Python Design

Functions

• Do one thing and do it well
  • Should be at the same level of abstraction
• Command-query separation: should either make some change or return data
• Only request information it needs
• Keep the number of paramters minimal
  • An indication of how focused the function is
• Don't create an use an object in the same place
  • Use dependency injection
• Don't use flag arguments
  • can mean the funcrion should be two distinct functions to handle each case
• Functions are objects, so can be passed around

Classes

• Keep classes small
  • Typically either data focused (structure) or behaviour focussed (methods)
• Make classes east to use as possible
  • eg properties instead of methods
  • use @property, @cached_property or @lru_cache
  • __repr__ for developers, __str__ for users
• Use Dependency Injection
  • Don't create objects in the class, pass as arguments
• Make sure a class is really needed
  • Classes are useful for multiple instances. If only one is needed, could be just a module function.

Protocols

• Classes derived from Protcol define an interface with their attributes
• Duck typing means another type of object that has at least the same attributes as the interface can be passed to it
• Alternative to subclassing