Fleece is a JSON mapper for F#. It simplifies mapping from a Json library's JsonValue onto your types, and mapping from your types onto JsonValue.
The Json library could be System.Json, System.Text.Json, FSharp.Data's or NewtonSoft's Json.NET.
Its design is strongly influenced by Haskell's Aeson. Like Aeson, Fleece is designed around two typeclasses (in FSharpPlus style) ToJson and OfJson.
For example, given this data type:
typePerson={
Name:string
Age:int
Children:Person list}You can map it to JSON like this:
openFleeceopenFleece.OperatorstypePersonwithstatic memberToJson(x:Person)=
jobj ["name".= x.Name
"age".= x.Age
"children".= x.Children
]letp={ Person.Name ="John"
Age =44
Children =[{ Person.Name ="Katy"
Age =5
Children =[]}{ Person.Name ="Johnny"
Age =7
Children =[]}]}// Test with System.Text.JsonopenFleece.SystemTextJson
printfn "%s"(toJsonText p)And you can map it from JSON like this:
typePersonwithstatic memberOfJson json =match json with| JObject o ->letname= o .@"name"letage= o .@"age"letchildren= o .@"children"match name, age, children with| Decode.Success name, Decode.Success age, Decode.Success children ->
Decode.Success {
Person.Name = name
Age = age
Children = children
}| x -> Error <| Uncategorized (sprintf "Error parsing person: %A" x)| x -> Decode.Fail.objExpected x
letjohn:Person ParseResult = ofJsonText """{"name": "John", "age": 44, "children": [{"name": "Katy", "age": 5, "children": []}, {"name": "Johnny", "age": 7, "children": []}]}"""Though it's much easier to do this in a monadic or applicative way. For example, using FSharpPlus (which is already a dependency of Fleece):
openFSharpPlustypePersonwithstatic memberCreate name age children ={ Person.Name = name; Age = age; Children = children }static memberOfJson json =match json with| JObject o -> Person.Create <!>(o .@"name")<*>(o .@"age")<*>(o .@"children")| x -> Decode.Fail.objExpected x
or with applicatives:
openFSharpPlustypePersonwithstatic memberOfJson json =match json with| JObject o -> monad {let!name= o .@"name"and! age = o .@"age"and! children = o .@"children"return{
Person.Name = name
Age = age
Children = children
}}| x -> Decode.Fail.objExpected xOr you can use the Choice monad/applicative in FSharpx.Extras instead, if you prefer.
You can see more examples in the EdmundsNet project.
For types that deserialize to Json Objets, typically (but not limited to) records, you can alternatively use codecs and have a single method which maps between fields and values.
typePerson={ name :string * string
age :int option
children:Person list }withstatic memberget_Codec()=fun f l a c ->{ name =(f, l); age = a; children = c }<!> jreq "firstName"(Some <<fun x -> fst x.name)<*> jreq "lastName"(Some <<fun x -> snd x.name)<*> jopt "age"(fun x -> x.age)// Optional fields can use 'jopt'<*> jreq "children"(fun x -> Some x.children)|> ofObjCodec
letjohn:Person ParseResult = ofJsonText """{"name": "John", "age": 44, "children": [{"name": "Katy", "age": 5, "children": []}, {"name": "Johnny", "age": 7, "children": []}]}"""Discriminated unions can be modeled with alternatives:
typeShape=| Rectangle ofwidth:float*length:float| Circle ofradius:float| Prism ofwidth:float*float*height:floatwithstatic memberget_Codec()=(Rectangle <!> jreq "rectangle"(function Rectangle (x, y)-> Some (x, y)|_-> None))<|>(Circle <!> jreq "radius"(function Circle x -> Some x |_-> None))<|>(Prism <!> jreq "prism"(function Prism (x, y, z)-> Some (x, y, z)|_-> None))|> ofObjCodecor using the jchoice combinator:
typeShapewithstatic memberJsonObjCodec=
jchoice
[
Rectangle <!> jreq "rectangle"(function Rectangle (x, y)-> Some (x, y)|_-> None)
Circle <!> jreq "radius"(function Circle x -> Some x |_-> None)
Prism <!> jreq "prism"(function Prism (x, y, z)-> Some (x, y, z)|_-> None)]|> ofObjCodec
But codecs for both types can easily be written with the codec computation expressions
typePerson={ name :string * string
age :int option
children:Person list }withstatic memberget_Codec()=
codec {let!f= jreq "firstName"(Some <<fun x -> fst x.name)and! l = jreq "lastName"(Some <<fun x -> snd x.name)and! a = jopt "age"(fun x -> x.age)// Optional fields can use 'jopt'and! c = jreq "children"(fun x -> Some x.children)return{ name =(f, l); age = a; children = c }}|> ofObjCodec
typeShape=| Rectangle ofwidth:float*length:float| Circle ofradius:float| Prism ofwidth:float*float*height:floatwithstatic memberget_Codec()=
codec {
Rectangle <!> jreq "rectangle"(function Rectangle (x, y)-> Some (x, y)|_-> None)
Circle <!> jreq "radius"(function Circle x -> Some x |_-> None)
Prism <!> jreq "prism"(function Prism (x, y, z)-> Some (x, y, z)|_-> None)}|> ofObjCodec
What's happening here is that we're getting a Codec to/from a Json Object (not neccesarily a JsonValue) which Fleece is able to take it and fill the gap by composing it with a codec from JsonObject to/from JsonValue.
We can also do that by hand, we can manipulate codecs by using functions in the Codec module. Here's an example:
openSystem.TextopenFleece.SystemTextJson.OperatorstypePerson={ name :string * string
age :int option
children:Person list }withstatic memberJsonObjCodec:Codec<PropertyList<Fleece.SystemTextJson.Encoding>,Person>= codec {let!f= jreq "firstName"(Some <<fun x -> fst x.name)and! l = jreq "lastName"(Some <<fun x -> snd x.name)and! a = jopt "age"(fun x -> x.age)// Optional fields can use 'jopt'and! c = jreq "children"(fun x -> Some x.children)return{ name =(f, l); age = a; children = c }}letpersonBytesCodec=
Person.JsonObjCodec
|> Codec.compose jsonObjToValueCodec // this is the codec that fills the gap to/from JsonValue|> Codec.compose jsonValueToTextCodec // this is a codec between JsonValue and JsonText|> Codec.invmap (Encoding.UTF8.GetString: byte []-> string) Encoding.UTF8.GetBytes // This is a pair of of isomorphic functionsletp={ name ="John","Smith"; age = Some 42; children =[]}letbytePerson= Codec.encode personBytesCodec p
// val bytePerson : byte [] = [|123uy; 34uy; 102uy; 105uy; 114uy; 115uy; ... |]letp' = Codec.decode personBytesCodec bytePersonSo far we've seen how Fleece is capable of encoding/decoding by deriving automatically a codec from static members in the type.
But for those cases where we don't have control over the types (extension members won't be taken into account) we can explicitly specify combinators.
To do so, a set of the available functions exists, ending with the With suffix, which accepts a combinator as first parameter:
typeColor= Red | Blue | White
typeCar={
Id :string
Color :Color
Kms :int }letcolorDecoder=function| JString "red"-> Decode.Success Red | JString "blue"-> Decode.Success Blue | JString "white"-> Decode.Success White
| JString x as v -> Decode.Fail.invalidValue v ("Wrong color: "+ x)| x -> Decode.Fail.strExpected x
letcolorEncoder=function| Red -> JString "red"| Blue -> JString "blue"| White -> JString "white"letcolorCodec()= colorDecoder <-> colorEncoder
letcarCodec()=
codec {let!i= jreqWith Codecs.string "id"(fun x -> Some x.Id)and! c = jreqWith colorCodec "color"(fun x -> Some x.Color)and! k = jreqWith Codecs.int "kms"(fun x -> Some x.Kms)return{ Id = i; Color = c; Kms = k }}|> Codec.compose (Codecs.propList Codecs.id)letcar={ Id ="xyz"; Color = Red; Kms =0}letjsonCar:Fleece.SystemTextJson.Encoding = Codec.encode (carCodec ()) car
// val jsonCar: SystemTextJson.Encoding = {"id":"xyz","color":"red","kms":0}Json lenses allow to focus on a specific part of the json structure to perform operations like view, write and update.
For a quick reference have a look at this test file