Getting started

We're pleased that you are interested in working on Warehouse.

Your first pull request

After you set up your development environment and ensure you can run the tests and build the documentation (using the instructions in this document), take a look at our guide to the Warehouse codebase. Then, look at our open issues that are labelled "good first issue", find one you want to work on, comment on it to say you're working on it, then submit a pull request. Use our submitting patches documentation to help.

Setting up a development environment to work on Warehouse should be a straightforward process. If you have any difficulty, contact us so we can improve the process:

For bug reports or general problems, file an issue on GitHub;
For real-time chat with other PyPA developers, join #pypa-dev on Libera, or the PyPA Discord;
For longer-form questions or discussion, visit Discourse.

Detailed installation instructions

Get the Warehouse source code

First, fork and clone the Warehouse source code to work on it locally.

Fork the repository on GitHub and clone it to your local machine:

git clone git@github.com:YOUR-USERNAME/warehouse.git

Add a remote and regularly sync to make sure you stay up-to-date with our repository:

git remote add upstream https://github.com/pypi/warehouse.git
git checkout main
git fetch upstream
git merge upstream/main

Create your development environment

!!! note Warehouse development can be done using Makefile scripts which execute all developer actions inside Docker containers. You do not need to create a Python virtual environment.

Verify that you have Make installed

We use Make to build the docs and the Warehouse site. Verify that you have Make installed in your environment.

make --version

If you do not have Make installed, consult your operating system documentation on how to install make.

Do you want to build the docs or the Warehouse site?

If you wish to build the entire Warehouse site, use Docker and follow the instructions below. However, if you are only contributing to the user or development documentation, then you can skip setting up Docker below and use Make instead. To build the docs, follow the instructions here.

Why Docker?

Docker simplifies development environment set up.

Warehouse uses Docker and Docker Compose to automate setting up a "batteries included" development environment. The Dockerfile and docker-compose.yml files include all the required steps for installing and configuring all the required external services of the development environment.

Installing Docker

Install Docker Engine

The best experience for building Warehouse on Windows 10 is to use the Windows Subsystem for Linux (WSL) in combination with both Docker for Windows and Docker for Linux. Follow the instructions for both platforms.

Verifying Docker installation

Check that Docker is installed: docker -v

Install Docker Compose

Install Docker Compose using the Docker-provided installation instructions.

!!! note Docker Compose will be installed by Docker for Mac and Docker for Windows automatically.

Verifying Docker Compose installation

Check that Docker Compose is installed: docker compose version

Verifying the necessary ports are available

Warehouse needs access to a few local ports in order to run, namely ports 80, 5433, and 9000. You should check each of these for availability with the lsof command.

For example, checking port 80:

sudo lsof -i:80 | grep LISTEN

If the port is in use, the command will produce output, and you will need to determine what is occupying the port and shut down the corresponding service. Otherwise, the port is available for Warehouse to use, and you can continue.

Alternately, you may set the WEB_PORT environment variable for docker compose to use instead. An example:

export WEB_PORT=8080
make ...

# or inline:
WEB_PORT=8080 make ...

Building the Warehouse Container

Once you have Docker and Docker Compose installed, run:

make build

in the repository root directory.

This will pull down all of the required docker containers, build Warehouse and run all of the needed services. The Warehouse repository will be mounted inside the Docker container at /opt/warehouse/src/. After the initial build, you should not have to run this command again.

Running the Warehouse container and services

You have to start the Docker services that make up the Warehouse application.

!!! tip These services need ~4 GB of RAM dedicated to Docker to work. This is more than the default setting of the Docker Engine of 2 GB. Thus, you need to increase the memory allocated to Docker in Docker Preferences (on Mac by moving the slider to 4 GB in the GUI) or Docker Settings (on Windows by editing the config file found at C:\Users\<USER>\AppData\Local\Docker\wsl).

If you are using Linux, you may need to configure the maximum map count to get
the `opensearch` up and running. According to the
[documentation](https://opensearch.org/docs/2.15/install-and-configure/install-opensearch/index/#important-settings)
this can be set temporarily:

```shell
# sysctl -w vm.max_map_count=262144
```

or permanently by modifying the `vm.max_map_count` setting in your
`/etc/sysctl.conf`.

Also check that you have more than 5% disk space free, otherwise
opensearch will become read only. See `flood_stage` in the
[opensearch disk allocation docs](https://opensearch.org/docs/latest/install-and-configure/configuring-opensearch/cluster-settings/#cluster-level-routing-and-allocation-settings).

Once make build has finished, run the command:

make serve

This command will:

ensure the db is prepared,
run migrations,
load some example data from Test PyPI, and
index all the data for the search database.
start up the containers needed to run Warehouse

After the initial build process, you will only need this command each time you want to startup Warehouse locally.

make serve will produce output for a while, and will not exit. Eventually the output will cease, and you will see a log message indicating that either the web service has started listening:

warehouse-web-1   | [2022-12-26 19:27:12 +0000] [1] [INFO] Starting gunicorn 20.1.0
warehouse-web-1   | [2022-12-26 19:27:12 +0000] [1] [INFO] Listening at: http://0.0.0.0:8000 (1)
warehouse-web-1   | [2022-12-26 19:27:12 +0000] [1] [INFO] Using worker: sync
warehouse-web-1   | [2022-12-26 19:27:12 +0000] [7] [INFO] Booting worker with pid: 7

or that the static container has finished compiling the static assets:

warehouse-static-1  |
warehouse-static-1  | webpack 5.75.0 compiled with 1 warning in 6610 ms

or maybe something else.

Bootstrapping the TUF Metadata Repository

To enable PyPI Index Signing (PEP 458), you have to first bootstrap the TUF metadata repository.

make inittuf

You should see the following line at the bottom of the output:

Bootstrap completed using `dev/rstuf/bootstrap.json`. 🔐 🎉

This command sends a static bootstrap payload to the RSTUF API. The payload includes the TUF trust root for development and other configuration.

By calling this API, RSTUF creates the TUF metadata repository, installs the TUF trust root for development, and creates the initial set of TUF metadata.

!!! note The RSTUF API is exposed only for development purposes and will not be available in production. Currently, no upload hooks or automatic metadata update tasks are configured to interact with RSTUF.

Take a look at the [RSTUF API documentation](https://repository-service-tuf.readthedocs.io/en/stable/guide/general/usage.html#adding-artifacts)
to see how you can simulate artifact upload or removal, and how they affect
the TUF metadata repository:

* RSTUF API: http://localhost:8001
* TUF Metadata Repository: http://localhost:9001/tuf-metadata/

Resetting the development database

make resetdb

This command will fully reset the development database.

Viewing Warehouse in a browser

At this point all the services are up, and web container is listening on port 80. It's accessible at http://localhost:80/.

!!! note If you are using docker-machine on an older version of macOS or Windows, the warehouse application might be accessible at https://<docker-ip>:80/ instead. You can get information about the docker container with docker-machine env

!!! note On Firefox, the logos might show up as black rectangles due to the Content Security Policy used and an implementation bug in Firefox (see this bug report for more info).

If you've set a different port via the WEB_PORT environment variable, use that port instead.

Logging in to Warehouse

You can log into warehouse at http://localhost:80/account/login/.

There are 4 accounts ready for you to use:

ewdurbin - Superuser, 3 email addresses (one verified), has projects
di - Superuser, 2 email addresses (both verified), has projects
dstufft - Superuser, 2 email addresses (one verified), has projects
miketheman - Regular user, 1 email address (not verified), has a project

The password for every account has been set to the string password.

Using different accounts will allow you to see different parts of the site, and have slightly different experiences.

Note that there are no Moderator accounts in the dev db. Any Superuser can change a user to a moderator if needed.

TOTP and Recovery Codes

To generate a TOTP token, run the following from your terminal:

make totp

Alternatively, you can scan the QR code below to add these accounts to your authenticator app:

{ width="100" }

The following static recovery codes are pre-generated for these users:

6ebc846aadf23e35
7283821faf191a33
68108e19d25e2eec
4e6a18adb880fbc1
f62627d29675725f
4cda895a133b4cc8
8678c6f0d9a1e6de
edc6ce3800c0fc94 -- burned

Email Verification

Auth verification emails are output to the console, or can be accessed from http://localhost:1080.

See Testing Emails for more information.

Logging in as users without 2FA

For users that are not listed above, once logged in with the password password, you must enroll in a form of Two-Factor Authentication (2FA). This is a requirement for all users.

One way to make this easier is to use a command-line tool like totp-cli to generate a TOTP 2FA code.

For example, to generate a code for any of the above users, we have a common Key set in the database for those users:

totp-cli instant <<< IU7UP3EMIPI7EBPQUUSEHEJUFNBIWOYG

This will emit a 6-digit code you can paste into the 2FA form.

For other accounts, you'll need to preserve the Key used to generate the TOTP code the next time you need to log in.

To be able to "forget" the initial Key, and use it like a TOTP app, create a storage and set a password, like so:

$ totp-cli add-token localhost <username>
Token: <paste Key from warehouse web interface here>
Password: <set a password, is unique to this totp storage>

Then you can retrieve the current TOTP code with:

$ totp-cli g localhost <username>
Password: <the password you set for the totp storage>

Keep in mind: If the database is ever reset, you'll need to re-enroll user accounts in 2FA.

Remove the existing TOTP token from storage with:

$ totp-cli delete localhost <username>
Password: <the password you set for the totp storage>

Stopping Warehouse and other services

In the terminal where make serve is running, you can use Control-C to gracefully stop all Docker containers, and thus the one running the Warehouse application.

Or, from another terminal, use make stop in the Warehouse repository root; that'll stop all the Docker processes with warehouse in the name.

What did we just do and what is happening behind the scenes?

The repository is exposed inside of the web container at /opt/warehouse/src/ and Warehouse will automatically reload when it detects any changes made to the code.

The example data located in dev/example.sql.xz is taken from Test PyPI and has been sanitized to remove anything private.

Running your developer environment after initial setup

You won't have to initialize the database after the first time you do so, and you will rarely have to re-run make build. Ordinarily, to access your developer environment, you'll:

make serve

View Warehouse in the browser at http://localhost:80/.

Debugging the webserver

If you would like to use a debugger like pdb that allows you to drop into a shell, you can use make debug instead of make serve.

Troubleshooting

Errors when executing `make build`

If you are using Ubuntu and invalid reference format error is displayed, you can fix it by installing Docker through Snap.
```
snap install docker
```

Errors when executing `make serve`

If the Dockerfile is edited or new dependencies are added (either by you or a prior pull request), a new container will need to be built. A new container can be built by running make build. This should be done before running make serve again.
If make serve hangs after a new build, you should stop any running containers and repeat make serve.
If using Windows Subsystem for Linux (WSL) and running code . results in an Exec format error, ensure that the VS Code WSL extension is installed and that the project is opened using "Open Folder in WSL" from Visual Studio Code.

Reinstalling the VS Code server in WSL may also resolve the issue.
To run Warehouse behind a proxy set the appropriate proxy settings in the Dockerfile.
If sqlalchemy.exec.OperationalError is displayed in localhost after make serve has been executed, shut down the Docker containers. When the containers have shut down, run make serve in one terminal window while running make initdb in a separate terminal window.

Errors when executing `make purge`

If make purge fails with a permission error, check ownership and permissions on warehouse/static. docker compose is spawning containers with docker. Generally on Linux that process is running as root. So when it writes files back to the file system as the static container does those are owned by root. So your docker daemon would be running as root, so your user doesn't have permission to remove the files written by the containers. sudo make purge will work.

Errors when executing `make initdb`

If make initdb fails with a timeout like:
```
urllib3.exceptions.ConnectTimeoutError: (<urllib3.connection.HTTPConnection object at 0x8beca733c3c8>, 'Connection to opensearch timed out. (connect timeout=30)')
```
you might need to increase the amount of memory allocated to docker, since opensearch wants a lot of memory (Dustin gives warehouse ~4GB locally). Refer to the tip under Running the Warehouse container and services section for more details.
If make initdb fails with an error like:
```
A fatal error has been detected by the Java Runtime Environment:
SIGILL (0x4) at pc=0x0000f819dfc67c5c, pid=25, tid=26
```
while building the warehouse-opensearch image, this is a known issue with OpenSearch in Apple M4 processors. A workaround (described in this comment) is to add the following line to dev/compose/opensearch/Dockerfile.yml (after the FROM line):
```
ENV _JAVA_OPTIONS=-XX:UseSVE=0
```

"no space left on device" when using `docker compose`

docker compose may leave orphaned volumes during teardown. If you run into the message "no space left on device", try running the following command (assuming Docker >= 1.9):

docker volume rm $(docker volume ls -qf dangling=true)

!!! note This will delete orphaned volumes as well as directories that are not volumes in /var/lib/docker/volumes

(Solution found and further details available at https://github.com/chadoe/docker-cleanup-volumes)

`make initdb` is slow or appears to make no progress

This typically occurs when Docker is not allocated enough memory to perform the migrations. Try modifying your Docker configuration to allow more RAM for each container, temporarily stop make_serve and run make initdb again.

`make initdb` complains about PostgreSQL Version

You built a Warehouse install some time ago and PostgreSQL has been updated. If you do not need the data in your databases, it might be best to just blow away your builds + docker containers and start again:

make purge
docker volume rm $(docker volume ls -q --filter dangling=true)

Compilation errors in non-Docker development

While Warehouse is designed to be developed in using Docker, you may have tried to install Warehouse's requirements in your system or virtual environment. This is discouraged as it can result in compilation errors due to your system not including libraries or binaries required by some of Warehouse's dependencies.

An example of such dependency is psycopg which requires PostgreSQL binaries and will fail if not present.

If there's a specific use case you think requires development outside Docker please raise an issue in Warehouse's issue tracker.

Disabling services locally

Some services, such as OpenSearch, consume a lot of resources when running locally, but might not always be necessary when doing local development.

To disable these locally, you can create a docker-compose.override.yaml file to override any settings in the docker-compose.yaml file. To individually disable services, modify their entrypoint to do something else:

services:
  opensearch:
    entrypoint: ["echo", "OpenSearch disabled"]

Note that disabling services might cause things to fail in unexpected ways.

This file is ignored in Warehouse's .gitignore file, so it's safe to keep in the root of your local repo.

See the annotated file docker-compose.override.yaml-sample for ideas.

Building Styles

Styles are written in the scss variant of Sass and compiled using webpack. They will be automatically built when changed when make serve is running.

Running the Interactive Shell

There is an interactive shell available in Warehouse which will automatically configure Warehouse and create a database session and make them available as variables in the interactive shell.

To run the interactive shell, simply run:

make shell

The interactive shell will have the following variables defined in it:

Variable	Description
`config`	The Pyramid `Configurator` object which has already been configured by Warehouse.
`db`	The SQLAlchemy ORM `Session` object which has already been configured to connect to the database.
`request`	A dummy Pyramid `Request` object which has already been configured to provide a request-like object. Does not autocommit, call `request.db.commit()` to commit changes.

To use the db object in the interactive shell, import the class you're planning to use. For example, if I wanted to use the User object, I would do this:

$ make shell
docker compose run --rm web python -m warehouse shell
Starting warehouse_redis_1 ...
...
(InteractiveConsole)
>>>
>>> from warehouse.accounts.models import User
>>> db.query(User).filter_by(username='test').all()
[User(username='test')]

You can also run the IPython shell as the interactive shell. To do so export the environment variable WAREHOUSE_IPYTHON_SHELL prior to running the make build step:

export WAREHOUSE_IPYTHON_SHELL=1

Now you will be able to run the make shell command to get the IPython shell.

Running Pyramid p-scripts

Pyramid ships with a set of p-scripts that let you inspect the running application's configuration. These are useful for investigating the tween chain, registered routes, and views.

Available scripts: ptweens, proutes, pviews, pshell, prequest, pdistreport. pserve is not supported — use make serve (gunicorn) instead.

To run them inside the web container, use development.ini:

docker compose run --rm web ptweens development.ini
docker compose run --rm web proutes development.ini
docker compose run --rm web pviews development.ini /
docker compose run --rm web pshell development.ini

prequest issues a request directly against the application (without needing a running server), which is handy for scripting or quick checks:

# Fetch homepage HTML with response headers
docker compose run --rm web prequest development.ini / -d

# Fetch JSON API for a project
docker compose run --rm web prequest development.ini /pypi/pip/json -d

!!! note The development.ini file is a minimal PasteDeploy configuration that delegates to the same configure() function used by gunicorn. All application settings still come from environment variables (loaded via dev/environment).

Running tests and linters

!!! note PostgreSQL 9.4 is required because of pgcrypto extension

The Warehouse tests are found in the tests/ directory and are designed to be run using make.

To run all tests, in the root of the repository:

make tests
make static_tests

This will run the tests with the supported interpreter as well as all of the additional testing that we require.

If you want to run a specific test, you can use the T variable:

T=tests/unit/i18n/test_filters.py make tests

!!! note By default, using the T variable disables testcase parallelization (due to runner startup time being greater than actual test time). To re-enable parallelization, you can pass explicit TESTARGS:

```shell
T=tests/unit/i18n/test_filters.py TESTARGS="-n auto" make tests
```

It also turns off test coverage reporting because it is almost guaranteed
to fail and add test time overhead. To re-enable the coverage report, you
can pass explicit `COVERAGE`:

```shell
T=tests/unit/i18n/test_filters.py COVERAGE=1 make tests
```

You can also add arguments to the test runner by using the TESTARGS variable:

TESTARGS="-vvv -x" make tests

This will pass the arguments -vvv and -x down to pytest.

This is useful in scenarios like passing a random seed to the test runner:

TESTARGS="--randomly-seed=1234" make tests

You can run the common tools (formaters, linting, translations) to make sure that a patch is ready by running:

make patch

You can run linters, programs that check the code, with:

make lint

Warehouse uses ruff for opinionated formatting and linting. You can reformat with:

make reformat

Building translations

Warehouse is translated into a number of different locales, which are stored in the warehouse/locale/ directory.

These translation files contain references to untranslated text in source code and HTML templates, as well as the translations which have been submitted by our volunteer translators.

When making changes to files with strings marked for translation, it's necessary to update these references any time source strings are change, or the line numbers of the source strings in the source files.

Use make to build the translations. For example:

make translations

What next?

Look at our open issues that are labelled "good first issue", find one you want to work on, comment on it to say you're working on it, then submit a pull request. Use our submitting patches documentation to help.

Talk with us

You can find us via a GitHub issue, #pypa or #pypa-dev on Libera, the PyPA Discord or Discourse, to ask questions or get involved. And you can meet us in person at packaging sprints.

Learn about Warehouse and packaging

Resources to help you learn Warehouse's context:

Uh oh!

FilesExpand file tree

getting-started.md

Latest commit

History