Targets and BUILD files

Metadata for your code.

Most goals require metadata about your code. For example, to run a test, you need to know about all the transitive dependencies of that test. You may also want to set a timeout on that test.

Targets are an addressable set of metadata describing your code.

For example:

  • shell_source and python_test describe first-party code
  • python_requirement describes third-party requirements
  • pex_binary and archive describe artifacts you'd like Pants to build

To reduce boilerplate, some targets also generate other targets:

  • python_tests -> python_test
  • shell_sources -> shell_source
  • go_mod -> go_third_party_package

BUILD files

Targets are defined in files with the name BUILD. For example:



Each target type has different fields, or individual metadata values. Run pants help $target to see which fields a particular target type has, e.g. pants help file. Most fields are optional and use sensible defaults. See Field default values for how you may override a fields default value.

All target types have a name field, which is used to identify the target. Target names must be unique within a directory.

You can autoformat BUILD files by enabling a BUILD file formatter by adding it to [GLOBAL].backend_packages in pants.toml (such as pants.backend.build_files.fmt.black or others). Then to format, run pants fmt '**/BUILD' or pants fmt :: (formats everything).

Environment variables

BUILD files are very hermetic in nature with no support for using import or other I/O operations. In order to have dynamic data in BUILD files, you may inject values from the local environment using the env() function. It takes the variable name and optional default value as arguments.

  description=env("DIST_DESC", "Set the `DIST_DESC` env variable to override this value."),

Target addresses

A target is identified by its unique address, in the form path/to/dir:name. The above example has the addresses helloworld/greet:tests and helloworld/greet:bin.

Addresses are used in the dependencies field to depend on other targets. Addresses can also be used as command-line arguments, such as pants fmt path/to:tgt.

(Both "generated targets" and "parametrized targets" have a variant of this syntax; see the below sections.)


Default for the name field

The name field defaults to the directory name. So, this target has the address helloworld/greet:greet.

# helloworld/greet/BUILD

You can refer to this target with either helloworld/greet:greet or the abbreviated form helloworld/greet.


Use //:tgt for the root of your repository

Addressed defined in the BUILD file at the root of your repository are prefixed with //, e.g. //:my_tgt.

source and sources field

Targets like python_test and resource have a source: str field, while target generators like python_tests and resources have a sources: list[str] field. This determines which source files belong to the target.

Values are relative to the BUILD file's directory. Sources must be in or below this directory, i.e. ../ is not allowed.

The sources field also supports _ and ** as globs. To exclude a file or glob, prefix with !. For example, ["_.py", "!exclude_*.py"] will include f.py but not exclude_me.py.

resource(name="logo", source="logo.png")



Be careful with overlapping source fields

It's legal to include the same file in the source / sources field for multiple targets.

When would you do this? Sometimes you may have conflicting metadata for the same source file, such as wanting to check that a Shell test works with multiple shells. Normally, you should prefer Pants's parametrize mechanism to do this. See the below section "Parametrizing Targets".

Often, however, it is not intentional when multiple targets own the same file. For example, this often happens when using ** globs, like this:

# project/BUILD

# project/subdir/BUILD

Including the same file in the source / sources field for multiple targets can result in two confusing behaviors:

  • File arguments will run over all owning targets, e.g. pants test path/to/test.ext would run both test targets as two separate subprocesses, even though you might only expect a single subprocess.
  • Pants will sometimes no longer be able to infer dependencies on this file because it cannot disambiguate which of the targets you want to use. You must use explicit dependencies instead. (For some blessed fields, like the resolve field, if the targets have different values, then there will not be ambiguity.)

You can run pants list path/to/file.ext to see all "owning" targets to check if >1 target has the file in its source field.

dependencies field

A target's dependencies determines which other first-party code and third-party requirements to include when building the target.

Usually, you leave off the dependencies field thanks to dependency inference. Pants will read your import statements and map those imports back to your first-party code and your third-party requirements. You can run pants dependencies path/to:target to see what dependencies Pants infers.

However, dependency inference cannot infer everything, such as dependencies on resource and file targets.

To add an explicit dependency, add the target's address to the dependencies field. This augments any dependencies that were inferred.


You only need to declare direct dependencies. Pants will pull in transitive dependencies—i.e. the dependencies of your dependencies—for you.


Relative addresses, :tgt

When depending on a target defined in the same BUILD file, you can simply use :tgt_name, rather than helloworld/greet:tgt_name, for example.

Addresses for generated targets also support relative addresses in the dependencies field, as explained in the "Target Generation" section below.


Ignore dependencies with ! and !!

If you don't like that Pants inferred a certain dependency—as reported by pants dependencies path/to:tgt—tell Pants to ignore it with !:


You can use the prefix !! to transitively exclude a dependency, meaning that even if a target's dependencies include the bad dependency, the final result will not include the value.

Transitive excludes can only be used in target types that conventionally are not dependend upon by other targets, such as pex_binary and python_test / python_tests. This is meant to limit confusion, as using !! in something like a python_source / python_sources target could result in surprising behavior for everything that depends on it. (Pants will print a helpful error when using !! when it's not legal.)

Field default values

As mentioned above in BUILD files, most target fields have sensible defaults. And it's easy to override those values on a specific target. But applying the same non-default value on many targets can get unwieldy, error-prone and hard to maintain. Enter __defaults__.

Alternative default field values are set using the __defaults__ BUILD file symbol, and apply to targets in the filesystem tree under that BUILD file's directory.

The defaults are provided as a dictionary mapping target types to the default field values. Multiple target types may share the same set of default field values, when grouped together in parentheses (as a Python tuple).

Use the all keyword argument to provide default field values that should apply to all targets.

The extend=True keyword argument allows to add to any existing default field values set by a previous __defaults__ call rather than replacing them.

Default fields and values are validated against their target types, except when provided using the all keyword, in which case only values for fields applicable to each target are validated. Use ignore_unknown_fields=True to ignore invalid fields.

This means, that it is legal to provide a default value for all targets, even if it is only a subset of targets that actually supports that particular field.


    # Provide default `tags` to all targets in this subtree, and skip black, where applicable.
    __defaults__(all=dict(tags=["example"], skip_black=True))

Subdirectories may override defaults from a parent BUILD file:

    # For `files` and `resources` targets, we want to use some other defaults.
      (files, resources): dict(tags=["example", "overridden"], description="Our assets")

Use the extend=True keyword to update defaults rather than replace them, for any given target.

    # Add a default description to all types, in addition to the inherited default tags.
    __defaults__(extend=True, all=dict(description="Add default description to the defaults."))

To reset any modified defaults, simply override with the empty dict:


Supporting optional plugin fields

Normally Pants presents an error message when attempting to provide a default value for a field that doesn't exist for the target. However, some fields comes from plugins, and to support disabling a plugin without having to remove any default values referencing any plugin fields it was providing, there is a ignore_unknown_fields option to use:

        # Defaults...

Extending field defaults

To add to a default value rather than replacing it, the current default value for a target field is available in the BUILD file using <target>.<field>.default. This allows you to augment a field's default value with much more precision. As an example, if you want to make the default sources for a python_sources target to work recursively you may specify a target augmenting the default sources field:

    f"{pattern[0] if pattern.startswith("!") else ""}**/{pattern.lstrip("!")}"
    for pattern in python_sources.sources.default

Target generation

To reduce boilerplate, Pants provides target types that generate other targets. For example:

  • files -> file
  • python_tests -> python_test
  • go_mod -> go_third_party_package

Usually, prefer these target generators. pants tailor :: will automatically add them for you.

Run pants help targets to see how the target determines what to generate. Targets for first-party code, like resources and python_tests, will generate one target for each file in their sources field.

    # Will generate two `python_source` targets.
    sources=["app.py", "util.py"],

(Usually, you can leave off the sources field. When possible, it defaults to all relevant files in the current directory.)

Typically, fields declared in the target generator will be inherited by each generated target. For example, if you set timeout=120 in a python_tests target, each generated python_test target will have timeout=120. You can instead use the overrides field for more granular metadata:

    # This applies to every generated target.
    # These only apply to the relevant generated targets.
        "dirutil_test.py": {"timeout": 30},
        ("osutil_test.py", "strutil_test.py"): {"timeout": 15},


Field default values for generated targets

Default field values apply to target generators, not to generated targets. For example, if you have:

__defaults__({python_test: {"timeout": 30}})

python_tests(sources=["test_*.py", "!test_special.py"])
python_test(name="special", source="test_special.py")

Then the default timeout value will only apply to the "special" python_test, not to any of the targets generated by the python_tests target.

To specify defaults for both generated and manually-written instances of a target, you must list the target generator in your __defaults__ as well:

__defaults__({(python_test, python_tests): {"timeout": 30}})

The address for generated targets depends if the generated target is for first-party code or not:

Generated target typeGenerated address syntax
First-party, e.g. python_source and filepath/to/file.ext:tgt_generator

Example: src/py/app.py:lib

The address always starts with the path to the file.

If the file lives in the same directory as the target generator and the target generator left off the name field, you can use just the file path. For example, src/py/app.py (without the :lib suffix).

If the file lives in a subdirectory of the target generator, the suffix will look like ../tgt_generator. For example, src/py/subdir/f.py:../lib, where the target generator is src/py:lib.

With the dependencies field, you can use relative addresses by prefixing the path with ./, so long as the path is in the same directory or below the current BUILD file. For example, ./app.py:lib rather than src/py/app.py:lib.
All other targets, e.g. go_third_party_packagepath/to:tgt_generator#generated_name

Example: 3rdparty/py:reqs#django

Run pants help $target_type on the target generator to see how it sets the generated name. For example, go_mod uses the Go package's name.

If the target generator left off the name field, you can leave it off for the generated address too, e.g. 3rdparty/py#django (without the :reqs portion).

With the dependencies field, you can use relative addresses to reference generated targets in the same BUILD file, e.g. :generator#generated_name instead of src/py:generated#generated_name. If the target generator uses the default name, you can simply use #generated_name.

Run pants list dir: in the directory of the target generator to see all generated target addresses, and pants peek dir: to see all their metadata.

You can use the address for the target generator as an alias for all of its generated targets. For example, if you have the files target assets:logos, adding dependencies=["assets:logos"]to another target will add a dependency on each generated file target. Likewise, if you have a python_tests target project:tests, then pants test project:tests will run on each generated python_test target.


Tip: one BUILD file per directory

Target generation means that it is technically possible to put everything in a single BUILD file.

However, we've found that it usually scales much better to use a single BUILD file per directory. Even if you start with using the defaults for everything, projects usually need to change some metadata over time, like adding a timeout to a test file or adding dependencies on resources.

It's useful for metadata to be as fine-grained as feasible, such as by using the overrides field to only change the files you need to. Fine-grained metadata is key to having smaller cache keys (resulting in more cache hits), and allows you to more accurately reflect the status of your project. We have found that using one BUILD file per directory encourages fine-grained metadata by defining the metadata adjacent to where the code lives.

pants tailor :: will automatically create targets that only apply metadata for the directory.

Parametrizing targets

It can be useful to create multiple targets describing the same entity, each with different metadata. For example:

  • Run the same tests with different interpreter constraints, e.g. Python 2 vs Python 3.
  • Declare that a file should work with multiple "resolves" (lockfiles).

The parametrize builtin creates a distinct target per parametrized field value. All values other than the parametrized field(s) are the same for each target. For example:

# Creates two targets:
#    example:tests@shell=bash
#    example:tests@shell=zsh

    shell=parametrize("bash", "zsh"),

If multiple fields are parametrized, a target will be created for each value in the Cartesian product, with , as the delimiter in the address. See the next example.

If the field value is not a string—or it is a string but includes spaces—you can give it an alias, like the interpreter_constraints field below:

# Creates four targets:
#    example:tests@interpreter_constraints=py2,resolve=lock-a
#    example:tests@interpreter_constraints=py2,resolve=lock-b
#    example:tests@interpreter_constraints=py3,resolve=lock-a
#    example:tests@interpreter_constraints=py3,resolve=lock-b

    interpreter_constraints=parametrize(py2=["==2.7.*"], py3=[">=3.6"]),
    resolve=parametrize("lock-a", "lock-b"),

The targets' addresses will have @key=value at the end, as shown above. Run pants list dir: in the directory of the parametrized target to see all parametrized target addresses, and pants peek dir: to see all their metadata.

Generally, you can use the address without the @ suffix as an alias to all the parametrized targets. For example, pants test example:tests will run all the targets in parallel. Use the more precise address if you only want to use one parameter value, e.g. pants test example:tests@shell=bash.

Parametrization can be combined with target generation. The @key=value will be added to the end of the address for each generated target. For example:

# Generates four `shunit2_test` targets:
#    example/test1.sh:tests@shell=bash
#    example/test1.sh:tests@shell=zsh
#    example/test2.sh:tests@shell=bash
#    example/test2.sh:tests@shell=zsh
# Also creates two `shunit2_tests` target
# generators, which can be used as aliases
# to their generated targets:
#    example:tests@shell=bash
#    example:tests@shell=zsh
# Generally, you can still use `example:tests`
# without the `@` suffix as an alias to all the
# created targets.

    sources=["test1.sh", "test2.sh"],
    shell=parametrize("bash", "zsh"),

You can combine parametrize with the overrides field to set more granular metadata for generated targets:

# Generates three `shunit2_test` targets:
#    example/test1.sh:tests
#    example/test2.sh:tests@shell=bash
#    example/test2.sh:tests@shell=zsh
# The `shunit2_tests` target generator
# `example:tests` can be used as an alias
# to all 3 created targets.

    sources=["test1.sh", "test2.sh"],
        "test2.sh": {"shell": parametrize("bash", "zsh")},