User's Guide to Bazel

Bazel overview

To run Bazel, go to your base workspace directory or any of its subdirectories and type bazel.

  % bazel help
                             [Bazel release bazel-<version>]
  Usage: bazel <command> <options> ...

  Available commands:
    analyze-profile     Analyzes build profile data.
    build               Builds the specified targets.

    canonicalize-flags  Canonicalize Bazel flags.
    clean               Removes output files and optionally stops the server.

    help                Prints help for commands, or the index.

    info                Displays runtime info about the bazel server.

    fetch                Fetches all external dependencies of a target.
    mobile-install      Installs apps on mobile devices.

    query               Executes a dependency graph query.

    run                 Runs the specified target.
    shutdown            Stops the Bazel server.
    test                Builds and runs the specified test targets.
    version             Prints version information for Bazel.

  Getting more help:
    bazel help <command>
                     Prints help and options for <command>.
    bazel help startup_options
                     Options for the JVM hosting Bazel.
    bazel help target-syntax
                     Explains the syntax for specifying targets.
    bazel help info-keys
                     Displays a list of keys used by the info command.

The bazel tool performs many functions, called commands; users of CVS and Subversion will be familiar with this "Swiss army knife" arrangement. The most commonly used one is of course bazel build. You can browse the online help messages using bazel help.

Client/server implementation

The Bazel system is implemented as a long-lived server process. This allows it to perform many optimizations not possible with a batch-oriented implementation, such as caching of BUILD files, dependency graphs, and other metadata from one build to the next. This improves the speed of incremental builds, and allows different commands, such as build and query to share the same cache of loaded packages, making queries very fast.

When you run bazel, you're running the client. The client finds the server based on the output base, which by default is determined by the path of the base workspace directory and your userid, so if you build in multiple workspaces, you'll have multiple output bases and thus multiple Bazel server processes. Multiple users on the same workstation can build concurrently in the same workspace because their output bases will differ (different userids). If the client cannot find a running server instance, it starts a new one. The server process will stop after a period of inactivity (3 hours, by default).

For the most part, the fact that there is a server running is invisible to the user, but sometimes it helps to bear this in mind. For example, if you're running scripts that perform a lot of automated builds in different directories, it's important to ensure that you don't accumulate a lot of idle servers; you can do this by explicitly shutting them down when you're finished with them, or by specifying a short timeout period.

The name of a Bazel server process appears in the output of ps x or ps -e f as bazel(dirname), where dirname is the basename of the directory enclosing the root your workspace directory. For example:

  % ps -e f
  16143 ?        Sl     3:00 bazel(src-jrluser2) -server -Djava.library.path=...

This makes it easier to find out which server process belongs to a given workspace. (Beware that with certain other options to ps, Bazel server processes may be named just java.) Bazel servers can be stopped using the shutdown command.

You can also run Bazel in batch mode using the --batch startup flag. This will immediately shut down the process after the command (build, test, etc.) has finished and not keep a server process around.

When running bazel, the client first checks that the server is the appropriate version; if not, the server is stopped and a new one started. This ensures that the use of a long-running server process doesn't interfere with proper versioning.

.bazelrc, the Bazel configuration file, the --bazelrc=file option, and the --config=value option

Bazel accepts many options. Typically, some of these are varied frequently (e.g. --subcommands) while others stay the same across several builds (e.g. --package_path). To avoid having to specify these constant options every time you do a build or run some other Bazel command, Bazel allows you to specify options in a configuration file.

Bazel looks for an optional configuration file in the location specified by the --bazelrc=file option. If this option is not specified then, by default, Bazel looks for the file called .bazelrc in one of two directories: first, in your base workspace directory, then in your home directory. If it finds a file in the first (workspace-specific) location, it will not look at the second (global) location.

The --bazelrc=file option must appear before the command name (e.g. build).

The option --bazelrc=/dev/null effectively disables the use of a configuration file. We strongly recommend that you use this option when performing release builds, or automated tests that invoke Bazel.

Aside from the configuration file described above, Bazel also looks for a master configuration file next to the binary, in the workspace at tools/bazel.rc or system-wide at /etc/bazel.bazelrc. These files are here to support installation-wide options or options shared between users.

Like all UNIX "rc" files, the .bazelrc file is a text file with a line-based grammar. Lines starting # are considered comments and are ignored, as are blank lines. Each line contains a sequence of words, which are tokenized according to the same rules as the Bourne shell. The first word on each line is the name of a Bazel command, such as build or query. The remaining words are the default options that apply to that command. More than one line may be used for a command; the options are combined as if they had appeared on a single line. (Users of CVS, another tool with a "Swiss army knife" command-line interface, will find the syntax familiar to that of .cvsrc.)

Startup options may be specified in the .bazelrc file using the command startup. These options are described in the interactive help at bazel help startup_options.

Options specified in the command line always take precedence over those from a configuration file. In configuration files, lines for a more specific command take precedence over lines for a less specific command (e.g. the 'test' command inherits all the options from the 'build' command, so a 'test --foo=bar' line takes precedence over a 'build --foo=baz' line, regardless of which configuration files these two lines are in) and lines equally specific for which command they apply have precedence based on the configuration file they are in, with the user-specific configuration file taking precedence over the master one.

Options may include words other than flags, such as the names of build targets, etc; these are always prepended to the explicit argument list provided on the command-line, if any.

Common command options may be specified in the .bazelrc file using the command common.

In addition, commands may have :name suffixes. These options are ignored by default, but can be pulled in through the --config=name option, either on the command line or in a .bazelrc file. The intention is that these bundle command line options that are commonly used together, for example --config=memcheck.

Note that some config sections are defined in the master bazelrc file. To avoid conflicts, user-defined sections should start with the '_' (underscore) character.

The command named import is special: if Bazel encounters such a line in a .bazelrc file, it parses the contents of the file referenced by the import statement, too. Options specified in an imported file take precedence over ones specified before the import statement, options specified after the import statement take precedence over the ones in the imported file, and options in files imported later take precedence over files imported earlier.

Here's an example ~/.bazelrc file:

  # Bob's Bazel option defaults

  startup --batch --host_jvm_args=-XX:-UseParallelGC
  import /home/bobs_project/bazelrc
  build --show_timestamps --keep_going --jobs 600
  build --color=yes
  query --keep_going

  build:memcheck --strip=never --test_timeout=3600

Building programs with Bazel

The build command

The most important function of Bazel is, of course, building code. Type bazel build followed by the name of the target you wish to build. Here's a typical session:

  % bazel build //foo
  ____Loading package: foo
  ____Loading package: bar
  ____Loading package: baz
  ____Loading complete.  Analyzing...
  ____Building 1 target...
  ____[0 / 3] Executing Genrule //bar:helper_rule
  ____[1 / 3] Executing Genrule //baz:another_helper_rule
  ____[2 / 3] Building foo/foo.bin
  Target //foo:foo up-to-date:
  ____Elapsed time: 9.905s

Bazel prints the progress messages as it loads all the packages in the transitive closure of dependencies of the requested target, then analyzes them for correctness and to create the build actions, finally executing the compilers and other tools of the build.

Bazel prints progress messages during the execution phase of the build, showing the current build step (compiler, linker, etc.) that is being started, and the number of completed over total number of build actions. As the build starts the number of total actions will often increase as Bazel discovers the entire action graph, but the number will usually stabilize within a few seconds.

At the end of the build Bazel prints which targets were requested, whether or not they were successfully built, and if so, where the output files can be found. Scripts that run builds can reliably parse this output; see --show_result for more details.

Typing the same command again:

  % bazel build //foo
  ____Found 1 target...
  ____Building complete.
  Target //foo:foo up-to-date:
  ____Elapsed time: 0.280s

we see a "null" build: in this case, there are no packages to re-load, since nothing has changed, and no build steps to execute. (If something had changed in "foo" or some of its dependencies, resulting in the reexecution of some build actions, we would call it an "incremental" build, not a "null" build.)

Before you can start a build, you will need a Bazel workspace. This is simply a directory tree that contains all the source files needed to build your application. Bazel allows you to perform a build from a completely read-only volume.

Setting up a --package_path

Bazel finds its packages by searching the package path. This is a colon separated ordered list of bazel directories, each being the root of a partial source tree.

To specify a custom package path using the --package_path option:

  % bazel build --package_path %workspace%:/some/other/root

Package path elements may be specified in three formats:

  1. If the first character is /, the path is absolute.
  2. If the path starts with %workspace%, the path is taken relative to the nearest enclosing bazel directory.
    For instance, if your working directory is /home/bob/clients/bob_client/bazel/foo, then the string %workspace% in the package-path is expanded to /home/bob/clients/bob_client/bazel.
  3. Anything else is taken relative to the working directory.
    This is usually not what you mean to do, and may behave unexpectedly if you use Bazel from directories below the bazel workspace. For instance, if you use the package-path element ., and then cd into the directory /home/bob/clients/bob_client/bazel/foo, packages will be resolved from the /home/bob/clients/bob_client/bazel/foo directory.

If you use a non-default package path, we recommend that you specify it in your Bazel configuration file for convenience.

Bazel doesn't require any packages to be in the current directory, so you can do a build from an empty bazel workspace if all the necessary packages can be found somewhere else on the package path.

Example: Building from an empty client

  % mkdir -p foo/bazel
  % cd foo/bazel
  % bazel build --package_path /some/other/path //foo

Specifying targets to build

Bazel allows a number of ways to specify the targets to be built. Collectively, these are known as target patterns. The on-line help displays a summary of supported patterns:

% bazel help target-syntax

Target pattern syntax

The BUILD file label syntax is used to specify a single target. Target
patterns generalize this syntax to sets of targets, and also support
working-directory-relative forms, recursion, subtraction and filtering.

Specifying a single target:

  //foo/bar:wiz     The single target '//foo/bar:wiz'.
  foo/bar/wiz       Equivalent to:
                      '//foo/bar/wiz:wiz' if foo/bar/wiz is a package,
                      '//foo/bar:wiz' if foo/bar is a package,
                      '//foo:bar/wiz' otherwise.
  //foo/bar         Equivalent to '//foo/bar:bar'.

Specifying all rules in a package:

  //foo/bar:all       Matches all rules in package 'foo/bar'.

Specifying all rules recursively beneath a package:

  //foo/...:all     Matches all rules in all packages beneath directory 'foo'.
  //foo/...           (ditto)

  By default, directory symlinks are followed when performing this recursive traversal, except
  those that point to under the output base (for example, the convenience symlinks that are created
  in the root directory of the workspace) But we understand that your workspace may intentionally
  contain directories with unusual symlink structures that you don't want consumed. As such, if a
  directory has a file named
  in that directory won't be followed when evaluating recursive target patterns.

Working-directory relative forms:  (assume cwd = 'workspace/foo')

  Target patterns which do not begin with '//' are taken relative to
  the working directory.  Patterns which begin with '//' are always

  ...:all           Equivalent to  '//foo/...:all'.
  ...                 (ditto)

  bar/...:all       Equivalent to  '//foo/bar/...:all'.
  bar/...             (ditto)

  bar:wiz           Equivalent to '//foo/bar:wiz'.
  :foo              Equivalent to '//foo:foo'.

  bar               Equivalent to '//foo/bar:bar'.
  foo/bar           Equivalent to '//foo/foo/bar:bar'.

  bar:all           Equivalent to '//foo/bar:all'.
  :all              Equivalent to '//foo:all'.

Summary of target wildcards:

  :all,             Match all rules in the specified packages.
  :*, :all-targets  Match all targets (rules and files) in the specified
                      packages, including ones not built by default, such
                      as _deploy.jar files.

Subtractive patterns:

  Target patterns may be preceded by '-', meaning they should be
  subtracted from the set of targets accumulated by preceding
  patterns. (Note that this means order matters.) For example:

    % bazel build -- foo/... -foo/contrib/...

  builds everything in 'foo', except 'contrib'.  In case a target not
  under 'contrib' depends on something under 'contrib' though, in order to
  build the former bazel has to build the latter too. As usual, the '--' is
  required to prevent '-f' from being interpreted as an option.

Whereas labels are used to specify individual targets, e.g. for declaring dependencies in BUILD files, Bazel's target patterns are a syntax for specifying multiple targets: they are a generalization of the label syntax for sets of targets, using wildcards. In the simplest case, any valid label is also a valid target pattern, identifying a set of exactly one target.

foo/... is a wildcard over packages, indicating all packages recursively beneath directory foo (for all roots of the package path). :all is a wildcard over targets, matching all rules within a package. These two may be combined, as in foo/...:all, and when both wildcards are used, this may be abbreviated to foo/....

In addition, :* (or :all-targets) is a wildcard that matches every target in the matched packages, including files that aren't normally built by any rule, such as _deploy.jar files associated with java_binary rules.

This implies that :* denotes a superset of :all; while potentially confusing, this syntax does allow the familiar :all wildcard to be used for typical builds, where building targets like the _deploy.jar is not desired.

In addition, Bazel allows a slash to be used instead of the colon required by the label syntax; this is often convenient when using Bash filename expansion. For example, foo/bar/wiz is equivalent to //foo/bar:wiz (if there is a package foo/bar) or to //foo:bar/wiz (if there is a package foo).

Many Bazel commands accept a list of target patterns as arguments, and they all honor the prefix negation operator `-'. This can be used to subtract a set of targets from the set specified by the preceding arguments. (Note that this means order matters.) For example,

  bazel build foo/... bar/...

means "build all targets beneath foo and all targets beneath bar", whereas

  bazel build -- foo/... -foo/bar/...

means "build all targets beneath foo except those beneath foo/bar". (The -- argument is required to prevent the subsequent arguments starting with - from being interpreted as additional options.)

It's important to point out though that subtracting targets this way will not guarantee that they are not built, since they may be dependencies of targets that weren't subtracted. For example, if there were a target //foo:all-apis that among others depended on //foo/bar:api, then the latter would be built as part of building the former.

Targets with tags=["manual"] will not be included in wildcard target patterns (..., :*, :all, etc). You should specify such test targets with explicit target patterns on the command line if you want Bazel to build/test them.

Fetching external dependencies

By default, Bazel will download and symlink external dependencies during the build. However, this can be undesirable, either because you'd like to know when new external dependendencies are added or because you'd like to "prefetch" dependencies (say, before a flight where you'll be offline). If you would like to prevent new dependencies from being added during builds, you can specify the --fetch=false flag. Note that this flag only applies to repository rules that do not point to a directory in the local file system. Changes, for example, to local_repository, new_local_repository and Android SDK and NDK repository rules will always take effect regardless of the value --fetch .

If you disallow fetching during builds and Bazel finds new external dependencies, your build will fail.

You can manually fetch dependencies by running bazel fetch. If you disallow during-build fetching, you'll need to run bazel fetch:

  1. Before you build for the first time.
  2. After you add a new external dependency.
Once it has been run, you should not need to run it again until the WORKSPACE file changes.

fetch takes a list of targets to fetch dependencies for. For example, this would fetch dependencies needed to build //foo:bar and //bar:baz:

$ bazel fetch //foo:bar //bar:baz

To fetch all external dependencies for a workspace, run:

$ bazel fetch //...

You do not need to run bazel fetch at all if you have all of the tools you are using (from library jars to the JDK itself) under your workspace root. However, if you're using anything outside of the workspace directory then you will need to run bazel fetch before running bazel build.

Build configurations and cross-compilation

All the inputs that specify the behavior and result of a given build can be divided into two distinct categories. The first kind is the intrinsic information stored in the BUILD files of your project: the build rule, the values of its attributes, and the complete set of its transitive dependencies. The second kind is the external or environmental data, supplied by the user or by the build tool: the choice of target architecture, compilation and linking options, and other toolchain configuration options. We refer to a complete set of environmental data as a configuration.

In any given build, there may be more than one configuration. Consider a cross-compile, in which you build a //foo:bin executable for a 64-bit architecture, but your workstation is a 32-bit machine. Clearly, the build will require building //foo:bin using a toolchain capable of creating 64-bit executables, but the build system must also build various tools used during the build itself—for example tools that are built from source, then subsequently used in, say, a genrule—and these must be built to run on your workstation. Thus we can identify two configurations: the host configuration, which is used for building tools that run during the build, and the target configuration (or request configuration, but we say "target configuration" more often even though that word already has many meanings), which is used for building the binary you ultimately requested.

Typically, there are many libraries that are prerequisites of both the requested build target (//foo:bin) and one or more of the host tools, for example some base libraries. Such libraries must be built twice, once for the host configuration, and once for the target configuration.
Bazel takes care of ensuring that both variants are built, and that the derived files are kept separate to avoid interference; usually such targets can be built concurrently, since they are independent of each other. If you see progress messages indicating that a given target is being built twice, this is most likely the explanation.

Bazel uses one of two ways to select the host configuration, based on the --distinct_host_configuration option. This boolean option is somewhat subtle, and the setting may improve (or worsen) the speed of your builds.


When this option is false, the host and request configurations are identical: all tools required during the build will be built in exactly the same way as target programs. This setting means that no libraries need to be built twice during a single build, so it keeps builds short. However, it does mean that any change to your request configuration also affects your host configuration, causing all the tools to be rebuilt, and then anything that depends on the tool output to be rebuilt too. Thus, for example, simply changing a linker option between builds might cause all tools to be re-linked, and then all actions using them reexecuted, and so on, resulting in a very large rebuild. Also, please note: if your host architecture is not capable of running your target binaries, your build will not work.

If you frequently make changes to your request configuration, such as alternating between -c opt and -c dbg builds, or between simple- and cross-compilation, we do not recommend this option, as you will typically rebuild the majority of your codebase each time you switch.

--distinct_host_configuration=true (default)

If this option is true, then instead of using the same configuration for the host and request, a completely distinct host configuration is used. The host configuration is derived from the target configuration as follows:

  • Use the same version of Crosstool (--crosstool_top) as specified in the request configuration, unless --host_crosstool_top is specified.
  • Use the value of --host_cpu for --cpu (default: k8).
  • Use the same values of these options as specified in the request configuration: --compiler, --use_ijars, --java_toolchain, If --host_crosstool_top is used, then the value of --host_cpu is used to look up a default_toolchain in the Crosstool (ignoring --compiler) for the host configuration.
  • Use optimized builds for C++ code (-c opt).
  • Generate no debugging information (--copt=-g0).
  • Strip debug information from executables and shared libraries (--strip=always).
  • Place all derived files in a special location, distinct from that used by any possible request configuration.
  • Suppress stamping of binaries with build data (see --embed_* options).
  • All other values remain at their defaults.

There are many reasons why it might be preferable to select a distinct host configuration from the request configuration. Some are too esoteric to mention here, but two of them are worth pointing out.

Firstly, by using stripped, optimized binaries, you reduce the time spent linking and executing the tools, the disk space occupied by the tools, and the network I/O time in distributed builds.

Secondly, by decoupling the host and request configurations in all builds, you avoid very expensive rebuilds that would result from minor changes to the request configuration (such as changing a linker options does), as described earlier.

That said, for certain builds, this option may be a hindrance. In particular, builds in which changes of configuration are infrequent (especially certain Java builds), and builds where the amount of code that must be built in both host and target configurations is large, may not benefit.

Correct incremental rebuilds

One of the primary goals of the Bazel project is to ensure correct incremental rebuilds. Previous build tools, especially those based on Make, make several unsound assumptions in their implementation of incremental builds.

Firstly, that timestamps of files increase monotonically. While this is the typical case, it is very easy to fall afoul of this assumption; syncing to an earlier revision of a file causes that file's modification time to decrease; Make-based systems will not rebuild.

More generally, while Make detects changes to files, it does not detect changes to commands. If you alter the options passed to the compiler in a given build step, Make will not re-run the compiler, and it is necessary to manually discard the invalid outputs of the previous build using make clean.

Also, Make is not robust against the unsuccessful termination of one of its subprocesses after that subprocess has started writing to its output file. While the current execution of Make will fail, the subsequent invocation of Make will blindly assume that the truncated output file is valid (because it is newer than its inputs), and it will not be rebuilt. Similarly, if the Make process is killed, a similar situation can occur.

Bazel avoids these assumptions, and others. Bazel maintains a database of all work previously done, and will only omit a build step if it finds that the set of input files (and their timestamps) to that build step, and the compilation command for that build step, exactly match one in the database, and, that the set of output files (and their timestamps) for the database entry exactly match the timestamps of the files on disk. Any change to the input files or output files, or to the command itself, will cause re-execution of the build step.

The benefit to users of correct incremental builds is: less time wasted due to confusion. (Also, less time spent waiting for rebuilds caused by use of make clean, whether necessary or pre-emptive.)

Build consistency and incremental builds

Formally, we define the state of a build as consistent when all the expected output files exist, and their contents are correct, as specified by the steps or rules required to create them. When you edit a source file, the state of the build is said to be inconsistent, and remains inconsistent until you next run the build tool to successful completion. We describe this situation as unstable inconsistency, because it is only temporary, and consistency is restored by running the build tool.

There is another kind of inconsistency that is pernicious: stable inconsistency. If the build reaches a stable inconsistent state, then repeated successful invocation of the build tool does not restore consistency: the build has gotten "stuck", and the outputs remain incorrect. Stable inconsistent states are the main reason why users of Make (and other build tools) type make clean. Discovering that the build tool has failed in this manner (and then recovering from it) can be time consuming and very frustrating.

Conceptually, the simplest way to achieve a consistent build is to throw away all the previous build outputs and start again: make every build a clean build. This approach is obviously too time-consuming to be practical (except perhaps for release engineers), and therefore to be useful, the build tool must be able to perform incremental builds without compromising consistency.

Correct incremental dependency analysis is hard, and as described above, many other build tools do a poor job of avoiding stable inconsistent states during incremental builds. In contrast, Bazel offers the following guarantee: after a successful invocation of the build tool during which you made no edits, the build will be in a consistent state. (If you edit your source files during a build, Bazel makes no guarantee about the consistency of the result of the current build. But it does guarantee that the results of the next build will restore consistency.)

As with all guarantees, there comes some fine print: there are some known ways of getting into a stable inconsistent state with Bazel. We won't guarantee to investigate such problems arising from deliberate attempts to find bugs in the incremental dependency analysis, but we will investigate and do our best to fix all stable inconsistent states arising from normal or "reasonable" use of the build tool.

If you ever detect a stable inconsistent state with Bazel, please report a bug.

Sandboxed execution

Bazel uses sandboxes to guarantee that actions run hermetically1 and correctly. Bazel runs Spawns (loosely speaking: actions) in sandboxes that only contain the minimal set of files the tool requires to do its job. Currently sandboxing works on Linux 3.12 or newer with the CONFIG_USER_NS option enabled, and also on Mac OS 10.11 for newer.

Bazel will print a warning if your system does not support sandboxing to alert you to the fact that builds are not guaranteed to be hermetic and might affect the host system in unknown ways. To disable this warning you can pass the --ignore_unsupported_sandboxing flag to Bazel.

On some platforms such as Google Container Engine cluster nodes or Debian, user namespaces are deactivated by default due to security concerns. This can be checked by looking at the file /proc/sys/kernel/unprivileged_userns_clone: if it exists and contains a 0, then user namespaces can be activated with sudo sysctl kernel.unprivileged_userns_clone=1.

In some cases, the Bazel sandbox fails to execute rules because of the system setup. The symptom is generally a failure that output a message similar to namespace-sandbox.c:633: execvp(argv[0], argv): No such file or directory. In that case, try to deactivate the sandbox for genrules with --genrule_strategy=standalone and for other rules with --spawn_strategy=standalone. Also please report a bug on our issue tracker and mention which Linux distribution you're using so that we can investigate and provide a fix in a subsequent release.

1: Hermeticity means that the action only uses its declared input files and no other files in the filesystem, and it only produces its declared output files.

Deleting the outputs of a build

The clean command

Bazel has a clean command, analogous to that of Make. It deletes the output directories for all build configurations performed by this Bazel instance, or the entire working tree created by this Bazel instance, and resets internal caches. If executed without any command-line options, then the output directory for all configurations will be cleaned.

Recall that each Bazel instance is associated with a single workspace, thus the clean command will delete all outputs from all builds you've done with that Bazel instance in that workspace.

To completely remove the entire working tree created by a Bazel instance, you can specify the --expunge option. When executed with --expunge, the clean command simply removes the entire output base tree which, in addition to the build output, contains all temp files created by Bazel. It also stops the Bazel server after the clean, equivalent to the shutdown command. For example, to clean up all disk and memory traces of a Bazel instance, you could specify:

  % bazel clean --expunge

Alternatively, you can expunge in the background by using --expunge_async. It is safe to invoke a Bazel command in the same client while the asynchronous expunge continues to run. Note, however, that this may introduce IO contention.

The clean command is provided primarily as a means of reclaiming disk space for workspaces that are no longer needed. However, we recognize that Bazel's incremental rebuilds might not be perfect; clean may be used to recover a consistent state when problems arise.

Bazel's design is such that these problems are fixable; we consider such bugs a high priority, and will do our best fix them. If you ever find an incorrect incremental build, please file a bug report. We encourage developers to get out of the habit of using clean and into that of reporting bugs in the tools.

Phases of a build

In Bazel, a build occurs in three distinct phases; as a user, understanding the difference between them provides insight into the options which control a build (see below).

Loading phase

The first is loading during which all the necessary BUILD files for the initial targets, and their transitive closure of dependencies, are loaded, parsed, evaluated and cached.

For the first build after a Bazel server is started, the loading phase typically takes many seconds as many BUILD files are loaded from the file system. In subsequent builds, especially if no BUILD files have changed, loading occurs very quickly.

Errors reported during this phase include: package not found, target not found, lexical and grammatical errors in a BUILD file, and evaluation errors.

Analysis phase

The second phase, analysis, involves the semantic analysis and validation of each build rule, the construction of a build dependency graph, and the determination of exactly what work is to be done in each step of the build.

Like loading, analysis also takes several seconds when computed in its entirety. However, Bazel caches the dependency graph from one build to the next and only reanalyzes what it has to, which can make incremental builds extremely fast in the case where the packages haven't changed since the previous build.

Errors reported at this stage include: inappropriate dependencies, invalid inputs to a rule, and all rule-specific error messages.

The loading and analysis phases are fast because Bazel avoids unnecessary file I/O at this stage, reading only BUILD files in order to determine the work to be done. This is by design, and makes Bazel a good foundation for analysis tools, such as Bazel's query command, which is implemented atop the loading phase.

Execution phase

The third and final phase of the build is execution. This phase ensures that the outputs of each step in the build are consistent with its inputs, re-running compilation/linking/etc. tools as necessary. This step is where the build spends the majority of its time, ranging from a few seconds to over an hour for a large build. Errors reported during this phase include: missing source files, errors in a tool executed by some build action, or failure of a tool to produce the expected set of outputs.


The following sections describe the options available during a build. When --long is used on a help command, the on-line help messages provide summary information about the meaning, type and default value for each option.

Most options can only be specified once. When specified multiple times, the last instance wins. Options that can be specified multiple times are identified in the on-line help with the text 'may be used multiple times'.

Options that affect how packages are located

See also the --show_package_location option.


This option specifies the set of directories that are searched to find the BUILD file for a given package.


This option specifies a comma-separated list of packages which Bazel should consider deleted, and not attempt to load from any directory on the package path. This can be used to simulate the deletion of packages without actually deleting them.

Error checking options

These options control Bazel's error-checking and/or warnings.

--check_constraint constraint

This option takes an argument that specifies which constraint should be checked.

Bazel performs special checks on each rule that is annotated with the given constraint.

The supported constraints and their checks are as follows:

  • public: Verify that all java_libraries marked with constraints = ['public'] only depend on java_libraries that are marked as constraints = ['public'] too. If bazel finds a dependency that does not conform to this rule, bazel will issue an error.


If this option is set to false, visibility checks are demoted to warnings. The default value of this option is true, so that by default, visibility checking is done.


The experimental_action_listener option instructs Bazel to use details from the action_listener rule specified by label to insert extra_actions into the build graph.


The experimental_extra_action_filter option instructs Bazel to filter the set of targets to schedule extra_actions for.

This flag is only applicable in combination with the --experimental_action_listener flag.

By default all extra_actions in the transitive closure of the requested targets-to-build get scheduled for execution. --experimental_extra_action_filter will restrict scheduling to extra_actions of which the owner's label matches the specified regular expression.

The following example will limit scheduling of extra_actions to only apply to actions of which the owner's label contains '/bar/':

% bazel build --experimental_action_listener=//test:al //foo/... \

--output_filter regex

The --output_filter option will only show build and compilation warnings for targets that match the regular expression. If a target does not match the given regular expression and its execution succeeds, its standard output and standard error are thrown away. This option is intended to be used to help focus efforts on fixing warnings in packages under development. Here are some typical values for this option:

--output_filter= Show all output.
--output_filter='^//(first/project|second/project):' Show the output for the specified packages.
--output_filter='^//((?!(first/bad_project|second/bad_project):).)*$' Don't show output for the specified packages.
--output_filter=DONT_MATCH_ANYTHING Don't show output.


When this option is enabled, visible analysis warnings (as specified by the output filter) are treated as errors, effectively preventing the build phase from starting. This feature can be used to enable strict builds that do not allow new warnings to creep into a project.

Flags options

These options control which options Bazel will pass to other tools.

--copt gcc-option

This option takes an argument which is to be passed to gcc. The argument will be passed to gcc whenever gcc is invoked for preprocessing, compiling, and/or assembling C, C++, or assembler code. It will not be passed when linking.

This option can be used multiple times. For example:

  % bazel build --copt="-g0" --copt="-fpic" //foo

will compile the foo library without debug tables, generating position-independent code.

Note that changing --copt settings will force a recompilation of all affected object files. Also note that copts values listed in specific cc_library or cc_binary build rules will be placed on the gcc command line after these options.

Warning: C++-specific options (such as -fno-implicit-templates) should be specified in --cxxopt, not in --copt. Likewise, C-specific options (such as -Wstrict-prototypes) should be specified in --conlyopt, not in copt. Similarly, gcc options that only have an effect at link time (such as -l) should be specified in --linkopt, not in --copt.

--host_copt gcc-option

This option takes an argument which is to be passed to gcc for source files that are compiled in the host configuration. This is analogous to the --copt option, but applies only to the host configuration.

--host_cxxopt gcc-option

This option takes an argument which is to be passed to gcc for source files that are compiled in the host configuration. This is analogous to the --cxxopt option, but applies only to the host configuration.

--conlyopt gcc-option

This option takes an argument which is to be passed to gcc when compiling C source files.

This is similar to --copt, but only applies to C compilation, not to C++ compilation or linking. So you can pass C-specific options (such as -Wno-pointer-sign) using --conlyopt.

Note that copts parameters listed in specific cc_library or cc_binary build rules will be placed on the gcc command line after these options.

--cxxopt gcc-option

This option takes an argument which is to be passed to gcc when compiling C++ source files.

This is similar to --copt, but only applies to C++ compilation, not to C compilation or linking. So you can pass C++-specific options (such as -fpermissive or -fno-implicit-templates) using --cxxopt. For example:

  % bazel build --cxxopt="-fpermissive" --cxxopt="-Wno-error" //foo/cruddy_code

Note that copts parameters listed in specific cc_library or cc_binary build rules will be placed on the gcc command line after these options.

--linkopt linker-option

This option takes an argument which is to be passed to gcc when linking.

This is similar to --copt, but only applies to linking, not to compilation. So you can pass gcc options that only make sense at link time (such as -lssp or -Wl,--wrap,abort) using --linkopt. For example:

  % bazel build --copt="-fmudflap" --linkopt="-lmudflap" //foo/buggy_code

Build rules can also specify link options in their attributes. This option's settings always take precedence. Also see cc_library.linkopts.

--strip (always|never|sometimes)

This option determines whether Bazel will strip debugging information from all binaries and shared libraries, by invoking the linker with the -Wl,--strip-debug option. --strip=always means always strip debugging information. --strip=never means never strip debugging information. The default value of --strip=sometimes means strip iff the --compilation_mode is fastbuild.

  % bazel build --strip=always //foo:bar

will compile the target while stripping debugging information from all generated binaries.

Note that if you want debugging information, it's not enough to disable stripping; you also need to make sure that the debugging information was generated by the compiler, which you can do by using either -c dbg or --copt -g.

Note also that Bazel's --strip option corresponds with ld's --strip-debug option: it only strips debugging information. If for some reason you want to strip all symbols, not just debug symbols, you would need to use ld's --strip-all option, which you can do by passing --linkopt=-Wl,--strip-all to Bazel.

--stripopt strip-option

An additional option to pass to the strip command when generating a *.stripped binary. The default is -S -p. This option can be used multiple times.

Note that --stripopt does not apply to the stripping of the main binary with --strip=(always|sometimes).

--fdo_instrument profile-output-dir

The --fdo_instrument option enables the generation of FDO (feedback directed optimization) profile output when the built C/C++ binary is executed. For GCC, the argument provided is used as a directory prefix for a per-object file directory tree of .gcda files containing profile information for each .o file.

Once the profile data tree has been generated, the profile tree should be zipped up, and provided to the --fdo_optimize=profile-zip Bazel option to enable the FDO optimized compilation.

For the LLVM compiler the argument is also the directory under which the raw LLVM profile data file(s) is dumped, e.g. --fdo_instrument=/path/to/rawprof/dir/.

The options --fdo_instrument and --fdo_optimize cannot be used at the same time.

--fdo_optimize profile-zip

The --fdo_optimize option enables the use of the per-object file profile information to perform FDO (feedback directed optimization) optimizations when compiling. For GCC, the argument provided is the zip file containing the previously-generated file tree of .gcda files containing profile information for each .o file.

Alternatively, the argument provided can point to an auto profile identified by the extension .afdo.

Note that this option also accepts labels that resolve to source files. You may need to add an exports_files directive to the corresponding package to make the file visible to Bazel.

For the LLVM compiler the argument provided should point to the indexed LLVM profile output file prepared by the llvm-profdata tool, and should have a .profdata extension.

The options --fdo_instrument and --fdo_optimize cannot be used at the same time.

--lipo (off|binary)

The --lipo=binary option enables LIPO (Lightweight Inter-Procedural Optimization). LIPO is an extended C/C++ optimization technique that optimizes code across different object files. It involves compiling each C/C++ source file differently for every binary. This is in contrast to normal compilation where compilation outputs are reused. This means that LIPO is more expensive than normal compilation.

This option only has an effect when FDO is also enabled (see the --fdo_instrument and --fdo_options). Currently LIPO is only supported when building a single cc_binary rule.

Setting --lipo=binary implicitly sets --dynamic_mode=off.

--lipo_context context-binary

Specifies the label of a cc_binary rule that was used to generate the profile information for LIPO that was given to the --fdo_optimize option.

Specifying the context is mandatory when --lipo=binary is set. Using this option implicitly also sets --linkopt=-Wl,--warn-unresolved-symbols.


If enabled, each gold-invoked link of a C++ executable binary will output a symbol counts file (via the --print-symbol-counts gold option). For each linker input, the file logs the number of symbols that were defined and the number of symbols that were used in the binary. This information can be used to track unnecessary link dependencies. The symbol counts file is written to the binary's output path with the name [targetname].sc.

This option is disabled by default.

--jvmopt jvm-option

This option allows option arguments to be passed to the Java VM. It can be used with one big argument, or multiple times with individual arguments. For example:

  % bazel build --jvmopt="-server -Xms256m" java/com/example/common/foo:all

will use the server VM for launching all Java binaries and set the startup heap size for the VM to 256 MB.

--javacopt javac-option

This option allows option arguments to be passed to javac. It can be used with one big argument, or multiple times with individual arguments. For example:

  % bazel build --javacopt="-g:source,lines" //myprojects:prog

will rebuild a java_binary with the javac default debug info (instead of the bazel default).

The option is passed to javac after the Bazel built-in default options for javac and before the per-rule options. The last specification of any option to javac wins. The default options for javac are:

  -source 8 -target 8 -encoding UTF-8

Note that changing --javacopt settings will force a recompilation of all affected classes. Also note that javacopts parameters listed in specific java_library or java_binary build rules will be placed on the javac command line after these options.


This javac option enables extra correctness checks. Any problems found will be presented as errors. Either -extra_checks or -extra_checks:on may be used to force the checks to be turned on. -extra_checks:off completely disables the analysis. When this option is not specified, the default behavior is used.

--strict_java_deps (default|strict|off|warn|error)

This option controls whether javac checks for missing direct dependencies. Java targets must explicitly declare all directly used targets as dependencies. This flag instructs javac to determine the jars actually used for type checking each java file, and warn/error if they are not the output of a direct dependency of the current target.

  • off means checking is disabled.
  • warn means javac will generate standard java warnings of type [strict] for each missing direct dependency.
  • default, strict and error all mean javac will generate errors instead of warnings, causing the current target to fail to build if any missing direct dependencies are found. This is also the default behavior when the flag is unspecified.

Semantics options

These options affect the build commands and/or the output file contents.

--compilation_mode (fastbuild|opt|dbg) (-c)

This option takes an argument of fastbuild, dbg or opt, and affects various C/C++ code-generation options, such as the level of optimization and the completeness of debug tables. Bazel uses a different output directory for each different compilation mode, so you can switch between modes without needing to do a full rebuild every time.

  • fastbuild means build as fast as possible: generate minimal debugging information (-gmlt -Wl,-S), and don't optimize. This is the default. Note: -DNDEBUG will not be set.
  • dbg means build with debugging enabled (-g), so that you can use gdb (or another debugger).
  • opt means build with optimization enabled and with assert() calls disabled (-O2 -DNDEBUG). Debugging information will not be generated in opt mode unless you also pass --copt -g.

--cpu cpu

This option specifies the target CPU architecture to be used for the compilation of binaries during the build.

Note that a particular combination of crosstool version, compiler version, libc version, and target CPU is allowed only if it has been specified in the currently used CROSSTOOL file.

--host_cpu cpu

This option specifies the name of the CPU architecture that should be used to build host tools.

--fat_apk_cpu cpu[,cpu]*

The CPUs to build C/C++ libraries for in the transitive deps of android_binary rules. Other C/C++ rules are not affected. For example, if a cc_library appears in the transitive deps of an android_binary rule and a cc_binary rule, the cc_library will be built at least twice: once for each CPU specified with --fat_apk_cpu for the android_binary rule, and once for the CPU specified with --cpu for the cc_binary rule.

The default is armeabi-v7a.

One .so file will be created and packaged in the APK for each CPU specified with --fat_apk_cpu. The name of the .so file will be the name of the android_binary rule prefixed with "lib", e.g., if the name of the android_binary is "foo", then the file will be

Note that an Android-compatible crosstool must be selected. If an android_ndk_repository rule is defined in the WORKSPACE file, an Android-compatible crosstool is automatically selected. Otherwise, the crostool can be selected using the --android_crosstool_top or --crosstool_top flags.


The --experimental_skip_static_outputs option causes all statically-linked C++ binaries to not be output in any meaningful way.

If you set this flag, you must also set --distinct_host_configuration. It is also inherently incompatible with running tests — don't use it for that. This option is experimental and may go away at any time.

--per_file_copt [+-]regex[,[+-]regex]...@option[,option]...

When present, any C++ file with a label or an execution path matching one of the inclusion regex expressions and not matching any of the exclusion expressions will be built with the given options. The label matching uses the canonical form of the label (i.e //package:label_name). The execution path is the relative path to your workspace directory including the base name (including extension) of the C++ file. It also includes any platform dependent prefixes. Note, that if only one of the label or the execution path matches the options will be used.

Notes: To match the generated files (e.g. genrule outputs) Bazel can only use the execution path. In this case the regexp shouldn't start with '//' since that doesn't match any execution paths. Package names can be used like this: --per_file_copt=base/.*\.pb\.cc@-g0. This will match every file under a directory called base.

This option can be used multiple times.

The option is applied regardless of the compilation mode used. I.e. it is possible to compile with --compilation_mode=opt and selectively compile some files with stronger optimization turned on, or with optimization disabled.

Caveat: If some files are selectively compiled with debug symbols the symbols might be stripped during linking. This can be prevented by setting --strip=never.

Syntax: [+-]regex[,[+-]regex]...@option[,option]... Where regex stands for a regular expression that can be prefixed with a + to identify include patterns and with - to identify exclude patterns. option stands for an arbitrary option that is passed to the C++ compiler. If an option contains a , it has to be quoted like so \,. Options can also contain @, since only the first @ is used to separate regular expressions from options.

Example: --per_file_copt=//foo:.*\.cc,-//foo:file\.cc@-O0,-fprofile-arcs adds the -O0 and the -fprofile-arcs options to the command line of the C++ compiler for all .cc files in //foo/ except

--dynamic_mode mode

Determines whether C++ binaries will be linked dynamically, interacting with the linkstatic attribute on build rules.


  • auto: Translates to a platform-dependent mode; default for linux and off for cygwin.
  • default: Allows bazel to choose whether to link dynamically. See linkstatic for more information.
  • fully: Links all targets dynamically. This will speed up linking time, and reduce the size of the resulting binaries.
  • off: Links all targets in mostly static mode. If -static is set in linkopts, targets will change to fully static.

--fission (yes|no|[dbg][,opt][,fastbuild])

Enables Fission, which writes C++ debug information to dedicated .dwo files instead of .o files, where it would otherwise go. This substantially reduces the input size to links and can reduce link times.

When set to [dbg][,opt][,fastbuild] (example: --fission=dbg,fastbuild), Fission is enabled only for the specified set of compilation modes. This is useful for bazelrc settings. When set to yes, Fission is enabled universally. When set to no, Fission is disabled universally. Default is dbg.


If this flag is set, any -static options in linkopts of cc_* rules BUILD files are ignored. This is only intended as a workaround for C++ hardening builds.


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"). Default is disabled.

Note that dynamically linked binaries (i.e. --dynamic_mode fully) generate PIC code regardless of this flag's setting. So this flag is for cases where users want PIC code explicitly generated for static links.


Selects whether to perform resource shrinking for android_binary rules. Sets the default for the shrink_resources attribute on android_binary rules; see the documentation for that rule for further details. Defaults to off.

--custom_malloc malloc-library-target

When specified, always use the given malloc implementation, overriding all malloc="target" attributes, including in those targets that use the default (by not specifying any malloc).

--crosstool_top label

This option specifies the location of the crosstool compiler suite to be used for all C++ compilation during a build. Bazel will look in that location for a CROSSTOOL file and uses that to automatically determine settings for --compiler.

--host_crosstool_top label

If not specified, bazel uses the value of --crosstool_top to compile code in the host configuration, i.e., tools run during the build. The main purpose of this flag is to enable cross-compilation.

--apple_crosstool_top label

The crosstool to use for compiling C/C++ rules in the transitive deps of objc_*, ios__*, and apple_* rules. For those targets, this flag overwrites --crosstool_top.

--android_crosstool_top label

The crosstool to use for compiling C/C++ rules in the transitive deps of android_binary rules. This is useful if other targets in the build require a different crosstool. The default is to use the crosstool generated by the android_ndk_repository rule in the WORKSPACE file. See also --fat_apk_cpu.

--compiler version

This option specifies the C/C++ compiler version (e.g. gcc-4.1.0) to be used for the compilation of binaries during the build. If you want to build with a custom crosstool, you should use a CROSSTOOL file instead of specifying this flag.

Note that only certain combinations of crosstool version, compiler version, libc version, and target CPU are allowed.

--glibc version

This option specifies the version of glibc that the target should be linked against. If you want to build with a custom crosstool, you should use a CROSSTOOL file instead of specifying this flag. In that case, Bazel will use the CROSSTOOL file and the following options where appropriate:

Note that only certain combinations of crosstool version, compiler version, glibc version, and target CPU are allowed.

--android_sdk label

This option specifies the Android SDK/platform toolchain and Android runtime library that will be used to build any Android-related rule. The Android SDK will be automatically selected if an android_sdk_repository rule is defined in the WORKSPACE file.

--java_toolchain label

This option specifies the label of the java_toolchain used to compile Java source files.

--javabase (path|label)

This option sets the label or the path of the base Java installation to use for running JavaBuilder, SingleJar, for bazel run and bazel test, and for Java binaries built by java_binary and java_test rules. A path must be to a JDK or JRE directory that contains bin/java. The various "Make" variables for Java (JAVABASE, JAVA, JAVAC and JAR) are derived from this option.

This does not select the Java compiler that is used to compile Java source files. The compiler can be selected by settings the --java_toolchain option.

Build strategy options

These options affect how Bazel will execute the build. They should not have any significant effect on the output files generated by the build. Typically their main effect is on the speed on the build.

--spawn_strategy strategy

This option controls where and how commands are executed.

  • standalone causes commands to be executed as local subprocesses.
  • sandboxed causes commands to be executed inside a sandbox on the local machine. This requires that all input files, data dependencies and tools are listed as direct dependencies in the srcs, data and tools attributes. This is the default on systems that support sandboxed execution.

--genrule_strategy strategy

This option controls where and how genrules are executed.

  • standalone causes genrules to run as local subprocesses.
  • sandboxed causes genrules to run inside a sandbox on the local machine. This requires that all input files are listed as direct dependencies in the srcs attribute, and the program(s) executed are listed in the tools attribute. This is the default for Bazel on systems that support sandboxed execution.

--local_genrule_timeout_seconds seconds

Sets a timeout value for local genrules with the given number of seconds.

--jobs n (-j)

This option, which takes an integer argument, specifies a limit on the number of jobs that should be executed concurrently during the execution phase of the build. The default is 200.

Note that the number of concurrent jobs that Bazel will run is determined not only by the --jobs setting, but also by Bazel's scheduler, which tries to avoid running concurrent jobs that will use up more resources (RAM or CPU) than are available, based on some (very crude) estimates of the resource consumption of each job. The behavior of the scheduler can be controlled by the --ram_utilization_factor option.

--progress_report_interval n

Bazel periodically prints a progress report on jobs that are not finished yet (e.g. long running tests). This option sets the reporting frequency, progress will be printed every n seconds.

The default is 0, that means an incremental algorithm: the first report will be printed after 10 seconds, then 30 seconds and after that progress is reported once every minute.

--ram_utilization_factor percentage

This option, which takes an integer argument, specifies what percentage of the system's RAM Bazel should try to use for its subprocesses. This option affects how many processes Bazel will try to run in parallel. The default value is 67. If you run several Bazel builds in parallel, using a lower value for this option may avoid thrashing and thus improve overall throughput. Using a value higher than the default is NOT recommended. Note that Bazel's estimates are very coarse, so the actual RAM usage may be much higher or much lower than specified. Note also that this option does not affect the amount of memory that the Bazel server itself will use.

--local_resources availableRAM,availableCPU,availableIO

This option, which takes three comma-separated floating point arguments, specifies the amount of local resources that Bazel can take into consideration when scheduling build and test activities. Option expects amount of available RAM (in MB), number of CPU cores (with 1.0 representing single full core) and workstation I/O capability (with 1.0 representing average workstation). By default Bazel will estimate amount of RAM and number of CPU cores directly from system configuration and will assume 1.0 I/O resource.

If this option is used, Bazel will ignore --ram_utilization_factor.

This option, which is currently enabled by default, specifies whether the runfiles symlinks for tests and cc_binary targets should be built in the output directory. Using --nobuild_runfile_links can be useful to validate if all targets compile without incurring the overhead for building the runfiles trees. Within Bazel's output tree, the runfiles symlink tree is typically rooted as a sibling of the corresponding binary or test.

When tests (or applications) are executed, their run-time data dependencies are gathered together in one place, and may be accessed by the test using paths of the form $TEST_SRCDIR/workspace/packagename/filename. The "runfiles" tree ensures that tests have access to all the files upon which they have a declared dependence, and nothing more. By default, the runfiles tree is implemented by constructing a set of symbolic links to the required files. As the set of links grows, so does the cost of this operation, and for some large builds it can contribute significantly to overall build time, particularly because each individual test (or application) requires its own runfiles tree.

The --build_runfile_links flag controls the construction of the tree of symbolic links (for C++ applications and tests only). The reasons only C++ non-test rules are affected are numerous and subtle: C++ builds are more likely to be slower due to runfiles; no C++ host tools (tools that run during the build) need their runfiles, so this option can be used by the host configuration; and other rules (notably Python) need their runfiles for other purposes besides test execution.


When this option is enabled, Bazel will discard the analysis cache right before execution starts, thus freeing up additional memory (around 10%) for the execution phase. The drawback is that further incremental builds will be slower.

--[no]keep_going (-k)

As in GNU Make, the execution phase of a build stops when the first error is encountered. Sometimes it is useful to try to build as much as possible even in the face of errors. This option enables that behavior, and when it is specified, the build will attempt to build every target whose prerequisites were successfully built, but will ignore errors.

While this option is usually associated with the execution phase of a build, it also effects the analysis phase: if several targets are specified in a build command, but only some of them can be successfully analyzed, the build will stop with an error unless --keep_going is specified, in which case the build will proceed to the execution phase, but only for the targets that were successfully analyzed.


This option changes the way java_library targets are compiled by Bazel. Instead of using the output of a java_library for compiling dependent java_library targets, Bazel will create interface jars that contain only the signatures of non-private members (public, protected, and default (package) access methods and fields) and use the interface jars to compile the dependent targets. This makes it possible to avoid recompilation when changes are only made to method bodies or private members of a class.

Note that using --use_ijars might give you a different error message when you are accidentally referring to a non visible member of another class: Instead of getting an error that the member is not visible you will get an error that the member does not exist.

Note that changing the --use_ijars setting will force a recompilation of all affected classes.


This option enables interface shared objects, which makes binaries and other shared libraries depend on the interface of a shared object, rather than its implementation. When only the implementation changes, Bazel can avoid rebuilding targets that depend on the changed shared library unnecessarily.

Output selection options

These options determine what to build or test.


This option causes the execution phase of the build to occur; it is on by default. When it is switched off, the execution phase is skipped, and only the first two phases, loading and analysis, occur.

This option can be useful for validating BUILD files and detecting errors in the inputs, without actually building anything.


If specified, Bazel will build only what is necessary to run the *_test and test_suite rules that were not filtered due to their size, timeout, tag, or language. If specified, Bazel will ignore other targets specified on the command line. By default, this option is disabled and Bazel will build everything requested, including *_test and test_suite rules that are filtered out from testing. This is useful because running bazel test --build_tests_only foo/... may not detect all build breakages in the foo tree.


This option causes Bazel not to perform a build, but merely check whether all specified targets are up-to-date. If so, the build completes successfully, as usual. However, if any files are out of date, instead of being built, an error is reported and the build fails. This option may be useful to determine whether a build has been performed more recently than a source edit (e.g. for pre-submit checks) without incurring the cost of a build.

See also --check_tests_up_to_date.


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: for each source filename, one rule that depends on it will be built. For C++ and Java sources, rules in the same language space are preferentially chosen. For multiple rules with the same preference, the one that appears first in the BUILD file is chosen. An explicitly named target pattern which does not reference a source file results in an error.


The --save_temps option causes temporary outputs from gcc to be saved. These include .s files (assembler code), .i (preprocessed C) and .ii (preprocessed C++) files. These outputs are often useful for debugging. Temps will only be generated for the set of targets specified on the command line.

Note that our implementation of --save_temps does not use gcc's -save-temps flag. Instead, we do two passes, one with -S and one with -E. A consequence of this is that if your build fails, Bazel may not yet have produced the ".i" or ".ii" and ".s" files. If you're trying to use --save_temps to debug a failed compilation, you may need to also use --keep_going so that Bazel will still try to produce the preprocessed files after the compilation fails.

The --save_temps flag currently works only for cc_* rules.

To ensure that Bazel prints the location of the additional output files, check that your --show_result n setting is high enough.

--build_tag_filters tag[,tag]*

If specified, Bazel will build only targets that have at least one required tag (if any of them are specified) and does not have any excluded tags. Build tag filter is specified as comma delimited list of tag keywords, optionally preceded with '-' sign used to denote excluded tags. Required tags may also have a preceding '+' sign.

--test_size_filters size[,size]*

If specified, Bazel will test (or build if --build_tests_only is also specified) only test targets with the given size. Test size filter is specified as comma delimited list of allowed test size values (small, medium, large or enormous), optionally preceded with '-' sign used to denote excluded test sizes. For example,

  % bazel test --test_size_filters=small,medium //foo:all
  % bazel test --test_size_filters=-large,-enormous //foo:all

will test only small and medium tests inside //foo.

By default, test size filtering is not applied.

--test_timeout_filters timeout[,timeout]*

If specified, Bazel will test (or build if --build_tests_only is also specified) only test targets with the given timeout. Test timeout filter is specified as comma delimited list of allowed test timeout values (short, moderate, long or eternal), optionally preceded with '-' sign used to denote excluded test timeouts. See --test_size_filters for example syntax.

By default, test timeout filtering is not applied.

--test_tag_filters tag[,tag]*

If specified, Bazel will test (or build if --build_tests_only is also specified) only test targets that have at least one required tag (if any of them are specified) and does not have any excluded tags. Test tag filter is specified as comma delimited list of tag keywords, optionally preceded with '-' sign used to denote excluded tags. Required tags may also have a preceding '+' sign.

For example,

  % bazel test --test_tag_filters=performance,stress,-flaky //myproject:all

will test targets that are tagged with either performance or stress tag but are not tagged with the flaky tag.

By default, test tag filtering is not applied. Note that you can also filter on test's size and local tags in this manner.

--test_lang_filters lang[,lang]*

Specifies a comma-separated list of test languages for languages with an official *_test rule the (see build encyclopedia for a full list of these). Each language can be optionally preceded with '-' to specify excluded languages. The name used for each language should be the same as the language prefix in the *_test rule, for example, cc, java or sh.

If specified, Bazel will test (or build if --build_tests_only is also specified) only test targets of the specified language(s).

For example,

  % bazel test --test_lang_filters=cc,java foo/...

will test only the C/C++ and Java tests (defined using cc_test and java_test rules, respectively) in foo/..., while

  % bazel test --test_lang_filters=-sh,-java foo/...

will run all of the tests in foo/... except for the sh_test and java_test tests.

By default, test language filtering is not applied.


Specifies a filter that the test runner may use to pick a subset of tests for running. All targets specified in the invocation are built, but depending on the expression only some of them may be executed; in some cases, only certain test methods are run.

The particular interpretation of filter-expression is up to the test framework responsible for running the test. It may be a glob, substring, or regexp. --test_filter is a convenience over passing different --test_arg filter arguments, but not all frameworks support it.

Verbosity options: options that control what Bazel prints

These options control the verbosity of Bazel's output, either to the terminal, or to additional log files.

--explain logfile

This option, which requires a filename argument, causes the dependency checker in bazel build's execution phase to explain, for each build step, either why it is being executed, or that it is up-to-date. The explanation is written to logfile.

If you are encountering unexpected rebuilds, this option can help to understand the reason. Add it to your .bazelrc so that logging occurs for all subsequent builds, and then inspect the log when you see an execution step executed unexpectedly. This option may carry a small performance penalty, so you might want to remove it when it is no longer needed.


This option increases the verbosity of the explanations generated when the --explain option is enabled.

In particular, if verbose explanations are enabled, and an output file is rebuilt because the command used to build it has changed, then the output in the explanation file will include the full details of the new command (at least for most commands).

Using this option may significantly increase the length of the generated explanation file and the performance penalty of using --explain.

If --explain is not enabled, then --verbose_explanations has no effect.

--profile file

This option, which takes a filename argument, causes Bazel to write profiling data into a file. The data then can be analyzed or parsed using the bazel analyze-profile command. The Build profile can be useful in understanding where Bazel's build command is spending its time.


This option causes Bazel to output package-loading progress messages. If it is disabled, the messages won't be shown.


This option causes progress messages to be displayed; it is on by default. When disabled, progress messages are suppressed.

--show_progress_rate_limit n

This option causes bazel to display only one progress message per n seconds, where n is a real number. If n is -1, all progress messages will be displayed. The default value for this option is 0.03, meaning bazel will limit the progress messages to one per every 0.03 seconds.

--show_result n

This option controls the printing of result information at the end of a bazel build command. By default, if a single build target was specified, Bazel prints a message stating whether or not the target was successfully brought up-to-date, and if so, the list of output files that the target created. If multiple targets were specified, result information is not displayed.

While the result information may be useful for builds of a single target or a few targets, for large builds (e.g. an entire top-level project tree), this information can be overwhelming and distracting; this option allows it to be controlled. --show_result takes an integer argument, which is the maximum number of targets for which full result information should be printed. By default, the value is 1. Above this threshold, no result information is shown for individual targets. Thus zero causes the result information to be suppressed always, and a very large value causes the result to be printed always.

Users may wish to choose a value in-between if they regularly alternate between building a small group of targets (for example, during the compile-edit-test cycle) and a large group of targets (for example, when establishing a new workspace or running regression tests). In the former case, the result information is very useful whereas in the latter case it is less so. As with all options, this can be specified implicitly via the .bazelrc file.

The files are printed so as to make it easy to copy and paste the filename to the shell, to run built executables. The "up-to-date" or "failed" messages for each target can be easily parsed by scripts which drive a build.

--subcommands (-s)

This option causes Bazel's execution phase to print the full command line for each command prior to executing it.

  >>>>> # //examples/cpp:hello-world [action 'Linking examples/cpp/hello-world']
  (cd /home/jrluser/.cache/bazel/_bazel_jrluser/4c084335afceb392cfbe7c31afee3a9f/bazel && \
    exec env - \
    /usr/bin/gcc -o bazel-out/local_linux-fastbuild/bin/examples/cpp/hello-world -B/usr/bin/ -Wl,-z,relro,-z,now -no-canonical-prefixes -pass-exit-codes -Wl,-S -Wl,@bazel-out/local_linux-fastbuild/bin/examples/cpp/hello-world-2.params)

Where possible, commands are printed in a Bourne shell compatible syntax, so that they can be easily copied and pasted to a shell command prompt. (The surrounding parentheses are provided to protect your shell from the cd and exec calls; be sure to copy them!) However some commands are implemented internally within Bazel, such as creating symlink trees. For these there's no command line to display.

See also --verbose_failures, below.


This option causes Bazel's execution phase to print the full command line for commands that failed. This can be invaluable for debugging a failing build.

Failing commands are printed in a Bourne shell compatible syntax, suitable for copying and pasting to a shell prompt.


This option controls whether stamping is enabled for rule types that support it. For most of the supported rule types stamping is enabled by default (e.g. cc_binary). By default, stamping is disabled for all tests. Specifying --stamp does not force affected targets to be rebuilt, if their dependencies have not changed.

Stamping can be enabled or disabled explicitly in BUILD using the stamp attribute of certain rule types, please refer to the build encyclopedia for details. For rules that are neither explicitly or implicitly configured as stamp = 0 or stamp = 1, the --[no]stamp option selects whether stamping is enabled. Bazel never stamps binaries that are built for the host configuration, regardless of the stamp attribute.

Miscellaneous options

Changes the prefix of the generated convenience symlinks. The default value for the symlink prefix is bazel- which will create the symlinks bazel-bin, bazel-testlogs, and bazel-genfiles.

If the symbolic links cannot be created for any reason, a warning is issued but the build is still considered a success. In particular, this allows you to build in a read-only directory or one that you have no permission to write into. Any paths printed in informational messages at the conclusion of a build will only use the symlink-relative short form if the symlinks point to the expected location; in other words, you can rely on the correctness of those paths, even if you cannot rely on the symlinks being created.

Some common values of this option:

  • Suppress symlink creation: --symlink_prefix=/ will cause Bazel to not create or update any symlinks, including the bazel-out and bazel-<workspace> symlinks. Use this option to suppress symlink creation entirely.

  • Reduce clutter: --symlink_prefix=.bazel/ will cause Bazel to create symlinks called bin (etc) inside a hidden directory .bazel.

--platform_suffix string

Adds a suffix to the configuration short name, which is used to determine the output directory. Setting this option to different values puts the files into different directories, for example to improve cache hit rates for builds that otherwise clobber each others output files, or to keep the output files around for comparisons.


Temporary flag for testing bazel default visibility changes. Not intended for general use but documented for completeness' sake.

Using Bazel for releases

Bazel is used both by software engineers during the development cycle, and by release engineers when preparing binaries for deployment to production. This section provides a list of tips for release engineers using Bazel.

Significant options

When using Bazel for release builds, the same issues arise as for other scripts that perform a build, so you should read the scripting section of this manual. In particular, the following options are strongly recommended:

These options (q.v.) are also important:

  • --package_path
  • --symlink_prefix: for managing builds for multiple configurations, it may be convenient to distinguish each build with a distinct identifier, e.g. "64bit" vs. "32bit". This option differentiates the bazel-bin (etc.) symlinks.

Running tests with Bazel

To build and run tests with bazel, type bazel test followed by the name of the test targets.

By default, this command performs simultaneous build and test activity, building all specified targets (including any non-test targets specified on the command line) and testing *_test and test_suite targets as soon as their prerequisites are built, meaning that test execution is interleaved with building. Doing so usually results in significant speed gains.

Options for bazel test

--cache_test_results=(yes|no|auto) (-t)

If this option is set to 'auto' (the default) then Bazel will only rerun a test if any of the following conditions applies:

  • Bazel detects changes in the test or its dependencies
  • the test is marked as external
  • multiple test runs were requested with --runs_per_test
  • the test failed.

If 'no', all tests will be executed unconditionally.

If 'yes', the caching behavior will be the same as auto except that it may cache test failures and test runs with --runs_per_test.

Note that test results are always saved in Bazel's output tree, regardless of whether this option is enabled, so you needn't have used --cache_test_results on the prior run(s) of bazel test in order to get cache hits. The option only affects whether Bazel will use previously saved results, not whether it will save results of the current run.

Users who have enabled this option by default in their .bazelrc file may find the abbreviations -t (on) or -t- (off) convenient for overriding the default on a particular run.


This option tells Bazel not to run the tests, but to merely check and report the cached test results. If there are any tests which have not been previously built and run, or whose tests results are out-of-date (e.g. because the source code or the build options have changed), then Bazel will report an error message ("test result is not up-to-date"), will record the test's status as "NO STATUS" (in red, if color output is enabled), and will return a non-zero exit code.

This option also implies --check_up_to_date behavior.

This option may be useful for pre-submit checks.


This option tells Bazel to explicitly warn the user if a test's timeout is significantly longer then the test's actual execution time. While a test's timeout should be set such that it is not flaky, a test that has a highly over-generous timeout can hide real problems that crop up unexpectedly.

For instance, a test that normally executes in a minute or two should not have a timeout of ETERNAL or LONG as these are much, much too generous. This option is useful to help users decide on a good timeout value or sanity check existing timeout values.

Note that each test shard is allotted the timeout of the entire XX_test target. Using this option does not affect a test's timeout value, merely warns if Bazel thinks the timeout could be restricted further.


By default, all tests are run to completion. If this flag is disabled, however, the build is aborted on any non-passing test. Subsequent build steps and test invocations are not run, and in-flight invocations are canceled. Do not specify both --notest_keep_going and --keep_going.

--flaky_test_attempts attempts

This option specifies the maximum number of times a test should be attempted if it fails for any reason. A test that initially fails but eventually succeeds is reported as FLAKY on the test summary. It is, however, considered to be passed when it comes to identifying Bazel exit code or total number of passed tests. Tests that fail all allowed attempts are considered to be failed.

By default (when this option is not specified, or when it is set to "default"), only a single attempt is allowed for regular tests, and 3 for test rules with the flaky attribute set. You can specify an integer value to override the maximum limit of test attempts. Bazel allows a maximum of 10 test attempts in order to prevent abuse of the system.

--runs_per_test [regex@]number

This option specifies the number of times each test should be executed. All test executions are treated as separate tests (e.g. fallback functionality will apply to each of them independently).

The status of a target with failing runs depends on the value of the --runs_per_test_detects_flakes flag:

  • If absent, any failing run causes the entire test to fail.
  • If present and two runs from the same shard return PASS and FAIL, the test will receive a status of flaky (unless other failing runs cause it to fail).

If a single number is specified, all tests will run that many times. Alternatively, a regular expression may be specified using the syntax regex@number. This constrains the effect of --runs_per_test to targets which match the regex (e.g. "--runs_per_test=^//pizza:.*@4" runs all tests under //pizza/ 4 times). This form of --runs_per_test may be specified more than once.


If this option is specified (by default it is not), Bazel will detect flaky test shards through --runs_per_test. If one or more runs for a single shard fail and one or more runs for the same shard pass, the target will be considered flaky with the flag. If unspecified, the target will report a failing status.

--test_summary output_style

Specifies how the test result summary should be displayed.

  • short prints the results of each test along with the name of the file containing the test output if the test failed. This is the default value.
  • terse like short, but even shorter: only print information about tests which did not pass.
  • detailed prints each individual test case that failed, not only each test. The names of test output files are omitted.
  • none does not print test summary.

--test_output output_style

Specifies how test output should be displayed:

  • summary shows a summary of whether each test passed or failed. Also shows the output log file name for failed tests. The summary will be printed at the end of the build (during the build, one would see just simple progress messages when tests start, pass or fail). This is the default behavior.
  • errors sends combined stdout/stderr output from failed tests only into the stdout immediately after test is completed, ensuring that test output from simultaneous tests is not interleaved with each other. Prints a summary at the build as per summary output above.
  • all is similar to errors but prints output for all tests, including those which passed.
  • streamed streams stdout/stderr output from each test in real-time.


This option causes the Java virtual machine of a java test to wait for a connection from a JDWP-compliant debugger before starting the test. This option implies --test_output=streamed.


By default this option is enabled, causing test times and other additional information (such as test attempts) to be printed to the test summary. If --noverbose_test_summary is specified, test summary will include only test name, test status and cached test indicator and will be formatted to stay within 80 characters when possible.

--test_tmpdir path

Specifies temporary directory for tests executed locally. Each test will be executed in a separate subdirectory inside this directory. The directory will be cleaned at the beginning of the each bazel test command. By default, bazel will place this directory under Bazel output base directory. Note that this is a directory for running tests, not storing test results (those are always stored under the bazel-out directory).

--test_timeout seconds OR --test_timeout seconds,seconds,seconds,seconds

Overrides the timeout value for all tests by using specified number of seconds as a new timeout value. If only one value is provided, then it will be used for all test timeout categories.

Alternatively, four comma-separated values may be provided, specifying individual timeouts for short, moderate, long and eternal tests (in that order). In either form, zero or a negative value for any of the test sizes will be substituted by the default timeout for the given timeout categories as defined by the page Writing Tests. By default, Bazel will use these timeouts for all tests by inferring the timeout limit from the test's size whether the size is implicitly or explicitly set.

Tests which explicitly state their timeout category as distinct from their size will receive the same value as if that timeout had been implicitly set by the size tag. So a test of size 'small' which declares a 'long' timeout will have the same effective timeout that a 'large' tests has with no explicit timeout.

--test_arg arg

Passes command-line options/flags/arguments to each test process. This option can be used multiple times to pass several arguments, e.g. --test_arg=--logtostderr --test_arg=--v=3.

--test_env variable=value OR --test_env variable

Specifies additional variables that must be injected into the test environment for each test. If value is not specified it will be inherited from the shell environment used to start the bazel test command.

The environment can be accessed from within a test by using System.getenv("var") (Java), getenv("var") (C or C++),


This specifies a prefix that the test runner will insert in front of the test command before running it. The command-prefix is split into words using Bourne shell tokenization rules, and then the list of words is prepended to the command that will be executed.

If the first word is a fully qualified label (i.e. starts with //) it is built. Then the label is substituted by the corresponding executable location that is prepended to the command that will be executed along with the other words.

Some caveats apply:

  • The PATH used for running tests may be different than the PATH in your environment, so you may need to use an absolute path for the --run_under command (the first word in command-prefix).
  • stdin is not connected, so --run_under can't be used for interactive commands.


        --run_under='/usr/bin/strace -c'
        --run_under='/usr/bin/valgrind --quiet --num-callers=20'

Test selection

As documented under Output selection options, you can filter tests by size, timeout, tag, or language. A convenience general name filter can forward particular filter args to the test runner.

Other options for bazel test

The syntax and the remaining options are exactly like bazel build.

Running executables with Bazel

The bazel run command is similar to bazel build, except it is used to build and run a single target. Here is a typical session:

  % bazel run -- java/myapp:myapp --arg1 --arg2
  Welcome to Bazel
  INFO: Loading package: java/myapp
  INFO: Loading package: foo/bar
  INFO: Loading complete.  Analyzing...
  INFO: Found 1 target...
  Target //java/myapp:myapp up-to-date:
  INFO: Elapsed time: 0.638s, Critical Path: 0.34s

  INFO: Running command line: bazel-bin/java/myapp:myapp --arg1 --arg2
  Hello there

Bazel closes stdin, so you can't use bazel run if you want to start an interactive program or pipe data to it.

Note the use of the --. This is needed so that Bazel does not interpret --arg1 and --arg2 as Bazel options, but rather as part of the command line for running the binary. (The program being run simply says hello and prints out its args.)

Options for bazel run


This has the same effect as the --run_under option for bazel test (see above), except that it applies to the command being run by bazel run rather than to the tests being run by bazel test and cannot run under label.

Executing tests

bazel run can also execute test binaries, which has the effect of running the test, but without the setup documented on the page Writing Tests, so that the test runs in an environment closer to the current shell environment. Note that none of the --test_* arguments have an effect when running a test in this manner.

Querying the dependency graph with Bazel

Bazel includes a query language for asking questions about the dependency graph used during the build. The query tool is an invaluable aid to many software engineering tasks.

The query language is based on the idea of algebraic operations over graphs; it is documented in detail in Bazel Query Reference. Please refer to that document for reference, for examples, and for query-specific command-line options.

The query tool accepts several command-line option. --output selects the output format. --[no]keep_going (disabled by default) causes the query tool to continue to make progress upon errors; this behavior may be disabled if an incomplete result is not acceptable in case of errors.

The --[no]host_deps option, enabled by default, causes dependencies on "host configuration" targets to be included in the dependency graph over which the query operates.

The --[no]implicit_deps option, enabled by default, causes implicit dependencies to be included in the dependency graph over which the query operates. An implicit dependency is one that is not explicitly specified in the BUILD file but added by bazel.

Example: "Show the locations of the definitions (in BUILD files) of all genrules required to build all the tests in the PEBL tree."

  bazel query --output location 'kind(genrule, deps(kind(".*_test rule", foo/bar/pebl/...)))'

Miscellaneous Bazel commands and options

The help command

The help command provides on-line help. By default, it shows a summary of available commands and help topics, as shown in the Bazel overview section above. Specifying an argument displays detailed help for a particular topic. Most topics are Bazel commands, e.g. build or query, but there are some additional help topics that do not correspond to commands.

--[no]long (-l)

By default, bazel help [topic] prints only a summary of the relevant options for a topic. If the --long option is specified, the type, default value and full description of each option is also printed.

The shutdown command

Bazel server processes (see Client/server implementation) may be stopped by using the shutdown command. This command causes the Bazel server to exit as soon as it becomes idle (i.e. after the completion of any builds or other commands that are currently in progress). Bazel servers stop themselves after an idle timeout, so this command is rarely necessary; however, it can be useful in scripts when it is known that no further builds will occur in a given workspace.

shutdown accepts one option, --iff_heap_size_greater_than n, which requires an integer argument (in MB). If specified, this makes the shutdown conditional on the amount of memory already consumed. This is useful for scripts that initiate a lot of builds, as any memory leaks in the Bazel server could cause it to crash spuriously on occasion; performing a conditional restart preempts this condition.

The info command

The info command prints various values associated with the Bazel server instance, or with a specific build configuration. (These may be used by scripts that drive a build.)

The info command also permits a single (optional) argument, which is the name of one of the keys in the list below. In this case, bazel info key will print only the value for that one key. (This is especially convenient when scripting Bazel, as it avoids the need to pipe the result through sed -ne /key:/s/key://p:

Configuration-independent data

  • release: the release label for this Bazel instance, or "development version" if this is not a released binary.
  • workspace the absolute path to the base workspace directory.
  • install_base: the absolute path to the installation directory used by this Bazel instance for the current user. Bazel installs its internally required executables below this directory.
  • output_base: the absolute path to the base output directory used by this Bazel instance for the current user and workspace combination. Bazel puts all of its scratch and build output below this directory.
  • execution_root: the absolute path to the execution root directory under output_base. This directory is the root for all files accessible to commands executed during the build, and is the working directory for those commands. If the workspace directory is writable, a symlink named bazel-<workspace> is placed there pointing to this directory.
  • output_path: the absolute path to the output directory beneath the execution root used for all files actually generated as a result of build commands. If the workspace directory is writable, a symlink named bazel-out is placed there pointing to this directory.
  • server_pid: the process ID of the Bazel server process.
  • command_log: the absolute path to the command log file; this contains the interleaved stdout and stderr streams of the most recent Bazel command. Note that running bazel info will overwrite the contents of this file, since it then becomes the most recent Bazel command. However, the location of the command log file will not change unless you change the setting of the --output_base or --output_user_root options.
  • used-heap-size, committed-size, max-heap-size: reports various JVM heap size parameters. Respectively: memory currently used, memory currently guaranteed to be available to the JVM from the system, maximum possible allocation.
  • gc-count, gc-time: The cumulative count of garbage collections since the start of this Bazel server and the time spent to perform them. Note that these values are not reset at the start of every build.
  • package_path: A colon-separated list of paths which would be searched for packages by bazel. Has the same format as the --package_path build command line argument.

Example: the process ID of the Bazel server.

% bazel info server_pid

Configuration-specific data

These data may be affected by the configuration options passed to bazel info, for example --cpu, --compilation_mode, etc. The info command accepts all the options that control dependency analysis, since some of these determine the location of the output directory of a build, the choice of compiler, etc.

  • bazel-bin, bazel-testlogs, bazel-genfiles: reports the absolute path to the bazel-* directories in which programs generated by the build are located. This is usually, though not always, the same as the bazel-* symlinks created in the base workspace directory after a successful build. However, if the workspace directory is read-only, no bazel-* symlinks can be created. Scripts that use the value reported by bazel info, instead of assuming the existence of the symlink, will be more robust.
  • The complete "Make" environment. If the --show_make_env flag is specified, all variables in the current configuration's "Make" environment are also displayed (e.g. CC, GLIBC_VERSION, etc). These are the variables accessed using the $(CC) or varref("CC") syntax inside BUILD files.

Example: the C++ compiler for the current configuration. This is the $(CC) variable in the "Make" environment, so the --show_make_env flag is needed.

  % bazel info --show_make_env -c opt COMPILATION_MODE

Example: the bazel-bin output directory for the current configuration. This is guaranteed to be correct even in cases where the bazel-bin symlink cannot be created for some reason (e.g. you are building from a read-only directory).

The version command

The version command prints version details about the built Bazel binary, including the changelist at which it was built and the date. These are particularly useful in determining if you have the latest Bazel, or if you are reporting bugs. Some of the interesting values are:

  • changelist: the changelist at which this version of Bazel was released.
  • label: the release label for this Bazel instance, or "development version" if this is not a released binary. Very useful when reporting bugs.

The mobile-install command

The mobile-install command installs apps to mobile devices. Currently only Android devices running ART are supported. See bazel mobile-install for more information.

Note that this command does not install the same thing that bazel build produces: Bazel tweaks the app so that it can be built, installed and re-installed quickly. This should, however, be mostly transparent to the app.

The following options are supported:


If set, Bazel tries to install the app incrementally, that is, only those parts that have changed since the last build. This cannot update resources referenced from AndroidManifest.xml, native code or Java resources (i.e. ones referenced by Class.getResource()). If these things change, this option must be omitted. Contrary to the spirit of Bazel and due to limitations of the Android platform, it is the responsibility of the user to know when this command is good enough and when a full install is needed. If you are using a device with Marshmallow or later, consider the --split_apks flag.


Whether to use split apks to install and update the application on the device. Works only with devices with Marshmallow or later. Note that the --incremental flag is not necessary when using --split_apks.


Starts the app in a clean state after installing. Equivalent to --start=COLD.


Waits for debugger to be attached before starting the app in a clean state after installing. Equivalent to --start=DEBUG.


How the app should be started after installing it. Supported start_types are:

  • NO Does not start the app. This is the default.
  • COLD Starts the app from a clean state after install.
  • WARM Preserves and restores the application state on incremental installs.
  • DEBUG Waits for the debugger before starting the app in a clean state after install.
Note that if more than one of --start=start_type, --start_app or --debug_app is set, the last value will be used.

--adb path

Indicates the adb binary to be used. The default is to use the adb in the Android SDK specified by --android_sdk.

--adb_arg arg

Extra arguments to adb. These come before the subcommand in the command line and are typically used to specify which device to install to. For example, to select the Android device or emulator to use:

% bazel mobile-install --adb_arg=-s --adb_arg=deadbeef
will invoke adb as
adb -s deadbeef install ...

--adb_jobs number

The number of instances of adb to use in parallel to update files on the device.

--incremental_install_verbosity number

The verbosity for incremental install. Set to 1 for debug logging to be printed to the console.

The analyze-profile command

The analyze-profile command analyzes data previously gathered during the build using --profile option. It provides several options to either perform analysis of the build execution or export data in the specified format.

The following options are supported:

  • --dump=text displays all gathered data in a human-readable format
  • --dump=raw displays all gathered data in a script-friendly format
  • --html generates an HTML file visualizing the actions and rules executed in the build, as well as summary statistics for the build
    • --html_details adds more fine-grained information on actions and rules to the HTML visualization
      • --html_histograms adds histograms for Skylark functions clicked in the statistics table. This will increase file size massively
      • --nochart hides the task chart from generated HTML
  • --combine combines multiple profile data files into a single report. Does not generate HTML task charts
  • --task_tree prints the tree of tasks matching the given regular expression
    • --task_tree_threshold skip tasks with duration less than threshhold, in milliseconds. Default is 50ms

See the section on Troubleshooting performance by profiling for format details and usage help.

The canonicalize-flags command

The canonicalize-flags command, which takes a list of options for a Bazel command and returns a list of options that has the same effect. The new list of options is canonical, i.e., two lists of options with the same effect are canonicalized to the same new list.

The --for_command option can be used to select between different commands. At this time, only build and test are supported. Options that the given command does not support cause an error.

Note that a small number of options cannot be reordered, because Bazel cannot ensure that the effect is identical.

Bazel startup options

The options described in this section affect the startup of the Java virtual machine used by Bazel server process, and they apply to all subsequent commands handled by that server. If there is an already running Bazel server and the startup options do not match, it will be restarted.

All of the options described in this section must be specified using the --key=value or --key value syntax. Also, these options must appear before the name of the Bazel command.


This option requires a path argument, which must specify a writable directory. Bazel will use this location to write all its output. The output base is also the key by which the client locates the Bazel server. By changing the output base, you change the server which will handle the command.

By default, the output base is derived from the user's login name, and the name of the workspace directory (actually, its MD5 digest), so a typical value looks like: /var/tmp/google/_bazel_jrluser/d41d8cd98f00b204e9800998ecf8427e. Note that the client uses the output base to find the Bazel server instance, so if you specify a different output base in a Bazel command, a different server will be found (or started) to handle the request. It's possible to perform two concurrent builds in the same workspace directory by varying the output base.

For example:

  % bazel --output_base /tmp/1 build //foo  &  bazel --output_base /tmp/2 build //bar

In this command, the two Bazel commands run concurrently (because of the shell & operator), each using a different Bazel server instance (because of the different output bases). In contrast, if the default output base was used in both commands, then both requests would be sent to the same server, which would handle them sequentially: building //foo first, followed by an incremental build of //bar.

We recommend you do not use NFS locations for the output base, as the higher access latency of NFS will cause noticeably slower builds.


By default, the output_base value is chosen to as to avoid conflicts between multiple users building in the same workspace directory. In some situations, though, it is desirable to build from a directory shared between multiple users; release engineers often do this. In those cases it may be useful to deliberately override the default so as to ensure "conflicts" (i.e., sharing) between multiple users. Use the --output_user_root option to achieve this: the output base is placed in a subdirectory of the output user root, with a unique name based on the workspace, so the result of using an output user root that is not a function of $USER is sharing. Of course, it is important to ensure (via umask and group membership) that all the cooperating users can read/write each others files.

If the --output_base option is specified, it overrides using --output_user_root to calculate the output base.

The install base location is also calculated based on --output_user_root, plus the MD5 identity of the Bazel embedded binaries.

You can also use the --output_user_root option to choose an alternate base location for all of Bazel's output (install base and output base) if there is a better location in your filesystem layout.


Specifies a startup option to be passed to the Java virtual machine in which Bazel itself runs. This can be used to set the stack size, for example:

  % bazel --host_jvm_args="-Xss256K" build //foo

This option can be used multiple times with individual arguments. Note that setting this flag should rarely be needed. You can also pass a space-separated list of strings, each of which will be interpreted as a separate JVM argument, but this feature will soon be deprecated.

That this does not affect any JVMs used by subprocesses of Bazel: applications, tests, tools, etc. To pass JVM options to executable Java programs, whether run by bazel run or on the command-line, you should use the --jvm_flags argument which all java_binary and java_test programs support. Alternatively for tests, use bazel test --test_arg=--jvm_flags=foo ....


This option causes the Java virtual machine to wait for a connection from a JDWP-compliant debugger before calling the main method of Bazel itself. This is primarily intended for use by Bazel developers.

(Please note that this does not affect any JVMs used by subprocesses of Bazel: applications, tests, tools, etc.)


This switch will cause bazel to be run in batch mode, instead of the standard client/server mode described above. Doing so provides more predictable semantics with respect to signal handling, job control, and environment variable inheritance, and is necessary for running bazel in a chroot jail.

Batch mode retains proper queueing semantics within the same output_base. That is, simultaneous invocations will be processed in order, without overlap. If a batch mode bazel is run on a client with a running server, it first kills the server before processing the command.

Bazel will run slower in batch mode, compared to client/server mode. Among other things, the build file cache is memory-resident, so it is not preserved between sequential batch invocations. Therefore, using batch mode often makes more sense in cases where performance is less critical, such as continuous builds.

--max_idle_secs n

This option specifies how long, in seconds, the Bazel server process should wait after the last client request, before it exits. The default value is 10800 (3 hours).

This option may be used by scripts that invoke Bazel to ensure that they do not leave Bazel server processes on a user's machine when they would not be running otherwise. For example, a presubmit script might wish to invoke bazel query to ensure that a user's pending change does not introduce unwanted dependencies. However, if the user has not done a recent build in that workspace, it would be undesirable for the presubmit script to start a Bazel server just for it to remain idle for the rest of the day. By specifying a small value of --max_idle_secs in the query request, the script can ensure that if it caused a new server to start, that server will exit promptly, but if instead there was already a server running, that server will continue to run until it has been idle for the usual time. Of course, the existing server's idle timer will be reset.


If enabled, Bazel will wait for other Bazel commands holding the server lock to complete before progressing. If disabled, Bazel will exit in error if it cannot immediately acquire the lock and proceed. Developers might use this in presubmit checks to avoid long waits caused by another Bazel command in the same client.

--io_nice_level n

Sets a level from 0-7 for best-effort IO scheduling. 0 is highest priority, 7 is lowest. The anticipatory scheduler may only honor up to priority 4. Negative values are ignored.


Use batch CPU scheduling for Bazel. This policy is useful for workloads that are non-interactive, but do not want to lower their nice value. See 'man 2 sched_setscheduler'. This policy may provide for better system interactivity at the expense of Bazel throughput.

Miscellaneous options


Controls whether Bazel announces command options read from the bazelrc file when starting up. (Startup options are unconditionally announced.)

--color (yes|no|auto)

This option determines whether Bazel will use colors to highlight its output on the screen.

If this option is set to yes, color output is enabled. If this option is set to auto, Bazel will use color output only if the output is being sent to a terminal and the TERM environment variable is set to a value other than dumb, emacs, or xterm-mono. If this option is set to no, color output is disabled, regardless of whether the output is going to a terminal and regardless of the setting of the TERM environment variable.

--config name

Selects additional config section from the rc files; for the current command, it also pulls in the options from command:name if such a section exists. Can be specified multiple times to add flags from several config sections. Expansions can refer to other definitions (i.e. expansions can be chained).

--curses (yes|no|auto)

This option determines whether Bazel will use cursor controls in its screen output. This results in less scrolling data, and a more compact, easy-to-read stream of output from Bazel. This works well with --color.

If this option is set to yes, use of cursor controls is enabled. If this option is set to no, use of cursor controls is disabled. If this option is set to auto, use of cursor controls will be enabled under the same conditions as for --color=auto.


If specified, a timestamp is added to each message generated by Bazel specifying the time at which the message was displayed.

Calling Bazel from scripts

Bazel can be called from scripts in order to perform a build, run tests or query the dependency graph. Bazel has been designed to enable effective scripting, but this section lists some details to bear in mind to make your scripts more robust.

Choosing the output base

The --output_base option controls where the Bazel process should write the outputs of a build to, as well as various working files used internally by Bazel, one of which is a lock that guards against concurrent mutation of the output base by multiple Bazel processes.

Choosing the correct output base directory for your script depends on several factors. If you need to put the build outputs in a specific location, this will dictate the output base you need to use. If you are making a "read only" call to Bazel (e.g. bazel query), the locking factors will be more important. In particular, if you need to run multiple instances of your script concurrently, you will need to give each one a different (or random) output base.

If you use the default output base value, you will be contending for the same lock used by the user's interactive Bazel commands. If the user issues long-running commands such as builds, your script will have to wait for those commands to complete before it can continue.

Server or no server?

By default, Bazel uses a long-running server process as an optimization; this behavior can be disabled using the --batch option. There's no hard and fast rule about whether or not your script should use a server, but in general, the trade-off is between performance and reliability. The server mode makes a sequence of builds, especially incremental builds, faster, but its behavior is more complex and prone to failure. We recommend in most cases that you use batch mode unless the performance advantage is critical.

If you do use the server, don't forget to call shutdown when you're finished with it, or, specify --max_idle_secs=5 so that idle servers shut themselves down promptly.

What exit code will I get?

Bazel attempts to differentiate failures due to the source code under consideration from external errors that prevent Bazel from executing properly. Bazel execution can result in following exit codes:

Exit Codes common to all commands:
  • 0 - Success
  • 2 - Command Line Problem, Bad or Illegal flags or command combination, or Bad Environment Variables. Your command line must be modified.
  • 8 - Build Interrupted but we terminated with an orderly shutdown.
  • 32 - External Environment Failure not on this machine.
  • 33 - OOM failure. You need to modify your command line.
  • 34 - Reserved for Google-internal use.
  • 35 - Reserved for Google-internal use.
  • 36 - Local Environmental Issue, suspected permanent.
  • 37 - Unhandled Exception / Internal Bazel Error.
  • 38 - Reserved for Google-internal use.
  • 40-44 - Reserved for errors in Bazel's command line launcher, that are not command line related. Typically these are related to bazel server being unable to launch itself.
Return codes for commands bazel build, bazel test.
  • 1 - Build failed.
  • 3 - Build OK, but some tests failed or timed out.
  • 4 - Build successful but no tests were found even though testing was requested.
For bazel run:
  • 1 - Build failed.
  • 6 - Run command failure. The executed subprocess returned a non-zero exit code. The actual subprocess exit code is given in stderr.
For bazel query:
  • 3 - Partial success, but the query encountered 1 or more errors in the input BUILD file set and therefore the results of the operation are not 100% reliable. This is likely due to a --keep_going option on the command line.
  • 7 - Command failure.

Future Bazel versions may add additional exit codes, replacing generic failure exit code 1 with a different non-zero value with a particular meaning. However, all non-zero exit values will always constitute an error.

Reading the .bazelrc file

By default, Bazel will read the .bazelrc file from the base workspace directory or the user's home directory. Whether or not this is desirable is a choice for your script; if your script needs to be perfectly hermetic (e.g. when doing release builds), you should disable reading the .bazelrc file by using the option --bazelrc=/dev/null. If you want to perform a build using the user's preferred settings, the default behavior is better.

Command log

The Bazel output is also available in a command log file which you can find with the following command:

% bazel info command_log

The command log file contains the interleaved stdout and stderr streams of the most recent Bazel command. Note that running bazel info will overwrite the contents of this file, since it then becomes the most recent Bazel command. However, the location of the command log file will not change unless you change the setting of the --output_base or --output_user_root options.

Parsing output

The Bazel output is quite easy to parse for many purposes. Two options that may be helpful for your script are --noshow_progress which suppresses progress messages, and --show_result n, which controls whether or not "build up-to-date" messages are printed; these messages may be parsed to discover which targets were successfully built, and the location of the output files they created. Be sure to specify a very large value of n if you rely on these messages.

Troubleshooting performance by profiling

The first step in analyzing the performance of your build is to profile your build with the --profile option.

The file generated by the --profile command is a binary file. Once you have generated this binary profile, you can analyze it using Bazel's analyze-profile command. By default, it will print out summary analysis information for each of the specified profile datafiles. This includes cumulative statistics for different task types for each build phase and an analysis of the critical execution path.

The first section of the default output describes an overview of the time spent on the different build phases:


Total launch phase time         6.00 ms    0.01%
Total init phase time            864 ms    1.11%
Total loading phase time       21.841 s   28.05%
Total analysis phase time       5.444 s    6.99%
Total preparation phase time     155 ms    0.20%
Total execution phase time     49.473 s   63.54%
Total finish phase time         83.9 ms    0.11%
Total run time                 77.866 s  100.00%

The following sections show the execution time of different tasks happening during a particular phase:


Total init phase time                     864 ms

Total time (across all threads) spent on:
              Type    Total    Count     Average
          VFS_STAT    2.72%        1     23.5 ms
      VFS_READLINK   32.19%        1      278 ms


Total loading phase time                21.841 s

Total time (across all threads) spent on:
              Type    Total    Count     Average
             SPAWN    3.26%      154      475 ms
          VFS_STAT   10.81%    65416     3.71 ms
SKYLARK_BUILTIN_FN   13.12%    45138     6.52 ms


Total analysis phase time                5.444 s

Total time (across all threads) spent on:
              Type    Total    Count     Average
     SKYFRAME_EVAL    9.35%        1     4.782 s
       SKYFUNCTION   89.36%    43332     1.06 ms


Total preparation time                    155 ms
Total execution phase time              49.473 s
Total time finalizing build              83.9 ms

Action dependency map creation           0.00 ms
Actual execution time                   49.473 s

Total time (across all threads) spent on:
              Type    Total    Count     Average
            ACTION    2.25%    12229     10.2 ms
       SKYFUNCTION    1.87%   236131     0.44 ms

The last section shows the critical path:

Critical path (32.078 s):
    Id        Time Percentage   Description
1109746     5.171 s   16.12%   Building [...]
1109745      164 ms    0.51%   Extracting interface [...]
1109744     4.615 s   14.39%   Building [...]
1109639     2.202 s    6.86%   Executing genrule [...]
1109637     2.00 ms    0.01%   Symlinking [...]
1109636      163 ms    0.51%   Executing genrule [...]
           4.00 ms    0.01%   [3 middleman actions]

You can use the following options to display more detailed information:

  • --dump=text

    This option prints all recorded tasks in the order they occurred. Nested tasks are indented relative to the parent. For each task, output includes the following information:

    [task type] [task description]
    Thread: [thread id]    Id: [task id]     Parent: [parent task id or 0 for top-level tasks]
    Start time: [time elapsed from the profiling session start]       Duration: [task duration]
    [aggregated statistic for nested tasks, including count and total duration for each nested task]
  • --dump=raw

    This option is most useful for automated analysis with scripts. It outputs each task record on a single line using '|' delimiter between fields. Fields are printed in the following order:

    1. thread id - integer positive number, identifies owner thread for the task
    2. task id - integer positive number, identifies specific task
    3. parent task id for nested tasks or 0 for root tasks
    4. task start time in ns, relative to the start of the profiling session
    5. task duration in ns. Please note that this will include duration of all subtasks.
    6. aggregated statistic for immediate subtasks per type. This will include type name (lower case), number of subtasks for that type and their cumulative duration. Types are space-delimited and information for single type is comma-delimited.
    7. task type (upper case)
    8. task description
    1|1|0|0|0||PHASE|Launch Bazel
    1|2|0|6000000|0||PHASE|Initialize command
    1|5|0|869955040|0||PHASE|Load packages
  • --html

    This option writes a file called <profile-file>.html in the directory of the profile file. Open it in your browser to see the visualization of the actions in your build. Note that the file can be quite large and may push the capabilities of your browser – please wait for the file to load.

    In most cases, the HTML output from --html is easier to read than the --dump output. It includes a Gantt chart that displays time on the horizontal axis and threads of execution along the vertical axis. If you click on the Statistics link in the top right corner of the page, you will jump to a section that lists summary analysis information from your build.

    • --html_details

      Additionally passing this option will render a more detailed execution chart and additional tables on the performance of built-in and user-defined Skylark functions. Beware that this increases the file size and the load on the browser considerably.

If Bazel appears to be hung, you can hit ctrl + \ or send Bazel a SIGQUIT signal (kill -3 $(bazel info server_pid)) to get a thread dump in the file $(bazel info output_base)/server/jvm.out.

Since you may not be able to run bazel info if bazel is hung, the output_base directory is usually the parent of the bazel-<workspace> symlink in your workspace directory.

当前网页内容, 由 大妈 ZoomQuiet 使用工具: ScrapBook :: Firefox Extension 人工从互联网中收集并分享;
若有不妥, 欢迎评注提醒:


订阅 substack 体验古早写作:

点击注册~> 获得 100$ 体验券: DigitalOcean Referral Badge

关注公众号, 持续获得相关各种嗯哼:


关于 ~ DebugUself with DAMA ;-)
公安备案号: 44049002000656 ...::