Development environment guide (Ubuntu)#

This guide will cover:

  • setting up and maintaining a development environment, including installing compilers and SciPy build dependencies;

  • creating a personal fork of the SciPy repository on GitHub;

  • using git to manage a local repository with development branches;

  • performing an in-place build of SciPy; and

  • creating a virtual environment that adds this development version of SciPy to the Python path

in Ubuntu. (Tested on 16.04, 18.04, and 20.04). Users running Windows can follow these instructions after setting up Windows Subsystem for Linux or an Amazon EC2 instance with Ubuntu 20.04. However, the instructions for setting up a development environment with Docker may be more reliable.

Note

This guide does not present the only way to set up a development environment; there are many valid choices of Python distribution, C/Fortran compiler, and installation options. The steps here can often be adapted for other choices, but we cannot provide documentation tailored for them.

This guide assumes that you are starting without an existing Python 3 installation, and that you are using pip. If you already have Python 3, you might want to uninstall it first to avoid ambiguity over which Python version is being used at the command line.

Building SciPy#

With pip#

Start with installing pip:

sudo apt install -y python3-pip

All further work should proceed in a virtual environment. Popular options include the standard library venv module or a separate virtualenv package. There are muliple third-party tutorials on how to set up a virtual environment, so we cover only briefly these two options here.

Note

We repeat: all work should happen in a virtual environment. Never use sudo pip.

Using virtualenv#

Install the virtualenvwrapper package:

python3 -m pip install virtualenvwrapper --user

Edit the .bashrc file to add some environment variables which are used internally by the virtualenvwrapper:

export WORKON_HOME=$HOME/virtualenvs
export VIRTUALENVWRAPPER_PYTHON=/usr/bin/python3
. $HOME/.local/bin/virtualenvwrapper.sh

Here we store the virtualenvs in a virtualenvs folder in the home directory. (you might need to create the folder manually).

Now open a new terminal window for the changes to the .bashrc to take effect.

Create a new virtual environment and activate it:

mkvirtualenv scipy-dev

Your command prompt now lists the name of your new environment, like so (scipy-dev)$. This means that the environment is active. If it is not, activate it manually with:

workon scipy-dev

Note mkvirtualenv and workon commands come from the virtualwrapper package.

Using the standard-library venv package#

Install the venv package:

sudo apt install -y python3-venv

Change the directory to your home folder and create a directory .venvs there. Create the virtualenvironment:

python3 -m venv scipy-dev

To activate the environment, use

source $HOME/.venvs/scipy-dev/bin/activate

Your command prompt now lists the name of your new environment, like so (scipy-dev)$.

(For the official docs for the venv package see https://docs.python.org/3/tutorial/venv.html).

Building SciPy#

Inside the scipy-dev environment, install the python-level dependencies:

python -m pip install numpy pytest cython pythran pybind11 meson ninja

Note that when the virtual environment is active, the system-wide names pip3 and python3 are aliased to pip and python, respectively.

Now that you have all external dependencies, navigate to the directory where you cloned the source code into. Download the submodules:

git submodule update --init

Finally, build SciPy and test it (this takes a while; subsequent rebuilds will be much faster):

python dev.py

With conda#

  1. Download, install, and test the latest release of the Anaconda Distribution of Python. In addition to the latest version of Python 3, the Anaconda Distribution includes dozens of the most popular Python packages for scientific computing, the conda package manager, and tools for managing virtual environments.

    If you’re installing using the terminal, be sure to follow the “Next Steps” listed after the installer finishes. You might also need to restart your terminal window or enter source ~/.bashrc for all the changes to take effect.

  2. (Optional) In a terminal window, enter conda list.
    This shows a list of all the Python packages that came with the Anaconda Distribution of Python. Note the latest released version of SciPy is among them; this is not the development version you are going to build and will be able to modify.

    Ideally, we’d like to have both versions, and we’d like to be able to switch between the two as needed. Virtual environments can do just that. With a few keystrokes in the terminal or even the click of an icon, we can enable or disable our development version. Let’s set that up.

    Note

    If conda is not a recognized command, try restarting your terminal. If

    it is still not recognized, please see “Should I add Anaconda to the macOS or Linux PATH?” in the Anaconda FAQ.

  3. Enter conda config --env --add channels conda-forge to tell Anaconda the source we want for our packages. Then enter conda create --name scipy-dev python=3.9 numpy pybind11 cython pythran pytest gfortran gxx sphinx pydata-sphinx-theme sphinx-panels matplotlib mypy git.
    This tells conda to create a virtual environment named scipy-dev (or another name that you prefer) with several packages.

    • numpy pybind11 cython pythran are four packages that SciPy depends on.

    • gfortran gxx are compilers used to build SciPy’s Fortran, C, and C++ source code.

    • pytest is needed for running the test suite.

    • sphinx, pydata-sphinx-theme, sphinx-panels and matplotlib are required to render the SciPy documentation.

    • mypy is a static type checker for Python. Consider using it.

    • git is a version control system used to download and manage the SciPy source code.

    Note that we’re installing SciPy’s build dependencies and some other software, but not SciPy itself.

    Note

    You could conda create an empty virtual environment first, then conda install the packages, but creating the virtual environment with all the packages you need is preferable to installing packages individually because it makes it easier for conda to solve the package dependencies optimally.

  4. You’re still in the base environment. Activate your new virtual environment by entering conda activate scipy-dev.
    If you’re working with an old version of conda, you might need to type source activate scipy-dev instead (see here). Note that you’ll need to have this virtual environment active whenever you want to work with the development version of SciPy.

  5. Browse to the SciPy repository on GitHub and create your own fork. You’ll need to create a GitHub account if you don’t already have one.

  6. Browse to your fork. Your fork will have a URL like https://github.com/mdhaber/scipy, except with your GitHub username in place of “mdhaber”.

  7. Click the big, green “Clone or download” button, and copy the “.git” URL to the clipboard. The URL will be the same as your fork’s URL, except it will end in “.git”.

  8. Create a folder for the SciPy source code in a convenient place on your computer. Navigate to it in the terminal.

  9. Enter the command git clone followed by your fork’s .git URL. Note that this creates in the terminal’s working directory a scipy folder containing the SciPy source code.

  10. In the terminal, navigate into the scipy root directory (e.g. cd scipy).

  11. Initialize git submodules: git submodule update --init.

  12. Do an in-place build: enter python3 setup.py build_ext --inplace.
    This will compile the C, C++, and Fortran code that comes with SciPy. We installed python3 with Anaconda. setup.py is a script in the root directory of SciPy, which is why you have to be in the SciPy root directory to call it. build_ext is a command defined in setup.py, and --inplace is an option we’ll use to ensure that the compiling happens in the SciPy directory you already have rather than the default location for Python packages. By building in-place, you avoid having to re-build SciPy before you can test changes to the Python code.

  13. Test the build: enter python3 runtests.py -v. runtests.py is another script in the SciPy root directory. It runs a suite of tests that make sure SciPy is working as it should, and -v activates the --verbose option to show all the test output. If the tests are successful, you now have a working development build of SciPy! You could stop here, but you would only be able to use this development build when the Python working directory is the SciPy root directory.

  14. Enter conda develop ., where . refers to the present directory.
    This will allow us to import the development version of SciPy in Python regardless of Python’s working directory.

  15. In a new terminal window, test your setup. If you activate your virtual environment (e.g. conda activate scipy-dev) and run Python code that imports from SciPy, any changes you make to the SciPy code should be reflected when the code runs. After deactivating the virtual environment (conda deactivate), Python imports from the version of SciPy installed by Anaconda. You can also check which version of SciPy you’re using by executing in Python:

    import scipy
    print(scipy.__version__)
    

    If you have successfully imported a development version of SciPy, the word dev will appear in the output, e.g.:

    1.6.0.dev0+be97f1a