Jsont
cookbookA few conventions and recipes to describe JSON data with Jsont
.
Jsont.t
valuesGiven an OCaml type t
its JSON type value should be called t_jsont
. If your type follows the M.t
module convention use M.jsont
.
Note that constructing Jsont.t
values has a cost. In particular when object descriptions are Jsont.Object.finish
ed a few checks are performed on the definition. Hence it's better to construct them as toplevel values or at least make sure you are not repeatedly constructing them dynamically in a tight loop.
null
valuesNullable JSON values are naturally mapped to ocaml option
types. The Jsont.option
combinator does exactly that.
It is also possible to map JSON null
s to a default value with Jsont.null
. This can then be combined with Jsont.any
to compose with other JSON types.
For example the following maps JSON null
s to ""
and JSON strings to string
on decoding. On encoding we unconditionally map back ""
to null
:
let string_null_is_empty =
let null = Jsont.null "" in
let enc = function "" -> null | _ -> Jsont.string in
Jsont.any ~dec_null:null ~dec_string:Jsont.string ~enc ()
See also Non-finite numbers and the tangentially related topic of Optional members.
JSON is utterly broken to interchange numbers reliably as the standard provides no constraints on their representation. Generally interopable implementations, in particular the most widely deployed and formally specified ECMAScript implementation, use IEEE 754 binary64
values to represent finite JSON numbers and null
values to represent non-finite one. This has the following consequences.
For representing integers by JSON numbers one is limited to the range [-253;253] which are the only integers represented precisely in IEEE 754 binary64
. If you want to serialize numbers beyond this range you need to represent them by a JSON string. These scheme can be seen in the wild:
Jsont.int_as_string
or Jsont.int64_as_string
can be used.Jsont.int
or Jsont.int64
can be used.Jsont.int8
, Jsont.uint8
, Jsont.int16
, etc. can be used.JSON numbers cannot represent IEEE 754 binary64
numbers: infinities and NaNs cannot be represented. The formally defined ECMAScript's JSON.stringify
function replaces these values by null
.
For this reason in Jsont
the domain of Jsont.Base.number
maps is JSON numbers or JSON null. In the decoding direction a null is mapped to Float.nan
and in the encoding direction any float not satisfying Float.is_finite
is mapped to a JSON null.
If you can agree with a third party on a better encoding, the Jsont.any_float
or Jsont.float_as_hex_string
provide lossless representations of IEEE 754 binary64
values in JSON.
The Jsont.map
combinator is a general map over Jsont.t
types. It should rather be used to alter the representation of existing Jsont.t
values. For transforming base types it is better to use the base maps of Jsont.Base
as more context is made available to the functions, notably when erroring.
A few simple JSON string transformers like Jsont.enum
or Jsont.binary_string
are provided.
If you need to devise your own maps from your own M.{of,to}_string
functions that return result
or raise Faiulre _
you can adapt them with these functions. For example:
let m_jsont =
let dec = Jsont.Base.dec_result M.result_of_string in
let enc = Jsont.Base.enc M.to_string in
Jsont.Base.string (Jsont.Base.map ~kind:"M.t" ~dec ~enc ())
let m_jsont' =
let dec = Jsont.Base.dec_failure M.of_string_or_failure in
let enc = Jsont.Base.enc M.to_string in
Jsont.Base.string (Jsont.Base.map ~kind:"M.t" ~dec ~enc ())
If you are dealing with result decoders you can also simply use Jsont.of_of_string
:
let m_jsont'' =
Jsont.of_of_string ~kind:"M.t" M.result_of_string ~enc:M.to_string
which is a shortcut for the m_jsont
written above.
JSON arrays can be directly mapped to OCaml lists, arrays, bigarray or bespoke low-dimensional tuples. If your JSON is an array of objects keyed by some identifier you may find Jsont.array_as_string_map
handy.
If none of that fits you can always devise your own Jsont.Array.map
.
Suppose our JSON object is:
{ "name": "Jane Doe"
"age": 56 }
We represent it with an OCaml record as follows:
module Person = struct
type t = { name : string; age : int }
let make name age = { name; age }
let name p = p.name
let age p = p.age
let jsont =
Jsont.Object.map ~kind:"Person" make
|> Jsont.Object.mem "name" Jsont.string ~enc:name
|> Jsont.Object.mem "age" Jsont.int ~enc:age
|> Jsont.Object.finish
end
JSON objects can be used as maps from strings to a single type of value (example). Such maps can be easily converted to OCaml as follows:
module String_map = Map.Make (String)
let map : ?kind:string -> 'a Jsont.t -> 'a String_map.t Jsont.t =
fun ?kind t ->
Jsont.Object.map ?kind Fun.id
|> Jsont.Object.keep_unknown (Jsont.Object.Mems.string_map t) ~enc:Fun.id
|> Jsont.Object.finish
Since the pattern is common this is directly exposed as Jsont.Object.as_string_map
.
By default members specified via Jsont.Object.mem
are mandatory and decoding errors if the member is absent.
For those cases where the member is optional a default dec_absent
value must be specified to use on decoding when absent. For encoding an enc_omit
function can be specified to determine whether the member should be omitted on encoding.
In the following example we use an option type to denote the potential absence of the age
member:
module Person_opt_age = struct
type t = { name : string; age : int option }
let make name age = { name; age }
let name p = p.name
let age p = p.age
let jsont =
Jsont.Object.map ~kind:"Person" make
|> Jsont.Object.mem "name" Jsont.string ~enc:name
|> Jsont.Object.mem "age" Jsont.(some int)
~dec_absent:None ~enc_omit:Option.is_none ~enc:age
|> Jsont.Object.finish
end
When absence is represented by None
like here the Jsont.Object.opt_mem
function can be used. It's stricly equivalent to the above but more concise.
In JSON objects maps, there are three different ways to handle object members that have not been declared by a Jsont.Object.mem
or Jsont.Object.opt_mem
.
By default Jsont.Object.map
skips unknown object members.
To error on unknown members use Jsont.Object.error_unknown
:
module Person_strict = struct
type t = { name : string; age : int; }
let make name age = { name; age }
let name p = p.name
let age p = p.age
let jsont =
Jsont.Object.map ~kind:"Person" make
|> Jsont.Object.mem "name" Jsont.string ~enc:name
|> Jsont.Object.mem "age" Jsont.int ~enc:age
|> Jsont.Object.error_unknown
|> Jsont.Object.finish
end
If a JSON data schema allows foreign members or to partially model an object, unknown members can be collected into a generic Jsont.Json.t
object and stored in an OCaml field by using Jsont.Object.keep_unknown
and Jsont.json_mems
:
module Person_keep = struct
type t = { name : string; age : int; unknown : Jsont.json ; }
let make name age unknown = { name; age; unknown }
let name p = p.name
let age p = p.age
let unknown v = v.unknown
let jsont =
Jsont.Object.map ~kind:"Person" make
|> Jsont.Object.mem "name" Jsont.string ~enc:name
|> Jsont.Object.mem "age" Jsont.int ~enc:age
|> Jsont.Object.keep_unknown Jsont.json_mems ~enc:unknown
|> Jsont.Object.finish
end
The value of the unknown
field can be further queried with other JSON types and Jsont.Json.decode
. It is also possible to define your own data structure to keep unknown members, see Jsont.Object.Mems
. See also Objects as key-value maps.
Sometimes JSON objects have a distinguished case member, called "type"
, "class"
or "version"
whose value define the rest of the object.
The Jsont.Object.Case
module handles this pattern. Each case is described by a Jsont.Object.map
object description and the Jsont.Object.case_mem
allows to chose between them according to the value of the case member.
In OCaml there are two main ways to represent these case objects. Either by an enclosing variant type with one case for each object kind:
type t = C1 of C1.t | C2 of C2.t | …
or with a record which holds common fields an a field that holds the cases:
type type' = C1 of C1.t | C2 of C2.t | …
type t = { type' : type'; … (* other common fields *) }
From Jsont
's perspective there is not much difference.
We show both modellings on a hypothetic Geometry
object which has a "name"
member and a "type"
string case member indicating whether the object is a "Circle"
or a "Rect"
. Except for the position of the name
field, not much changes in each modelling.
Using an enclosing variant type:
module Geometry_variant = struct
module Circle = struct
type t = { name : string; radius : float; }
let make name radius = { name; radius }
let name c = c.name
let radius c = c.radius
let jsont =
Jsont.Object.map ~kind:"Circle" make
|> Jsont.Object.mem "name" Jsont.string ~enc:name
|> Jsont.Object.mem "radius" Jsont.number ~enc:radius
|> Jsont.Object.finish
end
module Rect = struct
type t = { name : string; width : float; height : float }
let make name width height = { name; width; height }
let name r = r.name
let width r = r.width
let height r = r.height
let jsont =
Jsont.Object.map ~kind:"Rect" make
|> Jsont.Object.mem "name" Jsont.string ~enc:name
|> Jsont.Object.mem "width" Jsont.number ~enc:width
|> Jsont.Object.mem "height" Jsont.number ~enc:height
|> Jsont.Object.finish
end
type t = Circle of Circle.t | Rect of Rect.t
let circle c = Circle c
let rect r = Rect r
let jsont =
let circle = Jsont.Object.Case.map "Circle" Circle.jsont ~dec:circle in
let rect = Jsont.Object.Case.map "Rect" Rect.jsont ~dec:rect in
let enc_case = function
| Circle c -> Jsont.Object.Case.value circle c
| Rect r -> Jsont.Object.Case.value rect r
in
let cases = Jsont.Object.Case.[make circle; make rect] in
Jsont.Object.map ~kind:"Geometry" Fun.id
|> Jsont.Object.case_mem "type" Jsont.string ~enc:Fun.id ~enc_case cases
|> Jsont.Object.finish
end
Using a record with a type'
field:
module Geometry_record = struct
module Circle = struct
type t = { radius : float; }
let make radius = { radius }
let radius c = c.radius
let jsont =
Jsont.Object.map ~kind:"Circle" make
|> Jsont.Object.mem "radius" Jsont.number ~enc:radius
|> Jsont.Object.finish
end
module Rect = struct
type t = { width : float; height : float }
let make width height = { width; height }
let width r = r.width
let height r = r.height
let jsont =
Jsont.Object.map ~kind:"Rect" make
|> Jsont.Object.mem "width" Jsont.number ~enc:width
|> Jsont.Object.mem "height" Jsont.number ~enc:height
|> Jsont.Object.finish
end
type type' = Circle of Circle.t | Rect of Rect.t
let circle c = Circle c
let rect r = Rect r
type t = { name : string; type' : type' }
let make name type' = { name; type' }
let name g = g.name
let type' g = g.type'
let jsont =
let circle = Jsont.Object.Case.map "Circle" Circle.jsont ~dec:circle in
let rect = Jsont.Object.Case.map "Rect" Rect.jsont ~dec:rect in
let enc_case = function
| Circle c -> Jsont.Object.Case.value circle c
| Rect r -> Jsont.Object.Case.value rect r
in
let cases = Jsont.Object.Case.[make circle; make rect] in
Jsont.Object.map ~kind:"Geometry" make
|> Jsont.Object.mem "name" Jsont.string ~enc:name
|> Jsont.Object.case_mem "type" Jsont.string ~enc:type' ~enc_case cases
|> Jsont.Object.finish
end
If you are only interested in extracting data it may be useful to flatten some objects whose members are too nested for your needs.
For that just remember that nothing says that JSON objects cannot be mapped to OCaml functions. For examples to gather this kind of data for a group of person into a single record:
{
"info" : { "id" : 1, "name": "untitled" }
"persons" : [ … ]
}
You can use the following structure:
module Group = struct
type t = { id : int; name : string; persons : Person.t list }
let make id name persons = { id; name; persons }
let info_jsont =
Jsont.Object.map make
|> Jsont.Object.mem "id" Jsont.int
|> Jsont.Object.mem "name" Jsont.string
|> Jsont.Object.finish
let jsont =
Jsont.Object.map (fun k persons -> k persons)
|> Jsont.Object.mem "info" info_jsont
|> Jsont.Object.mem "persons" (Jsont.list Person.jsont)
|> Jsont.Object.finish
end
This however will not allow you to use jsont
to encode. If you wish to do so it's likely better to follow the JSON structure and hide the annoying access structure under an abstract type behind a nice API.
To describe recursive JSON values you need to define your description in a lazy
expression and use Jsont.rec'
to refer to the value you are defining. This results in the following structure:
let jsont : t Jsont.t =
let rec t = lazy ( … Jsont.rec' t … ) in
Lazy.force t
For example a tree encoded as a JSON object with:
{ "value": …,
"children": […] }
Is modelled by:
module Tree = struct
type 'a t = Node of 'a * 'a t list
let make v children = Node (v, children)
let value (Node (v, _)) = v
let children (Node (_, children)) = children
let jsont value_type =
let rec t = lazy
(Jsont.Object.map ~kind:"Tree" make
|> Jsont.Object.mem "value" value_type ~enc:value
|> Jsont.Object.mem "children" (Jsont.list (Jsont.rec' t)) ~enc:children
|> Jsont.Object.finish)
in
Lazy.force t
end
The topojson.ml
and geojson.ml
examples in the source repository provide more extensive examples of recursive definition.