I contain examples in C# and F# of functional programming.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
FExamples Complex tests with Try Aug 7, 2018
LanguageExtExamples
.gitignore
LanguageExtExamples.sln
README.md

README.md

Elevated Examples

I contain examples in C# and F# of functional programming. I wrote some example code in F# and then tried to use the same patterns to write the C# code. The C# code makes use of LanguageExt, a functional helper library for C#.

Background

Each test case has a workflow that goes something like Get data -> Mutate data -> Validate -> Log -> Map to different type. The idea is that there are different elevated types that need to be mapped between to allow for chaining. This is meant to be very practical with only as much theory as needed to understand the code.

Why use C# for functional programming(FP)

First of all FP is not an all or nothing choice. Michael Feathers wrote about using a hybrid approach back in 2012 and others have even further back. Some would even argue that Actor model coupled with FP may be closer to what Alan Kay initially envisaged anyway.

For me the main benefit is the minimization of moving parts, and so an increase in predictability of the system. This is done in a few ways. The first one is purity of functions. By minimizing the functions that mutate state, and pushing those to the outside, your system becomes both predictable and testable.

Next an emphasis on honest argument and return types makes functions very explicit about what they do. OOP values encapsulation but often that encapsulation means information on intent and consequence is lost. This makes it hard to reason and predict the behavior of a system. A return result of Result<MyType option,string> tells me this function could throw an error, so it is probably an IO type of function. It also tells me the data is optional so it might return no data. Finally the error case will give me a string describing the error. That is explicit, and I can code accordingly without knowing the internal details. With a C# method that throws exceptions I need to know about the internal details to handle the failure modes. That is actually poor encapsulation.

Related to this but not mentioned explicitly is the idea of immutability. Getting a new instance back rather than mutating state avoids many weird unintended side-effects. Unfortunately this is fairly painful in C# currently.

class MyType
{
    public int Nr { get; private set; }
    public MyType(int nr) => Nr = nr;
    public MyType With(int nr) => new MyType(nr);
}

Examples

  • Example of an immutable C# type
  • Example of an immutable C# list
  • flip function parameter order to allow currying
  • curry (partial application) functions so only a single parameter needed so the can be chained together
  • Composing a workflow with Apply or pipe |> (Note I don't mean composing as F# >> but as a concept)
  • Usage of Tap (or tee) to change a unit returning function into a pass-through function
  • mapError throwing an exception and LangExt Try

Overview

Although not necessary, I defined some function types in F# that formed the definition of the main functions chained together in my use cases. These were useful even when defining the C# implementations.

type GetMyTypeFn = int -> Result<MyType option,string>
type SetFn = Result<MyType option,string> -> int -> Result<MyType,string>
type ValidatePositiveFn = Result<MyType,string> -> Result<MyType,string>
type LogFn = Result<MyType,string> -> Result<MyType,string>
type ConvertFn = Result<MyType,string> -> Result<MyTypeDescriptor,string>

The final happy-path workflow in F# then looks like this:

let setTo2 = (set |> flip) 2
let r = get(1) |> setTo2 |> validate |> convert |> tapLog

and in C#:

Func<int, Func<OptionalResult<MyType>, Result<MyType>>> currySet = curry(flip(Set));
var set2 = currySet(2);
var result =
    Get(1)
    .Apply(set2)
    .Apply(Validate)
    .Apply(Convert);

As you can see from the definition of SetFn, the int parameter comes after the elevated Result type which would be passed through when chaining. So to use partial application we need to flip the parameter order.

A note on Elevated types (Return)

Return functions elevate normal values to the elevated world.

So the first thing to point out is that Elevated Types is not an official thing. Scott introduces the idea on fsharpforfunandprofit.com to avoid using functional programming terms that can initially be quite overwhelming.

Basically they are types that have some sort of state. The 2 we deal with here are Option and Result.

  • Option represents something where data might not be present. Return functions for Option are Some and None.
  • Result represents a return type that might be data but instead could also be an error. In F# to lift a value you use Ok and Error. Note: in the examples I will often shorten Result<MyType,string> to Result<MyType> to keep things concise.

Here is an example in F# of a function that takes in Result<MyType option,string> and returns Result<MyType,string>

//Result<MyType option,string> -> Result<MyType,string>
let errorIfNone r =
    match r with
    | Ok (Some x) -> Ok x
    | Ok None -> Error "Not found"
    | Error s -> Error s

So it checks if there is no data and converts that None case to an Error.

A final note on elevated types in general. It is best to stay in the elevated world as much as possible within your application. If you keep dropping back to normal values to work with them you will experience a lot of friction. Instead of getting values out and working with them it is better to use the techniques outlined in this example to use functions to manipulate the wrapped values within the elevated world. This will not always be possible but it is more often than you initially would think. It does require a change in mindset. Instead of get a piece of data and issuing imperative commands that manipulate it you use functions to declare what you would like to happen and then hand those functions off appropriately.

See here for further reading on return

Using Map

map allows you to apply a function to the normal value inside the elevated type. In the example below we define a function f that take a MyType and transforms it into MyTypeDescriptor.

//Result<MyType> -> Result<MyTypeDescriptor>
let convert : ConvertFn = fun r ->
    //MyType -> MyTypeDescriptor
    let f (x:MyType) = x.Nr |> toString |> MyTypeDescriptor.create
    Result.map f r

So the function signature is MyType -> MyTypeDescriptor. Calling Result.map with this function and a data value with type Result<MyType> will return a value with type Result<MyTypeDescriptor>. Thus we have mapped a value from one type to another while staying in the same elevated world.

Do you see Map is the same as LINQ Select?

See here for further reading on map

Using Apply

apply is a little different. Apply is used when you have an elevated value and an elevated function. Applying the function which is in terms of elevated types yields an elevated value out. So in the snippet:

...
.Apply(Validate)
.Apply(Convert)

Validate has the signature Result<MyType> -> Result<MyType> and as we saw earlier Convert has Result<MyType> -> Result<MyTypeDescriptor>.

Validate input: Result<MyType> output: Result<MyType>

Convert input: Result<MyType> output: Result<MyTypeDescriptor>

As you can see, the output of Validate matches the input of Convert. So Convert is a function in elevated Result that will take the output value of Validate and output the result of that function when applied. All this in the elevated world of Result.

See here for further reading on apply

Using Bind

bind allows us to cross between worlds, moving from normal world to elevated. Where map used a function that operated in the normal world like in the Convert example of MyType -> MyTypeDescriptor, bind uses a function that crosses worlds eg. MyType -> Result<MyType>.

//MyType -> Result<MyType>
let validateMyTypeIsPositiveR x = if validateMyTypeIsPositive x then Ok x else Error "Number should not be negative"

//Result<MyType> -> Result<MyType>
let validate r = Result.bind validateMyTypeIsPositiveR r

So here validateMyTypeIsPositiveR is a function that lifts a normal type to an elevated one

Do you see Bind is the same as LINQ SelectMany?

See here for further reading on bind

Summary of Map, Apply, and Bind

Map: map internal value to another value
Bind: For function that takes normal value and returns an elevated type
Apply: For function that takes an elevated value and can return anything

Example from ComplexTests.cs

[Fact]
public void Use_Bind_Instead_Of_ToTry_To_Return_A_Different_Elevated_Type()
{
    var toTry = fun<TryOption<Person>, Try<Person>>(opt => opt.Match(
            Some: x => Try(x),
            None: () => Try<Person>(new ArgumentNullException()),
            Fail: ex => Try<Person>(ex)
        ));

    var updatedPerson =
        people
        .Fetch("Bob")
        // map: on an E(x) takes in the normal wrapped value x and returns a normal value y. Result is transformed value E(y).
        .Map(p => Person.UpdateName(p, "Bobby"))
        // apply: on an E(x) takes in E(x) and returns whatever. Useful for changing elevated types or passing to function that accepts E(x)
        .Apply(toTry)
        //bind: on an E(x) takes in the normal wrapped value x and returns a E(y). Useful when function takes in normal value but returns same elevated value.
        .Bind(p => people.Update("Bob", p))
        .Try();
    var updated = ElevatedTypesUnsafeHelpers.ExtractUnsafe(people.Fetch("Bobby").Try());
    Assert.Equal("Bobby", updated.Name);
}

Curry, Adapters, Tee/Tap, and Error handling

Often the input and output of function calls don't line up and you need to do some extra work to get types to match up.

Partial Application and Currying

If functions have more than 1 input you can use partial application to apply values to the function and get a new function back with that value baked in. Earlier we had the example of the set function. The set function has the signature Result<MyType option> -> int -> Result<MyType>. Partial application works from the first parameter so we first call flip on set. Note that partial application works automatically with F# if not all parameters are supplied.

//set:Result<MyType option> -> int -> Result<MyType>
let flippedSet = flip set //int -> Result<MyType option> -> Result<MyType>
let set2 = flippedSet 2//Result<MyType option> -> Result<MyType> with the 2 now baked in

In the C# example things are a little busier because C# does not support it so we use some functions from the LanguageExt library to help

//flip so int is in correct place to curry, then curry the function with value of 2
Func<int, Func<OptionalResult<MyType>, Result<MyType>>> currySet = curry(flip(Set));
var set2 = currySet(2);

Adapters

In the examples above I chose to use Set, one of the main workflow parts to map from Result<MyType option> to change to Result<MyType>. That probably isn't the best but luckily internally it just uses an adapter function. So I can reuse that. A workflow that does not use set could then look something like this:

//adapters
let errorIfNone r =
    match r with
    | Ok (Some x) -> Ok x
    | Ok None -> Error "Not found"
    | Error s -> Error s

//workflow
let r = get(1) |> errorIfNone |> validate |> convert |> tapLog

errorIfNone is an adapter from the optional result to the non-optional result.

Tee/Tap

When chaining these workflows having a function that returns unit (think of it as a functional void except it is a value) isn't very useful. void functions are usually something that changes state. You can continue to chain calls together though by defining a function that does what needs to be done and then just returns the input parameter. That function is often called tee or tap.

let tap f x = //or tee
    f x |> ignore
    x

//normal logging function
let logObj a = printf "%A" a

//pass-through logging
let tapLog a = tap logObj a
static T1 Tap<T1>(Action<T1> f, T1 x) { f(x); return x; }

As you can see we can use tap as an adapter that gives us a function that can be chained.

Exception handling

TODO

Lists

TODO