Cucumber

Cucumber executes Gherkin scenarios against the system, which makes the agreed specification and the regression suite the same artefact.

It exists in every stack this hub covers — Cucumber-JVM for Spring Boot, Quarkus, EAP and the Android domain module; cucumber-js for Angular and Node.

Where to attach it

At the use case port. Not at the browser, not at the HTTP controller, not at the database.

That single decision is what separates a Cucumber suite people keep from one they delete in year two. Attached to the port:

• the scenario exercises real business behaviour, with the adapters stubbed;
• it runs in milliseconds, so it can run on every push;
• it does not break when a button moves, because it never knew about the button.

The stubs behind those ports come from Microcks where the dependency is an API, so the scenario runs against the contract rather than against someone’s idea of it.

The two ways teams ruin it

Writing the scenarios alone. Gherkin written by one analyst after the conversation is transcription with a runtime cost. The scenarios have to come out of example mapping, where the room agreed on the examples — otherwise the collaboration benefit, which is the whole return, is never collected and only the maintenance burden arrives.

Driving the UI with it. Given I click the button with id submit-btn is a UI test wearing a costume. It is unreadable to the business, brittle for the developers, and slow for everyone. If a journey needs driving through the browser, that is Playwright’s job and it does not need Gherkin to do it.

Step reuse, and its limit

Reusing step definitions is good until a step grows its fourth optional parameter and a regular expression nobody can read. At that point the step library has become a private DSL with no documentation, and writing a new scenario requires archaeology.

The usable rule: steps are reused when the domain concept is the same, not when the sentence looks similar.

When not to use it

If nobody outside the development team ever opens the scenarios, Cucumber is costing a step-definition layer and buying nothing. Write the same examples as plain acceptance tests in the team’s own test framework and keep the conversation that produced them — the conversation was always the valuable half.

This is a real and common outcome, and choosing it is not a failure.