270 likes | 474 Views
Using Tests as Agile Documentation. Brian Button Principal Consultant. Introduction. Welcome to my experiment! Not concrete yet, still taking shape Trying to prove an idea In Agile world, there is the claim that the tests are the documentation. What would it take to make that true???.
E N D
Using Tests as Agile Documentation Brian Button Principal Consultant
Introduction • Welcome to my experiment! • Not concrete yet, still taking shape • Trying to prove an idea • In Agile world, there is the claim that the tests are the documentation. • What would it take to make that true???
Who is the Audience? • Who is reading the documentation? • Programmers (task oriented) • Just want to get in and get out • Maintainers (holistic understanding) • Need to understand design • Use tests as backup during maintenance • Evaluators (kicking the tires) • Setup, overview, simple tasks
My Target Audience • This talk with focus on the Programmers using the library • Most basic need • If this need is not met, tests will not suffice • Also focusing on projects where code is the output
What is the Problem? • We all need documentation • But its expensive to: • Create • Maintain • Written as separate activity by writers or programmers • And it needs to change as requirements change
What Can We Do? • Nothing • Limit changes to keep down costs • Embrace Change
Status Quo • Don’t change • Accept that documentation costs lots of money • We’re all comfortable with that • But costs keep increasing, budgets keep shrinking, projects keep getting shorter • Probably not the right answer
Limit Changes • Let documentation costs lead design decisions • Huh??? • Leads to missed opportunities • After all, we’re building software, not manuals • Also, probably not the way to go
The Agile Solution • Embrace changes • Because they’re gonna happen • Adapt documentation process to allow for easy changes • Leverage an existing, changing asset to keep documentation creation and maintenance costs low
Unit Tests?? • Means different things to different people • My definition: • Specific, targeted tests that focus on one path through code under test • Not system-wide or complex transactions • Simple, method-level tests
Example – String Immutability public void ConvertingStringToUpperCaseCreatesNewStringInstance() { string lowerCaseString = "brian"; string upperCaseString = lowerCaseString.ToUpper(); Assert.AreEqual("brian", lowerCaseString); Assert.AreEqual("BRIAN", upperCaseString); }
Preconditions to Adoption • Assumes Agile Development practices • Must have Unit Tests for Everything • Tests must be run all the time • Tests must pass 100% all the time • Assumes leeway and understanding from audience • It’s gonna be different!
Agile Documentation • But if they’re open to it, we can really wow them • We can create • Simple statements of facts • Give simple usage examples • Document all behavior explicitly • Guarantee code and docs are in sync
Standard Programming Docs • Documentation for Stack in Visual Studio • Class level • Gives overview, class hierarchy, example • Example takes some effort to read and is incomplete • Peek() • Gives overview, exception, more specific example • Example gets worse as method gets more complex
Agile Docs for Stack public class StackBehaviorFixture { public void StackIsEmptyAtCreation() public void StackNotEmptyAfterItemPushed() public void StackChangesBackToEmptyAfterPushAndPop() public void ValuePushedOntoStackIsSameAsValuePoppedOffStack() public void MultipleItemsPushedOntoStackArePoppedOffInReverseOrder() public void PopFromEmptyStackThrowsInvalidOperationException() public void PeekDoesNotRemoveItemFromStack() public void PeekWillReturnItemPushedOnStack() public void PeekWillAlwaysReturnTheSameItem() public void PeekWillAlwaysReturnMostRecentPushedItem() public void PeekOnEmptyStackThrowsInvalidOperationException() public void NullItemsCanBePushedOntoStack() public void PopReturnsNullItemOffStack() public void PeekReturnsNullItemOffStack() }
Agile Docs for Stack • Each test is statement of fact • Human readable • Simple usage example • Would a Test List be better?
Creating Agile Documentation • Generate a Test List • List of candidate tests to flesh out interface and functionality. • Grows over time as you learn more • Task oriented • Flows from simple to complex • Implement tests
Anatomy of a Good Test • Strong names, simplicity, clarity • 3 A’s public void PeekDoesNotRemoveItemFromStack() { Stack stack = new Stack(); stack.Push("foo"); stack.Peek(); Assert.IsFalse(stack.IsEmpty); }
What’s Missing? • What versus Why • High-level UML diagrams • Guidance on where to look
(Some) Documentation == Necessary Evil • Unit tests tell what not why • For complex classes or subsystems, some level of written documentation is required • But not at same level of detail as before • Overview, design decisions, key UML diagrams • Unlikely to change as details change
Test Map • On complex project, there are lots of tests • Key question is how to find right tests easily • Is this harder than with regular documentation? • Lots of paper, web pages, etc, versus lots of test • Creation of a test map would go a long way to helping
What is a Test Map • Idea of my own creation • It simply maps from a method being tested to those tests that exercise that method • Creates a manual page that lists methods and has hotlinks to tests that exercise them • Use of good test names will help programmer find right test to look at
Test Maps and Automation • Test Map would be hard to keep in sync • Creation should be automated [Test] [TestFor("StackExample.Stack.Push")] [TestFor("StackExample.Stack.Pop")] public void ValuePushedOntoStackIsSameAsValuePoppedOffStack() { Stack stack = new Stack(); stack.Push("fred"); object poppedValue = stack.Pop(); Assert.AreEqual("fred", poppedValue); }
Downside of Markups • Has to be maintained manually • Failing test won’t point you to incorrect markups • Error prone • Drives up cost • Lessens value • Where is the balancing point?
Conclusion • Tests can replace much written documentation • Some written docs and UML still needed • Remaining documentation less change-prone • Requires right audience and context • Requires different mindset while writing tests