Demystifying Jeremy's Interfaces

At the last Elm Online Meetup, there were two great talks:

  • “Plug-and-play design systems for elm applications” from Georges Boris
  • “Thought experiment: Hiding implementation types in Elm” from Jeremy H. Brown

After the meeting, Jeremy posted a link to the slides of his talk on the Elm Slack and wrote:

I believe this talk got the most WTFs I have ever gotten from a talk, so I consider it a total success :slight_smile:

I don’t want to offend anyone, so from now one I don’t use the three-letter-abbreviation but the symbol :interrobang: “to express or describe outraged surprise, recklessness, confusion, or bemusement” (according to Merriam-Webster).

Watching Jeremy’s talk, I as well thought:

:interrobang:

There are many, many things I don’t understand, and I’m perfectly fine with it, either because they just are way over my head, or because I don’t feel the need to invest enough time and energy to get a better understanding. But there are things for which an inner voice keeps telling me:

Pit, given what you know, you really should be able to understand this.

Jeremy’s code to hide implementation types behind interfaces was one of those things, so I sat down trying to understand what was going on. To be honest, the title of this post is clickbait. The mystic of his code still remains, but at least I’ve been able to concentrate it to a single one-line function, to find an alternative way to structure the code, and to fix a flaw in the implementation.

I love reading about the thought process of others tackling a problem, so in the next days (weeks?), I’ll keep adding comments to this post describing the steps I took trying to understand Jeremy’s interfaces.

I’ll add links to each one of my comments right here like in a table of contents, so feel free to add your comments, too, if you like.

Analyzing the Code

Telling a Story

Recap

Motivation: A List of Different Things

(This introduction will be too easy for most of you. Sorry for this. We’ll come back to this example after looking at more interesting code.)


On the “#beginners” channel of the Elm Slack, people often ask questions like:

How can I create a list of different things, for example a list with strings and integers?

In other languages it isn’t a problem to define a type like string|number (TypeScript) and create a list of it, but not so in Elm. The obvious answer for Elm is:

You have to wrap them in a custom type.

For example like in

type Thing
    = AString String
    | AnInt Int

Now we can create a list of Things:

myListOfThings : List Thing
myListOfThings =
    [ AString "one", AnInt 1, AnInt 12 ]

Fine. But why might one want to create a list of different things? What can I do with such a list?

One possible reason could be:

To be able to perform the same actions on the elements, independent of their underlying type.


Let’s say we want to get the “sizes” of the Things in our list. If we had a function like

thingSize : Thing -> Int

we could simply use List.map:

List.map thingSize myListOfThings

Such a function is easy to implement: we just use a case and then handle each type in its own branch:

thingSize : Thing -> Int
thingSize thing =
    case thing of
        AString string ->
            stringSize string

        AnInt int ->
            intSize int

For the “size” of a String we choose its length:

stringSize : String -> Int
stringSize string =
    String.length string

For an Integer, we choose the number of bits it takes to represent the number:

intSize : Int -> Int
intSize int =
    if int < 2 then
        1

    else
        1 + floor (logBase 2 (toFloat int))

Now we can get the sizes of the things in our list:

List.map thingSize myListOfThings
--> [ 3, 1, 4 ]

What about modifying a wrapped type? Say we want to “double” our Things:

thingDouble : Thing -> Thing

Again, we use a case and call the type-specific functions, but in this case we have to post-process the type-specific results with the appropriate wrapping constructor:

thingDouble : Thing -> Thing
thingDouble thing =
    case thing of
        AString string ->
            stringDouble string |> AString

        AnInt int ->
            intDouble int |> AnInt

“Doubling” a string and an int:

stringDouble : String -> String
stringDouble string =
    string ++ string

intDouble : Int -> Int
intDouble int =
    2 * int

Now we can double the list elements all at once and get their new sizes:

myListOfThings
    |> List.map thingDouble
    |> List.map thingSize
--> [ 6, 2, 5 ]

So far, so good. What if we want to support Bools too?

We have to add another subtype to the wrapper type:

type Thing
    = AString String
    | AnInt Int
    | ABool Bool

Now we can add wrapped Bools to the list:

myListOfThings : List Thing
myListOfThings =
    [ AString "one", AnInt 1, AnInt 12, ABool True ]

We also need a size and a double function for Bools. Here are two contrived implementations just to see whether something is happening:

boolSize : Bool -> Int
boolSize bool =
    if bool then
        1

    else
        0


boolDouble : Bool -> Bool
boolDouble bool =
    not bool

And we need to use the Bool functions in the wrapper functions:

thingSize : Thing -> Int
thingSize thing =
    case thing of
        AString string ->
            stringSize string

        AnInt int ->
            intSize int

        ABool bool ->
            boolSize bool


thingDouble : Thing -> Thing
thingDouble thing =
    case thing of
        AString string ->
            stringDouble string |> AString

        AnInt int ->
            intDouble int |> AnInt

        ABool bool ->
            boolDouble bool |> ABool

Now we can get the sizes of a list containing Bools as well:

List.map thingSize myListOfThings
--> [ 3, 1, 4, 1 ]

And we can “double” them and get the new sizes, too:

myListOfThings
    |> List.map thingDouble
    |> List.map thingSize
--> [ 6, 2, 5, 0 ]

So, to recap: what do we have to do to add yet another wrapped type, for example Char?

  1. Implement the type-specific functions (charSize, charDouble)
  2. Add the new subtype to the wrapper type
  3. Add a new case branch to the wrapper functions (thingSize, thingDouble)

There’s no way in Elm to get around task #1. We cannot automatically “derive” size and double functions for a new type. But what if we could omit tasks #2 and #3? Sounds magical?

It is, and it needs magic, the magic of Jeremy’s Interface module!

Stay tuned…

2 Likes

I also had an experiment with Jeremy’s idea. When I saw the talk title “Hiding implementation types in Elm”, I instantly thought - existential types for Elm, that sounds interesting.

Existential types are what makes object oriented style programming possible. Since they allow you to ‘hide’ the implementation type of something, allowing multiple implementations to have the exact same type.

I had an example of where I had previously used a poor-mans-typeclass in Elm. The idea is that I needed some kind of buffer that elements can be put into, and later retrieved. The retrievel order can vary - it could be FIFO, or LIFO or some kind of priority ordering based on the properties of the data elements in the buffer.

Originally I wanted to expose this buffer specification, so that users can provide their own implementations, or use mine and mix and match with other aspects of the API. The API is for doing classical AI search, where changing the buffer implementation gives breadth-first, or depth-first or some kind of ordered search (Artificial Intelligence: A Modern Approach, Chapter 3 - Solving Problems by Searching).

Here is the buffer from the current version of the package:

{-| Naming is a bit odd here, `orelse` might be better called `put` or 
`push` or something. It is called `orelse` in this implementation 
because the node part of a search space being saved for examination
at a later time, so it is an or-else possibility in the search.
-}
type alias Buffer state buffer =
    { orelse : Node state -> buffer -> buffer -- Pushes a node onto the buffer.
    , head : buffer -> Maybe ( Node state, buffer ) -- Tries to get a node from the buffer.
    , init : List (Node state) -> buffer -- Creates a buffer from a list of nodes.
    }

In this version, I had to expose the type of the buffers underlying implementation, as Elm does not support existential types. In the above Node state is the type of the search nodes, and buffer is the type of the buffer implementation, which could be a List, or an Array, or a Heap and so on.

Buffer implementations here:
https://github.com/the-sett/ai-search/blob/master/src/Search.elm#L276
https://github.com/the-sett/ai-search/blob/master/src/Search.elm#L293
https://github.com/the-sett/ai-search/blob/master/src/Search.elm#L304

The thing I got excited about Jeremy’s way of hiding a type by threading it through a continuation, was that it could let me hide this existential buffer type.

Here is what that looks like:

type Buffer state
    = Buffer (BufferIF state)


type alias BufferIF state =
    { orelse : Node state -> Buffer state
    , head : Maybe ( Node state, Buffer state )
    }

And an example implementation of the FIFO is:

import Existential as E exposing (..)

fifo : List (Node state) -> Buffer state
fifo =
    impl BufferIF
        |> wrap (\raise rep n -> raise (n :: rep))
        |> wrap
            (\raise rep ->
                case rep of
                    [] ->
                        Nothing

                    x :: xs ->
                        Just ( x, raise xs )
            )
        |> map Buffer
        |> init (\raise rep -> raise rep)

Where the Existential module contains the impl, wrap, add, map and init functions from the talk slides. Code for it in this Gist Type hiding in Elm by using continuations. · GitHub.

Buffer is now a slightly better poor-mans-typeclass. A middle-class-mans-typeclass perhaps? :grin:

A very neat idea, now I could have a list of AI search strategies which all have the same type even when different buffer implementations are being used. For example, maybe I want to try out several AI search strategies in parallel, now it is possible to iterate over them in a list.

My concern about this technique would be does it have a performance overhead? I always believed that continuations in javascript can be slow, but have never measured it or recall any numbers on the subject. An AI search should be able to examine millions of nodes, as search spaces tend to grow exponentially. Chess is the classical example of using this kind of search, nowadays superceded by deep learning machines.

So a TODO for me is to benchmark the two styles of buffers. Some more experiments with this idea to report on too…

1 Like

Thank you, Rupert, for your buffer example. You are a little bit further in your code than I am here, but I’ll catch up in this part.


Jeremy’s Original Code

So, after the (maybe not so interesting) introduction, let’s proceed straight to the example from Jeremy’s talk. I recommend reading his slides first, because I don’t repeat his introduction here.

First, he defines two mutually recursive types:

type Counter
    = Counter CounterRecord


type alias CounterRecord =
    { up : Int -> Counter
    , down : Int -> Counter
    , value : Int
    }

This defines an “interface” named Counter together with the functions up, down, and value, all three combined in the CounterRecord.

There certainly are similarities between Jeremy’s Counter and my introductory Thing example:

Jeremy Pit
Exposed type Counter Thing
Exposed functions up, down, value thingSize, thingDouble
Underlying types Int, List () String, Int, Char

But there are differences. The most important one: in Jeremy’s example, you don’t see the underlying types at all. In fact, I mentioned the underlying types Int and List () just because Jeremy used them in his examples, as we’ll see soon, but there could be many more. The definition of the “interface” is totally independent of the actual “implementations”!


Great, but now the question is: how can I create such a Counter?

This is where the magic begins. Here’s Jeremy’s code to create two kinds of Counters:

intCounter : Int -> Counter
intCounter =
    impl CounterRecord
        |> wrap (\raise rep n -> raise (rep + n))
        |> wrap (\raise rep n -> raise (rep - n))
        |> add identity
        |> map Counter
        |> init (\raise rep -> raise rep)


listCounter : Int -> Counter
listCounter =
    impl CounterRecord
        |> wrap (\raise rep n -> raise (List.repeat n () ++ rep))
        |> wrap (\raise rep n -> raise (List.drop n rep))
        |> add List.length
        |> map Counter
        |> init (\raise n -> raise (List.repeat n ()))

Both functions have the same signature: Int -> Counter. They return a Counter with the given Int as the starting value. An intCounter allegedly uses the type Int internally to represent the counter state, a listCounter uses the type List (), with the length of the list representing the counter state. We’ll look at the details of these functions in the next parts…

Here are the helper functions that make it all possible:

impl : t -> (raise -> rep -> t)
impl constructor =
    \_ _ -> constructor


wrap : (raise -> rep -> t) -> (raise -> rep -> (t -> q)) -> (raise -> rep -> q)
wrap method pipeline raise rep =
    method raise rep |> pipeline raise rep


add : (rep -> t) -> (raise -> rep -> (t -> q)) -> (raise -> rep -> q)
add method pipeline raise rep =
    method rep |> pipeline raise rep


map : (a -> b) -> (raise -> rep -> a) -> (raise -> rep -> b)
map op pipeline raise rep =
    pipeline raise rep |> op


init : ((rep -> sealed) -> flags -> output) -> ((rep -> sealed) -> rep -> sealed) -> flags -> output
init initRep pipeline flags =
    let
        raise : rep -> sealed
        raise rep =
            pipeline raise rep
    in
    initRep raise flags

As I’ve already written, at this point in the talk my reaction was simply:

:interrobang:


Before trying to understand how this code works, I wanted to check whether it works at all.

Sidemark: the German translation of “to understand” is “begreifen”. This word contains another German verb: “greifen”, which means “to grab” or “to take hold of”. I like this connection, because in order to understand something it often helps if I can grab it, touch it and play with it.

We have functions to create Counters, but the functions in the CounterRecord look a little bit unfamiliar. For example, for a function like up, which increases a Counter, I’d expect a signature like

up : Int -> Counter -> Counter

but in the CounterRecord we have the function

up : Int -> Counter

At this place, it helps to have some OO knowledge. In an object-oriented programming language, for example in Kotlin, we could define a Counter interface like this:

interface Counter {
    fun up(n: Int): Counter
    fun down(n: Int): Counter
    val value: Int
}

Note that here we have exactly the same function signatures as in Jeremy’s CounterRecord!

We continue with the OO example: Whenever I have an object which “implements” the Counter interface, then I can call the defined methods on this object. For example, in Kotlin, I could write a function to increment a Counter like this:

fun increment(counter: Counter): Counter =
    counter.up(1)

The function increment gets an object named counter which implements the Counter interface, and returns a new Counter object. Internally, it calls the up method on the counter object using the counter.up(1) syntax.

In Elm, the increment function could be defined (surprisingly similar) like this:

increment : Counter -> Counter
increment (Counter counterRecord) =
    counterRecord.up 1

Takeaway: Whenever I would call a method on an object implementing an interface in the OO world, in Elm, with Jeremy’s interfaces, I destructure the value to get the record with the functions and then call the appropriate one.

This allows me to implement some more Elm-like Counter functions:

up : Int -> Counter -> Counter
up n (Counter counterRecord) =
    counterRecord.up n


down : Int -> Counter -> Counter
down n (Counter counterRecord) =
    counterRecord.down n


value : Counter -> Int
value (Counter counterRecord) =
    counterRecord.value

And now I can start to play with Jeremy’s code.

Let’s create a list of different Counter types with different starting values:

myListOfCounters : List Counter
myListOfCounters =
    [ intCounter 1, listCounter 2, intCounter 11, listCounter 12 ]

Now we can check whether they were created with the correct initial values:

List.map value myListOfCounters
--> [ 1, 2, 11, 12 ]

Yeah, it seems to work!

Let’s increment them and then get the new values:

myListOfCounters
    |> List.map (up 3)
    |> List.map value
--> [ 4, 5, 14, 15 ]

And now decrement:

myListOfCounters
    |> List.map (down 3)
    |> List.map value
--> [ -2, 0, 8, 9 ]

Hmm. At first glance, the second value looks suspicious. “2 - 3” should be “-1” and not “0”. At second glance, the second and fourth Counters have been created with the function listCounter, so they internally use a value of type List (), and the length of the list is used to represent the current state of the counter. Since the length of a list can’t be negative, the implementation of listCounter isn’t able to represent negative values. The lowest value we can get is “0”.


Great. The code works as expected!

In case you want to play with the code, too, Jeremy posted an Ellie with the code from his slides, and here’s an Ellie with my test code so far.

In the next part, we’ll start to get a better understanding of Jeremy’s code as we begin to modify it a little bit.

3 Likes

Good that you are taking an interest too and exploring the idea, I look forward to seeing how you modified the original set of functions.

I would like to point out a limitation of this pattern, as compared with typical OO languages. In Java say, you can do this:

public class MyClass
{
    private int field;
    public equals(Object other) {
        if (o instanceof MyClass) { return (field == other.field); }
        return false;
    }
}

(Hope that is right, haven’t touched Java for a while…) Even though the field is private, within the MyClass implementation it is possible to look inside another instance of MyClass and access the field.

It is interesting to try to do this with this Elm pattern, assuming the int field is not exposed in the interface in any way. As the hidden representation is kept inside the continuation, it is only possible to get to it from inside. It is not possible to get to the private field inside another instance and the Elm typechecker will prevent you from ever doing that.

Attempt in Elm looks like:

type MyClass = MyClass MyClassIF

type alias MyClassIF = 
    { equals : MyClass -> Bool}

myClass = 
      impl MyClassIF
        |> add (\rep other -> other.field == rep.field )
        |> map MyClass
        |> init (\raise rep -> raise rep)

and the compiler gives the error:

This function cannot handle the argument sent through the (|>) pipe:

40|       impl MyClassIF
41|         |> add (\rep other -> other.field == rep.field )
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
The argument is:

    raise -> { b | field : a } -> (MyClass -> Bool) -> MyClassIF

But (|>) is piping it to a function that expects:

    raise -> { b | field : a } -> ({ c | field : a } -> Bool) -> q

Useful to understand the limits of this technique - the access rules are a bit more black and white than many OO languages typically allow. It would also be interesting to see how these things are dealt with in the OO calculi such as Ob<:, which were popular around the late 90s as accademic tools for exploring the semantics of OO. A Theory of Objects | SpringerLink

1 Like

First: Merry Christmas to everyone!

Second: Rupert, thank you once more for your interesting comparison between Jeremy’s interfaces and Java classes. I tested your example in IntelliJ, and indeed it works (after some minor corrections). You’re right that there’s no way to get at the internals of an “interface” in Elm. I’ll come back to this in a later post.


W r t a - Rename, Type alias, Applicative functor

Let’s start to look at the code in more detail.

Counter is just one example for the usage of Jeremy’s interfaces. In his talk, Jeremy also used the technique with pages of an SPA application, and Rupert used it with his buffer implementation. So, first, I’ll try to hide most of the Counter-specific things.

We can simply rename the types and the Counter functions:

What Old New
Main type of the interface Counter T
Type constructor Counter C
Record with the operations CounterRecord O
Function to create an Int counter intCounter intT
Function to create a List () counter listCounter listT

This gives us

type T
    = C O


type alias O =
    { up : Int -> T
    , down : Int -> T
    , value : Int
    }


intT : Int -> T
intT =
    impl O
        |> wrap (\raise rep n -> raise (rep + n))
        |> wrap (\raise rep n -> raise (rep - n))
        |> add identity
        |> map C
        |> init (\raise rep -> raise rep)


listT : Int -> T
listT =
    impl O
        |> wrap (\raise rep n -> raise (List.repeat n () ++ rep))
        |> wrap (\raise rep n -> raise (List.drop n rep))
        |> add List.length
        |> map C
        |> init (\raise n -> raise (List.repeat n ()))

Let’s examine the steps of the pipelines in intT and listT. In order to do so, I’ll change every line but the first into a comment and let my editor (VS Code) tell me the resulting type:

intT : (Int -> T) -> Int -> (Int -> T) -> (Int -> T) -> Int -> O
intT =
    impl O
        -- |> wrap (\raise rep n -> raise (rep + n))
        -- |> wrap (\raise rep n -> raise (rep - n))
        -- |> add identity
        -- |> map C
        -- |> init (\raise rep -> raise rep)

Next, I uncomment the second line, note the resulting type, and so on. In the end, this gives us

intT : Int -> T
intT =
    impl O
        -- (Int -> T) -> Int -> (Int -> T) -> (Int -> T) -> Int -> O
        |> wrap (\raise rep n -> raise (rep + n))
        -- (Int -> T) -> Int -> (Int -> T) -> Int -> O
        |> wrap (\raise rep n -> raise (rep - n))
        -- (Int -> T) -> Int -> Int -> O
        |> add identity
        -- (Int -> T) -> Int -> O
        |> map C
        -- (Int -> T) -> Int -> T
        |> init (\raise rep -> raise rep)
        -- Int -> T


listT : Int -> T
listT =
    impl O
        -- (List () -> T) -> List () -> (Int -> T) -> (Int -> T) -> Int -> O
        |> wrap (\raise rep n -> raise (List.repeat n () ++ rep))
        -- (List () -> T) -> List () -> (Int -> T) -> Int -> O
        |> wrap (\raise rep n -> raise (List.drop n rep))
        -- (List () -> T) -> List () -> Int -> O
        |> add List.length
        -- (List () -> T) -> List () -> O
        |> map C
        -- (List () -> T) -> List () -> T
        |> init (\raise n -> raise (List.repeat n ()))
        -- Int -> T

In theses types, there’s a repeating pattern, which Jeremy named “the pipeline shape” in his talk:

(Int     -> T) -> Int     -> ...
(List () -> T) -> List () -> ...

Let’s define a type alias for it. Int and List () are the hidden types used to represent the internal state of the counter. Jeremy names them rep. I’ll stick with one-letter names and use r. For the type T I’ll use the type variable t, and for the remaining “…” an innocent looking a. For the type alias itself I’d choose the name ⁉️, but once again I use a one-letter name. This gives me the following type alias:

type alias W r t a =
    (r -> t) -> r -> a

Note 1: Using type aliases for functions can make function signatures harder to understand, especially if the functions return those type aliases. In this case, it helps me to better understand the roles of Jeremy’s magic internal functions.

Note 2: In Jeremy’s code, the function (r -> t) is named raise, the second r is named rep. So whenever we see (raise -> rep -> xyz), we can replace it with W r t xyz.

Using the new type alias and renaming the type variables, the function signatures can be changed like this:

impl : t -> (raise -> rep -> t)
-->
impl : a -> W r t a
wrap : (raise -> rep -> t) -> (raise -> rep -> (t -> q)) -> (raise -> rep -> q)
-->
wrap : W r t a -> W r t (a -> b) -> W r t b
add : (rep -> t) -> (raise -> rep -> (t -> q)) -> (raise -> rep -> q)
-->
add : (r -> a) -> W r t (a -> b) -> W r t b
map : (a -> b) -> (raise -> rep -> a) -> (raise -> rep -> b)
-->
map : (a -> b) -> W r t a -> W r t b
init : ((rep -> sealed) -> flags -> output) -> ((rep -> sealed) -> rep -> sealed) -> flags -> output
-->
init : ((r -> t) -> i -> t) -> W r t t -> i -> t

Remark: the new signatures are not as general as the old ones, but they match the actual usage, so the new code still compiles.

Ok, the new signature of the wrap function looks familiar! It’s the well-known andMap function! You can find andMap functions in many packages, even for basic Elm types:

Maybe.Extra.andMap       : Maybe a    -> Maybe (a -> b)    -> Maybe b
Result.Extra.andMap      : Result e a -> Result e (a -> b) -> Result e b
Json.Decode.Extra.andMap : Decoder a  -> Decoder (a -> b)  -> Decoder b

wrap                     : W r t a    -> W r t (a -> b)    -> W r t b

W r t a is what in other functional languages is called an “Applicative Functor”.

To be an Applicative Functor, a type has to have two other functions, besides an andMap function. One is a map function, and we indeed have it. Jeremy named it already in the Applicative Functor jargon:

map : (a -> b) -> W r t a -> W r t b

The second missing function for an Applicative Functor is called pure, and it is right there, under another name:

impl : a -> W r t a

In Elm, there are many different names for such a function:

Maybe.Just :          a -> Maybe a
Result.Ok  :          a -> Result e a
Json.Decode.succeed : a -> Decoder a

So why not rename some of the internal functions? What about

impl --> succeed
wrap --> andMap
add  --> andAdd

I chose to rename add, too, because its signature looks very similar to that of andMap, and I therefore wanted to have a name that resembles andMap.

While renaming the functions, I’ll also rename the function parameters. Here’s the result:

succeed : a -> W r t a
succeed a _ _ =
    a


andMap : W r t a -> W r t (a -> b) -> W r t b
andMap rtra rtrab rt r =
    rtrab rt r (rtra rt r)


andAdd : (r -> a) -> W r t (a -> b) -> W r t b
andAdd ra rtrab rt r =
    rtrab rt r (ra r)


map : (a -> b) -> W r t a -> W r t b
map ab rtra rt r =
    ab (rtra rt r)


init : ((r -> t) -> i -> t) -> W r t t -> i -> t
init rtit rtrt i =
    let
        rt : r -> t
        rt r =
            rtrt rt r
    in
    rtit rt i

Why the … did I choose completely meaningless and unreadable parameter names?

Well, for the same reason that I renamed the original types. If I only have to look at the types and not at names like “method”, “pipeline”, and “raise”, it helps me to understand the effect of these functions.

The names I chose are not completely meaningless, of course. Let’s look at andMap:

andMap : W r t a -> W r t (a -> b) -> W r t b
andMap rtra rtrab rt r =
    rtrab rt r (rtra rt r)

The first parameter is of type W r t a, which is an abbreviation of (r -> t) -> r -> a. If you join these letters, you get the name I chose for the first parameter: “rtra”.

The second parameter is of type W r t (a -> b) or (r -> t) -> r -> (a -> b), giving me “rtrab”.

The function result is W r t b or (r -> t) -> r -> b. If a function like andMap returns another function like here, we can remove the parentheses around the function result:

andMap :
    ((r -> t) -> r -> a)
    -> ((r -> t) -> r -> (a -> b))
    -> ((r -> t) -> r -> b)

is the same as

andMap :
    ((r -> t) -> r -> a)
    -> ((r -> t) -> r -> (a -> b))
    -> (r -> t)
    -> r
    -> b

So, we could as well say that andMap takes two more parameters: one of type (r -> t), named “rt”, and the last one of type r named “r”. The function finally returns a value of type b.

This notation helps me to visually check whether the types are correct. Applying a function a -> b (which would be named “ab”) to a value of type a (named “a”) yields a value of type b. In my notation it means that the type of “ab a” is “b”. To get the resulting type (“b”), I only need to remove the name of the argument (“a”) from the beginning of the function name (“ab”). Here’s how this works in the case of andMap:

andMap : W r t a -> W r t (a -> b) -> W r t b
andMap rtra rtrab rt r =
    rtrab rt r (rtra rt r)
    --------    -------
       rab   r     ra   r
       -------     ------
          ab         a
          ------------
                b

It’s no surprise that this code type checks “visually”, too. This way of naming the parameters might help me later when I start to rearrange the code.


Using the new function and parameter names, I think I finally understand all the internal functions, with one exception: init. I don’t understand how it works, but at least I can visually verify that the types are correct. Here’s the function again:

init : ((r -> t) -> i -> t) -> W r t t -> i -> t
init rtit rtrt i =
    let
        rt : r -> t
        rt r =
            rtrt rt r
    in
    rtit rt i

The second parameter of type W r t t or (r -> t) -> r -> t seems to be especially important. The let clause somehow transforms this to a function of type r -> t, which is exactly what is needed as the parameter “raise” in the intT and listT functions. It is a function that takes an r (also called “rep”, a value of the internal type like Int or List ()) and transforms it to the type t, in our case to the main type T.

It’s still mysterious to me…


How do the user-supplied functions look like after the renaming? I’ll leave the type comments in the code, now using the type alias:

intT : Int -> T
intT =
    succeed O
        -- W Int T ((Int -> T) -> (Int -> T) -> Int -> O)
        |> andMap (\raise rep n -> raise (rep + n))
        -- W Int T ((Int -> T) -> Int -> O)
        |> andMap (\raise rep n -> raise (rep - n))
        -- W Int T (Int -> O)
        |> andAdd identity
        -- W Int T O
        |> map C
        -- W Int T T
        |> init (\raise rep -> raise rep)
        -- Int -> T


listT : Int -> T
listT =
    succeed O
        -- W (List ()) T ((Int -> T) -> (Int -> T) -> Int -> O)
        |> andMap (\raise rep n -> raise (List.repeat n () ++ rep))
        -- W (List ()) T ((Int -> T) -> Int -> O)
        |> andMap (\raise rep n -> raise (List.drop n rep))
        -- W (List ()) T (Int -> O)
        |> andAdd List.length
        -- W (List ()) T O
        |> map C
        -- W (List ()) T T
        |> init (\raise n -> raise (List.repeat n ()))
        -- Int -> T

If you know Parser or Decoder pipelines, at least the first steps should be understandable now:

  • We start with succeed O, where O is the record constructor function, with expects values for the up, down and value fields.

  • We then use andMap and andAdd to supply the values for the up, down, and value fields. What we get after these steps is a W r T O value, with r being the internal type, either Int or List ().

  • With map C this is changed to a value of type W r T T, which is exactly the special type mentioned above, which lets init do its magic to internally generate a function of type r -> T.

  • In both cases, init finally gives us a function Int -> T (this Int is the starting value for the counter).


Let’s take a final look at the parameters of andMap, andAdd and init:

In the counter example, andMap always gets a function with 3 parameters: “raise”, the type-raising function of type r -> T, “rep”, the internal representation of the current counter state of type r, and “n”, the Int parameter of the up and down functions.

In these functions, we calculate the result (again of type r). For example, in the down function, the values for the new state are rep - n and List.drop n rep, respectively. In the end, the “raise” function is applied to the result, so that we get back a value of type T.


andAdd is similar to andMap. In this example, it takes a function of type r -> Int, which returns the current counter state. Here it isn’t necessary to change the type of the result, so the functions don’t need a “raise” parameter.

To make it even more explicit and more similar to the andMap calls, I would change the andAdd lines:

        |> andAdd identity
-->
        |> andAdd (\rep -> rep)
        |> andAdd List.length
-->
        |> andAdd (\rep -> List.length rep)

Finally, init gets a function with 2 parameters: again “raise”, the type-raising function of type r -> T, and a second parameter, named “rep” in the intT function, and “n” in the listT function. In both cases it is the starting value for the counter of type Int, therefore I would name the parameter “n” in both cases.

In these functions, we create an internal representation of the starting value (of type r), and then apply the “raise” function to the result, so that we get back a value of type T, just like in the andMap cases.


With the minor changes mentioned above, the functions now look like this:

intT : Int -> T
intT =
    succeed O
        |> andMap (\raise rep n -> raise (rep + n))
        |> andMap (\raise rep n -> raise (rep - n))
        |> andAdd (\rep -> rep)
        |> map C
        |> init (\raise n -> raise n)


listT : Int -> T
listT =
    succeed O
        |> andMap (\raise rep n -> raise (List.repeat n () ++ rep))
        |> andMap (\raise rep n -> raise (List.drop n rep))
        |> andAdd (\rep -> List.length rep)
        |> map C
        |> init (\raise n -> raise (List.repeat n ()))

At least for me, it is now much easier to understand what we are doing here.


In the next part, I’ll start to change the structure of the code.

2 Likes

Also worth noting that we don’t really need add, its just a special case of wrap. Neat in the sense that it reduces the whole thing down to just the essential map, andMap, succeed and pure.

-- Shapes with bounding boxes.
type Shape = Shape ShapeIF

type alias ShapeIF msg =
    { bbox : BoundingBox2d Unitless Local
    }

-- Using add
shape bboxFn = 
    impl ShapeIF
        |> add (\rep -> bboxFn rep)
        |> map Shape
        |> init (\raise rep -> raise rep)

-- Using wrap
shape bboxFn = 
    impl ShapeIF
        |> wrap (\raise rep -> bboxFn rep) -- Didnt need the raise
        |> map Shape
        |> init (\raise rep -> raise rep)
2 Likes

Change #1: Remove the Record Constructor Function

In this part, we’ll start to look at the code from different angles: it could be from the point of view of the author of the interface technique, or from a user’s point of view.

While lying in my bed this morning, pondering about what to write in this part, I thought: why don’t you introduce some fictitious characters for those different viewpoints? It felt interesting, so let’s try it. So far, I have been simply analyzing the code. From now on, I’ll be more like telling a story. (I split the table of contents at the beginning of this series into the appropriate sections.) So let’s start our story. Of course, any similarities with existing persons are purely coincidental :wink:


The first character is Jeremy. Jeremy is a real type system wizard. He manages the magic internal functions we’ve analyzed in the last part. Jeremy packaged the functions and the type alias W in a neat module called Interface:

module Interface exposing (W, andAdd, andMap, init, map, succeed)

The next character, let’s name him Rupert, is a user of Jeremy’s Interface module. He’s very versed in applying the interface technique to various use cases, be it buffers, geometric shapes, or recently counters. Here’s his Counter code again:

import Interface as IF


type Counter
    = Counter CounterRecord


type alias CounterRecord =
    { up : Int -> Counter
    , down : Int -> Counter
    , value : Int
    }


intCounter : Int -> Counter
intCounter =
    IF.succeed CounterRecord
        |> IF.andMap (\raise rep n -> raise (rep + n))
        |> IF.andMap (\raise rep n -> raise (rep - n))
        |> IF.andAdd (\rep -> rep)
        |> IF.map Counter
        |> IF.init (\raise n -> raise n)


listCounter : Int -> Counter
listCounter =
    IF.succeed CounterRecord
        |> IF.andMap (\raise rep n -> raise (List.repeat n () ++ rep))
        |> IF.andMap (\raise rep n -> raise (List.drop n rep))
        |> IF.andAdd (\rep -> List.length rep)
        |> IF.map Counter
        |> IF.init (\raise n -> raise (List.repeat n ()))

(Note that we re-introduced the original type and function names, because as a user, Rupert’s main interest is the domain-specific logic.)

Overall, Rupert is fine with this code, but there’s something that bothers him. Lue, a friend of Rupert, keeps telling him:

Rupert, you never should use these record type alias constructor functions, never ever!

Rupert understands his friend’s concern. He knows that, because of the problematic usage of the CounterRecord function in the IF.succeed lines, the intCounter and listCounter implementations are dependent on the field order in CounterRecord. Rupert is responsible enough to not change the field order without good reason, but he is not the only one working on the code. For example, there’s his new colleague, Pit. Rupert knows that Pit loves to touch and to modify every single piece of code, whether he understands it or not, just to get a better feeling for it.

There’s no doubt that one day Pit will find the CounterRecord definition. If he changes the field order to something like

type alias CounterRecord =
    { value : Int
    , up : Int -> Counter
    , down : Int -> Counter
    }

it would be no problem, because the Elm compiler would show him where the other code had to be changed, too. But if he just swaps the up and down fields as in

type alias CounterRecord =
    { down : Int -> Counter
    , up : Int -> Counter
    , value : Int
    }

the code would still compile, and maybe Pit would check in his change, despite the fact that their app wouldn’t work as before.

So Rupert looks at the implementation of the listCounter function again, trying to understand how the problematic CounterRecord constructor function is used, and whether he can do anything against it. He looks at the types that every step in the pipeline returns, just as we did in the last part of the series, and he notices that the type after the IF.andAdd call looks interesting:

listCounter : Int -> Counter
listCounter =
    IF.succeed CounterRecord
        |> IF.andMap (\raise rep n -> raise (List.repeat n () ++ rep))
        |> IF.andMap (\raise rep n -> raise (List.drop n rep))
        |> IF.andAdd (\rep -> List.length rep)
        -- IF.W (List ()) Counter CounterRecord
        |> IF.map Counter
        |> IF.init (\raise n -> raise (List.repeat n ()))

The type is

IF.W (List ()) Counter CounterRecord

which is just a shorthand for

(List () -> Counter) -> List () -> CounterRecord

Here, CounterRecord is the record type, not the problematic record constructor function his friend Lue is talking about. Rupert says to himself:

Couldn’t I directly create such a value myself?

The type says, that it needs to be a function taking two parameters:

  • a function (typically called “raise” in the listCounter code) which can turn the internal representation of the counter state, in this case List (), into the main Counter type
  • the current counter state (normally called “rep”), in the form of the internally used List ()

The function should return a CounterRecord. So Rupert starts with:

firstFourSteps : (List () -> Counter) -> List () -> CounterRecord
firstFourSteps raise rep =
    ...

He needs to return a record with the fields up, down, and value, and suddenly the function body just flows from his fingers:

firstFourSteps : (List () -> Counter) -> List () -> CounterRecord
firstFourSteps raise rep =
    { up = \n -> raise (List.repeat n () ++ rep)
    , down = \n -> raise (List.drop n rep)
    , value = List.length rep
    }

This function type-checks! Rupert doesn’t want to keep the code in the form of a top-level function, so he just takes the function body and uses it to replace the first four steps of the pipeline:

listCounter : Int -> Counter
listCounter =
    (\raise rep ->
        { up = \n -> raise (List.repeat n () ++ rep)
        , down = \n -> raise (List.drop n rep)
        , value = List.length rep
        }
    )
        |> IF.map Counter
        |> IF.init (\raise n -> raise (List.repeat n ()))

He runs a few tests, and the code works! Rupert calls Jeremy, the author of the Interface module, and tells him what he changed in his code and why he did it.


Jeremy, being the type system wizard he is, quickly recognizes the potential of Rupert’s new way to use the Interface module. He not only could completely remove the succeed, andMap, and andAdd functions, but even go one step further: the next step in the user’s pipeline would always be

        |> IF.map C

with C being the constructor function of the type

type T
    = C O

(As a type system wizard, Jeremy prefers to use the abstract names…)

Jeremy is a friendly wizard who always wants the best for the users of his module, so he’ll save his users from having to write the IF.map step every time. He removes the map function from the exposes list of the Interface module, too, and exposes a new function instead. For this function he chooses a name which is still well-known to his users. In a previous version of the module, the success function has been named impl to signal the start of the implementation of an interface.

This is the signature of the new function:

impl : (o -> t) -> W r t o -> W r t t

Jeremy really is a genius! This is the same signature as that of the previous map function, with slightly renamed type variables to better reflect the desired usage. By simply renaming the map function, he enables his users to replace the first lines of an interface pipeline with

    IF.impl C
        (\rt r ->
            ...
        )

He publishes a new version of the interface module and tells Rupert how to use it.


Here’s Rupert’s new code for the intCounter and listCounter functions:

intCounter : Int -> Counter
intCounter =
    IF.impl Counter
        (\raise rep ->
            { up = \n -> raise (rep + n)
            , down = \n -> raise (rep - n)
            , value = rep
            }
        )
        |> IF.init (\raise n -> raise n)


listCounter : Int -> Counter
listCounter =
    IF.impl Counter
        (\raise rep ->
            { up = \n -> raise (List.repeat n () ++ rep)
            , down = \n -> raise (List.drop n rep)
            , value = List.length rep
            }
        )
        |> IF.init (\raise n -> raise (List.repeat n ()))

I don’t know about you, but I very much like how this code reads. Here’s an Ellie with the actual code.


In the evening, Rupert meets Lue in a bar and proudly tells him that he not only managed to completely remove record alias constructors from their codebase, but also successfully added Lue’s elm-review rule which forbids record type alias constructors to the set of their project’s elm-review rules.

Later at home, he falls asleep relieved, knowing that even his colleague Pit can’t break the code simply by changing the order of the CounterRecord fields.


Meanwhile, Jeremy, the type system wizard, has a new idea for improving his Interface module even further…

2 Likes

Change #2: Redesigned Initialization

Happy new year! I’ll continue the story…


Rupert and Jeremy are having lunch, and obviously they also talk about interfaces. Rupert tells Jeremy:

You know that I have functions like

intCounter : Int -> Counter

and

listCounter : Int -> Counter

which take a starting value and create a Counter, appropriately backed by either an Int or a List (). But I noticed that the users of these functions don’t really need the starting value, because they always create counters starting from zero, yet. So I was thinking whether I could remove this parameter. In the current function implementations, the parameter isn’t declared explicitly, but if it would be, I would have to pass it as the last parameter to the init function of the Interface module:

listCounter : Int -> Counter
listCounter start =
    ...
        |> (\pipeline ->
                IF.init (\raise n -> raise (List.repeat n ()))
                    pipeline
                    start
           )

(I had to introduce the current value of the pipeline as an intermediate parameter.)

So I thought: what would I do, if I wanted to pass two or more parameters to the “initialization function” (the first parameter of IF.init), or no parameter at all?


Jeremy smiles and answers:

You know what, Rupert? Right after we changed the Interface module last time, I’ve been looking at the remaining code. The impl function looks OK to me, but I had the feeling that the init function could still be improved. Before I tell you what I’ve been thinking about, please do me a favour: could you briefly look at the usages of the init function in your code? I suppose that the “initialization functions” you pass to the init function all have the following shape:

  1. They create an initial value of the representation type. In my function signatures, it has the type r.
  2. As the last step, they pass that value to the “raise” function of type r -> t, which I pass as the first parameter to your initialization functions.

After lunch, Rupert skims through their code.

intCounter : Int -> Counter
intCounter =
    ...
        |> IF.init (\raise n -> raise n)
--                                    -
--   initial value of type Int: n
--   passed to "raise"


listCounter : Int -> Counter
listCounter =
    ...
        |> IF.init (\raise n -> raise (List.repeat n ()))
--                                     ----------------
--   initial value of type List (): List.repeat n ()
--   passed to "raise"


fifo : List (Node state) -> Buffer state
fifo =
    ...
        |> IF.init (\raise rep -> raise rep)
--                                      ---
--   initial value of type List (Node state): rep
--   passed to "raise"


shape bboxFn = 
    ...
        |> IF.init (\raise rep -> raise rep)
--                                      ---
--   initial value: rep
--   passed to "raise"

He calls Jeremy and tells him: You’re right. They all have the shape you supposed. How did you know that?


Jeremy answers:

The initialization functions, which you pass as the first parameter to the init function, have to have the signature (r -> t) -> i -> t (in my abstract type notation). This means that they have to return a value of type t, in the counter example a value of type Counter.

How can you create such a value? The only way to do this is by calling the magic “raise” function, which takes a value of the internal representation type r. So, in your initialization functions, you first have to create this internal value and then pass it to the “raise” function.

After I recognized this, I knew that I could remove this last step (calling the “raise” function) from the user-supplied initialization functions and move it into the Interface module. This would make the init function look like

init : (i -> r) -> W r t t -> i -> t
init ir rtrt i =
    let
        rt : r -> t
        rt r =
            rtrt rt r
    in
    rt (ir i)

Now the initialization functions have the simpler type i -> r. They don’t get the “raise” function anymore, and just have to create an r value from the given i value. I’ll then call the former “raise” function, internally called “rt”, in the init function.

We can go even further: why should the init function be responsible to take the i value from the user and then directly pass it to the user-supplied initialization function? I don’t do anything else with the i value. Why shouldn’t the user perform the ir i call, which I do on the last line, in her own code and only give me the resulting r value? This would reduce the init function to

init : r -> W r t t -> t
init r rtrt =
    let
        rt : r -> t
        rt r_ =
            rtrt rt r_
    in
    rt r

An example for its usage:

listCounter : Int -> Counter
listCounter start =
    impl Counter
        ...
        |> init (List.repeat start ())

But why stop here? Syntactically, all the init function does in the “impl |> init” pipeline is to get another parameter of type r. If the impl function would take this parameter itself, we wouldn’t need a pipeline at all! We could combine the code of the impl and the init functions into one.

Before:

impl : (o -> t) -> W r t o -> W r t t
impl ot rtro rt r =
    ot (rtro rt r)


init : r -> W r t t -> t
init r rtrt =
    let
        rt : r -> t
        rt r_ =
            rtrt rt r_
    in
    rt r

After:

impl : (o -> t) -> W r t o -> r -> t
impl ot rtro r =
    let
        rtrt : W r t t
        rtrt rt_ r_ =
            ot (rtro rt_ r_)

        rt : r -> t
        rt r_ =
            rtrt rt r_
    in
    rt r

If we inline the internal rtrt function and remove the outer “r” parameter (to reduce shadowing), we get

impl : (o -> t) -> W r t o -> r -> t
impl ot rtro =
    let
        rt : r -> t
        rt r =
            (\rt_ r_ -> ot (rtro rt_ r_)) rt r
    in
    rt

We can simplify the last line in the internal rt function to get the final version:

impl : (o -> t) -> W r t o -> r -> t
impl ot rtro =
    let
        rt : r -> t
        rt r =
            ot (rtro rt r)
    in
    rt

In fact, in the next version of the Interface module, I’ll even remove the W type alias, because in the meantime it is only used once. The signature of the impl function then will be

impl : (o -> t) -> ((r -> t) -> r -> o) -> (r -> t)

Oh, by the way, Rupert, did you notice that we don’t have a value of type i anymore in the signature? Coming back to your original question, this means that you are completely free to choose how you want to create the initial r value. You can take one value from your user as before, two or more values, or no value at all. Here’s a version of your listCounter function starting always at zero:

zeroListCounter : Counter
zeroListCounter =
    IF.impl Counter
        (\raise rep ->
            { up = \n -> raise (List.repeat n () ++ rep)
            , down = \n -> raise (List.drop n rep)
            , value = List.length rep
            }
        )
        []

Note how you now simply pass the initial [] value of type List () (representing the counter value “zero”) as the last parameter to the impl function?


Rupert nods and says: it always feels extremely satisfying if you manage to improve your code just by removing parts of it, doesn’t it?


You can find the actual code in this Ellie. In the next part, we’ll be adding some code again…

2 Likes

So you have reduced the initial set of functions down to just a single function? Pretty neat. :clap:

impl : (o -> t) -> ((r -> t) -> r -> o) -> (r -> t)
impl ot rtro =
    let
        rt : r -> t
        rt r =
            ot (rtro rt r)
    in
    rt
1 Like

Do you think that using continuations to chain state like this, could lead to this compiler bug? Elm seems able to form bad closures when returning a lambda function. So far I have not run into any issues whilst exploring this.

1 Like

Hmm, I didn’t know these issues. But aren’t they caused by an error in Elm’s tail call optimization? If this is the cause, then I don’t think that Jeremy’s interface technique is affected by this problem, because it doesn’t use tail recursive functions, or does it?

Maybe the problem only shows up with the specific combination of TCO and returning a lambda - I don’t know. I get a little suspicious sometimes if you push the Elm compiler to do something weird thats all. Might take a little time this week to investigate this bug more fully and understand exactly what is needed to cause it.

Change #3: Separate Interface from Implementation

Remember that the characters in the story are purely fictional…


Time has passed, and Rupert is using interfaces in more and more places. He even wrote a blog post about it. Reflected as he is, he notices that his role in the interface business has shifted recently. He is mainly defining the interfaces, leaving the implementation to others, for example to Pit, who loves to use Lists for everything, just as in the listCounter example.

Just like Jeremy managed to hide the “raise” function from the users of his init function in the last part, Rupert would love to hide the “raise” function from the folks implementing his interfaces.

Applying the “raise” function is something which has to be repeated in each interface implementation:

intCounter : Int -> Counter
intCounter start =
    impl Counter
        (\raise rep ->
            { up = \n -> raise (rep + n)
            , down = \n -> raise (rep - n)
            , value = rep
            }
        )
        start


listCounter : Int -> Counter
listCounter start =
    impl Counter
        (\raise rep ->
            { up = \n -> raise (List.repeat n () ++ rep)
            , down = \n -> raise (List.drop n rep)
            , value = List.length rep
            }
        )
        (List.repeat start ())

Wouldn’t it be nice, if he could somehow hide the call of the “raise” function, so that the code would look more like

intCounter : Int -> Counter
intCounter start =
    impl Counter
        (\rep ->
            { up = \n -> rep + n
            , down = \n -> rep - n
            , value = rep
            }
        )
        start


listCounter : Int -> Counter
listCounter start =
    impl Counter
        (\rep ->
            { up = \n -> List.repeat n () ++ rep
            , down = \n -> List.drop n rep
            , value = List.length rep
            }
        )
        (List.repeat start ())

As always in a situation like this, he calls Jeremy and tells him about his new idea. Jeremy is skeptical at first, but promises to get back to Rupert.

Indeed, just a few days later, Jeremy calls back and says:

Jeremy

Rupert, I’m sorry, but this isn’t something I can do for you, because there’s no general way to define this behavior for each and every interface. But if you are willing to tell me which fields of the operations record need a “raise” call and which don’t, then I can change the impl function in a way that lets you completely hide the “raise” function from the implementers of your interfaces.

Of course, Rupert would gladly provide this information to the Interface module, so Jeremy sends him the new version of the module and guides him in the process to transform Rupert’s existing code to the new version.

Jeremy

OK Rupert, let’s start: first, you need to add a type parameter to the operations record. And could you rename the operations record from CounterRecord to, say, CounterOperations? In my mind they are always interface “operations”…

Rupert is fine with the name change. After all, he doesn’t know yet how the code will look like in the end, so he trusts Jeremy that the new name will be well chosen. The new types look like this:

type Counter
    = Counter (CounterOperations Counter)


type alias CounterOperations t =
    { up : Int -> t
    , down : Int -> t
    , value : Int
    }

Jeremy

Now you have to use a new function from the Interface module. The new version of the module doesn’t expose an impl function anymore, but a new interface method. You’ll see why I renamed the function, soon.

Jeremy’s new method has the following signature:

interface :
    (ot -> t)
    -> ((r -> t) -> or -> ot)
    -> (r -> or)
    -> (r -> t)

Rupert has worked with Jeremy’s code long enough to easily map the abstract type names to his concrete use cases. Here’s the mapping for the Counter example:

Abstract Concrete
t Counter
r hidden type, for example Int or List ()
ot CounterOperations Counter
or CounterOperations r

Rupert

Ok, Jeremy, I can translate the types to the counter example. What should I do with the function?


Jeremy

You as the interface designer should publish the interface function partially applied with the first two parameters.

Oh-K… Rupert looks at the first two parameters in more detail.

Rupert

The first parameter of type ot -> t (in my example CounterOperations Counter -> Counter) is the constructor function (Counter).


Jeremy

Right. I can tell you a little bit about the second parameter of type (r -> t) -> or -> ot.

It is a function that takes the well-known “raise” function of type r -> t and an operations record, parameterized with the type r (in your example CounterOperations r).

The function should return the same operations record, but now parameterized with the type t (in your example CounterOperations Counter).

So, basically, this is just a map function on the operations record.


Rupert

OK, a map function seems to be easy.


Jeremy

If I might add another naming proposal: I’d name the partially applied function “xxxInterface”.

Rupert writes the following counterInterface function:

counterInterface =
    interface Counter <|
        \raise ops ->
            { up = ops.up >> raise
            , down = ops.down >> raise
            , value = ops.value
            }

His IDE shows that the function signature is

counterInterface : (r -> CounterOperations r) -> (r -> Counter)

Rupert

Hey, I’m starting to understand how your new interface function helps to separate the interface definition from the implementation:

The second parameter tells the Interface module that the up and down fields of the CounterOperations record need a “raise” call, but the value field doesn’t. This is exactly the information I wanted to hide from the interface implementers.

Plus, I think I also understand why you named the function “interface”: the interface designers use it to define their interfaces.

So what can the interface implementers do with my counterInterface function?


Jeremy

They need to provide the next parameter, the one with type r -> CounterOperations r. This implements the counter operations for one specific internal representation type.


Rupert

No problem. I can easily extract the logic from the previous implementations for Int and List ().

As these will be the interface implementations, I’d probably name them counterImplInt and counterImplList. What do you think?


Jeremy

The names sound great. Go ahead.

Rupert writes the following implementations:

counterImplInt : Int -> CounterOperations Int
counterImplInt rep =
    { up = \n -> rep + n
    , down = \n -> rep - n
    , value = rep
    }


counterImplList : List () -> CounterOperations (List ())
counterImplList rep =
    { up = \n -> List.repeat n () ++ rep
    , down = \n -> List.drop n rep
    , value = List.length rep
    }

Rupert

Yes! This is exactly what I wanted to achieve: the implementers just have to be concerned with their internal type and nothing else. No need to call a “raise” function. All they need to do is to implement the interface operations using their internal type.


Jeremy

You’ve got it.

Now look again at the signature of the interface function. The last remaining part is r -> t.

When the implementers pass one of the “implementation” functions to the provided “interface” function, they get a function of type r -> t, which lets them create “objects” implementing the interface.


Rupert

Let’s just do this for the listCounter example. I really want to see the final result.

He writes:

listCounter : Int -> Counter
listCounter start =
    counterInterface counterImplList <|
        List.repeat start ()

Rupert

It compiles!

What do you think? Should I reverse the order? Let’s see…

intCounter : Int -> Counter
intCounter start =
    start
        |> counterInterface counterImplInt

Rupert

I’ll have to play with this a little bit more. At least the code is very modular now. Thank you very much, Jeremy.


Jeremy

I’m glad I could help you.

If you’re not sure about the “look-and-feel” of the code: there are many more options to structure the code. The only important thing is that you somehow collect the three values needed by the interface function and that you know how to combine them to get the r -> t function in the end.


Rupert

Ok, let me summarize the three values…

Rupert writes:

Type Contents
ot → t Constructor function
(r → t) → or → ot “map” function for the operations record
r → or Implementation of the interface operations

Rupert

Done. But how could I structure the code differently?


Jeremy

I’ll show you an example soon, but let’s first look at the implementation of the interface function to see how these three values can be combined:

interface : (ot -> t) -> ((r -> t) -> or -> ot) -> (r -> or) -> (r -> t)
interface ott rtorot ror =
    let
        rt : r -> t
        rt r =
            ott (rtorot rt (ror r))
    in
    rt

With the abstract names representing the types, it’s easy to visually type-check the code. The implementation of this new function looks very similar to the implementation of the previous impl function. Jeremy’s magic shines again!


Jeremy

OK, with that out of the way, here’s another possible syntax / DSL for the counter interface example:

The Counter and CounterOperations types stay the same as above.

The interface designer could supply her information in this way:

counterOperations : IF2.Operations r Counter (CounterOperations r) (CounterOperations Counter)
counterOperations =
    IF2.defineOperations <|
        \raise ops ->
            { up = ops.up >> raise
            , down = ops.down >> raise
            , value = ops.value
            }

Jeremy

I’m using an opaque type Operations here which could be defined in an Interface2 module and a function defineOperations to create such a value. I can show you their implementation later.

I’m sure you see what kind of value this type holds?


Rupert

Of course. It’s the “map” function for the CounterOperations record used to “raise” the parameterized record from the internal representation type to the Counter type.


Jeremy

Right. The interface implementers could use this to add their knowledge, too:

counterImplInt : IF2.Implementation Int Counter (CounterOperations Int) (CounterOperations Counter)
counterImplInt =
    IF2.implement counterOperations <|
        \rep ->
            { up = \n -> rep + n
            , down = \n -> rep - n
            , value = rep
            }


counterImplList : IF2.Implementation (List ()) Counter (CounterOperations (List ())) (CounterOperations Counter)
counterImplList =
    IF2.implement counterOperations <|
        \rep ->
            { up = \n -> List.repeat n () ++ rep
            , down = \n -> List.drop n rep
            , value = List.length rep
            }

Jeremy

Again, I’m using an opaque type Implementation and a function implement from a possible Interface2 module.


Rupert

Nice. Here you add the type-specific implementations of the interface operations.

But didn’t we miss the first value that is needed, the type constructor function?


Jeremy

I left it for the last part where we put all three things together. If the constructor function is named like the interface type, in your case Counter for both the type and the constructor, then the following code reads nicely:

intCounter : Int -> Counter
intCounter start =
    IF2.createInstanceOf Counter <|
        IF2.implementedBy counterImplInt <|
            IF2.fromValue <|
                start


listCounter : Int -> Counter
listCounter start =
    IF2.createInstanceOf Counter <|
        IF2.implementedBy counterImplList <|
            IF2.fromValue <|
                List.repeat start ()

Jeremy

I know that this is a very contrived example, but I wanted to show you what is possible.


Rupert

Ok, ok. This reads very nicely, indeed. I’m not sure I like the syntax with the many backward pipe operators, but I understand what you wanted to teach me: that we have many ways to design the interface API.

Can you show me the implementation of those types and functions?


Jeremy

Sure. Here they are:

type Operations r t or ot
    = Operations ((r -> t) -> or -> ot)


type Implementation r t or ot
    = Implementation
        { rtorot : (r -> t) -> or -> ot
        , ror : r -> or
        }


type Instance r t or ot
    = Instance
        { rtorot : (r -> t) -> or -> ot
        , ror : r -> or
        , r : r
        }


type Value r
    = Value r


defineOperations :
    ((r -> t) -> or -> ot)
    -> Operations r t or ot
defineOperations =
    Operations


implement :
    Operations r t or ot
    -> (r -> or)
    -> Implementation r t or ot
implement (Operations rtorot) ror =
    Implementation
        { rtorot = rtorot
        , ror = ror
        }


createInstanceOf :
    (ot -> t)
    -> Instance r t or ot
    -> t
createInstanceOf ott (Instance inst) =
    let
        rt : r -> t
        rt r =
            ott (inst.rtorot rt (inst.ror r))
    in
    rt inst.r


implementedBy :
    Implementation r t or ot
    -> Value r
    -> Instance r t or ot
implementedBy (Implementation impl) (Value r) =
    Instance
        { rtorot = impl.rtorot
        , ror = impl.ror
        , r = r
        }


fromValue :
    r
    -> Value r
fromValue =
    Value

Rupert

Oh, wow. I have to look at this in more detail, but I think I see how you use the types to gradually collect the needed values and finally put them all together in the… wait… the createInstanceOf function.

I’m sure I could change the DSL syntax, for example, to use forward pipe operators now.

Once again: thank you, Jeremy. You gave me a lot to play with.


Jeremy

Once again: I’m glad I could help. Feel free to ask me if you get stuck.


Is this the end of the story? A happy end?

Maybe… maybe not…

Here’s an Ellie with the new code and another one with the DSL like syntax.

1 Like

What are the practical benefits of doing it this way versus Jeremeny’s original code? Is it simply so that implementors of interfaces do not have to use raise?

My other comment is that I find this highly unreadable. Its all the rt rtor or r t oror stuff, both in type var name and fields. I know the exact choice of names for these things is hard to come up with, and that you chose these names to represent how the parts compose together, but I think the original set of names was also good. I mean rep is the representation of the data structure, raise raises the representation into its higher level hidden form, and so on. I can speak these names and attach meaning to them, which helps me a lot when building up a mental picture of what is going on. I might have a go at doing some renaming on it and see if I can get it back to something a bit more sane…

-- JEREMY'S MAGIC INTERFACE MODULE


type Operations r t or ot
    = Operations ((r -> t) -> or -> ot)


type Implementation r t or ot
    = Implementation
        { rtorot : (r -> t) -> or -> ot
        , ror : r -> or
        }


type Instance r t or ot
    = Instance
        { rtorot : (r -> t) -> or -> ot
        , ror : r -> or
        , r : r
        }


type Value r
    = Value r


defineOperations :
    ((r -> t) -> or -> ot)
    -> Operations r t or ot
defineOperations =
    Operations


implement :
    Operations r t or ot
    -> (r -> or)
    -> Implementation r t or ot
implement (Operations rtorot) ror =
    Implementation
        { rtorot = rtorot
        , ror = ror
        }


createInstanceOf :
    (ot -> t)
    -> Instance r t or ot
    -> t
createInstanceOf ott (Instance inst) =
    let
        rt : r -> t
        rt r =
            ott (inst.rtorot rt (inst.ror r))
    in
    rt inst.r


implementedBy :
    Implementation r t or ot
    -> Value r
    -> Instance r t or ot
implementedBy (Implementation impl) (Value r) =
    Instance
        { rtorot = impl.rtorot
        , ror = impl.ror
        , r = r
        }


fromValue :
    r
    -> Value r
fromValue =
    Value
1 Like

Hi Rupert, thank you once more for your valuable feedback :+1:

My other comment is that I find this highly unreadable. Its all the rt rtor or r t oror stuff, both in type var name and fields.

I’m sorry if it gave the impression that I would recommend the current state of the implementation. You’re absolutely right that Jeremy’s names are a much better fit. In fact, here’s the version I used in my experiments:

interface : (opsTyp -> typ) -> ((rep -> typ) -> opsRep -> opsTyp) -> (rep -> opsRep) -> (rep -> typ)
interface constructor mapOps impl =
    let
        repTyp : rep -> typ
        repTyp rep =
            constructor (mapOps repTyp (impl rep))
    in
    repTyp

(You could rename repTyp to raise to remove the last “abstract” name.)

But up to this point, I used the abstract names only. They helped me personally to come up with other possibilities to structure the code. Only when I felt I had reached the end of my experiments, I thought about better names.

But in this series of posts, I’m not finished yet, so I’m still using the abstract names.

What are the practical benefits of doing it this way versus Jeremeny’s original code? Is it simply so that implementors of interfaces do not have to use raise?

Yeah, until now, this is the only reason. But remember, I’m not finished yet…

I see, you are using these strange names as mathematical tools to explore the possibilities, very clever. I will hold off doing any renaming then, until you have finished.

Back to the Real World

I love happy ends, so it’s a perfect time to end the story of Jeremy, Rupert, and Pit.


The last thing for me to do is to apply the new interface technique to my introductory example: a list of different things. If you remember, I wanted to create a list with Strings, Ints, and Bools:

myListOfThings : List Thing
myListOfThings =
    [ AString "one", AnInt 1, AnInt 12, ABool True ]

and then be able to apply functions like

thingSize : Thing -> Int

thingDouble : Thing -> Thing

to the list elements. The problem with the original implementation was that, in order to add yet another wrapped type, for example Char, we needed to

  1. Implement the type-specific functions (charSize, charDouble)
  2. Add the new subtype to the wrapper type
  3. Add a new case branch to the wrapper functions (thingSize, thingDouble)

I promised that it’s possible to omit tasks #2 and #3, and this is exactly what Jeremy’s magic interface technique enables us to achieve.


First, we have to define the interface type and the operations:

type Thing
    = Thing (ThingOperations Thing)


type alias ThingOperations t =
    { size : Int
    , double : t
    }

I add some convenience functions for the users of the Thing type:

thingSize : Thing -> Int
thingSize (Thing ops) =
    ops.size


thingDouble : Thing -> Thing
thingDouble (Thing ops) =
    ops.double

As the designer of the interface, I have to provide:

  • the constructor function: Thing
  • a map function for the operations record:
thingOps : (r -> t) -> ThingOperations r -> ThingOperations t
thingOps raise ops =
    { size = ops.size
    , double = raise ops.double
    }

As an implementer of the interface, I have to implement the operations for my internal representation type:

thingImplString rep =
    { size = stringSize rep
    , double = stringDouble rep
    }


thingImplInt rep =
    { size = intSize rep
    , double = intDouble rep
    }


thingImplBool rep =
    { size = boolSize rep
    , double = boolDouble rep
    }

Using the three parts (constructor, map for operations record, implementation of operations), I can provide functions to create the various kinds of things:

aString : String -> Thing
aString =
    IF.createInstanceOf Thing thingOps thingImplString


anInt : Int -> Thing
anInt =
    IF.createInstanceOf Thing thingOps thingImplInt


aBool : Bool -> Thing
aBool =
    IF.createInstanceOf Thing thingOps thingImplBool

I really like the look-and-feel of the code :star_struck:

(Note that I used the interface function from the last part of the story, but renamed it to createInstanceOf. I’m still experimenting with the Interface API…)


Now I can put on the interface user’s hat and create a list of different things:

myListOfThings : List Thing
myListOfThings =
    [ aString "one", anInt 1, anInt 12, aBool True ]

Instead of the constructor functions AString, AnInt, and ABool from the former wrapper type, I now use the instance creation functions aString, anInt, and aBool. Nice and easy!

I create a small test program (here’s an Ellie with the code):

main : Html msg
main =
    myListOfThings
        |> List.map thingSize
        |> Debug.toString
        |> Html.text

I start elm reactor, navigate to my source file, and get:

Initialization Error

RangeError: Out of memory


:interrobang:


What ??? Has all this just been a fairy tale ???

No, no, no. I need a happy end! The story has to be continued…

Did it blow the stack?

It’s an infinite recursion, and depending on how you run the code you get different errors (for example, in elm-test, I get a JavaScript error).

FYI: unfortunately, this would happen in Jeremy’s original code, too.

Stay tuned :wink: