IAC 2024 - IA Fast Track to Search Focused AI Solutions
Lessons learned with Bdd: a tutorial
1. Lessons Learned With
BDD
Thinking and Automating with Domain
Specific Languages
Alan Richardson
@eviltester
www.compendiumdev.co.uk
www.eviltester.com
www.seleniumsimplified.com
alan@compendiumdev.co.uk
2. Scope
● Part 1 - BDD
○ BDD Opinionated Overview
○ Lessons Learned
○ Exercises for BDD with no tools
● Part 2 - BDD & Tools
○ Cucumber overview
○ FitNesse overview
○ Custom DSL Overview
● Part 3 -
○ Q&A + Summaries
Warning: pretty much everything in here can
be argued with. BDD emphasises
communication. So feel free to disagree.
4. The Brief
Lessons Learned with BDD - Thinking and Automating with Domain Specific Languages
Starting with the the tools for BDD and Acceptance Testing has become increasingly simple. To learn tools like
Cucumber and FitNesse to create tests you can now read books and articles and watch countless videos to help.
But using the tools is only part of the battle.
Over years of working with Agile teams, and using various tools to help Business Analysts collaborate with
Development staff, writing high level tests in a human readable language, I've learned that most people do not
realise that they are writing a Domain Specific Language.
Each of the tools provides a different mechanism for modeling the domain and writing the language. But
fundamentally we are creating a language unique to our domain. And if we do not recognise this, or harness this,
then we make our life harder.
So in this session I want to look at some of the BDD and Acceptance Testing tool mechanisms e.g. tables or
Gherkin, with the additional modeling perspective of a high level language that models the domain.
Business Analysts are trained in modeling domains. Testers are trained in identifying ambiguity. Developer are
trained in refactoring to domain objects. These skills exist in the development team, we just have to harness
them.
In this session we will explore what this means, and how it changes our approach to writing BDD. This session
does not require access to a computer. Tools will be demonstrated. But tests will be written on paper and
discussed by the group to explore how to use BDD tools and and languages to more effectively model the
domain.
5. Alan Richardson
● Has used BDD, ATDD, DDD using tools
such as FitNesse, Cucumber, Domain
Specific Languages
● Domains - finance, gambling,
pharmaceuticals, media
● Has done it well, and badly
● Still gets confused by other people so makes
up his own models
6. My 'BDD' Messages
● The definition of BDD changes, so you
'should' own and define it for your domain
● Tools are not the answer, but neither is
'communication'
● I think of BDD as a process of modelling
communication across multiple domains
● Modelling is a domain specific activity, use
DSLs and DDD to improve BDD
● Expect to make mistakes and receive no
external 'instant fit' answers
(but then I think of
everything as a
process)
Don't get 'Hung Up'
8. BDD relates to communication
... but I think it is the worst communicated
'thing' in Software Development at the moment
BDD, ATDD, Specification By Example, BD
Design, ... Development, 'Real Options', "for
Communication (not development)", Deliberate
Discovery, Feature Injection
9. Plenty of platitudes for what it is not
● "it isn't about testing",
● "it isn't the tool",
● "it isn't just for the development team"
● "if must involve the business"
● "it isn't about the 'how'"
● BDD isn't ATDD
● "it isn't about the automation"
● "It isn't just for the business"
And most of these are probably true, but don't
necessarily help you understand what to do.
10. There are universal platitudes of
"isness"
● It is about communication
● It is for bridging the gaps
● It brings the business and development
closer together
Again these are true, but don't help, you do it.
12. The Definition 'Problem' with
understanding 'BDD'
● 'Definition' has changed over the years
○ ..D used to be for Development
■ because of TDD gaps and flow
■ ATDD not using abstraction layers
○ now is more like Design, but is still development
13. The Tool 'Problem' with
understanding 'BDD'
● Same tools used for BDD, ATDD, Spec By
Example
○ so what is the differentiator?
● BDD is not the tool, but the tool changes the
way you do BDD
● Tools enforce a syntax which BDD does not
require
○ Gherkin e.g. Given, When, Then
○ Tables
17. Domain Specific Language
A shared understanding requires a
shared language, which we agree
over time.
The language 'should' become ubiquitous: in
verbal discussion, in written communication, in
the code, in the tests.
18. Language
● Grammar
● Syntax
● Vocabulary
A simple model:
● Vocabulary and Syntax allow automation
● Grammar unites these with domain rules
19. BDD, ATDD, etc. Nuances over time
● BDD, ATDD, Specification by Example etc. -
different ways of looking at the same
process:
○ Moving from desire to reality
○ Maintaining as many options as possible
○ Allowing us to commit late
○ Allow us to experiment
○ Building a Shared Understanding
○ Checking we are adding value
○ Checking we are building it right
○ Each 'name' attempts to expose an additional
nuance
● Not the same things at all
20. Not the same things at all but BDD is
winning, because...
● When people talk about 'this' they now say
BDD more than any other term
● BDD has changed its definition to stay on
top
● Modelling is hard, so it is easy to point at
other peoples BDD and say 'this is wrong',
then explain the principles again rather than
offer a 'right' answer
21. Principles of this 'BDD' thing
● Communication
● Ubiquitous Language
● 'should' - does it? should it? how? when?
● Declarative vs Procedural
● Concise examples illustrate, explain &
explore
● Use explanatory text
● Model multiple levels
● Living spec - review for durability
● Review for waste
23. Some possible lenses or views if I
have to use the existing terms
● DDD (inside out)
○ are we using the right language? Are we working at
the right level in the domain?
● BDD (outside in)
○ what behaviour should this thing exhibit, how do I
know?
● ATDD (what else?)
○ what else should it do? how would it handle X? how
did we get it wrong?
● Specification by Example (examples?)
○ Do we have enough examples? Too many? The
right examples? Will this make sense later? Adds
Context
● Real Options (commitment)
24. Gherkin: Given x When y Then z
A ubiquitious 'language' (DSL) for analysis, that
made communication easier to automate:
● Feature, Scenario,
● Given,
○ a particular context (preconditions)
● When,
○ a particular sequence of events happens
● Then,
○ observable outcomes
● Examples
○ A set of examples
We add the
Grammar and
Vocabulary of the
Domain around this
Syntax
25. Opinionated Summary
● I don't care what you call it.
● I don't really care how you do it.
● But if I'm involved, and I see waste or
inefficiency, then I use all the lenses to look
at how you model the domain in a DSL and
make suggestions accordingly
● Many times I see people using one lense
above others, and rarely using, or
recognising their use of, a Domain Specific
Language
27. Lessons Learned
● Don't be tool driven
● 'Agree' your domain language
● Change your language
● Building a DSL is hard
● Refine your existing models
28. Don't be tool driven
● When you start, expect to be tool driven
● Fitnesse
○ Script tables
○ Essentially keyword driven automation
○ Because the testers can't code
○ Reuse of 'chunks' of FitNesse pages
○ Refactoring nightmare
● Eventually used for
○ Database setup and comparison
○ A GUI for the data
○ But Spreadsheets would have been better (have you
ever tried to edit FitNesse tables?)
29. Why not be tool driven?
● You're not fixing your communication
problem, you're adding a distraction
● New tools won't save you
● Tool introduces a new layer of abstraction
which may not help communication
● Tool can slow down communication
● Tool might impact your options
○ may not make change easy
○ may not make options easy
30. Do 'agree' your Domain language
● We just wrote down what the BAs said
○ including all the inconsistencies
● Different nouns to mean the same entity
● Different verbs to mean the same process
(sometimes a verb, sometimes a verb
phrase)
31. Expect your domain language to
change
● Because you will discover and learn more as
you try to express and model the domain
● The Domain is not restricted to the business
○ e.g. as soon as you commit to building a web app,
the language of the web becomes part of your
domain
● Do refactor your language
○ if later you decide that 'visitor' is better than 'a user
who is not logged in' then change it - everywhere
32. Building a DSL is hard
● Many people find modelling a DSL hard
○ I didn't realise this until we had people try to do it
● I think because:
○ people are not used to analysing language
○ people are not used to building new languages
○ people don't want to share/own the language
○ people don't feel confident with the written word
○ they are overloaded 'this' is new, the tool is new,
communication is new
○ Gherkin == BDD
33. Refine your existing models
● We kept adding more stuff
● Refine your models to remain relevant and
revealing
○ move acceptance criteria into functional tests as you
generalise and expand the scenarios
■ e.g. after committing to a web app move the web
acceptance criteria from BDD into functional tests
and make the BDD more general again
○ join scenarios and given when thens into a more
generic statement - cover detail in lower level tests
● Try to keep the BDD small by reworking new
parts of the domain into the existing model
34. How to make DSL construction
easier?
● You can limit the DSL modelling options to
the Vocabulary and the Grammar if you use
Gherkin
36. What is the DSL here?
Given a registered user
When the customer logs in
Then the account holder can view their
personal details
Critique?
37. Exercise: How might it read?
5 mins, make it better
● Given a registered user
● When the customer logs in
● Then the account holder can view their
personal details
39. How might it read?
● Are the following:
○ Right or wrong?
○ Good or Bad?
○ Better or worse?
○ Does it provide 'real options'
○ Supported by examples?
40. How might it read?
● A logged in user can view their account
details
Critique?
41. How might it read?
● A logged in user can view their account
details
○ the language encapsulates the context and event
○ this may be harder to automated
42. How might it read?
● Given a logged in user,
● when they view their details,
● then they can view their details
Critique?
43. How might it read?
● Given a logged in user,
● when they view their details,
● then they can view their details
○ context encapsulates details so event written at a
different level,
○ Gherkin forces a redundant 'then'
■ does it have to be redundant?
44. How might it read?
● Given a logged in user, when the user views
their details, then the user can see account
details are correct
Critique?
45. How might it read?
● Given a logged in user, when the user views
their details, then the user can see account
details are correct
○ provides more of a 'why'
○ reveals lack of options
■ there exist other ways that a user could check
the correctness of their account details
46. Right, Wrong?
● Context and Domain specific
● Domain has to include:
○ Business Domain
○ Technical
○ Also automation domain - how much of what we said
could be automated?
47. Is login an appropriate example?
● Is login obvious enough that this is a poor
example?
● Do we need communication around login?
49. BDD does not require a tool
Do some BDD on the Tracks application.
GTD - we think it is going to come back in style.
Pretend we are going to create a new version
of the Tracks app and use BDD to help us.
50. Product Owner (me) Demo of App
Demo of Tracks:
- default admin user created during setup
- login, logout
- create & delete user
- amend details, only I can amend admin email
- create a next action on review pages: home,
project edit, tickler, starred
- projects are always personal
- CRUD actions on next action, project, context
51. Exercise: Pick a few of the areas and
'BDD' them
● Whatever format you want
● I will answer questions as the SME
● Small groups
● Debrief in 20 mins
54. Demo Scope
● cucumber-jvm added easily with maven
● feature files as resources
● run feature files from ide
● step definitions in class
● regex in annotation to find steps
○ supports params
● JUnit runner for CI and report creation
● IDE support for syntax, quick fix and code
completion
● Gherkin - loose syntax checking
● "Scenario Outline" for examples
55. Lesson Learned: Cucumber-JVM is
not BDD
● Cucumber-JVM is an implementation of
Cucumber tool
● Gherkin is a DSL
● Cucumber allows us to map Gherkin text file
statements to code so we can 'execute' a
Gherkin document
● Cucumber means we don't have to write a
parser or lexer or the reporting tools
● If we don't pull out the elements of the tool
then we can't consider options and
alternatives to help us do BDD
56. Lesson Learned: Don't add all your
code into step definitions
● Tempting to add all 'test' implementation
code into step definitions
● But you aren't modelling the 'domain' if you
do that, you are using a tool
● Build abstraction layers. Model the domains
● Creates more options for your
implementation approach
● Supports re-use of automation code across
multiple test layers
● Because "BDD is not about testing"
57. Use Multiple Abstraction Layers
● Domain
● Physical Implementation World
○ Page Objects, Page Components, Sites, etc.
● Workflows
● Feature Files
○ Step Definitions
● BDD
These help increase flexibility for changing,
restrict commitment to fewer layers
58. Lesson Learned: Multiple
Automation Levels
● Functional Testing in conjunction with BDD
helps create multiple abstraction layers
○ refactoring
○ DRY
● Choices about what to model at what level
○ e.g. don't have to do everything in 'BDD' tool
● In this demo I had existing functional tests
which flushed out the page objects prior to
the Scenario implementation
59. Lesson Learned: Refactor your
automation code
● Too easy to add new code in, or copy paste
bits between steps
● Refactor, Refactor, Refactor
● Write clean code
60. About Cucumber-JVM
● Java Implementation of Cucumber
● Available via maven so easy to add to
project
● Annotate Java code with @Given etc. allows
sharing steps between scenarios
● IDEs (IntelliJ 12) are cucumber aware so:
○ can run feature files directly and create steps
Alt+Enter
○ code completion and syntax highlighting
● JUnit runner for CI
● Uses lots of reflection to work
61. Minimal Gherkin Feature File
Feature:
Scenario:
* a statement to check
Supports non standard
modelling
e.g. non Given, When,
Then modelling
62. Typical Gherkin Feature File
Feature: short description of feature
Background: general info
about the feature
Scenario: a scenario description
Given: a given condition
When: a particular event happens
Then: we can observe that this happened
63. Lesson Learned: Use the tool
features
● Don't avoid tool features because you want
the ability to move between tools
○ e.g. cucumber @Before, @After, @tagging, etc.
● Using the tool ineffectively can force you to
move away from the tool
● You can move between tools using
abstraction layers
65. A DSL based on Tables
● Cucumber provides a Gherkin parser
● FitNesse provides a Table parser
● table name
○ a scenario
● columns
○ map on to examples
● setter Columns - add data in to fixture
● getter Columns? - pull data out of fixture
○ if no 'expected' value then these can be used for
execution documentation
66. Demo
● BDD table for login
● Code to implement fixture
○ reuses our page objects etc.
67. Pros & Cons Compared with
Cucumber-JVM
● Cons
○ Technically more painful to setup
○ May deter collaboration if users can edit on server
without pairing with dev or test
● Pros
○ Very rich documentation ability
■ but no-one ever seems to use this
○ Wiki based for collaboration with users
○ Easy to test out new examples from a server deploy
○ FitNesse seems to be receiving more love and
attention again so keep an eye on it
71. When you write your own
● Easiest to create abstraction layers that
'look' and 'act' like a DSL
● Readable code
● Supports refactoring, code completion
● Take this as far as you need to.
○ Can be as complicate or as simple as you like
73. Is it BDD when...
● Testers can write tests using the DSL
● Business can express intent using the DSL
● Moving from business intent to test can be
done collaboratively
74. Lessons Learned: Constant Reviews
● Constantly need to review the DSL code
● Patterns and understanding constantly
evolve so need to keep on top and
communicate across the team
● Review to help enforce 'good' programming
practices
75. Lessons Learned: 'Good'
Programming Practices Required
● Refactor
● Test
● DRY
● Single Responsibility
● Patterns
● Use the language well
● Leverage libraries where you can
76. Lessons Learned: Domain Objects
● Use DDD lessons and embody Domain
objects in your code
● Limit the extent of the objects to avoid
overlapping domains
79. Summary
● Don't get hung up on doing it 'right'
● You make BDD what it is, if its not working
for you, change it and fix it
● Use all the nuances: DDD, ATDD, Example,
Options, TDD to BDD
● BDD is not the tools, but the chosen tools
change your Domain and your BDD style
● Own your DSL
82. Recommended Books
● Domain Driven Design
○ by Eric Evans
● Growing Object Oriented Software
○ by Steve Freeman and Nat Pryce
● Specification by Example
○ by Gojko Adzic
● The Cucumber Book
○ by Matt Wynne and Aslak Hellesoy
● ATDD by Example
○ by Marcus Gartner
88. Some possible Levels to view the
communication process at
● Desire
● Contextualised
● Examples
● Explored
○ what did we miss? how did we get it wrong? is it
enough?
● Implementation (s)
Other people use different levels. So do I.