Skip to main content
Version: 2.9 (deprecated)

Source roots

Configuring Pants to understand your imports.


Go and Shell can skip this page

Go does have a notion of source roots: where your go.mod is located. However, that is handled automatically by Pants without you needing to follow this page.

Shell does not have any notion of source roots.

What are source roots?

Some project layouts use top-level folders for namespace purposes, but have the code live underneath. However, the code's imports will ignore these top-level folders, thanks to mechanisms like the $PYTHONPATH and the JVM classpath. Source roots are a generic equivalent of these concepts.

For example, given this Python project:

src
└── python
└── project
├── __init__.py
├── app.py
├── config
│ ├── __init__.py
│ └── prod.json
└── util
├── __init__.py
└── math.py

You would likely set PYTHONPATH=src/python and use imports like this:

from project.app import App
from project.util.math import add_two

pkgutil.get_data("project.config", "prod.json")

In the example above, src/python is a source root. So, when some code says from project.app import App, Pants can know that this corresponds to the code in src/python/project/app.py.

Configuring source roots

There are two ways to configure source roots:

  • Using patterns
  • Using marker files

You can mix and match between both styles. Run ./pants roots to see what Pants is using:

./pants roots
src/assets
src/python
src/rust

Configuring source roots using patterns

You can provide a set of patterns that match your source roots:

pants.toml
[source]
root_patterns = [
'/src/python',
'/test/python',
]

The / prefix means that the source root is located at the build root, so it will match src/python, but not project1/src/python.

You can leave off the / prefix to match any directory whose suffix matches a pattern. For example, root_patterns = ["src/python"] would consider all of these to be source roots, if they exist:

  • src/python
  • project1/src/python

You can use * as a glob. For example, root_patterns = ["/src/*"] would consider all of these to be source roots:

  • src/python
  • src/java
  • src/assets

Configuring no source roots

Many projects do not have any top-level folders used for namespacing.

For example, given this Python project:

project
├── __init__.py
├── app.py
├── config
│ ├── __init__.py
│ └── prod.json
└── util
├── __init__.py
└── math.py

You would likely not set PYTHONPATH and would still use imports like this:

from project.app import App
from project.util.math import add_two

pkgutil.get_data("project.config", "prod.json")

If you have no source roots, use this config:

pants.toml
[source]
root_patterns = ["/"]
Default source roots

The default value of the root_patterns config key is ["/", "src", "src/python", "src/py"].

These capture a range of common cases, including a source root at the root of the repository. If your source roots match these patterns, you don't need to explicitly configure them.

Configuring source roots using marker files

You can also denote your source roots using specially-named marker files. To do so, first pick a name (or multiple names) to use:

pants.toml
[source]
marker_filenames = ["SOURCE_ROOT"]

Then, place a file of that name in each of the source roots. The contents of those files don't matter. They can be empty.

For example, given this Python repo, where we have a setup.py for each distinct project:

.
├── server
│ ├── server
│ │ ├── __init__.py
│ │ └── app.py
│ └── setup.py
└── utils
├── setup.py
└── utils
├── __init__.py
├── math.py
└── strutil.py

We could use this config:

pants.toml
[source]
marker_filenames = ["setup.py"]

We can then run ./pants roots to find these source roots used:

./pants roots
server
utils

This means that Pants would work with these imports:

import server.app
from utils.strutil import capitalize

Whereas these imports are invalid:

import server.server.app
from utils.utils.strutil import capitalize

Examples

These project structures are all valid; Pants does not expect you to reorganize your codebase to use the tool.

src/<lang> setup

This setup is common in "polyglot" repositories: i.e. repos with multiple languages.

Project:

.
├── 3rdparty
│ ├── java
│ │ └── ivy.xml
│ └── python
│ └── requirements.txt
├── src
│ ├── java
│ │ └── org
│ │ └── pantsbuild
│ │ └── project
│ │ ├── App.java
│ │ └── util
│ │ └── Math.java
│ └── python
│ └── project
│ ├── __init__.py
│ ├── app.py
│ ├── config
│ │ ├── __init__.py
│ │ └── prod.json
│ └── util
│ ├── __init__.py
│ └── math.py
└── test
└── python
└── project
├── __init__.py
└── util
├── __init__.py
└── test_math.py

While we have tests in a separate source root here, it's also valid to have tests colocated with their src files.

Example imports:

# Python
from project.app import App
from project.util.test_math import test_add_2
// Java
import org.pantsbuild.project.App
import org.pantsbuild.project.util.Math

Config:

pants.toml
[source]
root_patterns = [
"/src/java",
"/src/python",
"/test/python",
]

Note that we organized our 3rdparty requirements in the top-level folders 3rdparty/python and 3rdparty/java, but we do not need to include them as source roots because we do not have any first-party code there.

Multiple top-level projects

Project:

This layout has lots of nesting; this is only one possible way to organize the repository.

.
├── ads
│ └── py
│ └── ads
│ ├── __init__.py
│ ├── billing
│ │ ├── __init__.py
│ │ └── calculate_bill.py
│ └── targeting
│ ├── __init__.py
│ └── validation.py
├── base
│ └── py
│ └── base
│ ├── __init__.py
│ ├── models
│ │ ├── __init__.py
│ │ ├── org.py
│ │ └── user.py
│ └── util
│ ├── __init__.py
│ └── math.py
└── news
└── js
└── spa.js

Example imports:

import ads.billing.calculate_bill
from base.models.user import User
from base.util.math import add_two

Note that even though the projects live in different top-level folders, you are still able to import from other projects. If you would like to limit this, you can use ./pants dependees or ./pants dependencies in CI to track where imports are being used. See Project introspection.

Config:

Either of these are valid and they have the same result:

pants.toml
[source]
root_patterns = [
"/ads/py",
"/base/py",
"/new/js",
]
pants.toml
[source]
root_patterns = [
"py",
"js",
]

No source root

Warning: while this project structure is valid, it often does not scale as well as your codebase grows, such as adding new languages.

Project:

.
├── project
│ ├── __init__.py
│ ├── app.py
│ ├── config
│ │ ├── __init__.py
│ │ └── prod.json
│ └── util
│ ├── __init__.py
│ └── math.py
└── pyproject.toml

Example imports:

from project.app import App
from project.util.math import add_two

pkgutil.get_data("project.config", "prod.json")

Config:

Either of these are valid and they have the same result:

pants.toml
[source]
root_patterns = ["/"]
toml
[source]
marker_filenames = ["pyproject.toml"]