My impressions from Øredev

Yesterday I wrote a post called "My impressions from Øredev" on Dotway's blog. In summary, I had a great time there and I'm still very inspired from the experience.

Hope to see you there next year! :)

FizzBuzz in LINQ

Earlier this week I realized that you are able to define functions in LINQ queries. So I made myself a sweet little FizzBuzz in LINQ.

It's not pretty code, but I had fun writing it. Though, the code would be a somewhat nicer without all the type declarations everywhere. Hehe, I guess that if your code reminds you of The Obfuscated Code Contest, it's a pretty good sign that your code stinks. Clearly muda!

[TestMethod]
public void FizzBuzz()
{
  var lines = from x in Enumerable.Range(1, 100)
          let divides = new Func<int, int, bool>((denominator, nom) => nom % denominator == 0)
          let fizz = new Func<int, Tuple<int, string>>(
                    n => new Tuple<int, string>(n, divides(3, n) ? "Fizz" : ""))
          let buzz = new Func<Tuple<int, string>, Tuple<int, string>>(
                    tuple => new Tuple<int, string>(tuple.Item1, tuple.Item2 + 
                                 (divides(5, tuple.Item1) ? "Buzz" : "")))
          let returnNumberIfEmptyString = new Func<Tuple<int, string>, string>(
                    tuple => tuple.Item2 == "" ? tuple.Item1.ToString() : tuple.Item2)
          select returnNumberIfEmptyString(buzz(fizz(x)));

  foreach (var line in lines)
  {
     Debug.WriteLine(line);
  }
}

(Oh, forgot to say that I use the .NET 4.0 tuples for this post)

From for loop to anamorphism

Introduction

Sometimes you want to generate a sequence of objects. This is often done using a for loop:

[Test]
public void CreateListOfFoos()
    {
        var xs = new List<Foo>();
        for (int i = 0; i < 10; i++)
        {
            xs.Add(new Foo());
        }

        Assert.AreEqual(10, xs.Count());
    }

In this post, I will show you how this code can be made more general, ultimately turning it into an anamorphism over lists.

Generalizing constructed type

First, what we want to do is to be able to create something other than a Foo. We apply ExtractVariable once and we also extract the constructor call to a lambda:

[Test]
public void ExtractMethod()
{
    var times = 10;
    Func<Foo>newFoo = () => new Foo();

    var xs = CreateFooNumberOfTimes(times, newFoo);

    Assert.AreEqual(10, xs.Count());
}

private IEnumerable<Foo> CreateFooNumberOfTimes(int times, Func<Foo> newFoo)
{
    var xs = new List<Foo>();
    for (int i = 0; i < times; i++)
    {
        xs.Add(newFoo.Invoke());
    }
    return xs;
}

From the above code, it’s easy to generalize on the type created. Note that we could have skipped the “constructor lambda” and instead used the “where T : new()” constraint.

[Test]
public void GeneralizeTypeForFunc()
{

    var times = 10;
    Func<Foo> newFoo = () => new Foo();

    var xs = CreateTNumberOfTimes(times, newFoo);

    Assert.AreEqual(10, xs.Count());
}

private IEnumerable<T> CreateTNumberOfTimes<T>(int times, Func<T> newFoo)
{
    var xs = new List<T>();
    for (int i = 0; i < times; i++)
    {
        xs.Add(newFoo.Invoke());
    }
    return xs;
}

Generalizing type for accumulator

We have now parametrized Foo to T, but wouldn’t it be possible to parametrize from “int” to A, as well? Let’s begin with breaking out the int-specific code from the for loop:

[Test]
public void ExtractForLoopLogic()
{
    Func<Foo> newFoo = () => new Foo();

    int i = 0; // Will be modified lots of times
    Func<bool> expr = () => i < 10;
    Action inc = () => i++;

    var xs = CreateTUntil(newFoo, expr, inc);

    Assert.AreEqual(10, xs.Count());
}

private IEnumerable<T>CreateTUntil<T>(Func<T> newT, Func<bool> expr, Action inc)
{
    var xs = new List<T>();
    for (; expr.Invoke(); inc.Invoke())
    {
        xs.Add(newT.Invoke());
    }
    return xs;
}

As indicated in the code, the variable “i” will be modified the closure called in CreateTUntil. Bart de Smet calls this a cruel lambda. Except that it’s quite hard to understand a lambda that mutate its outer scope, refactoring code that’s using cruel lambdas can make your code go totally bananas!

Let’s rewrite the code to use a pure lambdas instead. To do this, we need to refactor the for loop to a while loop, since the first and third “parameters” to a for loop are statements (not pure). We also parametrize from “int” to type parameter “A” instead.

Going pure and a more general constructor

[Test]
public void ForLoopToWhileLoop()
{
    Func<Foo> newFoo = () => new Foo();
    Func<int,bool> expr = a => a < 10;
    Func<int,int> inc = i => i + 1;

    var xs = CreateTUntilUsingWhile(newFoo, 0, expr, inc);

    Assert.AreEqual(10, xs.Count());
}

// Now a pure method
private IEnumerable<T> CreateTUntilUsingWhile<T>
    (Func<T> newT, int init, Func<int,bool> expr, Func<int,int> inc)
{
    
    var xs = new List<T>();

    int i = init;
    
    while (expr.Invoke(i))
    {
        xs.Add(newT.Invoke());
        i = inc.Invoke(i);
    }
    return xs;
}

Now we don’t have any concrete types in the method signature, except bool, which I think is ok to have there at this point. But, as the observant reader might have noticed, the constructor can’t be called with a variable argument, i.e. the accumulated value. What we need to do is to “connect” the lambdas that generates values, like this:

[Test]
public void ArgumentForConstructor()
{
    Func<int,Result<Foo,int>> gen = n => new Result<Foo,int>(new Foo(), n + 1);
    Func<int,bool> expr = a => a < 10;
    
    var xs = GeneralizedCreateTWithArgsUntilUsingWhile(gen, 0, expr);

    Assert.AreEqual(10, xs.Count());
}

private IEnumerable<T> GeneralizedCreateTWithArgsUntilUsingWhile<T,A>
    (Func<a,Result><T,A>> gen, A init, Func<A,bool> expr)
{
    var xs = new List<T>();

    var i = init;

    while (expr.Invoke(i))
    {
        var result = gen.Invoke(i);

        xs.Add(result.Value);
        i = result.Accumulator;
    }

    return xs;
}

Going recursive

Now it’s up to the generating lambda to pass an argument to the constructor or not. What’s funny with this is that if we replace the while loop to a recursive call, we come pretty close to the definition of an anamorphism over lists in the introduction of Functional Programming with Bananas, Lenses, Envelopes and Barbed Wire by Erik Meijer et. al (link to postscript version).

[Test]
public void WhileLoopToRec()
{

    Func<int, Result<Foo, int>> gen = n => new Result<Foo, int>(new Foo(), n + 1);
    Func<int, bool> expr = a => a < 10;

    var xs = CreateTUntilUsingRec(gen, 0, expr);

    Assert.AreEqual(10, xs.Count());
}

private IEnumerable<T> CreateTUntilUsingRec<T,A>
    (Func<A, Result<T, A>> gen, A init, Func<A, bool> expr)
{
    
    if (!expr.Invoke(init))
        return new List<T>();

    var result = gen.Invoke(init);

    return (new List<T> {result.Value})
        .Concat(CreateTUntilUsingRec(gen, result.Accumulator, expr));

}

Abstracting away from bool

It seems that the only thing left to do is to abstract away the dependency on “bool” in CreateTUsingRec, but then we bump into a small problem,as you will see. What we do is to join the two lambdas into one and performing a null check inside the recursive function.

[Test]
public void MergeOnceMore()
{
    int? i = 0;
    Func<int, Result<int?, int>> gen = n => new Result<int?, int>(n < 10 ? i : null, n + 1);

    var xs = CreateTUntilUsingRecNullCheck(gen, 0);

    Assert.AreEqual(10, xs.Count());
}

private IEnumerable<T> CreateTUntilUsingRecNullCheck<T, A>
    (Func<A, Result<T, A>> gen, A init)
{
    
    var result = gen.Invoke(init);
    if (result.Value == null)
        return new List<T>();

    return (new List<T> { result.Value })
        .Concat(CreateTUntilUsingRecNullCheck(gen, result.Accumulator));

}

Writing it in F# instead

The problem with this solution is that we have lost the ability to generate ordinary non-nullable structs and that’s bad! We could easily solve the problem with writing our own MyNullable<T> which would  allow both classes and structs as instantiators of the type variable, but instead of doing that, I’ll show you something similar: the option type in F#.

let rec anamorphism n f = 
    match f n with
      | option.None -> [] 
      | option.Some (e,next) -> e::(anamorphism next f)

> anamorphism 0 (fun a -> if a < 5 then option.Some("Bar" + a.ToString(), a + 1) else option.None);;
val it : string list = ["Bar0"; "Bar1"; "Bar2"; "Bar3"; "Bar4"]

Here, we removed the Result class and used a pair instead, together with the Option type, which is essentially the same as Nullable<T>, but without the restriction on type.

Conslusion

Functional programming is perhaps more abstract than imperative programming, but it is also seems to be more general, at least in this case. This post has only showed an anamorphism over lists, which is the simple case. Look here if you want to see a more advanced example.

Fluent language in FsStory

A nice feature for a story runner is to be able to provide arguments in a story sentence, like this:

ATableWithNumberOfLegs 4

This example is not hard to understand, but there's some mental translation going on, since the order of the words is all screwed up. Let's try another one:

TheNumberOfLegsOfATableIs 4

Better, but still not good. First, it's longer than the previous example. Second, even if the grammar is correct, the word order is, well, unusual.. ;)

What about this?

ATableWith 4 Legs

Now we're talking!

Here's a bigger and more complete example:

[<Fact>]
let tableLegsScenario =
    given (ATableWith 4 Legs)
 |> whens (ICutOf 1 Leg)
 |> thens (ItHasOnly 3 LegsLeft)

Note: As far as I know, RSpec/Cucumber is the only story runner(s) that is able to use variables inside a story sentence.

The nice thing about this is that I didn't have to change FsStory itself to make it work. Not a single line. So, what's the trick?

If you split up a story sentence like this, you have to prepare the step definition code (the behind-the-scenes code) in a certain way:

let ATableWith = fun n _ -> ... do something here ...

..or if you prefer, without a lambda:

let ATableWith n _ = ... do something here ...

Then you have to define Legs:

let Legs = id

As you see, in this case, it was just a matter of discovering the usage, rather than implementing it in the language. Honestly, I had no idea of this usage when I started to write on FsStory. This is clearly one of the reasons why I like internal DSLs!

Exercise: How would you implement the step definition for the following story sentence?

given (ATableWith 4 LegsAnd 2 Chairs)

Cheers!

ImpossibleEstimation and Pomodoro Technique

Sometimes when using Pomodoro Technique I find it real difficult to estimate how long a particular activity will take, i.e., when locating a bug or find out why the webserver won't read my files.

In Pomodoro Technique, every activity should have a time estimate - how many Pomodori I think it will take. Though, this is sometimes impossible! "The problem is solved when I find the bug and since I don't know what the bug is related to, it's impossible to say how much time it will take to find it."

The solution: instead of just writing a number besides the activity, I use the less-than sign (<) before the number, indicating that I have a time-box for the activity, but that it might take less time. If I'm not done when the time-box is over, I have to ask a colleague to help me or ask my boss for extra resources - thus, escalating my problem. Then, I'm forced to have collected some data of the problem to help them to help me. The nice thing is that I still can have most of the benefits Pomodoro Technique gives me, i.e., increased focus when in a Pomodoro and possibility to get "the whole picture" during my breaks. The latter have proved to be an extra nice thing to have during hard problem-solving.

And, if a colleague comes to the rescue, we can construct another time-box, to know when to escalate it further, or at least to notify the team or the boss that we have some nasty problems at hand.

Yet, if I'm collecting metrics of my estimation skills, it is not a very good idea to track data for these bug-fixing Pomodori estimates. Put a "N/A" or a "-" in your records sheet and think for yourself: "Today was an exception, tomorrow will be a bug-free day." And don't forget to do your daily mind-map before you leave for home.

[Announce] FsStory, executable stories in F#

Since Claudio Perrone's talk on Øredev, I have been thinking about what his MisBehave would look like in F#. In his talk, Claudio also mentioned Cucumber, a story runner written in Ruby. My plan was to make a lightweight DSL for writing stories in F# code, with the story parts separated from the implementation parts.

Currently, FsStory enables the developer to write user story scenarios (in Given/When/Then form) in F# code, like this:

#light

open FsStoryRunner
open MutatedTurtleMovesImpl
open Xunit

(*
In order to impress my friends
As a .NET programmer
I want to draw funny fractal pictures
*)

[<fact>]
let MoveTurtleToPosition() =
 given ATurtle
 |> andGiven IsRotated90DegreesToTheRight
 |> whens (TurtleWalksSteps 9)
 |> thens (TurtleIsLocatedAt (0,9))
 |> endStory

Did you notice [<fact>] attribute just before the function definition? It is a xUnit.net specific attribute, telling xUnit.net that the function is a runnable test. So, why xUnit.net? Answer: xUnit.net is the currently the only test framework that runs static test methods, which is what F# functions compiles to.

Note: If you think that the story above is too low level to be a "good" user story, you're right, but it's just an example..

The "ATurtle", "IsRotated90DegreesToTheRight", "TurtleWalksSteps", etc, are functions that you have to implement yourself. What these functions do is not FsStory's business, except that they have the same type. It's a good thing to think about this in advance.

If you're testing something object oriented, i.e. a C# project, then you're probably have to let the functions have the type () -> (). That is, they take no argument and return void, in C# lingo. You'd also need a mutable variable to accomplish this.

#light

open Turtle
open FsxUnit.Syntax

let mutable turtle = new Turtle() // turtle must have type Turtle

let ATurtle () = turtle <- new Turtle()    // For reuse in same story let MovesOneStepForward () = turtle.Go() let IsMovedOneStepForward () = turtle.Position.X |> should equal 1 let RotationIs angle () = turtle.Direction |> should equal 0.0  

It's up to the developer what library she wants to use for her assertions. In this example, FsTest was used, but she could go for NUnit or NBehave or something else. I hadn't actually tried this and do not longer think this will work. Either use xUnit.net or FsTest (which is based on xUnit.net).

Another style is to work with immutable objects. One example of immutable objects are value objects, in DDD. Immutable objects correspond well to functional programming principles. Here is an example of an implementation of a scenario when an immutable object is used in the SUT (System Under Test).

let ATurtle () = new TurtleImmutable()
let IsRotated90DegreesToTheRight = fun (turtle : TurtleImmutable) -> turtle.Left()
let TurtleWalksSteps steps = fun (turtle : TurtleImmutable) -> turtle.GoSteps(steps)
let TurtleIsLocatedAt (x,y) = fun (turtle : TurtleImmutable) -> turtle.Position |>
                  should equal (new Position(x,y)) ; turtle

To clarify, all methods on the (immutable) turtle return a new turtle and that turtle is returned and then passed in as an argument to the next test function (by FsStory). As you might have spotted, the example uses a lambda, an anonymous functions (the "fun") instead of specifying an argument explicitly. It's a good thing to get a running story before actually implementing the logic and assertions. Using the function "id" (just returning the argument) on the right-hand side is very helpful for getting everything to run.

You can find FsStory at http://www.codeplex.com/fsstory.

BDD using NBehave + Rhino Mocks AMC

Update: I found an old blog post by Aslak Hellesøy (the main developer behind Cucumber) that touches on this subject.

Some time ago, I investigated what BDD is all about. In essence, it's TDD with a twist. For example, the word "test" implies that we're testing what someone (we?) already made, but TDD says we're going to write the tests before the actual implementation of the unit of code. Somewhat, the word "test" has a direction backwards, while "should" has a direction forward. Hence, "should" is more comprehensive to use when describing and specifying the future. Makes sense?

What I'd like to show you is a piece of code I wrote to see how NBehave's story runner could work with Rhino Mocks' Auto Mocking Container[1, 2].

[Story, Test]
public void GetMoneyWhenPassGO()
{
 // Set up and initialize
 var mocks = new MockRepository();
 var container = new Rhino.Testing.AutoMocking.AutoMockingContainer(mocks);
 container.Initialize();

 // Resolve and obtain references
 IPlayerTurn turn = container.Create<PlayerTurn>();
 IBoard board = container.Resolve<IBoard>();
 IPlayer player = container.Resolve<IPlayer>();
 turn.AddPlayers(new List<Player>() {player});

 // Story begins here
 var story = new Story("Player recieves money when passes 'GO'");

 story
     .AsA("Player")
     .IWant("to recieve money when I pass 'GO'")
     .SoThat("I can buy things that generate money");

 story
     .WithScenario("Normal play scenario")
     .Given("A board with 4 squares", () => Expect.Call(board.NumberOfSquares).Return(4))
     .And("a player near 'GO'", () => Expect.Call(board.GetIndexForPlayer(player)).Return(2))     
     .And("mockery has started", () => mocks.ReplayAll())
     .When("player passes 'GO'",() => turn.PlayerSequence(player, 3))
     .Then("the player earns $4000", () => player.AssertWasCalled(x => x.Credit(4000)));

}

When this test is run with ReSharper, the test passes and outputs the story with indentation. Nice! (Except the "mockery has started" part of the story..)

Now a question pops up: since we have only specified and tested the first (top-level) interaction, what about the rest of the interactions? I think this is a good question that leads us further down the rabbit hole.

A written user story comes from a dialogue with a person with domain knowledge. Hopefully, after some discussion, we have understood some of the moving parts of the domain problem the customer wants us to solve. Let's assume that we want to implement a single feature at a time, a couple of questions arise: should we start top-down or bottom-up? And how far "up" should we go, i.e. should we start with the UI or the domain model, if we choose a top-down approach?

If we choose to start with the domain model, then I think the above way of specifying the behavior looks nice. The key question is where the classes in the story come from originally. I have no easy answer for that. Of course they should originate from the domain problem, but how? "The model is the code - the code is the model", but it probably takes a while to "get it right". Maybe code like the above could help us to see if we have understood the problem in the first place?

Now back to the question: "what about the rest of the interactions"? We have ensured that the class which is in "the center" of the particular interaction chain (the player turn) lives in a faked world (a small board and a player near go) and we finally assert that when something happens (player passes 'GO') then some state has changed (the player gets money). The nice thing is that we now know more about what functionality the dependent classes should provide. For example, IBoard needs to have a method GetIndexForPlayer and if there is a class implementing that interface, then a NotImplementedException is probably thrown from that method, in order to compile. Next step could be to start thinking on that particular method and choose to either write a mocked unit test or an "ordinary" unit test.

Of course, real acceptance tests are also needed, but the purpose and scope of those tests are quite different. At least that's what I think.

What do you think?