5 releases (breaking)

0.5.0 May 14, 2024
0.4.0 Mar 7, 2024
0.3.0 Mar 7, 2024
0.2.0 Mar 5, 2024
0.1.0 Dec 19, 2023

#974 in Procedural macros

Download history 9/week @ 2024-02-25 360/week @ 2024-03-03 153/week @ 2024-03-10 4/week @ 2024-03-17 32/week @ 2024-03-31 141/week @ 2024-05-12 39/week @ 2024-05-19 10/week @ 2024-05-26 14/week @ 2024-06-02 6/week @ 2024-06-09

74 downloads per month
Used in narrative




An immensely simple library for story-driven development


WIP to the first release. Published to crates.io for reserve the name.


Narrative is a library dedicated to developing a whole or some part of software based on stories expressed in a Rust trait. Though its primary design is for end-to-end testing, its simplicity supports a variety of use cases.


  • Story-driven: Code respects story, not the other way around
  • Data-driven: Enabling stories to include structured data
  • No additional tooling: Eliminating the need for extra installation and learning
  • Leverage existing ecosystem: Rich experience with less implementation
  • Zero runtime cost: Stories are processed at compile time


Key terms in this library are:

  • Story: a sequence of steps, which is written as a trait
  • Step: a single action or assertion in a story
  • Story Trait: a macro-generated trait that represents a story. it has a method for each step.
  • Story Context: a struct that holds the all related information or data of a story.
  • Story Env: a data structure that implements a story trait.


  1. Add narrative to your cargo dependencies.

  2. Write your first story as a trait.

#[narrative::story("This is my first story")]
trait MyFirstStory {
    #[step("Hi, I'm a user")]
    fn as_a_user();
    #[step("I have an apple", count = 1)]
    fn have_one_apple(count: u32);
    #[step("I have {count} orages", count = 2)]
    fn have_two_oranges(count: u32);
    #[step("I should have {total} fruits", total = 3)]
    fn should_have_three_fruits(total: u32);

Wow, it's neat!

  1. Implement the story in Rust.
pub struct MyFirstStoryImpl {
    apples: u8,
    oranges: u8,

impl MyFirstStory for MyFirstStoryImpl {
    type Error = ();

    fn as_a_user(&mut self) -> Result<(), Self::Error> {
        println!("Hi, I'm a user");

    fn have_one_apple(&mut self, count: u32) -> Result<(), Self::Error> {
        self.apples = count;

    fn have_two_oranges(&mut self, count: u32) -> Result<(), Self::Error> {
        self.oranges = count;

    fn should_have_three_fruits(&mut self, total: u32) -> Result<(), Self::Error> {
        assert_eq!(self.apples + self.oranges, total);

You may notice that the signature of the trait methods is a bit different from the declaration, but it's fine.

  1. Use the story in your code.
fn main() {
    let mut story = MyFirstStory { apples: 0, oranges: 0 };
    // You can run the story, and get the result.
    let story_result = story.run_all();
    // You can run the story step by step.
    for step in story.get_context().steps() {
        let step_result = step.run();

Subtle but Important Points

There are several points that you should know to use Narrative.

Async one is also defined automatically

Story doesn't have to use async keyword, and both sync and async version are defined automatically.

impl AsyncMyFirstStory for MyFirstStoryImpl {
    type Error = ();
    async fn as_a_user(&mut self) -> Result<(), Self::Error> {
        println!("Hi, I'm a user");
    async fn have_one_apple(&mut self, count: u32) -> Result<(), Self::Error> {
        self.apples = count;
    async fn have_two_oranges(&mut self, count: u32) -> Result<(), Self::Error> {
        self.oranges = count;
    async fn should_have_three_fruits(&mut self, total: u32) -> Result<(), Self::Error> {
        assert_eq!(self.apples + self.oranges, total);

Arguments of the step methods cannot be data structures not defined in standard library

It makes your stories truely independent from any implementation.

But, you can use trait coupled to the exact story as arguments

Rust's type system gives us a power to write correct codes without loosing productivity, and it's the same in writing stories (in Narrative). To achieve the benefits without adding any dependency to the story, we can define new struct or trait that strongly coupled to only the story, and use it as an associated type of the story trait.

Don't worry about the collision of the trait/struct names, it has a separate namespace than other stories.

#[narrative::story("This is my first story")]
trait MyFirstStory {
    fn data() {
        struct UserName(String);
        trait UserId {
            /// Generate a new user id with random uuid v4.
            fn new_v4() -> Self;

    const user_id: UserId = Self::UserId::new_v4();

    #[step("I'm a user with id: {id}", id = user_id, name = UserName("Alice".to_string()))]
    fn as_a_user(id: Self::UserId, name: UserName);

It's really weird for who knows correct Rust syntax, but it's the better one among alternative ideas to do the same thing, defining a new struct or trait in the same place.

You can forget about the actual implementation of a story while writing a story

We think that stories should not regard their actual implementations, so noisy details like async, &self, &mut self, and -> Result<(), Self::Error> are not required in the story definition. This surprising behavior can be mitigated by using "Implement missing members" feature of rust-analyzer.

Design Decisions

These decisions highlight Narrative's unique aspects, especially in comparison to Gauge, a well-known end-to-end testing framework.

Narrative is designed to implement stories exclusively in Rust, (though it can still be used for testing projects in other languages.)

Supporting other languages in Narrative would introduce a lot of complexity in its design, implementation, and usage. Narrative leverages Rust's core functionality and rust-analyzer to provide rich development experience. Rust wouldn't the best language for writing end-to-end tests for everyone, but, We believe that it still has advantages in this area, with a great compiler, robust, yet straightforward type system, and libraries from the vibrant community.

Users can dynamically get story context, so you can implement steps in other programming languages, and call them in dynamic way from Rust code:

fn execute_story(context: impl narrative::StoryContext) {
    for step in context.steps() {
        send_to_external_process(step.text(), step.arguments().map(|arg| Argument {
                name: arg.name(),
                ty: arg.ty(),
                debug: arg.debug(),
                json: step.serialize(serde_json::value::Serializer).unwrap(),

Narrative is a library, not a framework

Narrative has no test runner, no plugin system, nor no dedicated language server. Instead of being a framework, Narrative is a library that provides just a single macro to implement stories. It's just a small tie between a story to a plain Rust code. So, users can compose their own test runners or async runtime with stories, and can use the full of rust-analyzer's functionality.

Narrative itself doesn't provide any features other than the core functionality, declaring stories as traits and implementing them in Rust code. It lays the groundwork for the simplicity and extensibility of this library.

The followings are the missing features in Narrative, and they never be implemented in this library. But don't forget that you can do them by leveraging the core features.

  • step grouping
  • story grouping
  • test preparation and cleanup
  • table-driven tests
  • tags
  • screenshoting
  • retrying
  • parallelization
  • error reporting

Narrative uses a declaration of trait to write a story

In other words, a story is an interface and step implementation depends on it.

Gauge uses markdown, and it's a great format for writing specifications, documents, and stories while readable by non programmer. But, it's not the best format for expressing data in structured way. We think story is more like a data than a document, and it should be expressed in a structured way. With structured data, we can leverage the power of software in the processing of them. In Narrative, we use traits for expressing stories.

Using markdown for stories has another benefit, that is, it avoids the tight coupling between stories and the implementation. If stories depends on specific implementation, the story is not pure, and we loose many benefits of story-driven development. One of the benefits is that we, including non-programmer, can write stories freely without regard to the implementation, and it gives us a kind of agility to the development.

But, it's not the case in Narrative though it let you write stories in Rust. In Narrative, stories are written as traits, and it has no dependency to the implementation, and it's just a contract between the story and the implementation. Narrative would not loose the benefits of using markdown, on the contray, it would make the situation better.

Narrative explicitly separates the story and the implementation, and it forces the direction of the dependency. With markdown, we know that a story is the core of the development, but occasionally we forget it or have a kind of cognitive dissonance. It appeared to us as obvious experiences in the development, like, "we need to know defined tags in the implementation to write a correct story", "we have errors on the story editor if no step implementation", or "we failed to write the correct story because the steps chosen from the editor's suggestion are not implemented as we expect". In narrative, anyone can write stories anytime, and stories written can exist as valid real properties with no error even if implementation are completely undone.

The concept, a story is a contract to the implementation, makes the development process and logical dependency graph clean and simple, and although it requires a bit more effort to implement stories, it would give us a lot of benefits in the long run of the development.

Someone might think that writing or reading Rust traits is impossible or impractical to non-programmer, but we think it more optimistically. We are in the era of many people can read and write code with the help of the great tools and AIs, and, Personally, I believes clear codes wins documentation both for programmers and non-programmers, and I don't think non-programmers cannot read and write codes.

Narrative encourages no reusing of steps

We encourage you to write your stories in fresh mind every time without reusing existing steps, because we think stories should be self-contained. Being the situation comes with big wins described below.

Accessibility for Newcomers

It empowers story writers that are not familiar with the existing codebase. They don't need to know what steps already exist, to struggle with what steps to use, and to worry about whether the chosen step is implemented as they expect.

Contextual Clarity

Copying steps from other stories often leads to a mix-up of contexts, and making it not easy to decipher the key point of a story (without attaching proper aliases to common steps). While we tend to have many story have the same steps that shares the same context and implementation, it's challenging to maintain the coherency of sharing the same logic while we add, remove, modify the stories.

One downside of this approach is that stories could have inconsistency in the writing style among them, but it can be mitigated by organizing stories in the near each other with have the same contexts. It nudges writers to write stories in a consistent way.


Reusing steps or group of steps could be a source of complexity. It's nightmare to modify a step that is used by many stories without breaking them.

Fine-Grained Abstraction

A step is relatively large a unit for reuse or abstraction. Instead of sharing the whole a step, we should share code between stories. But it should be done by extracting common, story-agnostic, and atomic unit of logic. A step implementation should be a composition of such units, and it should not leak the story's context in the abstraction. For instance, if a step is about clicking a submit button, it might be implemented as a composition of atomic logic like find_element_by(id), click(element), and wait_for_page_load(), and not to leak the context like click_submit_button() or click_button("#submit").


~18K SLoC