Coding Conventions
Overall we take a pragmatic approach to coding conventions, trying to adhere to the practices set forth by the Django, Python, Java and Android communities.
PEP8
Code formatting, variable naming and layout should adhere to the PEP8 standard.
DRY - Do not Repeat Yourself
Code blocks should exist once and only once in the system, repetition should be avoided at all costs.
KISS - Keep It Simple Stupid
Overall the goal of code should be to communicate it’s intent first and foremost, optimizations for speed and or brevity should only be taken when absolutely necessary. In a system as complex as RapidPro it is critically important to only add complexity when absolutely necessary.
Naming Conventions
In general we attempt to keep the names of models and objects in the code the same as what is represented to the users, even if that requires significant refactors and renames as time goes on. This greatly reduces the cognitive load when trying to understand portions of the codebase. Names should err on the side of clarify over brevity.
Locality of Code
In general, try to keep related functionality together if at all possible. The bias should be towards having related views and forms together as opposed to grouping all forms and views together by type. This has led to some large files in the codebase which need to be refactored into smaller components of related functionality but this is still preferred over having one file containing all forms etc..
Convention over Configuration
RapidPro makes heavy use of Smartmin and as such heavily relies on convention over configuration. The use of CRUDL objects, automatic permission and URL naming greatly reduces the complexity of the system. As opportunity arises to create new conventions that yield similar simplifications, we should adopt them.
Code Reviews & Pull Requests
All committed code must be code reviewed via a formal pull request. Whether it is new feature work or bug fixes, all code should be written on a branch apart from master, then a pull request should be opened to review those changes before committing to master. Any code on the master branch is considered ready for live deployment.
Test Coverage
RapidPro is built as a hosted platform that continues to evolve on a daily basis. The two largest deployments of RapidPro deploy new software to their live servers two to three times per week. This is only possible due to having a very high coverage rate for our unit tests. All new functionality should have associated unit tests with full coverage and all bug fixes need to have an associated unit test demonstration the failure. Though this can feel like a tiring policy, it has allowed us to continue to evolve the platform and perform the necessary large refactors with confidence that the system remain stable.