Skip to content

A pattern for creating highly compact and light-weight general purpose finite state machines.

License

Notifications You must be signed in to change notification settings

BL1NDX3N0N/Discreet-State-Pattern

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Discrete State Pattern

A pattern for creating highly compact and light-weight general purpose finite state machines.

Two Flavors

Two example implementations of the pattern have been included. Basic and Minimal. Please do not use the minimal implementation in a production environment, it is meant to be a clever example.

Design

Simplicity is key. The state machine is comprised of two components. A state machine manager and the states. The state machine manager can be thought of as a scheduler, it recieves requests made by the states and processes said requests. The requests made by the states are transitions, which can either be to the same state (effectively restarting it), or to a new state. States never interact with each other directly, and it is the job of the state machine manager to enfore this seperation.

Four Principles

The pattern consists of four principles, and it is important that you follow them when implementing this pattern.

1. States are responsible for deciding the next transition.

The state machine manager should never be responsible for deciding which state it should transition to. This removes the need for switches and enumerations. The only added exception is mentioned in principle #3.

2. State machine manager is responsible for making transitions.

States themselves should never be allowed to make transitions directly. Instead, states are to request which states they would like the state machine manager to transition to by returning a pointer to said state. This ensures that control is always transferred back to the state machine manager, thus for, relieving pressure on the call stack.

3. State machine manager is responsible for state machine shutdown.

States themselves should never be allowed to shutdown the state machine. If there is no state for a state to transition to then that state should relay that to the state machine manager. The state machine manager will then be responsible for deciding if the state machine should shutdown, restart, or transition to a fallback/error state.

4. Discreteness

The internals of the state machine manager should never be exposed nor should the state machine manager store states outside of its scope. The state machine manager should operate as a sealed unit. However, the state machine manager may utilize global data to determine a valid fallback/error state or if the state machine should be shutdown. States are also free to utilize global data as well.

Adhering to these principles will ensure high scalability and maintainability across all of your models. Additionally, with some ingenuity, this pattern can also be used for deep recursion and will never flood the call stack.

About

A pattern for creating highly compact and light-weight general purpose finite state machines.

Topics

Resources

License

Stars

Watchers

Forks

Languages