#[typestate]

This library provides developers with a macro to design typestated objects.

[dependencies]
typestate = "0.9.0-rc2"

Compiler support: requires rustc 1.42+

• Introduction
  • Typestates in Rust
• Basic Guide
• Advanced Guide
  • Self-transitioning functions
  • Non-deterministic transitions
• Attributes
• Features
• Typestate Visualization
• Publications

Introduction

Are you frustrated with IllegalStateExceptions in Java?

Typestates allow you to define safe usage protocols for your objects. The compiler will help you on your journey and disallow errors on given states. You will no longer be able to try and read from closed streams.

#[typestate] builds on ideas from the state_machine_future crate. If typestates are so useful, why not use them with limit them to Futures?

Typestates in Rust

Typestates are not a new concept to Rust. There are several blog posts on the subject [1, 2, 3] as well as a chapter in The Embedded Rust Book.

In short, we can write typestates by hand, we add some generics here and there, declare them as a "state" and in the end we can keep living our lives with our new state machine.

This approach however is error-prone and verbose (especially with bigger automata). It also provides no guarantees about the automata, unless of course, you designed and tested the design previously.

As programmers, we want to automate this cumbersome job and to do so, we use Rust's powerful procedural macros!

Basic Guide

Consider we are tasked with building the firmware for a traffic light, we can turn it on and off and cycle between Green, Yellow and Red.

We first declare a module with the #[typestate] macro attached to it.

#[typestate]
mod traffic_light {}

This of course does nothing, in fact it will provide you an error, saying that we haven't declared an automaton.

And so, our next task is to do that. Inside our traffic_light module we declare a structure annotated with #[automaton].

#[automaton]
pub struct TrafficLight;

Our next step is to declare the states. We declare three empty structures annotated with "[state].

#[state] pub struct Green;
#[state] pub struct Yellow;
#[state] pub struct Red;

So far so good, however some errors should appear, regarding the lack of initial and final states.

To declare initial and final states we need to see them as describable by transitions. Whenever an object is created, the method that created leaves the object in the initial state. Equally, whenever a method consumes an object and does not return it (or a similar version of it), it made the object reach the final state.

With this in mind we can lay down the following rules:

• Functions that do not take a valid state (i.e. self) and return a valid state, describe an initial state.
• Functions that take a valid state (i.e. self) and do not return a valid state, describe a final state.

So we write the following function signatures:

fn turn_on() -> Red;
fn turn_off(self);

However, these are free functions, meaning that self relates to nothing. To attach them to a state we wrap them around a trait with the name of the state they are supposed to be attached to. So our previous example becomes:

trait Red {
    fn turn_on() -> Red;
    fn turn_off(self);
}

Before we go further, a quick review:

• The module is annotated with #[typestate] enabling the DSL.
• To declare the main automaton we attach #[automaton] to a structure.
• The states are declared by attaching #[state].
• State functions are declared through traits that share the same name.
• Initial and final states are declared by functions with a "special" signature.

Finally, we need to address how states transition between each other. An astute reader might have inferred that we can consume one state and return another, such reader would be 100% correct.

For example, to transition between the Red state and the Green we do:

trait Red {
    fn to_green(self) -> Green;
}

Building on this we can finish the other states:

pub trait Green {
    fn to_yellow(self) -> Yellow;
}

pub trait Yellow {
    fn to_red(self) -> Red;
}

pub trait Red {
    fn to_green(self) -> Green;
    fn turn_on() -> Red;
    fn turn_off(self);
}

And the full code becomes:

#[typestate]
mod traffic_light {
    #[automaton]
    pub struct TrafficLight {
        pub cycles: u64,
    }

    #[state] pub struct Green;
    #[state] pub struct Yellow;
    #[state] pub struct Red;

    pub trait Green {
        fn to_yellow(self) -> Yellow;
    }

    pub trait Yellow {
        fn to_red(self) -> Red;
    }

    pub trait Red {
        fn to_green(self) -> Green;
        fn turn_on() -> Red;
        fn turn_off(self);
    }
}

The code above will generate:

• Expand the main structure with a state: State field.
• A sealed trait which disallows states from being added externally.
• Traits for each state, providing the described functions.

Advanced Guide

There are some features which may be helpful when describing a typestate. There are two main features that weren't discussed yet.

Self-transitioning functions

Putting it simply, states may require to mutate themselves without transitioning, or maybe we require a simple getter. To declare methods for that purpose, we can use functions that take references (mutable or not) to self.

Consider the following example where we have a flag that can be up or not. We have two functions, one checks if the flag is up, the other, sets the flag up.

#[state] struct Flag {
    up: bool
}

impl Flag {
    fn is_up(&self) -> bool;
    fn set_up(&mut self);
}

As these functions do not change the typestate state, they transition back to the current state.

Non-deterministic transitions

Consider that a typestate relies on an external component that can fail, to model that, one would use Result<T>. However, we need our typestate to transition between known states, so we declare two things:

• An Error state along with the other states.
• An enum to represent the bifurcation of states.

#[state] struct Error {
    message: String
}

enum OperationResult {
    State, Error
}

Inside the enumeration there can only be other valid states and only Unit style variants are supported.

Attributes

This is the list of attributes that can be used along #[typestate]:

• #[typestate]: the main attribute macro, without attribute parameters.
• #[typestate(enumerate = "...")]: this option makes the macro generate an additional enum, the enum enables working with variables and structures "generic" to the state.
  • The parameter can be declared with or without a string literal, if declared with the string, that string will be used as identifier to the enum.
  • If the parameter is used with an empty string or without a string, the default behavior is to prepend an E to the
• #[typestate(state_constructors = "..."): this option generates basic constructors for states with fields.

Features

The cargo features you can enable:

• mermaid-docs will generate Mermaid.js state diagrams embedded in your documentation. This feature is set by default.
• export-dot will generate a .dot file of your state machine.
  • This feature can be customized through the following environment variables (taken from the DOT documentation):
    • DOT_PAD - Specifies how much, in inches, to extend the drawing area around the minimal area needed to draw the graph.
    • DOT_NODESEP - In dot, nodesep specifies the minimum space between two adjacent nodes in the same rank, in inches.
    • DOT_RANKSEP - In dot, sets the desired rank separation, in inches.
    • EXPORT_FOLDER - Declare the target folder for exported files.
• export-plantuml will generate a PlantUML state diagram (.uml file) of your state machine.
  • This feature can be customized through the following environment variables (taken from the PlantUML Hitchhiker's Guide)
    • PLANTUML_NODESEP - nodesep specifies the minimum space between two adjacent nodes in the same rank.
    • PLANTUML_RANKSEP - Sets the desired rank separation.
    • EXPORT_FOLDER - Declare the target folder for exported files.

Typestate Visualization

	LightBulb	SmartBulb
DOT
PlantUML

Publications

• Retrofitting Typestates into Rust (SBLP'21)