Code should avoid being "too clever." This means things like avoiding clever abuses of language (for instance, utilizing lack of closures in python to remove the need for a variable) and algorithms that cut corners in the interest of marginal time savings.
In general, we want to balance all needs for performance with needs for clarity and maintainability. Readability and maintainability are the key functionalities we need to emphasize, as this will ensure the code can remain useful beyond the original scope of the project.
Tests should be considered mandatory for new functionality and bug fixes to lab software. These tests should be part of automated testing suites.
For Python code, we use black to automatically format code. Let it do it's job and don't worry about formatting.
Documentation must be added with any new set of code, and should be documented in a way that is accessible to newcomers to a given project. Where possible, worked examples should be provided.
Comments should be considered encouraged, but not necessary. Code should be as self-describing as possible; where complex optimizations are present, comments should enable an individual to understand shortcuts and reasoning.
Whenever possible, citations to algorithms (whether from papers or URLs such as Stack Overflow) must be included in comments close to that section of the code. (Note that in the specific case of Stack Overflow, code and answers are subject to licensing restrictions, which in practice will likely not impact their utilization but which must be acknowledged.)
Whenever developing in Python, docstrings should be added to all public functions. These must at a minimum cover all parameters and expected output, along with a description of the function and citation to methods implemented.
All code generated in the lab should be open-source with a permissive (non-copyleft) license (MIT, BSD, etc). If contributions to upstream copyleft projects are made, those contributions can be licensed in accordance with the upstream project license.
Importantly, in the lab we should cultivate an atmosphere of respect for licensing terms and ensuring that we are at all times in compliance with those terms. Our preferred license is BSD 3-clause. For further discussion see Stodden (2008).
In general, the lab uses Python for most of its software. This can include web apps as well as analysis and visualization software like Fatiando a Terra. Experimenting with Julia and other languages is of course acceptable and even desirable. For optimized code that needs to be fast, the first preference is Numba, then Cython, and finally C.
In general, using external code in lab projects is a balance between a few factors. If a large external project provides a small amount of functionality that would be used in a lab project which is to be distributed widely, the difficulty of the external library to install and keep up with (i.e., API changes) must be accounted for versus the difficulty of either "vendoring" the functionality (license-permitting) or re-implementing. In general we want to use external code for functions that are either core functionality of our projects (numpy, lapack, etc) and for difficulty but non-core functions of our projects.
If the dependency is a small, lightweight dependency, typically it's better to utilize it than to re-implement the functionality. But, for difficult-to-install libraries or libraries where utilized functionality outweighs the overhead of installation, it's better to bring it up for a discussion.
When software is developed as a result of funded activity, the leadership directions will be set collaboratively during the proposal writing process and then subsequently between lab members and Leo. For projects which are externally developed, leadership for the project as a whole will be conducted 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", which can be conducted via the Slack chat.