Understand and create new YAML decoders.

Here you’ll find some useful information about the design of the YAML decoders.


Decoding the database is splitted into parsing the YAML files and decoding the result. Parsing is done by the Yaml_Parse module and the Yaml_Decode module controls the decoding of the various entry types. The Yaml module then combines all functions into a single exported function.

Decode Entry Types

Decoder structure

All decodable main types share a common decoder layout. All values only relevant while decoding are inside a Decode module at the same level as the type it is the decoder for.

(** Since this is the actual type, it's not really part of the decoding pipeline
    but I'll keep it here to make the definitions complete. *)
type t = {
  (* ... *)

module Decode = struct
  open Json.Decode
  open JsonStrict

  type translation = { (* ... language-specific properties *) }

  let translation json =
      (* ... decode language-specific properties *)

  type multilingual = {
    (* ... language-independent properties *)
    translations : translation TranslationMap.t;

  let multilingual json =
      (* ... decode language-independent properties *)
      translations =
        json |> field "translations" (TranslationMap.Decode.t translation);

  let make_assoc locale_order json =
    let open Option.Infix in
    json |> multilingual |> fun multilingual ->
    multilingual.translations |> TranslationMap.preferred locale_order
    <&> fun (translation : translation) ->
        (* ... merge language-independent and language-specific properties *)
      } )

Decoding types

Since language-specific values in the YAML files are always defined in dictionaries where the keys map to the locale identifiers, we always split the language-specific and the language-independent parts.

The language-specific part of a type is always defined as type translation. The corresponding translation decoder function is later used to decode the translation dictionary, also called translation map.

The language-independent part of a type is defined as type multilingual and usually has the property translations, which is the dictionary of all translations. The corresponding multilingual decoder function uses the translation decoder together with the translation map decoder from the TranslationMap module to fully decode the dictionary of translations.

Using a list of preferred locales, sorted by preference, (Locale.Order.t) we can retrieve the most preferred translation using TranslationMap.preferred and then combine the translation values and the multilingual values to form the final record type. Following OCaml conventions, this function is called make.

Often, in order to be used more easily, the make function returns a pair of the respective entry’s identifier and the entry itself to create a Map more easily. To hint at this specific use, the function is then called make_assoc. If a nested type is only decoded in a context of collection type (like a list), this decoder can be provided using a make variant as well (in this case, make_list).

If other decodable types are nested withing a type, the multilingual function usually takes the list of preferred locales as well to be able to pass it down to the nested types’ decoders.

Decode module visibility conventions

Only the make function or its variant should be exposed.

Naming conventions

  • The language-specific tyüe is always named translation.
  • The record containing all language-independent values as well as the translation map is called multilingual and the translation map is at the property translations (just like in the source).
  • A decoder for a specific type has the same name as the type.
  • The function resolving the translation map and returning the actual main type is called make.
  • A potential helper function that decodes the type as part of a collection is called make_c, where c is the name of the collection, e.g. make_list.