Awesome

Cucumber Compatibility Kit

The Cucumber Compatibility Kit (CCK) aims to validate a Cucumber implementation's support for the Cucumber Messages protocol.

For example, cucumber-ruby and cucumber-js are using it to make sure they emit well-formed Messages when executing the features from the kit.

Overview

The kit is composed of features, step definitions, and messages:

features, once executed, emit an exhaustive set of Messages as specified by the protocol
step definitions that are used in conjunction with the features to generate the Messages, and also serve as reference step definitions for implementations adopting the kit
Messages, serialized as .ndjson files, are the reference: a given feature from the kit, executed using its dedicated step definitions, must emit the corresponding Messages

The Messages from the kit are generated using fake-cucumber, which is used here as a reference implementation of the Messages protocol.

Adoption

To adopt the compatibility kit when writing a Cucumber implementation, an implementer must have some prerequisites, do some setup in the implementation project and then write a dynamic test that is fed by the kit.

Prerequisites

First, make sure your implementation is sufficiently testable - you should be able to execute a test run programatically for a given set of feature files and step definitions, and obtain the Messages output (via a formatter or some other means) to make assertions on.

Setup

Pull in the latest version of the CCK package for your platform. If there isn't one yet, consider just copying over the files you need from the devkit for now - you can figure out the publishing later once it's all working.

Next, for each feature in the CCK suite, write step definitions that work with your implementation - these should be equivalent to the reference step definitions from the devkit, but with the appropriate language, syntax etc.

NB: For some flavours of cucumber, how you write code to do certain actions can vary heavily! Please ensure that the steps are equivalent!

Testing

Now you can build your CCK test. For each feature in the CCK suite, it should:

Execute a test run, scoped to:

Only the feature file in question
Only the step definition file that you wrote for this feature

Capture the Messages output of that test run
Load the reference messages from the .ndjson file and assert that the output you captured matches it

Notes and exceptions

When we say "matches" above, that's heavily qualified - there are many individual fields that will vary from run to run, like generated identifiers, timestamps, durations and file URLs. You can freely exclude these fields from your assertion - the key thing to capture is the number, type and order<sup>*</sup> of the messages, and the content that would be consistent across runs. It's also not unheard of for implementations to fix up the order of messages, so it matches the CCK, although it's best to try to match it naturally if you can.

Bear in mind that if you, for example, omit or miss-type a step definition, the failure you get might be non-obvious e.g. a discrepancy between actual and expected messages. Many of the features in the CCK represent a test run that fails, so this is not something that should cause your test to fail, although you might find it useful to capture the stderr (or equivalent) output for debugging purposes.

A small minority of features also have an .arguments.txt that specify additional arguments/options that are pertinent to the feature - see retry for an example. You can adapt these how you see fit - they may or may not line up directly with your CLI options schema.

There may be some features in the CCK suite that cover functionality that your implementation doesn't have. For example, there's one for retry which only a subset of Cucumber implementations support. You can filter these out of your test until/unless you're ready to support them.

* Ordering

While the samples provide a strictly ordered set of messages. Messages themselves are only partially ordered. But this is rather hard to capture with approval testing.

For example when executing scenarios in parallel, the messages from different scenarios are allowed to interleave while still retaining their order for each scenario. Likewise, feature files may be processed in parallel. So while for a single feature file, the derived Source, GherkinDocument, Pickles, TestCase and TestStep, ect, ect may come in order, these may interleave with the same messages from another feature file.

Finally, some orderings are not specified at all. While TestRunStarted is typically the first message, it could also be StepDefinition, Source or another message without dependencies.

So at best we have the following guidelines:

{TestRun,TestCase,TestStep}Started messages are emitted before their {TestRun,TestCase,TestStep}Finished counterparts.
TestCase{Started,Finished} are bookended by TestRun{Started,Finished}.
TestStep{Started,Finished} are bookended by TestCase{Started,Finished}.
Attachment messages are bookended by TestStep{Started,Finished}.
In general, prior to emitting a message all messages it references or depends on should have been emitted.

When in doubt, keep in mind that that applications should be able to process the events in a streaming manner.