• Stars
    star
    138
  • Rank 264,439 (Top 6 %)
  • Language
    F#
  • License
    MIT License
  • Created over 6 years ago
  • Updated 3 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Create client-server Fable-Elmish apps keeping a single mindset

Elmish.Bridge

Formely Elmish.Remoting. This library creates a bridge between server and client using websockets so you can keep the same model-view-update mindset to create the server side model.

Available Packages:

Library Version
Elmish.Bridge.Client Nuget
Elmish.Bridge.DotnetClient Nuget
Elmish.Bridge.Suave Nuget
Elmish.Bridge.Giraffe Nuget

Why?

The MVU approach made programming client-side SPA really fun and fast for me. The idea of having a model updated in a predictable way, with a view just caring about what was in the model just clicked, and making a server-side with just consumable services made everything different and I felt like I was making two entirely different programs. I came up with this project so I can have the same fun when creating the server.

Drawbacks

You are now just passing messages between the server and the client. Stuff like authorization, headers or cookies are not there anymore. You have a stateful connection between the client while it's open, but if it gets disconnected the server will lose all the state it had. That can be mitigated by sending a message to the client when it reestabilishes the connection, so it knows that the server is there again. You can then send all the information that the server needs to feel like the connection was never broken in the first place.

tl;dr You have to relearn the server-side part

How to use it?

This is Elmish on server with a bridge to the client. You can learn about Elmish here. This assumes that you know how to use Elmish on the client-side.

Shared

I recommend to keep the messages and the endpoint on a shared file; you don't need to keep the model if you decide that you want the server and client to have different models.

// Messages processed on the server
type ServerMsg =
    |...
//Messages processed on the client
type ClientMsg =
    |...
module Shared =
    let endpoint = "/socket"

Client

What's different? Well, now you can send messages to and get messages from the server. For it, just send the message using Bridge.Send. Simple as that. But before that, you need to enable the bridge on your Elmish Program, preferably before anything else so it can inject the incoming messages at the right place.

open Elmish
open Elmish.Bridge

Program.mkProgram init update view
|> Program.withBridge Shared.endpoint
|> Program.withReact "elmish-app"

Warning: Because of a limitation on the reflection on Fable, you can't use primitives or generic types on the input of Bridge.Send/Bridge.NamedSend

Server

Now you can use the MVU approach on the server, minus the V. That still is just a client thing. For easy usage, the init and update function first argument is now a Dispatch<ClientMsg>. Just call that function with the client message and it will be sent there.

The server has a little more configuration to do, but that enables the client to be very simple to use while being compatible with the vanilla Elmish. A problem that exists is that the client can send any kind of message, not just the expected message that the update listens to. Because of that, you need register every type that you want to be listening and how to get to the original top-level message.

As of version 3.0, you won't need to register every mapping to the top-level type. The server will register every type in the union that has no ambiguity. For this example, you won't need to register them anymore as FirstInnerMsg is only mapped by First and SecondInnerMsg is only mapped by Second.

Imagine that you have the following model:

type FirstInnerMsg =
  | FIA
  | FIB
type SecondInnerMsg =
  | SIA
  | SIB
type OuterMsg = // That's the one the update expects
  | SomeMsg
  | First of FirstInnerMsg
  | Second of SecondInnerMsg

OuterMsg is registered by default, but to enable the client to do Bridge.Send FIA or Bridge.Send SIB all you need to do is:

let server =
  Bridge.mkServer Shared.endpoint init update
  |> Bridge.register First // First is a function (FirstInnerMsg -> OuterMsg)
  |> Bridge.register Second // Any ('a -> OuterMsg) will work
  |> Bridge.run (* server implementation here*)

As for the server implementation, there is one for Suave on the Elmish.Bridge.Suave package and another for Giraffe/Saturn on the Elmish.Bridge.Giraffe package.

  • Suave
open Elmish
open Elmish.Bridge

let server =
  Bridge.mkServer Shared.endpoint init update
  |> Bridge.run Suave.server

let webPart =
  choose [
    server
    Filters.path "/" >=> Files.browseFileHome "index.html"
  ]
startWebServer config webPart
  • Giraffe
open Elmish
open Elmish.Bridge

let server =
  Bridge.mkServer Shared.endpoint init update
  |> Bridge.run Giraffe.server

let webApp =
  choose [
    server
    route "/" >=> htmlFile "/pages/index.html" ]

let configureApp (app : IApplicationBuilder) =
  app
    .UseWebSockets()
    .UseGiraffe webApp

let configureServices (services : IServiceCollection) =
  services.AddGiraffe() |> ignore

WebHostBuilder()
  .UseKestrel()
  .Configure(Action<IApplicationBuilder> configureApp)
  .ConfigureServices(configureServices)
  .Build()
  .Run()
  • Saturn
open Elmish
open Elmish.Bridge

let server =
  Bridge.mkServer Shared.endpoint init update
  |> Bridge.run Giraffe.server

let webApp =
  choose [
    server
    route "/" >=> htmlFile "/pages/index.html" ]

let app =
  application {
    use_router webApp
    disable_diagnostics
    app_config Giraffe.useWebSockets
    url uri
  }

run app

ServerHub

WebSockets are even better when you use that communication power to send messages between clients. No library would be complete without help to broadcast messages or send a message to an specific client. That's where the ServerHub comes in! It is a very simple class that worries about keeping the information about all the connected clients so you don't have to.

Create a new ServerHub and register all the mappings that you will use by using:

let hub =
  ServerHub()
    .RegisterServer(First)
    .RegisterServer(Second)
    .RegisterClient(ClientMsg)

then you can use it on your server:

Bridge.mkServer Shared.endpoint init update
|> Bridge.withServerHub hub
|> Bridge.run server

It has three functions:

  • Broadcast(Server/Client)

    Send a message to anyone connected:

      hub.BroadcastClient (NewMessage "Hello, everyone!")
  • Send(Server/Client)If

    Send a message to anyone whose model satisfies a predicate.

      hub.SendClientIf (function {Gender = Female} -> true | _ -> false) (NewMessage "Hello, ladies!")
  • GetModels

    Gets a list with all models of all connected clients.

      let users = hub.GetModels() |> List.map (function {Name = n} -> n)
      hub.BroadcastClient (ConnectedUsers users)

These functions were enough when creating a simple chat, but let me know if you feel limited having only them!

Other configuration

Besides Program.withBridge, there is Program.withBridgeConfig that can configure some aspects:

  • As the endpoint is mandatory, create the BridgeConfig using Bridge.endpoint

  • If you need that endpoint to be relative instead of absolute, chain the config with Bridge.withUrlMode. More on that later.

  • For defining a time in seconds to reconnect, chain the config with Bridge.withRetryTime

  • For defining a message to be sent to the client when the connection is lost, chain the config with Bridge.withWhenDown

  • For defining a name for your bridge so you can use more than one, chain the config with Bridge.withName. More on that later.

  • For defining a mapping so the server can send a different message to the client, chain the config with Bridge.withMapping. More on that on the next section.

  • (Since 3.0) For defining a custom serializer so you aren't limited on sending JSON data through the socket, pass a custom serializer returning SerializerResult to withCustomSerializer. More on that later.

  • (Since 3.4) If you need to get some data before initializing the connection or you only want to enable it if your user does some action, like entering a chat page, you can pass the BridgeConfig to Bridge.asSubscription to get something that can be passed to Cmd.ofSub.

  • (Since 5.0) BridgeConfig now implements IDisposable, closing the connection when Dispose() is called. That is useful for using it with some React hooks that close resources when detaching the component.

Minimizing shared messages

You can share just a sub-set of the messages between client and server. The message type used by the first argument of init and update functions on the server is what will be sent, so on the client you can add a mapping from that type for the type used on the client's init/update.

Imagine you have the following model:

type BridgeAware =
  | Hello of string

type SomeOtherMessage =
  | InnerBridge of BridgeAware
  | ...

type TopLevel =
  | SomeMsg of SomeOtherMessage
  | ...

You can make the server functions like this:

let init (clientDispatch:Dispatch<BridgeAware>) () =
  clientDispatch (Hello "I came from the server!")
  someModel,someCmds

And configure the client like this:

Program.mkProgram init update view
|> Program.withBridgeConfig(
    Bridge.endpoint Shared.endpoint
    |> Bridge.withMapping
        (fun bridgeMsg ->
            bridgeMsg
            |> InnerBridge
            |> SomeMsg))
|> Program.withReact "elmish-app"

That way, only the BridgeAware type needs to be on the shared file

Configuring the endpoint

Sometimes you may need that path to be relative with the current URL. Maybe you want to define an external endpoint. For that, you can pass the following cases to Bridge.withUrlMode:

  • Replace: the default. Uses the current host and replaces the path with the endpoint defined.
  • Append: Appends the endpoint defined to the current URL.
  • Raw: Uses the defined path as a complete URL.
  • Calculated: Takes an extra function (string -> string -> string). The functions arguments are the current URL and the endpoint defined, returning the resulting URL.

As DotnetClient runs outside the browser the endpoint is assumed to be raw. Pay attention to the protocol and use ws:// or wss:// to connect.

Webpack caveat

When using the development mode of Webpack, usually a proxy is defined so the server calls can be redirected to the right place. That proxy doesn't work for websockets by default. To enable them, use ws: true when configuring the endpoint.

Example:

devServer: {
    proxy: {
      '/api/*': {
        target: 'http://localhost:' + port,
        changeOrigin: true
      },
      '/socket': {
        target: 'http://localhost:' + port,
        ws: true
     }
    },
    contentBase: "./public",
    hot: true,
    inline: true
  }

Named bridges

Sometimes you have more than one feature where the bridge can be useful. You can have a real-time notification when the user's order is approved and also have a chat so it can talk to support. These features have nothing in common, so you don't need to clutter your logic with all remote stuff you do.

When using a name, you can use the method Bridge.NamedSend that takes a name (defined with Bridge.withName) and the desired message tupled. There's a annoying behavior that prevents it to be curried and partially applied, but here is an workaround:

open Fable.Core
let inline chatMessage x = Bridge.NamedSend("Chat", x)
let inline notification x = Bridge.NamedSend("Notification", x)

then you can use it on your update:

  ...
  match msg with
  | ClientSentMessage msg ->
      chatMessage (NewMessage msg)
  ...

Don't forget to use the same name on the BridgeConfig:

Program.mkProgram init update view
|> Program.withBridgeConfig
    (Bridge.endpoint "/socket/chat"
    |> Bridge.withName "Chat"
    |> Bridge.withMapping ChatMessages)
|> Program.withBridgeConfig
    (Bridge.endpoint "/socket/notification"
    |> Bridge.withName "Notification"
    |> Bridge.withMapping NotificationMessages)
|> ...

Custom serialization (since 3.0)

If you want to use a different serialization for sending a type for the server, you can register a custom serialization that takes the desired input and maps it into a SerializerResult:

/// SerializerResult is defined as:
type SerializerResult =
    | Text of string
    | Binary of byte []

/// You can serialize a simple type like
type Action =
    | Increment
    | Decrement

/// Using a serializer with the signature (Action -> SerializerResult) like

let actionSerializer = function
    | Increment -> Text "+"
    | Decrement -> Text "-"

/// And then register it on the BridgeConfig
Program.mkProgram init update view
|> Program.withBridgeConfig
    (Bridge.endpoint "/socket/chat"
    |> Bridge.withCustomSerializer actionSerializer)
|> ...

/// As for the server, you need to use the BridgeDeserializer
type BridgeDeserializer<'server> =
    | Text of (string -> 'server)
    | Binary of (byte[] -> 'server)

/// For Action, that would be
let deserializer = Text (function "+" -> Increment | "-" -> Decrement)

/// Suppose your top level message is defined as that
type ServerMessage =
    | TheAction of Action
    | AnotherMessage

/// Then you can use it after registering it
Bridge.mkServer Shared.endpoint init update
|> Bridge.registerWithDeserializer TheAction deserializer
|> Bridge.run (...)

Elmish Cmd (since 3.1.1)

As an extension to the Cmd module, you can now use Cmd.bridgeSend and Cmd.namedBridgeSend to create a Cmd on your update and init functions. That is more alike other libraries that extends Elmish.

let init () =
    None, Cmd.bridgeSend (GiveMeTheModel)

By default, if you send a message while you are disconnected, nothing happens. If you want to dispatch a message to the loop in that case, you can instead use Cmd.bridgeSendOr or Cmd.namedBridgeSendOr to define a fallback message.

let init () =
    None, Cmd.bridgeSendOr (WhatsTheAnswer) (ItIs 42)

Reply channels (since 5.0.0)

If before you used some Bridge just to get some data from the server or vice versa, you probably used a message to encode the asking and another to encode the answering. Unless you had some sort of identifier for the round-trip, you'd have a bad time knowing which message came from each question. Reply channels aims to help on that specific case, but feel free to get creative!

For the shared part, now both Elmish.Bridge.Client and Elmish.Bridge.Server has a dependency on Elmish.Bridge.RPC. This package has a single record, IReplyChannel<'T>. It is inspired by the AsyncReplyChannel<'Reply> used on the MailboxProcessor when you call PostAndReply.

Suppose that before you had a:

type Server =
    | AskQuery of QueryParameters

type Client =
    | QueryAnswer of QueryResult

You could abstract that to a single message:

open Elmish.Bridge

type Server =
    | Query of QueryParameters * IReplyChannel<QueryResult>

Client code that before needed to use elmish messages

let update msg model =
    match msg with
    | DoQuery queryParameter ->
        model, Cmd.bridgeSend (AskQuery queryParameter)
    | Remote (QueryAnswer result) ->
        // do something with the result

can be refactored to instead be on async expressions:

let doQuery queryParameter =
  async {
    let! result = Bridge.AskServer(fun rc -> AskQuery(queryParameter, rc))
    /// do something with the result
  }

The client also has a Bridge.AskNamedServer for named Bridges.

For the side supposed to answer on the reply channel, the change is not that big:

let update clientDispatch msg model =
    match msg with
    | Remote(AskQuery queryParameter) ->
        let result = doQuery queryParameter
        clientDispatch (QueryAnswer result)
        model, Cmd.none

would turn into:

let update clientDispatch msg model =
    match msg with
    | Remote(AskQuery (queryParameter, replyChannel)) ->
        let result = doQuery queryParameter
        replyChannel.Reply result
        model, Cmd.none

The server can also ask the clients for values. The ServerHub was extended with the methods `

  • AskClient
    • takes
      • a Dispatch<'client>: same as the one on update
      • an IReplyChannel<'T> -> 'client: to build the client message
    • returns
      • an Async<'T>: with the client's response
  • AskAllClients
    • takes
      • an IReplyChannel<'T> -> 'client: to build the client message
      • an Dispatch<'client> -> Dispatch<'server> -> 'T -> unit: for processing the successful messages. You can use the client dispatch to send new messages to the client and use the server dispatch to add new messages to the server's update
      • an Dispatch<'client> -> Dispatch<'server> -> exn -> unit: same as above, but for exceptions
    • returns unit
  • AskAllClientsIf
    • takes
      • a model -> bool: filters so only clients that has something on their models receive the message
      • same as AskAllClients

Anything more?

This documentation is on an early stage, if you have any questions feel free to open an issue or PR so we can have it in a good shape. You can check a test project here. I hope you enjoy using it!