This tutorial walks you through some first steps with Pants.
It assumes you're already familiar with
basic Pants build concepts,
and that you're working in a source tree that already has
pants installed (such as
Pants's own repo:
The first time you run
pants, try it without arguments. This makes
Pants "bootstrap" itself, downloading and compiling things it needs:
$ ./pants goals
Now you're ready to invoke pants for more useful things.
You invoke pants with goals (like
bundle) and the targets to use (like
$ ./pants test src/python/myproject/example:tests
Goals (the "verbs" of Pants) produce new files from targets (the "nouns").
As a code author, you define your code's build targets in BUILD files. A build target might produce some output file[s]; it might have sources and/or depend on other build targets. There might be several BUILD files in the codebase; a target in one can depend on a target in another. Typically, a directory's BUILD file defines the target[s] whose sources are files in that directory.
Pants Command Line
Pants knows about goals ("verbs" like
test) and targets
(build-able things in your source code). A typical pants command-line
invocation looks like
$ ./pants test src/python/myproject/example:tests
Looking at the pieces of this we see
./ isn't a typo. A source tree that's been set up with Pants
build has a
pants executable in its top-level directory.
The first time you run
./pants, it might take a while: it will probably auto-update by
downloading the latest version.
test is a goal, a "verb" that Pants knows about. The
test goal runs tests and reports results.
Some goals are
gen (generate code from Thrift, Antlr, Protocol
run (run a binary), and
test (run tests and report results). Pants
knows that some of these goals depend on each other. E.g., in this
example, before it run tests, it must compile the code.
You can specify more than one goal on a command line. E.g., to run
tests and run a binary, we could have said
./pants test run ...
This is a build target, a "build-able" thing in your source code. To define these, you set up configuration files named
BUILD in your
source code file tree. (You'll see more about these later.)
Targets can depend on other targets. E.g., a test target normally depends on another target containing "library" code to test; to build and run the test code, Pants also first builds the library code.
You can specify more than one target on a command line. Pants will carry out its goals on all specified targets. E.g., you might use this to to run a few directories' worth of tests.
Pants produces files, both build outputs and intermediate files generated "along the way". These files live in directories under the top-level directory:
By default, build outputs go in the
dist/ directory. So far, you've
just run the
test goal, which doesn't output a file. But if you'd
instead invoked, for example, the
bundle goal on a
Pants would have populated this directory with many JVM
Intermediate files go in the
.pants.d/ directory. You don't want to
rely on files in there; if the Pants implementation changes, it's likely
to change how it uses intermediate files. You don't want to edit/delete
files in there; you may confuse Pants. But if you want to peek at some
generated code, the code is probably in here somewhere.
Multiple Goals, Multiple Targets
You can specify multiple goals and multiple targets. Pants applies all the goals to all the targets, skipping things that wouldn't make sense. E.g., you could
rungoals to both run tests and run a binary.
- Specify both test and binary targets.
In this example, it doesn't make sense to run a binary target as a test, so Pants doesn't do that.
One tricky side effect of this is accidental goal-target mismatch: You
can invoke a goal that doesn't make sense for a target. E.g., you can
test goal on a target that's not actually a test target. Pants won't
complain. It knows that it should compile code before it tests it; it
will happily compile the build targets. If you're not watching closely,
you might see a lot of output scrolling past and think it was running
You can specify some details of Pants' actions by means of command-line options. E.g., you could
tell Pants to "fail fast" on the first
junit test failure instead of running and reporting all
junit tests like so:
$ ./pants test.junit --fail-fast src/java/com/myorg/myproject/example/hello:tests
test has become
test goal is made up of parts, or tasks:
test.specs. We want to specify a flag to the
test.junit task, so we
specify that part on the command line. (Pants still runs the other parts of the
The dotted notation tells Pants where to apply options.)
We entered the
--fail-fast flag after
test.junit but before the target. Command-line flags
for a goal or task go immediately after that goal or task.
You can specify options for more than one part of a goal. For example,
$ ./pants test.junit --fail-fast test.pytest --no-fail-fast src/java:: src/python::
--fail-fast flag affects
Pants has some global options, options not associated with just one goal. For example,
If you pass the global
-ldebug flag after the word
goal but before any particular goal or
task, you get verbose debug-level logging for all goals:
$ ./pants -ldebug test src/java/myproject/example/hello:tests 09:18:53 00:00 [main] (To run a reporting server: ./pants server) 09:18:53 00:00 [bootstrap] 09:18:54 00:01 [setup] 09:18:54 00:01 [parse]DEBUG] Located Distribution(u'/Library/Java/JavaVirtualMachines/jdk1.7.0_60.jdk/Contents/Home/bin', minimum_version=None, maximum_version=None jdk=False) for constraints: minimum_version None, maximum_version None, jdk False DEBUG] Selected protoc binary bootstrapped to: /Users/lhosken/.pants.d/bin/protobuf/mac/10.9/2.4.1/protoc DEBUG] Selected thrift binary bootstrapped to: /Users/lhosken/.pants.d/bin/thrift/mac/10.9/0.5.0-finagle/thrift ...lots of build output...
For details about the Pants command line, see Invoking Pants.
To get help about a Pants goal, invoke ./pants goalname -h. This lists command-line options for that goal. E.g.,
$ ./pants test -h cache.test options: --[no-]cache-test-read (default: True) Read build artifacts from cache, if available. --[no-]cache-test-write (default: True) Write build artifacts to cache, if available. test.go options: Runs `go test` on Go packages. --[no-]test-go-remote (default: False) Enables running tests found in go_remote_libraries. --test-go-build-and-test-flags=<str> (default: ) Flags to pass in to `go test` tool. ...more test options... test.junit options: --test-junit-confs="['str1','str2',...]" (--test-junit-confs="['str1','str2',...]") ... (default: ['default']) Use only these Ivy configurations of external deps. --[no-]test-junit-skip (default: False) Skip running junit. --[no-]test-junit-fail-fast (default: False) Fail fast on the first test failure in a suite. ...many more test options...
test goal is made up of parts, or tasks such as
Command-line options apply to those tasks. The goal's help groups options by task. E.g., here, it
--test-go-build-and-test-flags option with
For a list of available goals,
For help with things that aren't goals (global options, other kinds of help), use
$ ./pants -h
If you want help diagnosing some strange Pants behavior, you might want verbose output.
To get this, instead of just invoking
./pants, set some environment variables and request
PEX_VERBOSE=5 ./pants -ldebug.
Pants allows you to also specify options in a config file. If you want to always run pants
with a particular option, you can configure it in a file at the root of the repo named
[default] jvm_options: ["-Xmx1g", "-Dfile.encoding=UTF8"] [test.junit] fail-fast: True
For more information on the
pants.ini file format, see
When we ran the
pants test goal, we told pants what target to build, but where are these
targets defined? Scattered around the source tree are
BUILD files. These
define targets. For example, this code snippet of
defines two targets: the app we ran and the binary that contains its code.
These targets are named
main (of type
jvm_app) and and
main-bin (of type
# Note that the target has no explicit name, so it defaults to the name # of the directory, in this case 'main'. jvm_app( basename = 'hello-example', dependencies = [ ':main-bin' ], bundles = [ bundle(relative_to='config', fileset=['config/*']) ] ) # The binary, the "runnable" part: jvm_binary(name = 'main-bin', dependencies = [ 'examples/src/java/org/pantsbuild/example/hello/greet', 'examples/src/resources/org/pantsbuild/example/hello', ], source = 'HelloMain.java', main = 'org.pantsbuild.example.hello.main.HelloMain', basename = 'hello-example', )
dependencies statements are interesting.
main-bin build target depends on other build targets;
dependencies lists those.
To build a runnable Java binary, we need to first compile its dependencies.
main-bin binary's dependency,
'examples/src/java/org/pantsbuild/example/hello/greet', is the address of
another target. Addresses look, roughly, like
path/to/dir:targetname. We can see this build
target in the
# Note that the target has no explicit name, so it defaults to the name # of the directory, in this case 'greet'. # It also has no explicit sources, so it defaults to the sources implied # by the target type, in this case "globs('*.java')". java_library( dependencies = , # A more realistic example would depend on other libs, # but this "hello world" is pretty simple. provides = artifact(org='org.pantsbuild.example', name='hello-greet', repo=public,), )
Pants uses dependency information to figure out how to build your code. You might find it useful for other purposes, too. For example, if you change a library's code, you might want to know which test targets depend on that library: you might want to run those tests to make sure they still work.
Anatomy of a
A target definition in a
BUILD file looks something like
java_library( name='util', dependencies = [ '3rdparty:commons-math', '3rdparty:thrift', 'src/java/com/myorg/myproject/example', ':base', ], sources=['*.java'], )
java_library is the target's type. Different target types
support different arguments. The following arguments are pretty common:
We use a target's name to refer to the target. If you omit the name, it defaults to the name of the directory the
BUILD file is in.
You use names on the command line to specify which targets to operate on.
You also use names in
BUILD files when one target refers to another,
List of things this target depends upon. If this target's code imports code that "lives" in other targets, list those targets here. If this target imports code that "lives" in
.eggs from elsewhere,
refer to them here.
List of source files, which must be under the directory tree rooted at the BUILD file's directory.
You can provide explicit file names and also use globs, e.g.
sources=['Foo.java', 'subdir/**/*.java']. To exclude a file or glob, prefix the value with
If you omit
sources in a target, Pants will attempt to apply a
sensible default that depends on the target type.
junit_tests will default to
java_library will default to
The Usual Commands
Make sure code compiles and tests pass:
test goal with the targets you're interested in. If they are
test targets, Pants runs the tests. If they aren't test targets, Pants
will still compile them since it knows it must compile before it can
$ ./pants test src/python:: src/java::
Run a binary
Use pants to execute a binary target. Compiles the code first if it is not up to date.
$ ./pants run src/python/myproject/example:my_script
Get the list of goals:
$ ./pants goals
Get help for one goal, e.g., test:
$ ./pants test -h
To learn more about working with Python projects, see the Python Tutorial.
To learn more about working with Java/JVM projects, see the Java Tutorial