The WhatsHap source code is on Bitbucket. WhatsHap is developed in Python 3, Cython and C++.

Development installation

For development, make sure that you install Cython and tox. We also recommend using a virtualenv. This sequence of commands should work:

git clone
cd whatshap
python3 -m venv venv
source venv/bin/activate
pip install -e .[dev]

The last command installs also all the development dependencies, such as Cython. Use only pip install -e . to omit those.

Next, you can run WhatsHap like this:

whatshap --help

Development installation when using Conda

If you are using Bioconda, it is convenient to develop WhatsHap in a separate environment:

conda create -n whatshap-dev python=3.6 pysam PyVCF pyfaidx xopen Cython pytest sphinx-issues
source activate whatshap-dev
git clone
cd whatshap
pip install -e .

The last command installs WhatsHap into your Conda environment named whatshap-dev. So when executing whatshap you will run the latest version you just cloned.

Running tests

While in the virtual environment, you can run the tests for the current Python version like this:


Whenever you change any Cython code (.pyx files), you need to re-run the pip install -e . step in order to compile it.

Optionally, to run tests for all supported Python versions, you can run tox. It creates separate virtual environments for each Python version, installs WhatsHap into it and then runs the tests. It also tests documentation generation with sphinx. Run it like this:


If tox is installed on the system, you do not need to be inside a virtual environment for this. However, you need to have all tested Python versions installed on the system! See the instructions below for how to do this on Ubuntu.

Installing other Python versions in Ubuntu

Ubuntu comes with one default Python 3 version, and in order to test WhatsHap with older or newer Python versions, follow the instructions for enabling the “deadsnakes” repository. After you have done so, ensure you have the following packages:

sudo apt install build-essential python-software-properties

Then get and install the desired Python versions. Make sure you install the -dev package. For example, for Python 3.4:

sudo apt update
sudo apt install python3.4-dev


Here is one way to get a backtrace from gdb (assuming the bug occurs while running the tests):

$ gdb python3
(gdb) run -m pytest

After you get a SIGSEGV, let gdb print a backtrace:

(gdb) bt

Wrapping C++ classes

The WhatsHap phasing algorithm is written in C++, as are many of the core data structures such as the “Read” class. To make the C++ classes usable from Python, we use Cython to wrap the classes. All these definitions are spread across multiple files. To add new attributes or methods to an existing class or to add a new class, changes need to be made in different places.

Let us look at the “Read” class. The following places in the code may need to be changed if the Read class is changed or extended:

  • src/read.cpp: Implementation of the class (C++).

  • src/read.h: Header with the class declaration (also normal C++).

  • whatshap/cpp.pxd: Cython declarations of the class. This repeats – using the Cython syntax this time – a subset of the information from the src/read.h file. This duplication is required because Cython cannot read .h files (it would need a full C++ parser for that).

    Note that the cpp.pxd file contains definitions for all the .h headers. (It would be cleaner to have them in separate .pxd files, but this leads to problems when linking the compiled files.)

  • whatshap/core.pxd: This contains declarations of all Cython classes wrapping C++ classes. Note that the class Read in this file has the same name as the C++ class, but that it is not the same as the C++ one! The distinction is made by prefixing the C++ class with cpp., which is the name of the module in which it is declared in (that is, the C++ class Read is declared in cpp.pxd). The wrapping (Cython) class Read stores the C++ class in an attribute named thisptr. If you add a new class, it needs to be added to this file. If you only modify an existing one, you probably do not need to change this file.

  • whatshap/core.pyx: The Cython implementation of the wrapper classes. Again, the name Read by itself is the Python wrapper class and cpp.Read is the name for the C++ class.

Before adding yet more C++ code, which then requires extra code for wrapping it, consider writing an implementation in Cython instead. See readselect.pyx, for example, which started out as a Python module and was then transferred to Cython to make it faster. Here, the Cython code is not merely a wrapper, but contains the implementation itself.

Writing documentation

Documentation is located in the doc/ subdirectory. It is written in reStructuredText format and is translated by Sphinx into HTML format.

Documentation is hosted on Read the Docs. In theory, it is built automatically whenever a commit is made. The documentation in the master branch should be visible at and documentation for the most recent released version should be visible at However, the connection between Bitbucket and Read the Docs has never worked well in this particular project, so builds actually need to be triggered manually until we have solved this problem.

To generate documentation locally, ensure that you installed sphinx and add-ons necessary to build documantation (running pip install -e .[dev] will take care of this). Then go into the doc/ directory and run make. You can then open doc/_build/html/index.html in your browser. The theme that is used is a bit different from the one the Read the Docs uses.

Making a release

If this is the first time you attempt to upload a distribution to PyPI, create a configuration file named .pypirc in your home directory with the following contents:

index-servers =


See also this blog post about getting started with PyPI. In particular, note that a % in your password needs to be doubled and that the password must not be put between quotation marks even if it contains spaces.

  1. Set the correct version number in the changelog. Ensure that the list of changes is up-to-date.

  2. Ensure you have no uncommitted changes in the working copy.

  3. Run tox, ensuring all tests pass.

  4. Tag the current commit with the version number (there must be a v prefix):

    git tag v0.1
  5. Create a distribution (.tar.gz file), ensuring that the auto-generated version number in the tarball is as you expect it:

    python3 sdist
  6. Build wheels (.whl files). Docker needs to be installed for this:

  7. Upload the distribution to PyPI:

    twine upload dist/whatshap-x.yz{.tar.gz,-*.whl}

    You may need to install the twine tool to run this command.

  8. Push the tag:

    git push --tags
  9. Update the bioconda recipe. It is probly easiest to edit the recipe via the web interface and send in a pull request. Ensure that the list of dependencies (the requirements: section in the recipe) is in sync with the file.

    Since this is just a version bump, the pull request does not need a review by other bioconda developers. As soon as the tests pass and if you have the proper permissions, it can be merged directly.

If something went wrong, fix the problem and follow the above instructions again, but with an incremented revision in the version number. That is, go from version x.y to x.y.1. Do not change a version that has already been uploaded.

Adding a new subcommand

Follow the instructions in whatshap/