End-to-end functional tests that can run in milliseconds Having Our Cake & Eating It Nat Pryce

A familiar story...

We needed a new approach Developers responsible for all test automation QA responsible for manual, exploratory & qualitative testing Automated functional tests written in terms of the domain model, not the user interface

The Hexagonal architecture AKA Ports & Adaptors HTTP service f Adaptor layer maps between technical domain & application domain model, but performs no business logic.

The domain model can be tested in isolation Test In-memory storage

Invert the Hexagonal architecture to run the same tests at a wider scale HTTP service Test driver f -1 Test f

Virtual Machine Virtual Machine Virtual Machine Similarly, at yet wider scales... Test driver f -1 Test Load balancer f Firewall

Service cluster Service cluster Front-end cluster Front-end cluster Service cluster ...and wider Test driver (g∘f) -1 Test Front-end cluster Web browser f g

Data centre Data centre Data centre ...and wider Test driver Test f (g∘f) -1 g Web browser CDN

Scenario Turn the Screenplay metaphor up to eleven! Actor Role Tasks Domain Model Behind the scenes Visible in test code System under test Production System Interface Plays Acts upon Acts upon Creates Creates Performs Defined in terms of Defines

Example @Test fun `An editor can request an amendment`() { val scenario = theProduction.newScenario() val alice = scenario.newAuthor() val ed = scenario.newEditorialStaffMember() val submissionDetails = exampleSubmission.withSingleAuthor(alice) val initialSubmission = alice.hasSubmitted(submissionDetails) val amendment = ed.canRequestAmendmentOf(initialSubmission) alice.canReview(amendment, expectedContents = submissionDetails.withStatus(IN_PROGRESS)) alice.canUpdate(amendment, SetTitle("a new title"), SubmitToPeerReview) ed.canSeeInThePeerReviewSystem(amendment, status = RECEIVED) }

Some benefits

Can trade off speed against scale The same test code can run... ● Quickly against the domain model ● More slowly against a deployed system

Tests are fast to write ● Concise ● Type safe ● Null safe ● Play well with the IDE: navigation, auto-complete, refactoring, etc.

Tests are easy to read ● Concise ● No user-interface details ● No technical details

No duplication ● within test code ● between test & production code

Some challenges

1. Unfamiliar: How to approach TDD? Inside-out? Outside-in?

1. Unfamiliar: How to approach TDD? Inside-out? Outside-in? Both!

2. Combinatorial explosion of mappings Add Author Remove Author Set Affiliation Set Title & Abstract Submit to Review Domain Model Service API Browser (with JS) Browser (no JS)

2. Apply the Lingua Franca Pattern! Add Author Remove Author Set Affiliation Set Title & Abstract Submit to Review Domain Model Service API Browser (with JS) Browser (no JS) User Actions

3. Multiple possible mappings Can map from action to user action in multiple ways ● Different ways to navigate through the user interface ● Different ways to upload files ● Keyboard vs mouse input ● JavaScript vs no JavaScript Approaches: ● Choose randomly on each run ○ A form of property test ● Ensure equivalence with unit tests and exercise only one in the functional tests

4. Is it worth the effort?

Q: Why system tests? What about the testing pyramid?

The Funnel o' Feedback "Developer UX" Type checking Unit tests Connector tests Service tests System tests Manual tests Refactor to prefer fast feedback and exhaustive coverage

Test-drive at the system scale

To test the system we need to... know what the system is doing know when it has stopped doing it know when the system has failed explain what has gone wrong restore the system to a good state

To test the system we need to... support know what the system is doing know when it has stopped doing it know when the system has failed explain what has gone wrong restore the system to a good state

The End