Write BDD tests in Markdown.

Writing BDD tests should be more comfortable than this, so why not use Markdown? It can look like this.

• it is a well supported document format, many tools like auto-formatters already exist
• it provide good tools to structure a hierarchical document
• it has support for embedding source code / JSON payloads, and even tables
• front matter can be used for feature-level configuration

Work on the original BDD e2e feature runner began in 2018, and the project has been proved very useful for testing cloud-native solutions. Read more about the original idea here. However, the implementation had some shortcomings. Especially understanding test results and the way state and retries were handled was not optimal. In addition was the old codebase itself not sufficiently covered with tests. Therefore this project was initiated in 2022, with four years of experience authoring and running tests. With a fresh set of eyes, the way to write test was complete changed from Gherkin to Markdown which called for releasing it as a standalone project.

• Demo of supported syntax
• Gherkin Rule keyword
• Mars Rover Kata (this demonstrates the Soon keyword which retries steps)
  Run: $(set -o pipefail && npx tsx examples/mars-rover/tests.ts | npx tsx reporter/console-cli.ts)
• Firmware UART log assertions (this demonstrates the use of the Context, which is a global object available to provide run-time settings to the test run, which replace placeholders in step titles and codeblocks.)
  Run: $(set -o pipefail && npx tsx examples/firmware/tests.ts | npx tsx reporter/console-cli.ts)

Let's have a look at this scenario:

# To Do List

## Create a new todo list item

Given I create a new task named `My item`

Then the list of tasks should contain `My item`

What if you are testing a todo list system, that is eventually consistent?

More specifically: creating a new task happens through a POST request to an API that returns a 202 Accepted status code.

The system does not guarantee that task you've just created is immediately available.

The Then assertion will fail, because it is executed immediately.

For testing eventual consistent systems, we need to either wait a reasonable enough time or retry the assertion.

However, if there are many similar assertions in your test suite will quickly add up to long run times.

Therefore the most efficient solution is to retry the assertion until it passes, or times out. This way a back-off algorithm can be used to wait increasing longer times and many tries during the test run will have the least amount of impact on the run time.

Implementing the appropriate way of retrying is left to the implementing step, however you are encourage to mark these eventual consisted steps using the Soon keyword.

By default the features are loaded in no particular order. You may attempt to order them using a naming convention, however this can enforce a forced ranking of all features, and over time files might need to get renamed to make room for new features.

In this project, features can specify a dependency to one or more other features in their front matter, and after parsing all features files, they will be sorted topologically.

Features can define their dependencies via the needs keyword:

---
needs:
  - First feature
---

# Second

## Scenario

Given this is the first step

This feature will be run after a feature with the name First feature

In addition, features can specify whether they should be run before all other features, or after all. Multiple keywords can have this flag, but dependencies will take precedence.

---
order: first
---

# Runs before all others

## Scenario

Given this is the first step

---
order: last
---

# Runs before all others

## Scenario

Given this is the first step

Features can be skipped, this will also skip all dependent and transiently dependent features.

---
run: never
---

# This feature never runs

## Scenario

Given this is the first step

Features can be run exclusively, this will also run all dependent and transiently dependent features. All other features not marked as run: only will be skipped.

---
run: only
---

# This feature runs, all other features are skipped

## Scenario

Given this is the first step

Variants (defined in the frontmatter of a feature) can be used to run the same feature in different variants. For every entry in variants, the feature file is run. (Example)

For JSON code-blocks there is a special notation to replace number placeholders, while still maintaining the JSON syntax and allow for formatters like prettier to format the code-block.

Given `v` is the number `42`

And I store this in `result`

```json
{ "foo": "$number{v}" }
```

Then `result` should match

```json
{ "foo": 42 }
```

It includes a markdown reporter, which will turn the suite result into markdown, suitable for displaying it as GitHub Actions job summaries.

Example: Mars Rover Report