haskell-to-elm

haskell-to-elm is a library that takes Haskell type definitions as input and generates matching Elm type definitions and JSON encoders and decoders that match Aeson's format.
The problem
Let's say we're building a web page with a Haskell backend and an Elm frontend.
We might have a Haskell type like this, that we pass to the frontend encoded as JSON. The JSON encoder is derived using the Aeson library.
data User = User { name :: Text , age :: Int } deriving (Generic, ToJSON)
We mirror the type on the Elm side and add a JSON decoder as follows:
type alias User = { name : String , age : Int } decoder : Decoder User decoder = Decode.map2 User (Decode.field "name" Decode.string) (Decode.field "age" Decode.int)
Now, let's say we want to change a field in the backend:
-- Haskell data User = User { name :: Text --, age :: Int , birthday :: Date -- <---- new! } deriving (Generic, ToJSON)
If we now run the application again, but forget to update the Elm code, the User decoder will fail at runtime in Elm.
The solution
haskell-to-elm solves this problem by letting us generate the Elm User type and decoder from the Haskell User type.
With haskell-to-elm as part of your build pipeline you can make sure that the frontend is always in sync with your backend, and get type errors in your frontend code when you change your backend types.
The companion library servant-to-elm also lets you generate Elm client libraries for your Servant APIs.
Basic usage
To generate code for the User type above, we first need to derive a bunch of class instances:
data User = User { name :: Text , age :: Int } deriving (Generic, Aeson.ToJSON, SOP.Generic, SOP.HasDatatypeInfo) instance HasElmType User where elmDefinition = Just $ deriveElmTypeDefinition @User defaultOptions "Api.User.User" instance HasElmDecoder Aeson.Value User where elmDecoderDefinition = Just $ deriveElmJSONDecoder @User defaultOptions Aeson.defaultOptions "Api.User.decoder" instance HasElmEncoder Aeson.Value User where elmEncoderDefinition = Just $ deriveElmJSONEncoder @User defaultOptions Aeson.defaultOptions "Api.User.encoder"
Then we can print the generated Elm code using the following code:
main :: IO () main = do let definitions = Simplification.simplifyDefinition <$> jsonDefinitions @User modules = Pretty.modules definitions forM_ (HashMap.toList modules) $ \(_moduleName, contents) -> print contents
Running main will print the following Elm code:
module Api.User exposing (..) import Json.Decode import Json.Decode.Pipeline import Json.Encode type alias User = { name : String, age : Int } encoder : User -> Json.Encode.Value encoder a = Json.Encode.object [ ("name" , Json.Encode.string a.name) , ("age" , Json.Encode.int a.age) ] decoder : Json.Decode.Decoder User decoder = Json.Decode.succeed User |> Json.Decode.Pipeline.required "name" Json.Decode.string |> Json.Decode.Pipeline.required "age" Json.Decode.int
In an actual project we would be writing the code to disk instead of printing it.
See this file for the full code with imports.
Parameterised types
Since version 0.3.0.0, haskell-to-elm supports generating code for types with type parameters.
For example, let's say we have the following Haskell type:
data Result e a = Err e | Ok a deriving (Generic, Aeson.ToJSON, SOP.Generic, SOP.HasDatatypeInfo)
We can derive the corresponding Elm type and JSON encoders and decoder definitions with the following code:
instance HasElmType Result where elmDefinition = Just $ deriveElmTypeDefinition @Result defaultOptions "Api.Result.Result" instance HasElmDecoder Aeson.Value Result where elmDecoderDefinition = Just $ deriveElmJSONDecoder @Result defaultOptions Aeson.defaultOptions "Api.Result.decoder" instance HasElmEncoder Aeson.Value Result where elmEncoderDefinition = Just $ deriveElmJSONEncoder @Result defaultOptions Aeson.defaultOptions "Api.Result.encoder"
For parameterised types we also have to add instances for how to handle the type when it's fully applied to type arguments. Like this:
instance (HasElmType a, HasElmType b) => HasElmType (Result a b) where elmType = Type.apps (elmType @Result) [elmType @a, elmType @b] instance (HasElmDecoder Aeson.Value a, HasElmDecoder Aeson.Value b) => HasElmDecoder Aeson.Value (Result a b) where elmDecoder = Expression.apps (elmDecoder @Aeson.Value @Result) [elmDecoder @Aeson.Value @a, elmDecoder @Aeson.Value @b] instance (HasElmEncoder Aeson.Value a, HasElmDecoder Aeson.Value b) => HasElmEncoder Aeson.Value (Result a b) where elmEncoder = Expression.apps (elmEncoder @Aeson.Value @Result) [elmEncoder @Aeson.Value @a, elmDecoder @Aeson.Value @b]
The rationale for having two instances of the classes for each type is that we both have to describe how the type is defined (with the unapplied instances), which generates parameterised definitions, and then we describe how to actually use those parameterised definitions with the applied instances.
These instances print the following code when run:
module Api.Result exposing (..) import Json.Decode import Json.Decode.Pipeline import Json.Encode type Result a b = Err a | Ok b encoder : (a -> Json.Encode.Value) -> (b -> Json.Encode.Value) -> Result a b -> Json.Encode.Value encoder a b c = case c of Err d -> Json.Encode.object [ ("tag" , Json.Encode.string "Err") , ("contents" , a d) ] Ok d -> Json.Encode.object [ ("tag" , Json.Encode.string "Ok") , ("contents" , b d) ] decoder : Json.Decode.Decoder a -> Json.Decode.Decoder b -> Json.Decode.Decoder (Result a b) decoder a b = Json.Decode.field "tag" Json.Decode.string |> Json.Decode.andThen (\c -> case c of "Err" -> Json.Decode.succeed Err |> Json.Decode.Pipeline.required "contents" a "Ok" -> Json.Decode.succeed Ok |> Json.Decode.Pipeline.required "contents" b _ -> Json.Decode.fail "No matching constructor")
Notice that the generated encoder and decoder are parameterised by the encoder and decoder for the type arguments.
See this file for the full code with imports.
Using DerivingVia to reduce boilerplate
We can use the DerivingVia extension to reduce some of the boilerplate that this library requires. This requires GHC version >= 8.8, because earlier versions had a bug that prevented it to work.
In this file we define a type called ElmType that we derive both the haskell-to-elm and Aeson classes through.
After having defined that type, the code for User is simply:
data User = User { _name :: Text , _age :: Int } deriving (Generic, SOP.Generic, SOP.HasDatatypeInfo) deriving (Aeson.ToJSON, Aeson.FromJSON, HasElmType, HasElmDecoder Aeson.Value, HasElmEncoder Aeson.Value) via ElmType "Api.User.User" User
This also means that we can ensure that we pass the same Aeson options to this library's Elm code generation functions and Aeson's JSON derivation functions, meaning that we don't risk mismatched JSON formats.
Roadmap
- Derive JSON encoders and generically
- Pretty-print the Elm AST
- Separate pretty printing from code generation: elm-syntax
- Generate Elm modules
- Servant client library generation: servant-to-elm
- Test that encoding and decoding round-trip: haskell-to-elm-test
- Support parameterised types
Libraries that use or are used by haskell-to-elm:
- elm-syntax defines Haskell ASTs for Elm's syntax, and lets us pretty-print it.
- servant-to-elm can be used to generate Elm client libraries from Servant APIs.
- haskell-to-elm-test does end-to-end testing of this library.
Others: