other rules. See https://github.com/bazelbuild/bazel/issues/12933.

This flag changes the default behavior so that __init__.py files are no longer automatically created in the runfiles of Python targets. Precisely, when a py_binary or py_test target has legacy_create_init set to "auto" (the default), it is treated as false if and only if this flag is set. See https://github.com/bazelbuild/bazel/issues/10076.

Controls if the autoloads (set by --incompatible_autoload_externally) are enabled in themain repository. When enabled the rules (or other symbols) that were previously part of Bazel need to have load statements. Use buildifier to add them.

If enabled, direct usage of the native Android rules is disabled. Please use the Starlark Android rules from https://github. com/bazelbuild/rules_android

No-op. Kept here for backwards compatibility.

If true, java_binary is always executable. create_executable attribute is removed.

Disable objc_library's custom transition and inherit from the top level target instead (No-op in Bazel)

List of flags for which the use in `select()` is disabled.

If set to true, rule attributes cannot set 'cfg = "host"'. Rules should set 'cfg = "exec"' instead.

If set to true, disable the ability to utilize the default provider via field syntax. Use provider-key syntax instead. For example, instead of using `ctx.attr.dep.files` to access `files`, utilize `ctx.attr.dep [DefaultInfo].files See https://github.com/bazelbuild/bazel/issues/9014 for details.

A comma-separated list of flags that cannot be used in transitions inputs or outputs.

If set to true, calling the deprecated ctx.resolve_tools API always fails. Uses of this API should be replaced by an executable or tools argument to ctx.actions.run or ctx.actions.run_shell.

If set to true, the default value of the `allow_empty` argument of glob() is False.

If true, disallow sdk_frameworks and weak_sdk_frameworks attributes in objc_library andobjc_import.

When true, Bazel no longer modifies command line flags used for linking, and also doesn't selectively decide which flags go to the param file and which don't. See https://github.com/bazelbuild/bazel/issues/7670 for details.

If true, Bazel will not enable 'host' and 'nonhost' features in the c++ toolchain (see https://github.com/bazelbuild/bazel/issues/7407 for more information).

Use toolchain resolution to select the Apple SDK for apple rules (Starlark and native)

If enabled, certain deprecated APIs (native.repository_name, Label. workspace_name, Label.relative) can be used.

If true, proto lang rules define toolchains from protobuf repository.

If true, enforce config_setting visibility restrictions. If false, every config_setting is visible to every target. See https://github. com/bazelbuild/bazel/issues/12932.

If enabled (or set to 'error'), fail if Starlark files are not UTF-8 encoded. If set to 'warning', emit a warning instead. If set to 'off', Bazel assumes that Starlark files are UTF-8 encoded but does not verify this assumption. Note that Starlark files which are not UTF-8 encoded can cause Bazel to behave inconsistently.

If true, exclusive tests will run with sandboxed strategy. Add `local` tag to force an exclusive test run locally

If enabled, targets that have unknown attributes set to None fail.

If true, runfiles of targets listed in the srcs attribute are available to targets that consume the filegroup as a data dependency.

In package_group's `packages` attribute, changes the meaning of the value "//..." to refer to all packages in the current repository instead of all packages in any repository. You can use the special value "public" in place of "//..." to obtain the old behavior. This flag requires that -- incompatible_package_group_has_public_syntax also be enabled.

Whether a target that provides an executable expands to the executable rather than the files in <code>DefaultInfo.files</code> under $(locations ...) expansion if the number of files is not 1.

This flag is a noop and scheduled for removal.

If true, the genfiles directory is folded into the bin directory.

When enabled, passing multiple `--modify_execution_info` flags is additive. When disabled, only the last flag is taken into account.

If set to true, disables the function `attr.license`.

If set, (used) source files are package private unless exported explicitly. See https://github.com/bazelbuild/proposals/blob/master/designs/2019-10-24- file-visibility.md

If true, then methods on <code>repository_ctx</code> that are passed a Label will no longer automatically watch the file under that label for changes even if <code>watch = "no"</code>, and <code>repository_ctx. path</code> no longer causes the returned path to be watched. Use <code>repository_ctx.watch</code> instead.

If set to true, disables the `outputs` parameter of the `rule()` Starlark function.

If true, make the default value true for alwayslink attributes in objc_library and objc_import.

In package_group's `packages` attribute, allows writing "public" or "private" to refer to all packages or no packages respectively.

When true, an error occurs when using the builtin py_* rules; instead the rule_python rules should be used. See https://github. com/bazelbuild/bazel/issues/17773 for more information and migration instructions.

Whether --remote_local_fallback applies to --remote_cache.

If true, Bazel will not link library dependencies as whole archive by default (see https://github.com/bazelbuild/bazel/issues/7362 for migration instructions).

If true, <code>--action_env=NAME=VALUE</code> will no longer affect repository rule and module extension environments.

This flag is a noop and scheduled for removal.

If enabled, string keys in dicts passed to select() in .bzl files are immediately resolved to Labels relative to the file instead of being interpreted relative to the BUILD file they are ultimately loaded from.

If set to true, the command parameter of actions.run_shell will only accept string

If true, simplify configurable rule attributes which contain only unconditional selects; for example, if ["a"] + select("//conditions: default", ["b"]) is assigned to a rule attribute, it is stored as ["a", "b"]. This option does not affect attributes of symbolic macros or attribute default values.

If set to true, deprecated ctx.build_file_path will not be available. ctx. label.package + '/BUILD' can be used instead.

If enabled, certain language-specific modules (such as `cc_common`) are unavailable in user .bzl files and may only be called from their respective rules repositories.

If true, Bazel uses an environment with a static value for PATH and does not inherit `LD_LIBRARY_PATH`. Use `--action_env=ENV_VARIABLE` if you want to inherit specific environment variables from the client, but note that doing so can prevent cross-user caching if a shared cache is used.

If true, strip action for executables will use flag -x, which does not break dynamic symbol resolution.

If specified, the value of the cpu constraint (`@platforms//cpu:cpu`) of the target platform is used to set the `$(TARGET_CPU)` make variable.

When true, Bazel will stringify the label @//foo:bar to @//foo:bar, instead of //foo:bar. This only affects the behavior of str(), the % operator, and so on; the behavior of repr() is unchanged. See https://github. com/bazelbuild/bazel/issues/15916 for more information.

When true, Bazel will no longer allow using cc_configure from @bazel_tools. Please see https://github.com/bazelbuild/bazel/issues/10134 for details and migration instructions.

If true, use the new implementation for cgroups. The old implementation only supports the memory controller and ignores the value of -- experimental_sandbox_limits.

This flag is a noop and scheduled for removal.

Does most of the work for dexing separately for each Jar file.

Adds a new repository with a local path in the form of `{repository name}={path}`. Thisonly takes effect with `--enable_bzlmod` and is equivalent to adding acorresponding `local_repository` to the root module's `MODULE.bazel` file via`use_repo_rule`. If the given path is an absolute path, it will be used as it is.If the given path is a relative path, it is relative to the current workingdirectory. If the given path starts with `%workspace%`, it is relative to theworkspace root, which is the output of `bazel info workspace`. If the given pathis empty, then remove any previous injections.

When coverage is enabled, specifies whether to consider instrumenting test rules. When set, test rules included by `--instrumentation_filter` are instrumented. Otherwise, test rules are always excluded from coverage instrumentation.

When coverage is enabled, only rules with names included by the specified regex-based filter will be instrumented. Rules prefixed with '-' are excluded instead. Note that only non-test rules are instrumented unless `--instrument_test_targets` is enabled.

Use interface shared objects if supported by the toolchain. All ELF toolchains currently support this setting.

Placeholder option so that we can tell in Blaze whether the spawn scheduler was enabled.

Unique identifier, in UUID format, for the command being run. If explicitly specified uniqueness must be ensured by the caller. The UUID is printed to stderr, the BEP and remote execution protocol.

Applies an invocation policy to the options to be canonicalized.

Enable checking for memory leaks in ios_test targets.

Minimum compatible iOS version for target simulators and devices. If unspecified, uses 'ios_sdk_version'.

Comma-separated list of architectures to build an ios_application with. The result is a universal binary containing all specified architectures.

Specifies the version of the iOS SDK to use to build iOS applications. If unspecified, uses the default iOS SDK version from 'xcode_version'.

Certificate name to use for iOS signing. If not set will fall back to provisioning profile. May be the certificate's keychain identity preference or (substring) of the certificate's common name, as per codesign's man page (SIGNING IDENTITIES).

The device to simulate when running an iOS application in the simulator, e. g. 'iPhone 6'. You can get a list of devices by running 'xcrun simctl list devicetypes' on the machine the simulator will be run on.

The version of iOS to run on the simulator when running or testing. This is ignored for ios_test rules if a target device is specified in the rule.

Additional options to pass to the J2ObjC tool.

Causes the Java virtual machine of a java test to wait for a connection from a JDWP-compliant debugger (such as jdb) before starting the test. Implies -test_output=streamed. Expands to: --test_arg=--wrapper_script_flag=--debug -- test_output=streamed --test_strategy=exclusive --test_timeout=9999 -- nocache_test_results

Generate dependency information (for now, compile-time classpath) per Java target.

Compile ijars directly from source.

The Java language version

The Java launcher to use when building Java binaries. If this flag is set to the empty string, the JDK launcher is used. The "launcher" attribute overrides this flag.

The Java runtime version

Additional options to pass to javac.

The number of concurrent jobs to run. Takes an integer, or a keyword ("auto", "HOST_CPUS", "HOST_RAM"), optionally followed by an operation ([- |*]<float>) eg. "auto", "HOST_CPUS*.5". Values must be between 1 and 5000. Values above 2500 may cause memory issues. "auto" calculates a reasonable default based on host resources.

Regex for overriding the matching logic for JDK21+ JVM heap memory collection. We are relying on volatile internal G1 GC implemenation details to get a clean memory metric, this option allows us to adapt to changes in that internal implementation without having to wait for a binary release. Passed to JDK Matcher.find()

Additional options to pass to the Java VM. These options will get added to the VM startup options of each java_binary target.

Continue as much as possible after an error. While the target that failed and those that depend on it cannot be analyzed, other prerequisites of these targets can be.

If false, Blaze will discard the inmemory state from this build when the build finishes. Subsequent builds will not have any incrementality with respect to this one.

Use this to suppress generation of the legacy `important_outputs` field in the `TargetComplete` event. `important_outputs` are required for Bazel to ResultStore/BTX integration.

Specifies a binary to use to generate the list of classes that must be in the main dex when compiling legacy multidex.

Deprecated, superseded by --incompatible_remove_legacy_whole_archive (see https://github.com/bazelbuild/bazel/issues/7362 for details). When on, use --whole-archive for cc_binary rules that have linkshared=True and either linkstatic=True or '-static' in linkopts. This is for backwards compatibility only. A better alternative is to use alwayslink=1 where required.

Additional option to pass to gcc when linking.

Number of parallel threads to use for the loading/analysis phase.Takes an integer, or a keyword ("auto", "HOST_CPUS", "HOST_RAM"), optionally followed by an operation ([-|*]<float>) eg. "auto", "HOST_CPUS*.5". "auto" sets a reasonable default based on host resources. Must be at least 1.

Set the number of resources available to Bazel. Takes in an assignment to a float or HOST_RAM/HOST_CPUS, optionally followed by [-|*]<float> (eg. memory=HOST_RAM*.5 to use half the available RAM). Can be used multiple times to specify multiple types of resources. Bazel will limit concurrently running actions based on the available resources and the resources required. Tests can declare the amount of resources they need by using a tag of the "resources:<resource name>:<amount>" format.

Time to wait between terminating a local process due to timeout and forcefully shutting it down.

The max number of local test jobs to run concurrently. Takes an integer, or a keyword ("auto", "HOST_CPUS", "HOST_RAM"), optionally followed by an operation ([-|*]<float>) eg. "auto", "HOST_CPUS*.5". 0 means local resources will limit the number of local test jobs to run concurrently instead. Setting this greater than the value for --jobs is ineffectual.

Specifies how and whether or not to use the lockfile. Valid values are `update` to use the lockfile and update it if there are changes, `refresh` to additionally refresh mutable information (yanked versions and previously missing modules) from remote registries from time to time, `error` to use the lockfile but throw an error if it's not up-to-date, or `off` to neither read from or write to the lockfile.

The logging level.

Additional option to pass to the LTO backend step (under -- features=thin_lto).

Additional option to pass to the LTO indexing step (under -- features=thin_lto).

Comma-separated list of architectures for which to build Apple macOS binaries.

Minimum compatible macOS version for targets. If unspecified, uses 'macos_sdk_version'.

Specifies the version of the macOS SDK to use to build macOS applications. If unspecified, uses the default macOS SDK version from 'xcode_version'.

Writes intermediate parameter files to output tree even when using remote action execution or caching. Useful when debugging actions. This is implied by --subcommands and --verbose_failures.

The maximum number of Starlark computation steps that may be executed by a BUILD file (zero means no limit).

When discarding the analysis cache due to a change in the build options, displays up to the given number of changed option names. If the number given is -1, all changed options will be displayed.

Specifies maximum per-test-log size that can be emitted when --test_output is 'errors' or 'all'. Useful for avoiding overwhelming the output with excessively noisy test output. The test header is included in the log size. Negative values imply no limit. Output is all or nothing.

If set, write memory usage data to the specified file at phase ends and stable heap to master log at end of build.

Tune memory profile's computation of stable heap at end of build. Should be and even number of integers separated by commas. In each pair the first integer is the number of GCs to perform. The second integer in each pair is the number of seconds to wait between GCs. Ex: 2,4,4,0 would 2 GCs with a 4sec pause, followed by 4 GCs with zero second pause

Use memprof profile.

The minimum OS version which your compilation targets.

Add or remove keys from an action's execution info based on action mnemonic. Applies only to actions which support execution info. Many common actions support execution info, e.g. Genrule, CppCompile, Javac, StarlarkAction, TestRunner. When specifying multiple values, order matters because many regexes may apply to the same mnemonic. Syntax: `regex=[+-]key,regex=[+-]key,...`. Examples: - `.*=+x,.*=-y,.*=+z` adds `x` and `z` to, and removes `y` from, the execution info for all actions. - `Genrule=+requires-x` adds `requires-x` to the execution info for all Genrule actions. - `(?!Genrule).*=-requires-x` removes `requires-x` from the execution info for all non-Genrule actions.

Specifies URLs under which the source URLs of Bazel modules can be found, in addition to and taking precedence over any registry-provided mirror URLs. This flag can be specified per-registry using the syntax `--module_mirrors=<registry>=<mirror1>[,<mirror2>,...]` (e.g., `--module_mirrors=https://bcr.bazel.build=https://mirror.example.com`). It can also be specified as a comma-separated list of mirror URLs that applies to all registries that don't have an explicit list (e.g., `--module_mirrors=https://mirror1,https: //mirror2`). Set this to an empty value to disable the use of any mirrors not specified by the registries. Later uses of this flag override earlier ones with the same (or no) registry. The default set of mirrors may change over time, but all downloads from mirrors are verified by hashes stored in the registry (and thus pinned by the lockfile).

The maximum depth of the graph internal to a depset (also known as NestedSet), above which the depset() constructor will fail.

If set, and compilation mode is set to 'dbg', define GLIBCXX_DEBUG, GLIBCXX_DEBUG_PEDANTIC and GLIBCPP_CONCEPT_CHECKS.

Whether to perform symbol and dead-code strippings on linked binaries. Binary strippings will be performed if both this flag and -- compilation_mode=opt are specified.

Specifies whether to generate a linkmap file.

If set, .d files emitted by clang will be used to prune the set of inputs passed into objc compiles.

Additional options to pass to gcc when compiling Objective-C/C++ source files.

When enabled, and with experimental_one_version_enforcement set to a non- NONE value, enforce one version on java_test targets. This flag can be disabled to improve incremental test performance at the expense of missing potential one version violations.

Specifies a binary to use to do dexing without sharding.

Only shows warnings and action outputs for rules with a name matching the provided regular expression.

A list of comma-separated output group names, each of which optionally prefixed by a `+` or a `-`. A group prefixed by `+` is added to the default set of output groups, while a group prefixed by `-` is removed from the default set. If at least one group is not prefixed, the default set of output groups is omitted. For example, `--output_groups=+foo,+bar` builds the union of the default set, foo, and bar, while `--output_groups=foo,bar` overrides the default set such that only foo and bar are built.

Override a module with a local path in the form of `{module name}={path}`. If the givenpath is an absolute path, it will be used as it is. If the given path is arelative path, it is relative to the current working directory. If the given pathstarts with `%workspace%`, it is relative to the workspace root, which is theoutput of `bazel info workspace`. If the given path is empty, then remove anyprevious overrides.

Override a repository with a local path in the form of `{repository name}={path}`,where the repository name can be either a canonical name or an apparent name from thepoint of view of the main repository.Note that if this is flag is used to override a module's repository, changes to theMODULE.bazel file will not be effective if the module is obtained from a registry. Use`--override_module` instead to for that purpose.If the given path is an absolute path, it will be used as it is. If the given path is arelative path, it is relative to the current working directory. If the given path startswith `%workspace%`, it is relative to the workspace root, which is the output of `bazelinfo workspace`. If the given path is empty, then remove any previous overrides.

A colon-separated list of where to look for packages. Elements beginning with '%workspace%' are relative to the enclosing workspace. If omitted or empty, the default is the output of 'bazel info default-package-path'.

Additional options to selectively pass to gcc when compiling certain files. This option can be passed multiple times. Syntax: regex_filter@option_1, option_2,...,option_n. Where regex_filter stands for a list of include and exclude regular expression patterns (Also see --instrumentation_filter). option_1 to option_n stand for arbitrary command line options. If an option contains a comma it has to be quoted with a backslash. Options can contain @. Only the first @ is used to split the string. Example: -- per_file_copt=//foo/.*\.cc,-//foo/bar\.cc@-O0 adds the -O0 command line option to the gcc command line of all cc files in //foo/ except bar.cc.

Additional options to selectively pass to LTO backend (under -- features=thin_lto) when compiling certain backend objects. This option can be passed multiple times. Syntax: regex_filter@option_1,option_2,..., option_n. Where regex_filter stands for a list of include and exclude regular expression patterns. option_1 to option_n stand for arbitrary command line options. If an option contains a comma it has to be quoted with a backslash. Options can contain @. Only the first @ is used to split the string. Example: --per_file_ltobackendopt=//foo/.*\.o,-//foo/bar\.o@-O0 adds the -O0 command line option to the LTO backend command line of all o files in //foo/ except bar.o.

Enable persistent Android dex and desugar actions by using workers. Expands to: --internal_persistent_android_dex_desugar -- strategy=Desugar=worker --strategy=DexBuilder=worker

Enable persistent Android resource processor by using workers. Expands to: --internal_persistent_busybox_tools -- strategy=AaptPackage=worker --strategy=AndroidResourceParser=worker -- strategy=AndroidResourceValidator=worker -- strategy=AndroidResourceCompiler=worker --strategy=RClassGenerator=worker --strategy=AndroidResourceLink=worker --strategy=AndroidAapt2=worker -- strategy=AndroidAssetMerger=worker -- strategy=AndroidResourceMerger=worker -- strategy=AndroidCompiledResourceMerger=worker -- strategy=ManifestMerger=worker --strategy=AndroidManifestMerger=worker -- strategy=Aapt2Optimize=worker --strategy=AARGenerator=worker -- strategy=ProcessDatabinding=worker -- strategy=GenerateDataBindingBaseClasses=worker

Enable persistent multiplexed Android dex and desugar actions by using workers. Expands to: --persistent_android_dex_desugar -- internal_persistent_multiplex_android_dex_desugar

Enable persistent multiplexed Android resource processor by using workers. Expands to: --persistent_android_resource_processor -- modify_execution_info=AaptPackage=+supports-multiplex-workers -- modify_execution_info=AndroidResourceParser=+supports-multiplex-workers -- modify_execution_info=AndroidResourceValidator=+supports-multiplex- workers --modify_execution_info=AndroidResourceCompiler=+supports- multiplex-workers --modify_execution_info=RClassGenerator=+supports- multiplex-workers --modify_execution_info=AndroidResourceLink=+supports- multiplex-workers --modify_execution_info=AndroidAapt2=+supports- multiplex-workers --modify_execution_info=AndroidAssetMerger=+supports- multiplex-workers --modify_execution_info=AndroidResourceMerger=+supports- multiplex-workers -- modify_execution_info=AndroidCompiledResourceMerger=+supports-multiplex- workers --modify_execution_info=ManifestMerger=+supports-multiplex- workers --modify_execution_info=AndroidManifestMerger=+supports-multiplex- workers --modify_execution_info=Aapt2Optimize=+supports-multiplex-workers --modify_execution_info=AARGenerator=+supports-multiplex-workers

Enable persistent and multiplexed Android tools (dexing, desugaring, resource processing). Expands to: --internal_persistent_multiplex_busybox_tools -- persistent_multiplex_android_resource_processor -- persistent_multiplex_android_dex_desugar

The location of a mapping file that describes which platform to use if none is set or which flags to set when a platform already exists. Must be relative to the main workspace root. Defaults to `platform_mappings` (a file directly under the workspace root).

Specifies a suffix to be added to the configuration directory.

The labels of the platform rules describing the target platforms for the current command.

Plugins to use in the build. Currently works with java_plugin.

When building a target //a:a, process headers in all targets that //a:a depends on (if header processing is enabled for the toolchain).

If set, profile Bazel and write data to the specified file. See https: //bazel.build/advanced/performance/json-trace-profile for more information.

Number of profiles to retain in the output base. If there are more than this number of profiles in the output base, the oldest are deleted until the total is under the limit.

Show the command progress in the terminal title. Useful to see what bazel is doing when having multiple terminal tabs.

The number of seconds to wait between reports on still running jobs. The default value 0 means the first report will be printed after 10 seconds, then 30 seconds and after that progress is reported once every minute. When `--curses` is enabled, progress is reported every second.

Specifies which version of ProGuard to use for code removal when building a Java binary.

Use Propeller profile information to optimize the build target.A propeller profile must consist of at least one of two files, a cc profile and a ld profile. This flag accepts a build label which must refer to the propeller profile input files. For example, the BUILD file that defines the label, in a/b/BUILD:propeller_optimize( name = "propeller_profile", cc_profile = "propeller_cc_profile.txt", ld_profile = "propeller_ld_profile.txt",) An exports_files directive may have to be added to the corresponding package to make these files visible to Bazel. The option must be used as: -- propeller_optimize=//a/b:propeller_profile

Absolute path name of cc_profile file for Propeller Optimized builds.

Absolute path name of ld_profile file for Propeller Optimized builds.

The label of the proto-compiler.

Whether to pass profile_path to the proto compiler.

The profile to pass to the proto compiler as profile_path. If unset, but -- proto_profile is true (the default), infers the path from --fdo_optimize.

Label of proto_lang_toolchain() which describes how to compile C++ protos

Label of proto_lang_toolchain() which describes how to compile j2objc protos

Label of proto_lang_toolchain() which describes how to compile Java protos

Label of proto_lang_toolchain() which describes how to compile JavaLite protos

Additional options to pass to the protobuf compiler.

An allowlist (package_group target) to use when enforcing -- incompatible_python_disallow_native_rules.

The absolute path of the Python interpreter invoked to run Python targets on the target platform. Deprecated; disabled by `-- incompatible_use_python_toolchains`.

By default, Bazel profiler will record only aggregated data for fast but numerous events (such as statting the file). If this option is enabled, profiler will record each event - resulting in more precise profiling data but LARGE performance hit. Option only has effect if --profile used as well.

If true and supported, instrumentation output is redirected to be written locally on a different machine than where bazel is running on.

Specifies the registries to use to locate Bazel module dependencies. The order is important: modules will be looked up in earlier registries first, and only fall back to later registries when they're missing from the earlier ones.

Whether to accept remotely cached action results.

If set, a JSON file is written to this location that contains a detailed log of the behavior of remote analysis caching. It's interpreted as a path relative to the current working directory.

If set to 'all', all local outputs referenced by BEP are uploaded to remote cache.If set to 'minimal', local outputs referenced by BEP are not uploaded to the remote cache, except for files that are important to the consumers of BEP (e.g. test logs and timing profile). bytestream:// scheme is always used for the uri of files even if they are missing from remote cache.Default to 'minimal'.

The hostname and instance name to be used in bytestream:// URIs that are written into build event streams. This option can be set when builds are performed using a proxy, which causes the values of --remote_executor and --remote_instance_name to no longer correspond to the canonical name of the remote execution service. When not set, it will default to "${hostname}/${instance_name}".

A URI of a caching endpoint. The supported schemas are http, https, grpc, grpcs (grpc with TLS enabled) and unix (local UNIX sockets). If no schema is provided Bazel will default to grpcs. Specify grpc://, http:// or unix: schema to disable TLS. See https://bazel.build/remote/caching

If true, uploading of action results to a disk or remote cache will happen in the background instead of blocking the completion of an action. Some actions are incompatible with background uploads, and may still block even when this flag is set.

If enabled, compress/decompress cache blobs with zstd when their size is at least --experimental_remote_cache_compression_threshold.

Specify a header that will be included in cache requests: --remote_cache_header=Name=Value. Multiple headers can be passed by specifying the flag multiple times. Multiple values for the same name will be converted to a comma-separated list.

Set the default exec properties to be used as the remote execution platform if an execution platform does not already set exec_properties.

Downloads all remote outputs to the local machine. This flag is an alias for --remote_download_outputs=all. Expands to: --remote_download_outputs=all

Does not download any remote build outputs to the local machine. This flag is an alias for --remote_download_outputs=minimal. Expands to: --remote_download_outputs=minimal

If set to 'minimal' doesn't download any remote build outputs to the local machine, except the ones required by local actions. If set to 'toplevel' behaves like 'minimal' except that it also downloads outputs of top level targets to the local machine. Both options can significantly reduce build times if network bandwidth is a bottleneck.

Force remote build outputs whose path matches this pattern to be downloaded, irrespective of --remote_download_outputs. Multiple patterns may be specified by repeating this flag.

Instead of downloading remote build outputs to the local machine, create symbolic links. The target of the symbolic links can be specified in the form of a template string. This template string may contain {hash} and {size_bytes} that expand to the hash of the object and the size in bytes, respectively. These symbolic links may, for example, point to a FUSE file system that loads objects from the CAS on demand.

Only downloads remote outputs of top level targets to the local machine. This flag is an alias for --remote_download_outputs=toplevel. Expands to: --remote_download_outputs=toplevel

Specify a header that will be included in remote downloader requests: --remote_downloader_header=Name=Value. Multiple headers can be passed by specifying the flag multiple times. Multiple values for the same name will be converted to a comma-separated list.

Specify a header that will be included in execution requests: --remote_exec_header=Name=Value. Multiple headers can be passed by specifying the flag multiple times. Multiple values for the same name will be converted to a comma-separated list.

The relative priority of actions to be executed remotely. The semantics of the particular priority values are server-dependent.

HOST or HOST:PORT of a remote execution endpoint. The supported schemas are grpc, grpcs (grpc with TLS enabled) and unix (local UNIX sockets). If no schema is provided Bazel will default to grpcs. Specify grpc:// or unix: schema to disable TLS.

If specified, a path to a file to log gRPC call related details. This log consists of a sequence of serialized com.google.devtools.build.lib.remote.logging.RemoteExecutionLog.LogEntry protobufs with each message prefixed by a varint denoting the size of the following serialized protobuf message, as performed by the method LogEntry.writeDelimitedTo(OutputStream).

Specify a header that will be included in requests: --remote_header=Name=Value. Multiple headers can be passed by specifying the flag multiple times. Multiple values for the same name will be converted to a comma-separated list.

Value to pass as instance_name in the remote execution API.

Whether to fall back to standalone local execution strategy if remote execution fails.

Deprecated. See https://github.com/bazelbuild/bazel/issues/7480 for details.

Limit the max number of concurrent requests per gRPC connection. By default the value is 100.

Limit the max number of concurrent connections to remote cache/executor. By default the value is 100. Setting this to 0 means no limitation. For HTTP remote cache, one TCP connection could handle one request at one time, so Bazel could make up to --remote_max_connections concurrent requests. For gRPC remote cache/executor, one gRPC channel could usually handle 100+ concurrent requests (controlled by -- remote_max_concurrency_per_connection), so Bazel could make around `-- remote_max_connections * 100` concurrent requests.

Choose when to print remote execution messages. Valid values are `failure`, to print only on failures, `success` to print only on successes and `all` to print always.

Connect to the remote cache through a proxy. Currently this flag can only be used to configure a Unix domain socket (unix:/path/to/socket).

The relative priority of remote actions to be stored in remote cache. The semantics of the particular priority values are server-dependent.

The maximum number of attempts to retry a transient error. If set to 0, retries are disabled.

The maximum backoff delay between remote retry attempts. Following units can be used: Days (d), hours (h), minutes (m), seconds (s), and milliseconds (ms). If the unit is omitted, the value is interpreted as seconds.

The maximum amount of time to wait for remote execution and cache calls. For the REST cache, this is both the connect and the read timeout. Following units can be used: Days (d), hours (h), minutes (m), seconds (s), and milliseconds (ms). If the unit is omitted, the value is interpreted as seconds.

Whether to upload locally executed action results to the remote cache if the remote cache supports it and the user is authorized to do so.

If set to true, Bazel will compute the hash sum of all remote downloads and discard the remotely cached values if they don't match the expected value.

Specifies the location of the repo contents cache, which contains fetched repo directories shareable across workspaces. An empty string as argument requests the repo contents cache to be disabled, otherwise the default of `{-- repository_cache}/contents` is used. Note that this means setting `--repository_cache=` would by default disable the repo contents cache as well, unless `--repo_contents_cache={some_path}` is also set.

Specifies the amount of time the server must remain idle before garbage collection happens to the repo contents cache.

Specifies the amount of time an entry in the repo contents cache can stay unused before it's garbage collected. If set to zero, only duplicate entries will be garbage collected.

Specifies additional environment variables to be available only for repository rules. Note that repository rules see the full environment anyway, but in this way variables can be set via command-line flags and <code>.bazelrc</code> entries. The special syntax <code>=NAME</code> can be used to explicitly unset a variable. The string <code>%bazel_workspace% </code> in a value will be replaced with the absolute path of the workspace as printed by <code>bazel info workspace</code>.

A list of additional repositories (beyond the hardcoded ones Bazel knows about) where autoloads are not to be added. This should typically contain repositories that are transitively depended on by a repository that may be loaded automatically (and which can therefore potentially create a cycle).

Specifies the cache location of the downloaded values obtained during the fetching of external repositories. An empty string as argument requests the cache to be disabled, otherwise the default of `{--output_user_root}/cache/repos/v1` is used.

If set, downloading using `ctx.download{,_and_extract}` is not allowed during repository fetching. Note that network access is not completely disabled; ctx.execute could still run an arbitrary executable that accesses the Internet.

If set to true, directories used by sandboxed non-worker execution may be reused to avoid unnecessary setup costs.

Prefix to insert before the executables for the `test` and `run` commands. If the value is `foo -bar`, and the execution command line is `test_binary - baz`, then the final command line is `foo -bar test_binary -baz`. This can also be a label to an executable target. Some examples are: - `valgrind` - `strace` - `strace -c` - `valgrind --quiet --num-callers=20` - `//package:target` - `//package:target --options`

Whether to run validation actions as part of the build. See [Validation Actions]. [Validation Actions]: https://bazel.build/extending/rules#validation_actions

Specifies number of times to run each test. If any of those attempts fail for anyreason, the whole test is considered failed. Normally the value specified isjust an integer.Example: `--runs_per_test=3` will run all tests 3 times.Alternate syntax: `regex_filter@runs_per_test`. Where `runs_per_test` stands foran integer value and `regex_filter` stands for a list of include and excluderegular expression patterns (Also see --instrumentation_filter).Example: `--runs_per_test=//foo/.*,-//foo/bar/.*@3` runs all tests in `//foo/` exceptthose under `//foo/bar` three times. This option can be passed multiple times. The mostrecently passed argument that matches takes precedence. If nothing matches,the test is only run once.

If true, any shard in which at least one run/attempt passes and at least one run/attempt fails gets a FLAKY status.

Add additional path pair to mount in sandbox.

Lets the sandbox create its sandbox directories underneath this path. Specify a path on tmpfs (like /run/shm) to possibly improve performance a lot when your build / tests have many input files. Note: You need enough RAM and free space on the tmpfs to hold output and intermediate files generated by running actions.

For sandboxed actions, disallow access to this path.

Enables debugging features for the sandboxing feature. This includes two things: first, the sandbox root contents are left untouched after a build; and second, prints extra debugging information on execution. This can help developers of Bazel or Starlark rules with debugging failures due to missing input files, etc.

Allow network access by default for actions; this may not work with all sandboxing implementations.

If true, a loopback device will be set up in the linux-sandbox network namespace for local actions.

Explicitly enable the creation of pseudoterminals for sandboxed actions. Some linux distributions require setting the group id of the process to 'tty' inside the sandbox in order for pseudoterminals to function. If this is causing issues, this flag can be disabled to enable other groups to be used.

Change the current hostname to 'localhost' for sandboxed actions.

Change the current username to 'nobody' for sandboxed actions.

For sandboxed actions, mount an empty, writable directory at this absolute path (if supported by the sandboxing implementation, ignored otherwise).

For sandboxed actions, make an existing directory writable in the sandbox (if supported by the sandboxing implementation, ignored otherwise).

If set, temporary outputs from gcc will be saved. These include .s files (assembler code), .i files (preprocessed C) and .ii files (preprocessed C++).

Dump a profile of serialized frontier bytes. Specifies the output path.