# 03.14.2022 - Elm / Building large Elm applications

Coming from React or other UI frameworks, when it comes to building large applications, you might have the habit of breaking it into smaller components.

This mental model does not fit in Elm, in fact, Component are discouraged in Elm. Instead, you are encouraged to stay as long as possible in a single file. As your code grows, if they reached the point where modifying or reading is challenged enough, you can start breaking them up into smaller functions, and composing them if needed.

The chapter 6 of the Programming Elm book outlined some strategies to do so.

## Splitting the View

The view can be split into multiple functions, and you can call each function for each case.

Personally, I think each subview function can take a smaller part of the model as its arguments, so rendering is much easier to organize.

## Simplifying Messages

Messages can be grouped to reduce duplication when handling updates, using parametrized message value.

For example:

Let’s say you are building an Elm app for a gas station, the user can select the gas type they want: Regular (87), Mid-grade (89), and Premium (91 or 93).

Initially, you would define each of the gas selections as an individual message, there will be duplication when updating the model:

-- message
type Msg
= SelectRegular

-- update
case msg of
SelectRegular -> { model | gasType = Regular }


Using the parameterized message, we only need one message SelectGas and the gas type would be a parameter, we can reduce code duplication in the update function:

-- message
type Msg
= SelectGas GasType

-- update
case msg of
SelectGas type -> { model | gasType = type }


## Use nested state

Organizing your application state as a flat object would make the update function handle too many responsibilities, making it complex to read and modify.

For example, instead of a big flat state:

type alias Model =
{ supplyLevel: Int
, supplyGasType: GasType
}


We can organize it as a nested state:

type alias Supply =
{ level: Int
, gasType: GasType
}

{ level: Int
, gasType: GasType
}

type alias Model =
{ supply: Supply
}


Using nested state would modularize your code better, the update function can be separated into smaller ones, each function will handle the smaller part of the state object.

For example, with the nested state, we split up the update function into
smaller functions for each type:

updateSupply : SupplyMsg -> Supply -> Supply
case msg of
...

case msg of
...

update : Msg -> Model -> Model
update msg model =
case msg of
SupplyMsg supplyMsg ->
{ model
| supply = updateSupply supplyMsg model.supply
}
{ model
}


But nested states will make it harder to handle, especially when you have multiple levels nested. The rule of thumb is, design states sparingly when possible. When you have to use a nested state, try not to go more than one level deep.

## Use extensible records

Another alternative to nested state is extensible records type. It is like an interface, defined in the syntax of type alias Something s where s is the type variable. Any record that contains all the fields defined in the extensible record, is an instance of that extensible record.

For example, if we have a type BlogPost, we can create an extensible
record type Reaction like this:

type alias BlogPost =
{ title: String
, content: String
, likes: Int
, dislikes: Int
}

type alias Reaction post =
{ post
| likes: Int
, dislikes: Int
}


Any instance of BlogPost is also an instance of Reaction, because
they all have the likes and dislikes fields.

On the other hand, any record that does not have the likes and
dislikes fields are not an instance of LikeCount.

We can use an extensible record to split up the update function into smaller ones. For example, to implement the post’s reaction feature, we can define a new ReactionMsg type for the message, and write a updateReaction function like this:

type ReactionMsg
= Like
| Dislike

updateReaction : ReactionMsg -> Reaction r -> Reaction r
updateReaction reactionMsg post =
case reactionMsg of
Like -> { post | likes = post.likes + 1 }
Dislike -> { post | dislikes = post.dislikes + 1 }


Then, the most important part is wiring up the ReactionMsg and updateReaction into the existing Msg type and update function:

type Msg
= ReactionMsg ReactionMsg
| OtherMsg OtherMsg
| ...

type Model =
{ post: BlogPost
}

update : Msg -> Model -> Model
update msg model =
case msg of
ReactionMsg reactionMsg ->
{ model
| post = ( updateReaction reactionMsg model.post )
}
...


The model.post record, which has the type BlogPost, can be passed into the updateReaction function because it has likes and dislikes fields, so it’s an instance of the Reaction type.

By using this characteristic of extensible record, we can modularize the different features of the application easily, while being able to keep the model as a flat structure, hence, there is no need for nested structure.

References