Software produced by the lab

The following are guidelines to be followed when producing software tools as part of your work at the lab.

Licensing

We value respect for licensing terms and ensuring that we comply with them at all times.
All code generated in the lab should be open-source with a permissive (non-copyleft) license (MIT, BSD, etc).
Our preferred license is BSD 3-clause.
If contributions to upstream copyleft (GPL, etc) projects are made, those contributions can be licensed in accordance with the upstream project license.
For further discussion see Stodden (2008).

When software is developed as a result of funded activity, the leadership directions will be set collaboratively during the proposal writing process and between lab members and Leo.
For externally developed projects, leadership is as decided by the broader community.
Decisions regarding strategic direction for software projects (including algorithmic strategies) will be made collaboratively.
For technical directions on lab-developed software, we operate by “lazy consensus”.

We strive to be good “citizens” of the broader open-source community.
We encourage lab members to contribute upstream to open-source projects.
Projects developed with external collaborators (e.g., Fatiando a Terra) should have their discussions held openly. This can help discourage the notion that projects developed by lab members are not community projects.
For more discussion about this topic, see Matt Turk’s paper Scaling a Code in the Human Dimension.

Use Python for most software.
Experimenting with Julia and other languages is acceptable and even desirable.
For code that needs to be optimized, the first preference is (in order): Numba, then Cython, and finally C.

Code should avoid being “too clever”, for example clever abuses of languages and algorithms that cut corners in the interest of marginal time savings.
Clarity and maintainability take precedence over performance to ensure that the code can remain useful beyond the original scope of the project.
Tests should be considered mandatory for new functionality and bug fixes. These tests should be part of automated testing suites.
Use black to automatically format Python code.
Documentation must be added with any new code and should be accessible to newcomers. Include examples whenever possible.
Comments (not including markdown cells in notebooks) are encouraged but not necessary. Code should be as self-describing as possible. Where complexity is inevitable, comments should allow anyone to understand the shortcuts and reasoning.
Include citations to algorithms whenever possible, including links to online resources.
Docstrings should be added to all functions and classes in Python code.