Design and coding standards for LambdaClass.
This will be quick and dirty list for now.
- make is the build tool. Consider these notes.
- Postgresql is the default database.
- Write tests.
- Favor integration tests over unit tests, but do write unit tests when the functions merit them.
- Do not write tests before you have solved the problem, you may waste time writing tests for the wrong implementation.
- Start with the simplest thing that could possibly work.
- Don't try to foresee future uses of your program: You aren't gonna need it.
- Write for humans: coding for computers is easy, but writing code that is understandable by another person is an art.
- Code and document in english, unless you have a very specific reason no to.
- Use meaningful, readable names for variables, functions and files. Don't try to save characters.
- Documentation is a sign of the quality of an API. It's easier to write it when the design is right.
- The less code you have the better. Deleted code is debugged code.
- Aim for simplicity, not performance. The latter is a by-product of the first.
- Only introduce optimizations if you have benchmarks that proof an improvement and that the improvement is relevant in the context of the program.
- Only introduce optimizations if they represent a concrete gain (e.g. cost savings, improved user experience).
- Follow the Zen of Python, regardless of the language you are using at the moment. English also counts as a language.
- Don't introduce dependencies prematurely. You must evaluate your requirements, maintenance and integration costs first.
- If you want to upgrade a dependency, test it first.
- Always lock your dependencies. Pin a specific version and a commit of a dependency, don't use the version at master.
- Use git and commit often, even in one-person projects.
- Use MIT license.
- Fill the
description
field at the top of the repo page. - Write a decent README.
- A good readme starts with a succint description (one or two sentences) and, when possible, a very short and illustrative example use. The rest of the details go after this header.
- Use continuous integration, most likely travis.org.
- Make a good balance of features vs maintenance. Maintenance details usually matter more than adding a lot of features.
- Use rebar3. Include the binary in the repository so it's not an external dependency and the tested version is used. rebar3 is not used directly but through make targets.
- Support the most recent Erlang version.
- When building libraries, try to make them both easily usable from the shell and easily configurable via application environment.
- Indent with two spaces.
- Avoid using header files (.hrl) [TODO ELABORATE].
- Supervised processes provide guarantees in their initialization phase, not a best effort. If you expect failure to happen on an external service, do not make its presence a guarantee of your system.
- Try to avoid
timer:sleep
on tests, ktn_task:wait_for_success can be a better option. More on this here. - Prefer maps to records.
- The Zen of Python is your bible.
- Use Python 3 in greenfield projects.
- Strive to migrate to Python 3 in non greenfield projects.
- Read this to understand environment hell in Python.
- Use pipenv to escape from environment hell in Python. You can even do this locally in projects that are set up to use virtualenv and virtualenvwrapper.
- If you also need to stick with a specific minor version of Python (e.g. Python 2.7.14) you can use pyenv in combination with pipenv.