Elm-inspired decoders for Ocaml

A combinator library for "decoding" JSON-like values into your own Ocaml types, inspired by Elm's Json.Decode and Json.Encode.


An Ocaml program having a JSON (or YAML) data source usually goes something like this:

  1. Get your data from somewhere. Now you have a string.
  2. Parse the string as JSON (or YAML). Now you have a Yojson.Basic.json, or maybe an Ezjsonm.value, or perhaps a Ocyaml.yaml.
  3. Decode the JSON value to an Ocaml type that's actually useful for your program's domain.

This library helps with step 3.

Getting started

Install one of the supported decoder backends:

For ocaml

opam install decoders-ezjsonm      # For Ezjsonm
opam install decoders-ocyaml       # For Ocyaml
opam install decoders-yojson       # For Yojson

For bucklescript

npm install --save-dev bs-decoders


Now we can start decoding stuff!

First, a module alias to save some keystrokes. In this guide, we'll parse JSON using Yojson's Basic variant.

utop # module D = Decoders_yojson.Basic.Decode;;
module D = Decoders_yojson.Basic.Decode

Let's set our sights high and decode an integer.

utop # D.decode_value (`Int 1);;
- : (int, error) result = Ok 1

Nice! We used decode_value, which takes a decoder and a value (in this case a Yojson.Basic.json) and... decodes the value.

utop # D.decode_value;;
- : 'a decoder -> value -> ('a, error) result = <fun>

For convenience we also have decode_string, which takes a string and calls Yojson's parser under the hood.

utop # D.decode_string "1";;
- : (int, error) result = Ok 1

What about a list of ints? Here's where the "combinator" part comes in.

utop # D.decode_string D.(list int) "[1,2,3]";;
- : (int list, error) result = Ok [1; 2; 3]


Ok, so what if we get some unexpected JSON?

utop # #install_printer D.pp_error;;
utop # D.decode_string D.(list int) "[1,2,true]";;
- : (int list, error) result =
Error while decoding a list: element 2: Expected an int, but got true

Complicated JSON structure

To decode a JSON object with many fields, we can use the bind operator (>>=) from the Infix module.

type my_user =
  { name : string
  ; age : int

let my_user_decoder : my_user decoder =
  let open D in
  field "name" string >>= fun name ->
  field "age" int >>= fun age ->
  succeed { name; age }

We can also use bind to decode objects with inconsistent structure. Say, for example, our JSON is a list of shapes. Squares have a side length, circles have a radius, and triangles have a base and a height.

[{ "shape": "square", "side": 11 },
 { "shape": "circle", "radius": 5 },
 { "shape": "triange", "base": 3, "height": 7 }]

We could represent these types in OCaml and decode them like this:

type shape =
  | Square of int
  | Circle of int
  | Triangle of int * int

let square_decoder : shape decoder =
  D.(field "side" int >>= fun s -> succeed (Square side))

let circle_decoder : shape decoder =
  D.(field "radius" int >>= fun r -> succeed (Circle r))

let triangle_decoder : shape decoder =
    field "base" int >>= fun b ->
    field "height" int >>= fun h ->
    succeed (Triangle (b, h))

let shape_decoder : shape decoder =
  let open D in
  field "shape" string >>= function
  | "square" -> square_decoder
  | "circle" -> circle_decoder
  | "triangle" -> triangle_decoder
  | _ -> fail "Expected a shape"

let decode_list (json_string : string) : (shape list, _) result =
  D.(decode_string (list shape_decoder) json_string)

Now, say that we didn't have the benefit of the "shape" field describing the type of the shape in our JSON list. We can still decode the shapes by trying each decoder in turn using the one_of combinator.

one_of takes a list of string * 'a decoder pairs and tries each decoder in turn. The string element of each pair is just used to name the decoder in error messages.

let shape_decoder_2 : shape decoder =
      [ ("a square", sqaure_decoder)
      ; ("a circle", circle_decoder)
      ; ("a triangle", triangle_decoder)

Generic decoders

Suppose our program deals with users and roles. We want to decode our JSON input into these types.

type role = Admin | User

type user =
  { name : string
  ; roles : role list

Let's define our decoders. We'll write a module functor so we can re-use the same decoders across different JSON libraries, with YAML input, or with Bucklescript.

module My_decoders(D : Decoders.Decode.S) = struct
  open D

  let role : role decoder =
    string >>= function
    | "ADMIN" -> succeed Admin
    | "USER" -> succeed User
    | _ -> fail "Expected a role"

  let user : user decoder =
    field "name" string >>= fun name ->
    field "roles" (list role) >>= fun roles ->
    succeed { name; roles }

module My_yojson_decoders = My_decoders(Decoders_yojson.Basic.Decode)

Great! Let's try them out.

utop # open My_yojson_decoders;;
utop # D.decode_string role {| "USER" |};;
- : (role, error) result = Ok User

utop # D.decode_string D.(field "users" (list user))
         {| {"users": [{"name": "Alice", "roles": ["ADMIN", "USER"]},
                       {"name": "Bob", "roles": ["USER"]}]}
- : (user list, error) result =
Ok [{name = "Alice"; roles = [Admin; User]}; {name = "Bob"; roles = [User]}]

Let's introduce an error in the JSON:

utop # D.decode_string D.(field "users" (list user))
         {| {"users": [{"name": "Alice", "roles": ["ADMIN", "USER"]},
                       {"name": "Bob", "roles": ["SUPER_USER"]}]}
- : (user list, error) result =
 in field "users":
   while decoding a list:
     element 1:
       in field "roles":
         while decoding a list:
           element 0: Expected a role, but got "SUPER_USER"

We get a nice pointer that we forgot to handle the SUPER_USER role.


After updating

npm version <newversion>
dune-release tag
git push
dune-release -p decoders,decoders-ezjsonm,decoders-yojson
AuthorMatt Bray <>
Issue Tracker
MaintainerMatt Bray <>
Source [http]
StatisticsInstalled 34 times last month.
No package is dependent