Hello 👋
I'm a maintainer of BDD testing tool. One of the popular requests I'm getting from consumers - to allow duplicate step definitions bound to different features. For example, I need to test an application that has "game" and "video-player" pages. Both pages have PLAY button in the interface. I write two scenarios:
game.feature
Given I have not started a game yet
When I click the PLAY button # <- duplicated step
Then the game begins
video-player.feature
Given I am watching a youtube video
When I click the PLAY button # <- duplicated step
Then the video plays
Step implementation for I click the PLAY button
is different for each feature.
Official Cucumber docs says it's an anti-pattern and step definitions should be globally unique. Proposed workaround is to modify step pattern to avoid the ambiguity. E.g.:
When('I click the PLAY button in game', ...);
When('I click the PLAY button in video player', ...);
That's annoying!
Existing solutions
Non of official Cucumber implementations supports duplicate steps. They report ambiguous step error once your step matches more than one step definition.
Cucumber plugin for Cypress introduced interesting feature called Paring. It allows to have duplicate step definitions paired to particular features via special configuration pattern. For example, having a files structure:
└── features/
├── steps/
│ ├── common.ts
│ ├── game.ts
│ └── video-player.ts
├── game.feature
└── video-player.feature
I can configure step paths with special keyword [filepath]
:
stepDefinitions: [
'features/steps/common.ts',
'features/steps/[filepath].ts', // <- pair steps to particular feature
]
During steps loading, [filepath]
will be replaced with actual feature name and these steps will be paired to the feature. Now it is possible to have separate step definitions I click the PLAY button
for "page" and "video-player".
Drawbacks
Although I like that pairing technique, I see two drawbacks:
You can't just define steps as a single string pattern, see a common mistake. You should make it more complex, splitting on common steps + pairing pattern steps.
Pairing can't be resolved without reading the configuration. That is mostly for tools like IDE extensions, for navigating to step definition by
cmd + click
. Currently, the most popular one does not support it, but hopefully will.
Proposed solution
While thinking about steps pairing in Cypress plugin, I've got another idea how it can be implemented. The solution is inspired by Next.js route groups.
We can introduce steps scope - a file or directory with name in parenthesis, e.g. (game)
or (video-player)
.
Step definitions inside scoped directory are applicable only to features inside that directory.
This is the only rule one should know to understand the approach.
Now we can define the file structure:
└── features/
├── steps/
│ └── common.ts
├── (game)/
│ ├── game.feature
│ └── steps.ts
└── (video-player)/
├── video-player.feature
└── steps.ts
-
(game)/steps.ts
are applied only togame.feature
-
(video-player)/steps.ts
are applied only tovideo-player.feature
-
steps/common.ts
are applied to both
The main advantage is that any tool or human can understand paring without reading configuration.
The configuration itself simply defines steps glob without any patterns:
stepDefinitions: 'features/**/*.ts'
Some projects have separate directories for features and steps. For such cases, the rule can be slightly enhanced:
Scoped step definitions are applicable only to features having that scope in the path.
Now the following structure is also possible:
└── features/
├── steps/
│ ├── common.ts
│ ├── (game).ts
│ └── (video-player).ts
├── (game).feature
└── (video-player).feature
- steps from
steps/(game).ts
will be applied only to(game).feature
, because feature path contains(game)
- steps from
steps/(video-player).ts
will be applied only to(video-player).feature
, because feature path contains(video-player)
- steps from
steps/common.ts
will be applied to both features, because there are no scoped directories in steps path
Such file structure explicitly shows how features are connected to steps.
Conclusion
I think, scoped duplicate steps are reasonable, especially for testing large applications. I haven't seen file-based solutions before and would appreciate any feedback from you. All of you have different projects with unique structure. Feel free to share, how that solution matches your setup.
Thanks in advance and happy testing ❤️
Top comments (2)
I don't think the concept should hold true.. cucumber was initially started as documentation/runner. We are trying to say that its purely a technical tool - most people would disagree if so.
Primarily gherkin serves as a easy way to describe steps and provide clarity, but also to replicate scenarios .. as well as orchestrate steps and provide data. Never solely to do efficient test creation.
We are breaking 2 paradigms - requirements clarity + accuracy of organization - by saying same description applies everywhere.Thats not how gherkin itself should be used.
How can I actually treat this..
-- GAME/Video second parameter just place holders. no real value.Clarifiers sort of..
and put them in a folder called common steps and import in feature wise steps. Step implementation and even lines of code will be common ..
So you mean the better approach is to define common step like this?
Isn't it more difficult to maintain such common function for all page?