Recommendations

Recommendations for object-oriented programming

No object orientation where F♯ is good

Type inference works better with function object than object.Member

Simple object hierarchy

❌ Avoid inheritance

✅ Prefer type Union and exhaustive pattern matching

Recursive types

This is particularly true for recursive types. You can define a fold function for them.

🔗 The "Recursive types and folds" series, F# for fun and profit

Structural equality

❌ Avoid class (reference equality by default)

✅ Prefer a Record or a Union

👌 Alternatively, consider Struct for performance purposes

❓ Consider custom structural equality for performance purposes 🔗 Custom Equality and Comparison in F#, Compositional IT

Object-oriented recommended use-cases

1. Encapsulate mutable state → in a class

2. Group features → in an interface

3. Expressive, user-friendly API → tuplified methods

4. F♯ API consumed in C♯ → member extensions

5. Dependency management → injection into the constructor

6. Tackle higher-order functions limits

Class to encapsulate mutable state

// 😕 Encapsulate mutable state in a closure → impure function → counter-intuitive ⚠️
let counter =
    let mutable count = 0
    fun () ->
        count <- count + 1
        count

let x = counter ()  // 1
let y = counter ()  // 2

// ✅ Encapsulate mutable state in a class
type Counter() =
    let mutable count = 0 // Private field
    member _.Next() =
        count <- count + 1
        count

Interface grouping features

let checkRoundTrip serialize deserialize value =
    value = (value |> serialize |> deserialize)
// val checkRoundTrip :
//   serialize:('a -> 'b) -> deserialize:('b -> 'a) -> value:'a -> bool
//     when 'a : equality

serialize and deserialize form a consistent group → Grouping them in an object makes sense

let checkRoundTrip serializer data =
    data = (data |> serializer.Serialize |> serializer.Deserialize)

💡 Prefer an interface to a Record (not possible with Fable.Remoting)

// ❌ Avoid: not a good use of a Record: unnamed parameters, structural comparison lost...
type Serializer<'T> = {
    Serialize: 'T -> string
    Deserialize: string -> 'T
}

// ✅ Recommended
type Serializer =
    abstract Serialize<'T> : value: 'T -> string
    abstract Deserialize<'T> : data: string -> 'T

• Parameters are named in the methods

• Object easily instantiated with an object expression

User-friendly API

// ✖️ Avoid                         // ✅ Favor
module Utilities =                  type Utilities =
    let add2 x y = x + y                static member Add(x,y) = x + y
    let add3 x y z = x + y + z          static member Add(x,y,z) = x + y + z
    let log x = ...                     static member Log(x, ?retryPolicy) = ...
    let log' x retryPolicy = ...

Advantages of OO implementation:

• Add method overloaded vs add2, add3 functions (2 and 3 = args count)

• Single Log method with retryPolicy optional parameter

🔗 F♯ component design guidelines - Libraries used in C♯

F♯ API consumed in C♯

Do not expose this type as is:

type RadialPoint = { Angle: float; Radius: float }

module RadialPoint =
    let origin = { Angle = 0.0; Radius = 0.0 }
    let stretch factor point = { point with Radius = point.Radius * factor }
    let angle (i: int) (n: int) = (float i) * 2.0 * System.Math.PI / (float n)
    let circle radius count =
        [ for i in 0..count-1 -> { Angle = angle i count; Radius = radius } ]

💡 To make it easier to discover the type and use its features in C♯

• Put everything in a namespace

• Augment type with the functionalities implemented in the companion module

namespace Fabrikam

type RadialPoint = {...}
module RadialPoint = ...

type RadialPoint with
    static member Origin = RadialPoint.origin
    static member Circle(radius, count) = RadialPoint.circle radius count |> List.toSeq
    member this.Stretch(factor) = RadialPoint.stretch factor this

👉 The API consumed in C♯ is ~equivalent to:

namespace Fabrikam
{
    public static class RadialPointModule { ... }

    public sealed record RadialPoint(double Angle, double Radius)
    {
        public static RadialPoint Origin => RadialPointModule.origin;

        public static IEnumerable<RadialPoint> Circle(double radius, int count) =>
            RadialPointModule.circle(radius, count);

        public RadialPoint Stretch(double factor) =>
            new RadialPoint(Angle@, Radius@ * factor);
    }
}

Dependency management

FP based technique

Parametrization of dependencies + partial application

• Small-dose approach: few dependencies, few functions involved

• Otherwise, quickly tedious to implement and to use

module MyApi =
    let function1 dep1 dep2 dep3 arg1 = doStuffWith dep1 dep2 dep3 arg1
    let function2 dep1 dep2 dep3 arg2 = doStuffWith' dep1 dep2 dep3 arg2

OO technique

Dependency injection

• Inject dependencies into the class constructor

• Use these dependencies in methods

👉 Offers a user-friendly API 👍

type MyParametricApi(dep1, dep2, dep3) =
    member _.Function1 arg1 = doStuffWith dep1 dep2 dep3 arg1
    member _.Function2 arg2 = doStuffWith' dep1 dep2 dep3 arg2

✅ Particularly recommended for encapsulating side-effects : → Connecting to a DB, reading settings...

Trap

Dependencies injected in the constructor make sense only if they are used throughout the class.

A dependency used in a single method indicates a design smell.

Advanced FP techniques

Dependency rejection = sandwich pattern

• Reject dependencies in Application layer, out of Domain layer

• Powerful and simple 👍

• ... when suitable ❗

Free monad + interpreter patter

• Pure domain

• More complex than the sandwich pattern but working in any case

• User-friendly through a dedicated computation expression

Reader monad

• Only if hidden inside a computation expression

...

🔗 Six approaches to dependency injection, F# for Fun and Profit, Dec 2020

Higher-order function limits

☝️ It's better to pass an object than a lambda as a parameter to a higher-order function when:

1. Lambda arguments are not explicit

❌ let test (f: float -> float -> string) =...

✅ Solution 1: type wrapping the 2 args float → f: Point -> string with type Point = { X: float; Y: float }

✅ Solution 2: interface + method for named parameters → type IPointFormatter = abstract Execute : x:float -> y:float -> string

2. Lambda is a command 'T -> unit

✅ Prefer to trigger a side-effect via an object → type ICommand = abstract Execute : 'T -> unit

3. Lambda: "really" generic!?

let test42 (f: 'T -> 'U) =
    f 42 = f "42"
// ❌ ^^     ~~~~
// ^^ Warning FS0064: This construct causes code to be less generic than indicated by the type annotations.
//                    The type variable 'T has been constrained to be type 'int'.
// ~~ Error FS0001: This expression was expected to have type 'int' but here has type 'string'
// 👉 `f: int -> 'U'` expected

✅ Solution: wrap the function in an object

type Func2<'U> =
    abstract Invoke<'T> : 'T -> 'U

let test42 (f: Func2<'U>) =
    f.Invoke 42 = f.Invoke "42"