Applying Structured Programming

In my previous post about Structured Programming I talked about that basic looping constructs, fold and so on are basically just still to powerful. In the sense of readability we should try to eliminate them with more specific ones. In this post i go through a toy example to show the various ways on how to refactor some code.

The Toy Example

Recently I had some conversation about code in a game and providing some kind of critical hit-chance in a game. The typical way on how to achieve that is actually easy. Let's assume that every attack of an player has a 16% chance to be critical. We only need to generate a random number between 0 and 99 (or 0 to 1) and test if that number is lower than 16 (or 0.16).

Let's assume we want to test if that really is true. We would just generate some random numbers. Test if that number is a critical hit. And either increase a critical hit variable or some normal hit variable. After 1000 tries we just calculate the average and see if we really have around 16% or around 160 hits.

Solution 1

Some very imperative code in F# could look like this.

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
11: 
12: 
let rng = System.Random()

let calculateChance chance =
    let mutable criticalHit = 0
    let mutable normalHit   = 0
    for i in 1 .. 1000 do
        let random = rng.Next(0, 100)
        if random < chance then
            criticalHit <- criticalHit + 1
        else
            normalHit <- normalHit + 1
    criticalHit, normalHit

Testing our code would look something like this.

1: 
2: 
3: 
4: 
5: 
let percentage x y = 100.0 / (x + y) * x

for i in 1..10 do
    let crit,normal = calculateChance 16
    printfn "Crit: %d Normal: %d Percentage: %f" crit normal (percentage (float crit) (float normal))

We now getting a result like this:

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
Crit: 1585 Normal: 8415 Percentage: 15.850000
Crit: 1638 Normal: 8362 Percentage: 16.380000
Crit: 1616 Normal: 8384 Percentage: 16.160000
Crit: 1603 Normal: 8397 Percentage: 16.030000
Crit: 1624 Normal: 8376 Percentage: 16.240000
Crit: 1667 Normal: 8333 Percentage: 16.670000
Crit: 1617 Normal: 8383 Percentage: 16.170000
Crit: 1639 Normal: 8361 Percentage: 16.390000
Crit: 1653 Normal: 8347 Percentage: 16.530000
Crit: 1613 Normal: 8387 Percentage: 16.130000

Actually we now have proven that the idea works, as around 16% of our attacks are now critical. But now let's actually look if we can improve our calculateChance function. As stated in the beginning. It is a toy example, so we usually wouldn't waste any time in improving this particular function. But by going through a toy example it can help to get the general concept on how to improve code.

Solution 2

In functional programming mutable variables are actually a little bit frowned upon. So let's try to eliminate our mutable variables by replacing our loop with recursion. Actually recursion is in that sense just looping and you can turn any kind of loop into recursion. The difference between looping and recursion is just what is explicit and what is implicit.

In Looping we implicitly move to the next step, and we can explicitly break/abort a loop. In recursion we implicitly abort, and we have to explicitly move to the next step, by calling our function recursively.

Mutable variables outside of a loop turn into function parameters. This way we can turn any kind of loop into a recursive function and eliminate mutable variables all together.

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
let calculateChance chance =
    let rec loop count criticalHit normalHit =
        if count < 1000 then
            let random = rng.Next(0,100)
            match random < chance  with
            | false -> loop (count+1) criticalHit (normalHit+1)
            | true  -> loop (count+1) (criticalHit+1) normalHit
        else
            criticalHit, normalHit
    loop 0 0 0

We now eliminated all mutable variables and now provided an inner recursive function that replaces our loop. On top of it i replaced the inner if check on random < chance with pattern-matching. This way it is easier to see the difference.

Either in both ways we will call loop again, but with an increment count. If random < change is false we have a normal hit, so we increase normalHit by one, otherwise we increase criticalHit by one.

We continue our recursive call as long we have count < 1000. But as soon that condition is false we end up with just criticalHit, normalHit that will return both variables as a tuple.

The question overall is. Is that version better as Solution 1?

Well it depends. We eliminated the mutable variables, but actually at least I am someone that has nothing against mutable variables in a limited scope. If you are a programmer that primarily uses an imperative language and are used to looping then you will probably prefer Solution 1. If you are in the state of learning functional programing you should try to replace looping constructs in this kind of way to get used to it. This is especially important for the later Solutions we will look at. If you are used to this, like me, you will probably not find the recursive version any harder to understand as the looping example.

So what is with Structured Programming? Did we replace a powerful construct with a less powerful construct? At that moment, no we didn't. Recursion is just as much powerful as looping. The funny part. The compiler will turn such kind of tail-recursion into a loop when compiled. That's also the reason why functional programmers names such inner recursive functions just loop.

So was our transformation into Solution 2 wasteful? Not really, that now leads to Solution 3

Solution 3

Once we have eliminated all kinds of mutable variables we end up with a recursive function that just loops. Our function takes some additional variables like count, criticalHit and normalHit as it's state. What we really have is a looping construct with just an accumulator. But wait. That's exactly what fold is about! The question that starts to beg is. Can we replace Solution 2 somehow by using a fold?

The answer is yes. But which kind of fold do we need? Or in other words, over what exactly do we loop? We don't loop over a data-structure like an Array or List. So what is it that we are looping over?

When we examine our code we could say we loop over count. But that isn't true. Our count is just there so we know when to end the looping. We really are looping over the rng.Next(0,100) call. We only need the count because that is an infinite sequence. But that actually answers our question. Let's create a seq that just returns an infinite sequence of random numbers.

1: 
2: 
let rng = System.Random()
let randomSeq min max = Seq.initInfinite (fun _ -> rng.Next(min, max))

Note that I'm defining rng outside the function. Instantiating a System.Random inside the function and calling randomSeq in short time-interval would otherwise lead to a RNG with the same seed. Once we have our Random Sequence it also becomes clearer what our count was before. We just take the amount of randoms we need from our infinite sequence. After that, we only need to transform what is left into a fold call.

1: 
2: 
3: 
4: 
5: 
6: 
7: 
8: 
let calculateChance chance =
    randomSeq 0 100
    |> Seq.take 1000
    |> Seq.fold (fun (criticalHit,normalHit) random ->
        if   random < chance 
        then (criticalHit+1),normalHit
        else criticalHit,(normalHit+1)
    ) (0,0)

Looking at the current Solution i think we made our first improvements to our code. At first we created a randomSeq. A sequence out of random can be helpful in many other places. Seq.take 1000 makes it clear that we just fetch 1000 random numbers from it. And after having those we use fold.

Now, our fold only contains the real logic. It just checks whether our random is a criticalHit or not and we create our new state from it.

But as stated before. fold is still something that we consider as powerful, is there a way to also eliminate fold?

Solution 4

We actually can eliminate fold. But for that we need to go and look back at what we are actually doing. What we really was interested in was the percentage if we really get the right amount of critical-hits this way. What we really need is to split it into two parts. We need some function that test whether we have a critical hit or not. We could turn it into true and false values. But later we need to turn those somehow into a formula to calculate the average.

But by thinking separetely about it we easily can recognise that we can easily achive both things in one step. By just turning a critical-hit into 1.0 and a normal hit into 0.0. By calculating the average we would automatically get a percentage that ranges bewteen 0.0 and 1.0. We could multiply it by 100.0 or we also could use 100.0 and 0.0 instead of 1.0 and 0.0.

1: 
2: 
3: 
4: 
5: 
let calculateChance chance =
    randomSeq 0 100
    |> Seq.take 1000
    |> Seq.map (fun x -> if x < chance then 100.0 else 0.0)
    |> Seq.average

And as a final note. We can replace a call to map followed by average with averageBy

1: 
2: 
3: 
4: 
let calculateChance chance =
    randomSeq 0 100
    |> Seq.take 1000
    |> Seq.averageBy (fun x -> if x < chance then 100.0 else 0.0)

Conclusion

When we compare the final code with what we started I think we have reached a good refactoring.

We started with:

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
11: 
12: 
13: 
14: 
15: 
16: 
17: 
18: 
let rng = System.Random()

let calculateChance chance =
    let mutable criticalHit = 0
    let mutable normalHit   = 0
    for i in 1 .. 1000 do
        let random = rng.Next(0, 100)
        if random < chance then
            criticalHit <- criticalHit + 1
        else
            normalHit <- normalHit + 1
    criticalHit, normalHit

let percentage x y = 100.0 / (x + y) * x

for i in 1..10 do
    let crit,normal = calculateChance 16
    printfn "Crit: %d Normal: %d Percentage: %f" crit normal (percentage (float crit) (float normal))

End we ended with

 1: 
 2: 
 3: 
 4: 
 5: 
 6: 
 7: 
 8: 
 9: 
10: 
11: 
let rng = System.Random()
let randomSeq min max = Seq.initInfinite (fun _ -> rng.Next(min, max))

let calculateChance chance =
    randomSeq 0 100
    |> Seq.take 1000
    |> Seq.averageBy (fun x -> if x < chance then 100.0 else 0.0)

for i in 1..10 do
    let percentage = calculateChance 16
    printfn "Percentage: %f" percentage

From a code perspective we didn't write more code. We even reduced the line count number. By rewriting we created a reusable randomSeq sequence that can provide us an infinite stream of random numbers. I also find the logic easier to understand.

  1. Initialize a randomSeq that return numbers from 0 to 99
  2. Take 1000 of those numbers
  3. Map them to 100.0 if it smaller than chance or 0.0 and calculate the average from those numbers.

As stated in the beginning. It was a toy program with what we started but overall we can see the ways in how we can continuously improve our code.

I don't think that just turning a loop in itself into a recursive function as you see in Solution 2 provides much benefit. But doing such a thing can help to later turn it into a fold call. You also can move directly from Solution 1 to Solution 3. But doing the intermediate Step can greatly help if you are not used in doing this kind of things.

Once you have a fold in it you can further think in how you can eliminate it. If there exists other functions that can replace a fold use them instead! But what happens if you don't find other more specific function as fold? Well just use it, it is there to be used. But if you found similarities between multiple different lambda functions that you pass into fold, you should look into how you can abstract the lambda into a reusable function.

module Main
val rng : System.Random

Full name: Main.rng
namespace System
Multiple items
type Random =
  new : unit -> Random + 1 overload
  member Next : unit -> int + 2 overloads
  member NextBytes : buffer:byte[] -> unit
  member NextDouble : unit -> float

Full name: System.Random

--------------------
System.Random() : unit
System.Random(Seed: int) : unit
val calculateChance : chance:int -> int * int

Full name: Main.calculateChance
val chance : int
val mutable criticalHit : int
val mutable normalHit : int
val i : int32
val random : int
System.Random.Next() : int
System.Random.Next(maxValue: int) : int
System.Random.Next(minValue: int, maxValue: int) : int
val percentage : x:float -> y:float -> float

Full name: Main.percentage
val x : float
val y : float
val crit : int
val normal : int
val printfn : format:Printf.TextWriterFormat<'T> -> 'T

Full name: Microsoft.FSharp.Core.ExtraTopLevelOperators.printfn
Multiple items
val float : value:'T -> float (requires member op_Explicit)

Full name: Microsoft.FSharp.Core.Operators.float

--------------------
type float = System.Double

Full name: Microsoft.FSharp.Core.float

--------------------
type float<'Measure> = float

Full name: Microsoft.FSharp.Core.float<_>
val loop : (int -> int -> int -> int * int)
val count : int
val criticalHit : int
val normalHit : int
val randomSeq : min:int -> max:int -> seq<int>

Full name: Main.randomSeq
val min : int
val max : int
module Seq

from Microsoft.FSharp.Collections
val initInfinite : initializer:(int -> 'T) -> seq<'T>

Full name: Microsoft.FSharp.Collections.Seq.initInfinite
val take : count:int -> source:seq<'T> -> seq<'T>

Full name: Microsoft.FSharp.Collections.Seq.take
val fold : folder:('State -> 'T -> 'State) -> state:'State -> source:seq<'T> -> 'State

Full name: Microsoft.FSharp.Collections.Seq.fold
val calculateChance : chance:int -> float

Full name: Main.calculateChance
val map : mapping:('T -> 'U) -> source:seq<'T> -> seq<'U>

Full name: Microsoft.FSharp.Collections.Seq.map
val x : int
val average : source:seq<'T> -> 'T (requires member ( + ) and member DivideByInt and member get_Zero)

Full name: Microsoft.FSharp.Collections.Seq.average
val averageBy : projection:('T -> 'U) -> source:seq<'T> -> 'U (requires member ( + ) and member DivideByInt and member get_Zero)

Full name: Microsoft.FSharp.Collections.Seq.averageBy
val percentage : float