OCaml Support - atdgen

Tutorial

What is atdgen?

Atdgen is a tool that derives OCaml boilerplate code from type definitions. Currently it provides support for:

  • JSON serialization and deserialization.
  • Biniou serialization and deserialization. Biniou is a binary format extensible like JSON but more compact and faster to process.
  • Convenience functions for creating and validating OCaml data.

What are the advantages of atdgen?

Atdgen has a number of advantages over its predecessor json-static which was based on Camlp4:

  • produces explicit interfaces which describe what is available to the user (.mli files).
  • produces readable OCaml code that can be easily reviewed (.ml files).
  • produces fast code, 3x faster than json-static.
  • runs fast, keeping build times low.
  • same ATD definitions can be used to generate code other than OCaml. See for instance atdj which generates Java classes for JSON IO. Auto-generating GUI widgets from type definitions is another popular use of annotated type definitions. The implementation of such code generators is facilitated by the atd library.

Prerequisites

This tutorial assumes that you are using atdgen version 1.5.0 or above. The following command tells you which version you are using:

$ atdgen -version
1.5.0

The recommended way of installing atdgen and all its dependencies is with opam:

$ opam install atdgen

Getting started

From now on we assume that atdgen 1.5.0 or above is installed properly.

$ atdgen -version
1.5.0

Type definitions are placed in a .atd file (hello.atd):

type date = {
  year : int;
  month : int;
  day : int;
}

Our handwritten OCaml program is hello.ml:

open Hello_t
let () =
  let date = { year = 1970; month = 1; day = 1 } in
  print_endline (Hello_j.string_of_date date)

We produce OCaml code from the type definitions using atdgen:

$ atdgen -t hello.atd     # produces OCaml type definitions
$ atdgen -j hello.atd     # produces OCaml code dealing with JSON

We now have _t and _j files produced by atdgen -t and atdgen -j respectively:

$ ls
hello.atd  hello.ml  hello_j.ml  hello_j.mli  hello_t.ml  hello_t.mli

We compile all .mli and .ml files:

$ ocamlfind ocamlc -c hello_t.mli -package atdgen
$ ocamlfind ocamlc -c hello_j.mli -package atdgen
$ ocamlfind ocamlopt -c hello_t.ml -package atdgen
$ ocamlfind ocamlopt -c hello_j.ml -package atdgen
$ ocamlfind ocamlopt -c hello.ml -package atdgen
$ ocamlfind ocamlopt -o hello hello_t.cmx hello_j.cmx hello.cmx -package atdgen -linkpkg

And finally we run our hello program:

$ ./hello
{"year":1970,"month":1,"day":1}

Source code for this section

Inspecting and pretty-printing JSON

Input JSON data:

$ cat single.json
[1234,"abcde",{"start_date":{"year":1970,"month":1,"day":1},
"end_date":{"year":1980,"month":1,"day":1}}]

Pretty-printed JSON can be produced with the ydump command:

$ ydump single.json
[
  1234,
  "abcde",
  {
    "start_date": { "year": 1970, "month": 1, "day": 1 },
    "end_date": { "year": 1980, "month": 1, "day": 1 }
  }
]

Multiple JSON objects separated by whitespace, typically one JSON object per line, can also be pretty-printed with ydump. Input:

$ cat stream.json
[1234,"abcde",{"start_date":{"year":1970,"month":1,"day":1},
"end_date":{"year":1980,"month":1,"day":1}}]
[1,"a",{}]

In this case the -s option is required:

$ ydump -s stream.json
[
  1234,
  "abcde",
  {
    "start_date": { "year": 1970, "month": 1, "day": 1 },
    "end_date": { "year": 1980, "month": 1, "day": 1 }
  }
]
[ 1, "a", {} ]

From an OCaml program, pretty-printing can be done with Yojson.Safe.prettify which has the following signature:

val prettify : string -> string

We wrote a tiny program that simply calls the prettify function on some predefined JSON data (file prettify.ml):

let json =
"[1234,\"abcde\",{\"start_date\":{\"year\":1970,\"month\":1,\"day\":1},
\"end_date\":{\"year\":1980,\"month\":1,\"day\":1}}]"

let () = print_endline (Yojson.Safe.prettify json)

We now compile and run prettify.ml:

$ ocamlfind ocamlopt -o prettify prettify.ml -package atdgen -linkpkg
$ ./prettify
[
  1234,
  "abcde",
  {
    "start_date": { "year": 1970, "month": 1, "day": 1 },
    "end_date": { "year": 1980, "month": 1, "day": 1 }
  }
]

Source code for this section

Inspecting biniou data

Biniou is a binary format that can be displayed as text using a generic command called bdump. The only practical difficulty is to recover the original field names and variant names which are stored as 31-bit hashes. Unhashing them is done by consulting a dictionary (list of words) maintained by the user.

Let’s first produce a sample data file tree.dat containing the biniou representation of a binary tree. In the same program we will also demonstrate how to render biniou data into text from an OCaml program.

Here is the ATD file defining our tree type (file tree.atd):

type tree = [
  | Empty
  | Node of (tree * int * tree)
]

This is our OCaml program (file tree.ml):

open Printf

(* sample value *)
let tree : Tree_t.tree =
  `Node (
    `Node (`Empty, 1, `Empty),
    2,
    `Node (
      `Node (`Empty, 3, `Empty),
      4,
      `Node (`Empty, 5, `Empty)
    )
  )

let () =
  (* write sample value to file *)
  let fname = "tree.dat" in
  Atdgen_runtime.Util.Biniou.to_file Tree_b.write_tree fname tree;

  (* write sample value to string *)
  let s = Tree_b.string_of_tree tree in
  printf "raw value (saved as %s):\n%S\n" fname s;
  printf "length: %i\n" (String.length s);

  printf "pretty-printed value (without dictionary):\n";
  print_endline (Bi_io.view s);

  printf "pretty-printed value (with dictionary):\n";
  let unhash = Bi_io.make_unhash ["Empty"; "Node"; "foo"; "bar" ] in
  print_endline (Bi_io.view ~unhash s)

Compilation:

$ atdgen -t tree.atd
$ atdgen -b tree.atd
$ ocamlfind ocamlopt -o tree \
    tree_t.mli tree_t.ml tree_b.mli tree_b.ml tree.ml \
    -package atdgen -linkpkg

Running the program:

$ ./tree
raw value (saved as tree.dat):
"\023\179\2276\"\020\003\023\179\2276\"\020\003\023\003\007\170m\017\002\023\003\007\170m\017\004\023\179\2276\"\020\003\023\179\2276\"\020\003\023\003\007\170m\017\006\023\003\007\170m\017\b\023\179\2276\"\020\003\023\003\007\170m\017\n\023\003\007\170m"
length: 75
pretty-printed value (without dictionary):
<#33e33622:
  (<#33e33622: (<#0307aa6d>, 1, <#0307aa6d>)>,
    2,
    <#33e33622:
      (<#33e33622: (<#0307aa6d>, 3, <#0307aa6d>)>,
        4,
        <#33e33622: (<#0307aa6d>, 5, <#0307aa6d>)>)>)>
pretty-printed value (with dictionary):
<"Node":
  (<"Node": (<"Empty">, 1, <"Empty">)>,
    2,
    <"Node":
      (<"Node": (<"Empty">, 3, <"Empty">)>,
        4,
        <"Node": (<"Empty">, 5, <"Empty">)>)>)>

Now let’s see how to pretty-print any biniou data from the command line. Our sample data are now in file tree.dat:

$ ls -l tree.dat
-rw-r--r-- 1 martin martin 75 Apr 17 01:46 tree.dat

We use the command bdump to render our sample biniou data as text:

$ bdump tree.dat
<#33e33622:
  (<#33e33622: (<#0307aa6d>, 1, <#0307aa6d>)>,
    2,
    <#33e33622:
      (<#33e33622: (<#0307aa6d>, 3, <#0307aa6d>)>,
        4,
        <#33e33622: (<#0307aa6d>, 5, <#0307aa6d>)>)>)>

We got hashes for the variant names Empty and Node. Let’s add them to the dictionary:

$ bdump -w Empty,Node tree.dat
<"Node":
  (<"Node": (<"Empty">, 1, <"Empty">)>,
    2,
    <"Node":
      (<"Node": (<"Empty">, 3, <"Empty">)>,
        4,
        <"Node": (<"Empty">, 5, <"Empty">)>)>)>

bdump remembers the dictionary so we don’t have to pass the -w option anymore (for this user on this machine). The following now works:

$ bdump tree.dat
<"Node":
  (<"Node": (<"Empty">, 1, <"Empty">)>,
    2,
    <"Node":
      (<"Node": (<"Empty">, 3, <"Empty">)>,
        4,
        <"Node": (<"Empty">, 5, <"Empty">)>)>)>

Source code for this section

Optional fields and default values

Although OCaml records do not support optional fields, both the JSON and biniou formats make it possible to omit certain fields on a per-record basis.

For example the JSON record { “x”: 0, “y”: 0 } can be more compactly written as {} if the reader knows the default values for the missing fields x and y. Here is the corresponding type definition:

type vector_v1 = { ~x: int; ~y: int }

~x means that field x supports a default value. Since we do not specify the default value ourselves, the built-in default is used, which is 0.

If we want the default to be something else than 0, we just have to specify it as follows:

type vector_v2 = {
  ~x <ocaml default="1">: int; (* default x is 1 *)
  ~y: int;                     (* default y is 0 *)
}

It is also possible to specify optional fields without a default value. For example, let’s add an optional z field:

type vector_v3 = {
  ~x: int;
  ~y: int;
  ?z: int option;
}

The following two examples are valid JSON representations of data of type vector_v3:

{ "x": 2, "y": 2, "z": 3 }  // OCaml: { x = 2; y = 2; z = Some 3 }
{ "x": 2, "y": 2 }          // OCaml: { x = 2; y = 2; z = None }

By default, JSON fields whose value is null are treated as missing fields. The following two JSON objects are therefore equivalent:

{ "x": 2, "y": 2, "z": null }
{ "x": 2, "y": 2 }

Note also the difference between ?z: int option and ~z: int option:

type vector_v4 = {
  ~x: int;
  ~y: int;
  ~z: int option;  (* no unwrapping of the JSON field value! *)
}

Here are valid values of type vector_v4, showing that it is usually not what is intended:

{ "x": 2, "y": 2, "z": [ "Some", 3 ] }
{ "x": 2, "y": 2, "z": "None" }
{ "x": 2, "y": 2 }

Smooth protocol upgrades

Problem: you have a production system that uses a specific JSON or biniou format. It may be data files or a client-server pair. You now want to add a field to a record type or to add a case to a variant type.

Both JSON and biniou allow extra record fields. If the consumer does not know how to deal with the extra field, the default behavior is to happily ignore it.

Adding or removing an optional record field

type t = {
  x: int;
  y: int;
}

Same .atd source file, edited:

type t = {
  x: int;
  y: int;
  ~z: int; (* new field *)
}
  • Upgrade producers and consumers in any order
  • Converting old data is not required nor useful

Adding a required record field

type t = {
  x: int;
  y: int;
}

Same .atd source file, edited:

type t = {
  x: int;
  y: int;
  z: int; (* new field *)
}
  • Upgrade all producers before the consumers
  • Converting old data requires special-purpose hand-written code

Removing a required record field

  • Upgrade all consumers before the producers
  • Converting old data is not required but may save some storage space (just read and re^write each record using the new type)

Adding a variant case

type t = [ A | B ]

Same .atd source file, edited:

type t = [ A | B | C ]
  • Upgrade all consumers before the producers
  • Converting old data is not required and would have no effect

Removing a variant case

  • Upgrade all producers before the consumers
  • Converting old data requires special^purpose hand^written code

Avoiding future problems

  • In doubt, use records rather than tuples because it makes it possible to add or remove any field or to reorder them.
  • Do not hesitate to create variant types with only one case or records with only one field if you think they might be extended later.

Data validation

Atdgen can be used to produce data validators for all types defined in an ATD file, based on user-given validators specified only for certain types. A simple example is:

type t = string <ocaml valid="fun s -> String.length s >= 8"> option

As we can see from this example, the validation function is specified using the annotation <ocaml valid="p">, where p is a predicate p : t -> bool, returning true when the value of type t is valid and false otherwise.

Calling atdgen -v on a file containing this specification will produce a validation function equivalent to the following implementation:

let validate_t path x =
  match x with
  | None -> None
  | Some x ->
      let msg = "Failed check by fun s -> String.length s >= 8" in
      if (fun s -> String.length s >= 8) x
      then None
      else Some {error_path = path; error_msg = msg}

Let’s consider this particular example as an illustration of the general shape of generated validation functions.

The function takes two arguments: the first, path, is a list indicating where the second, x, was encountered. As specified by our example .atd code above, x has type t option.

The body of the validation function does two things:

1. it checks the value of x against the validation function specified in our .atd file, namely, checking whether there is Some s, and verifying that s is at least 8 characters long if so 2. in the event that the validation check fails, it constructs an appropriate error record.

In general, generated validation functions for a type t have a type equivalent to validate_t : path -> t -> error option, where the path gives the current location in a data structure and the error is a record of the location of, and reason for, validation failure.

A return value of None indicates successful validation, while Some {error_path; error_msg} tells us where and why validation failed.

Let’s now consider a more realistic example with complex validators defined in a separate .ml file. We will define a data structure representing a section of a resume recording work experience. We will also define validation functions that can enforce certain properties to protect against errors and junk data.

In the course of this example, we will manually create the following 3 source files:

  • resume.atd: contains the type definitions with annotations
  • resume_util.ml: contains our handwritten validators
  • resume.ml: is our main program that creates data and checks it using our generated validation functions.

After generating additional code with atdgen, we will end up with the following OCaml modules:

  • Resume_t: generated into resume_t.ml by atdgen -t resume.atd, this provides our OCaml type definitions
  • Resume_util: written manually in resume_util.ml, this depends on Resume_t and provides validators we will use in resume.atd
  • Resume_v: generated into resume_v.ml by atdgen -v resume.atd, this depends on Resume_util and Resume_t and provides a validation function for each type
  • Resume_j: generated into resume_j.ml by atdgen -j resume.atd, this provides functions to serialize and deserialize data in and out of JSON.
  • Resume: written manually in resume.ml, this depends on Resume_v, and Resume_t, and makes use of the generated types and validation functions.

To begin, we specify type definitions for a data structure representing a resume in resume.atd:

type text = string <ocaml valid="Resume_util.validate_some_text">

type date = {
  year : int;
  month : int;
  day : int;
} <ocaml valid="Resume_util.validate_date">

type job = {
  company : text;
  title : text;
  start_date : date;
  ?end_date : date option;
} <ocaml valid="Resume_util.validate_job">

type work_experience = job list

We can now call atdgen -t resume.atd to generate our Resume_t module in resume_t.ml, providing our data types. Using these data types, we’ll define the following handwritten validators in resume_util.ml (note that we’ve already referred to these validators in resume.atd):

open Resume_t

let ascii_printable c =
  let n = Char.code c in
  n >= 32 && n <= 127

(*
  Check that string is not empty and contains only ASCII printable
  characters (for the sake of the example; we use UTF-8 these days)
*)
let validate_some_text s =
  s <> "" &&
    try
      String.iter (fun c -> if not (ascii_printable c) then raise Exit) s;
      true
    with Exit ->
      false

(*
  Check that the combination of year, month and day exists in the
  Gregorian calendar.
*)
let validate_date x =
  let y = x.year in
  let m = x.month in
  let d = x.day in
  m >= 1 && m <= 12 && d >= 1 &&
  (let dmax =
    match m with
        2 ->
          if y mod 4 = 0 && not (y mod 100 = 0) || y mod 400 = 0 then 29
          else 28
      | 1 | 3 | 5 | 7 | 8 | 10 | 12 -> 31
      | _ -> 30
  in
  d <= dmax)

(* Compare dates chronologically *)
let compare_date a b =
  let c = compare a.year b.year in
  if c <> 0 then c
  else
    let c = compare a.month b.month in
    if c <> 0 then c
    else compare a.day b.day

(* Check that the end_date, when defined, is not earlier than the start_date *)
let validate_job x =
  match x.end_date with
      None -> true
    | Some end_date ->
        compare_date x.start_date end_date <= 0

After we call atdgen -v resume.atd, the module Resume_v will be generated in resume_v.ml, providing the function validate_work_experience . We can then use this function, along with the generated Resume_j in the following program written in resume.ml:

let check_experience x =
  let is_valid = match Resume_v.validate_work_experience [] x with
    | None -> false
    | _ -> true
  in
  Printf.printf "%s:\n%s\n"
    (if is_valid then "VALID" else "INVALID")
    (Yojson.Safe.prettify (Resume_j.string_of_work_experience x))

let () =
  (* one valid date *)
  let valid = { Resume_t.year = 2000; month = 2; day = 29 } in
  (* one invalid date *)
  let invalid = { Resume_t.year = 1900; month = 0; day = 0 } in
  (* two more valid dates, created with Resume_v.create_date *)
  let date1 = { Resume_t.year = 2005; month = 8; day = 1 } in
  let date2 = { Resume_t.year = 2006; month = 3; day = 22 } in

  let job = {
    Resume_t.company = "Acme Corp.";
    title = "Tester";
    start_date = date1;
    end_date = Some date2;
  }
  in
  let valid_job = { job with Resume_t.start_date = valid } in
  let invalid_job = { job with Resume_t.end_date = Some invalid } in
  let valid_experience = [ job; valid_job ] in
  let invalid_experience = [ job; invalid_job ] in
  check_experience valid_experience;
  check_experience invalid_experience

Output:

VALID:
[
  {
    "company": "Acme Corp.",
    "title": "Tester",
    "start_date": { "year": 2005, "month": 8, "day": 1 },
    "end_date": { "year": 2006, "month": 3, "day": 22 }
  },
  {
    "company": "Acme Corp.",
    "title": "Tester",
    "start_date": { "year": 2000, "month": 2, "day": 29 },
    "end_date": { "year": 2006, "month": 3, "day": 22 }
  }
]
INVALID:
[
  {
    "company": "Acme Corp.",
    "title": "Tester",
    "start_date": { "year": 2005, "month": 8, "day": 1 },
    "end_date": { "year": 2006, "month": 3, "day": 22 }
  },
  {
    "company": "Acme Corp.",
    "title": "Tester",
    "start_date": { "year": 2005, "month": 8, "day": 1 },
    "end_date": { "year": 1900, "month": 0, "day": 0 }
  }

Source code for this section

Modularity: referring to type definitions from another ATD file

It is possible to define types that depend on types defined in other .atd files. The example below is self-explanatory.

part1.atd:

type t = { x : int; y : int }

part2.atd:

type t1 <ocaml from="Part1" t="t"> = abstract
    (*
      Imports type t defined in file part1.atd.
      The local name is t1. Because the local name (t1) is different from the
      original name (t), we must specify the original name using t=.
    *)

type t2 = t1 list

part3.atd:

type t2 <ocaml from="Part2"> = abstract

type t3 = {
  name : string;
  ?data : t2 option;
}

main.ml:

let v = {
  Part3_t.name = "foo";
  data = Some [
    { Part1_t.x = 1; y = 2 };
    { Part1_t.x = 3; y = 4 };
  ]
}

let () =
  Atdgen_runtime.Util.Json.to_channel Part3_j.write_t3 stdout v;
  print_newline ()

Output:

{"name":"foo","data":[{"x":1,"y":2},{"x":3,"y":4}]}

Source code for this section

Managing JSON configuration files

JSON makes a good format for configuration files because it is human-readable, easy to modify programmatically and widespread. Here is an example of how to use atdgen to manage config files.

  • Specifying defaults is done in the .atd file. See section [Optional fields and default values] for details on how to do that.
  • Auto-generating a template config file with default values: a sample value in the OCaml world needs to be created but only fields without default need to be specified.
  • Describing the format is achieved by embedding the .atd type definitions in the OCaml program and printing it out on request.
  • Loading a config file and reporting illegal fields is achieved using the JSON deserializers produced by atdgen -j. Option -j-strict-fields ensures the misspelled field names are not ignored but reported as errors.
  • Reindenting a config file is achieved by the pretty-printing function Yojson.Safe.prettify that takes a JSON string and returns an equivalent JSON string.
  • Showing implicit (default) settings is achieved by passing the -j-defaults option to atdgen. The OCaml config data is then serialized into JSON containing all fields, including those whose value is the default.

The example uses the following type definitions:

type config = {
  title : string;
  ?description : string option;
  ~timeout <ocaml default="10"> : int;
  ~credentials : param list
    <ocaml valid="fun l ->
                    l <> [] || failwith \"missing credentials\"">;
}

type param = {
  name : string
    <ocaml valid="fun s -> s <> \"\"">;
  key : string
    <ocaml valid="fun s -> String.length s = 16">;
}

Our program will perform the following actions:

$ ./config -template
{
  "title": "",
  "timeout": 10,
  "credentials": [ { "name": "foo", "key": "0123456789abcdef" } ]
}

$ ./config -format
type config = {
  title : string;
  ?description : string option;
  ~timeout <ocaml default="10"> : int;
  ~credentials : param list
    <ocaml valid="fun l ->
                    l <> [] || failwith \"missing credentials\"">;
}

type param = {
  name : string
    <ocaml valid="fun s -> s <> \"\"">;
  key : string
    <ocaml valid="fun s -> String.length s = 16">;
}

$ cat sample-config.json
{
  "title": "Example",
  "credentials": [
    {
      "name": "joeuser",
      "key": "db7c0877bdef3016"
    },
    {
      "name": "tester",
      "key": "09871ff387ac2b10"
    }
  ]
}

$ ./config -validate sample-config.json
{
  "title": "Example",
  "timeout": 10,
  "credentials": [
    { "name": "joeuser", "key": "db7c0877bdef3016" },
    { "name": "tester", "key": "09871ff387ac2b10" }
  ]
}

This is our demo.sh script that builds and runs our example program called config:

#! /bin/sh -e

set -x

# Embed the contents of the .atd file into our OCaml program
echo 'let contents = "\' > config_atd.ml
sed -e 's/\([\\"]\)/\\\1/g' config.atd >> config_atd.ml
echo '"' >> config_atd.ml

# Derive OCaml type definitions from .atd file
atdgen -t config.atd

# Derive JSON-related functions from .atd file
atdgen -j -j-defaults -j-strict-fields config.atd

# Derive validator from .atd file
atdgen -v config.atd

# Compile the OCaml program
ocamlfind ocamlopt -o config \
  config_t.mli config_t.ml config_j.mli config_j.ml config_v.mli config_v.ml \
  config_atd.ml config.ml -package atdgen -linkpkg

# Output a sample config
./config -template

# Print the original type definitions
./config -format

# Fail to validate an invalid config file
./config -validate bad-config1.json || :

# Fail to validate another invalid config file (using custom validators)
./config -validate bad-config3.json || :

# Validate, inject missing defaults and pretty-print
./config -validate sample-config.json

This is the hand-written OCaml program. It can be used as a start
point for a real-world program using a JSON config file:
open Printf

let param_template =
  (* Sample item used to populate the template config file *)
  {
    Config_v.name = "foo";
    key = "0123456789abcdef"
  }

let config_template =
  (*
    Records can be conveniently created using functions generated by
    "atdgen -v".
    Here we use Config_v.create_config to create a record of type
    Config_t.config. The big advantage over creating the record
    directly using the record notation {...} is that we don't have to
    specify default values (such as timeout in this example).
  *)
  Config_v.create_config ~title:"" ~credentials: [param_template] ()

let make_json_template () =
  (* Thanks to the -j-defaults flag passed to atdgen, even default
    fields will be printed out *)
  let compact_json = Config_j.string_of_config config_template in
  Yojson.Safe.prettify compact_json

let print_template () =
  print_endline (make_json_template ())

let print_format () =
  print_string Config_atd.contents

let validate fname =
  let x =
    try
      (* Read config data structure from JSON file *)
      let x = Atdgen_runtime.Util.Json.from_file Config_j.read_config fname in
      (* Call the validators specified by <ocaml valid=...> *)
      if not (Config_v.validate_config x) then
        failwith "Some fields are invalid"
      else
        x
    with e ->
      (* Print decent error message and exit *)
      let msg =
        match e with
            Failure s
          | Yojson.Json_error s -> s
          | e -> Printexc.to_string e
      in
      eprintf "Error: %s\n%!" msg;
      exit 1
  in
  (* Convert config to compact JSON and pretty-print it.
    ~std:true means that the output will not use extended syntax for
    variants and tuples but only standard JSON. *)
  let json = Yojson.Safe.prettify ~std:true (Config_j.string_of_config x) in
  print_endline json

type action = Template | Format | Validate of string

let main () =
  let action = ref Template in
  let options = [
    "-template", Arg.Unit (fun () -> action := Template),
    "
          prints a sample configuration file";

    "-format", Arg.Unit (fun () -> action := Format),
    "
          prints the format specification of the config files (atd format)";

    "-validate", Arg.String (fun s -> action := Validate s),
    "<CONFIG FILE>
          reads a config file, validates it, adds default values
          and prints the config nicely to stdout";
  ]
  in
  let usage_msg = sprintf "\
Usage: %s [-template|-format|-validate ...]
Demonstration of how to manage JSON configuration files with atdgen.
"
    Sys.argv.(0)
  in
  let anon_fun s = eprintf "Invalid command parameter %S\n%!" s; exit 1 in
  Arg.parse options anon_fun usage_msg;

  match !action with
      Template -> print_template ()
    | Format -> print_format ()
    | Validate s -> validate s

let () = main ()

The full source code for this section with examples can be inspected and downloaded here.

Integration with ocamldoc

Ocamldoc is a tool that comes with the core OCaml distribution. It uses comments within (** and *) to produce hyperlinked documentation (HTML) of module signatures.

Atdgen can produce .mli files with comments in the syntax supported by ocamldoc but regular ATD comments within (* and *) are always discarded by atdgen. Instead, <doc text=”…”> must be used and placed after the element they describe. The contents of the text field must be UTF8-encoded.

type point = {
  x : float;
  y : float;
  ~z
    <doc text="Optional depth, its default value is {{0.0}}.">
    : float;
}
  <doc text="Point with optional 3rd dimension.

OCaml example:
{{{
let p =
  { x = 0.5; y = 1.0; z = 0. }
}}}
">

is converted into the following .mli file with ocamldoc-compatible comments:

(**
  Point with optional 3rd dimension.

  OCaml example:

{v
let p =
  \{ x = 0.5; y = 1.0; z = 0. \}
v}
*)
type point = {
  x: float;
  y: float;
  z: float (** Optional depth, its default value is [0.0]. *)
}

The only two forms of markup supported by <doc text="..."> are {{}} for inline code and {{{}}} for a block of preformatted code.

Integration with build systems

OMake

We provide an Atdgen plugin for OMake. It simplifies the compilation rules to a minimum.

The plugin consists of a self-documented file to copy into a project’s root. The following is a sample OMakefile for a project using JSON and five source files (foo.atd, foo.ml, bar.atd, bar.ml and main.ml):

# require file Atdgen.om
include Atdgen

# OCaml modules we want to build
OCAMLFILES = foo_t foo_j foo bar_t bar_j bar main

Atdgen(foo bar, -j-std)
OCamlProgram(foobar, $(OCAMLFILES))

.DEFAULT: foobar.opt

.PHONY: clean
clean:
  rm -f *.cm[ioxa] *.cmx[as] *.[oa] *.opt *.run *~
  rm -f $(ATDGEN_OUTFILES)

Running omake builds the native code executable foobar.opt.

omake clean removes all the products of compilation including the .mli and .ml produced by atdgen.

GNU Make

We provide Atdgen.mk, a generic makefile that defines the dependencies and rules for generating OCaml .mli and .ml files from .atd files containing type definitions. The Atdgen.mk file contains its own documentation.

Here is a sample Makefile that takes advantage of OCamlMakefile:

.PHONY: default
default: opt

ATDGEN_SOURCES = foo.atd bar.atd
ATDGEN_FLAGS = -j-std
include Atdgen.mk

SOURCES = \
  foo_t.mli foo_t.ml foo_j.mli foo_j.ml \
  bar_t.mli bar_t.ml bar_j.mli bar_j.ml \
  hello.ml
RESULT = hello
PACKS = atdgen
# "include OCamlMakefile" must come after defs for SOURCES, RESULT, PACKS, etc.
include OCamlMakefile

.PHONY: sources opt all
sources: $(SOURCES)
opt: sources
        $(MAKE) native-code
all: sources
        $(MAKE) byte-code

make alone builds a native code executable from source files foo.atd, bar.atd and hello.ml. make clean removes generated files. make all builds a bytecode executable.

In addition to native-code, byte-code and clean, OCamlMakefile provides a number of other targets and options which are documented in OCamlMakefile’s README.

Ocamlbuild

There is an atdgen plugin for ocamlbuild.

Dune (formerly jbuilder)

Dune currently needs atdgen build rules specified manually. Given an example.atd, this will usually look like:

(rule
 (targets example_j.ml
          example_j.mli)
 (deps    example.atd)
 (action  (run atdgen -j -j-std %{deps})))

(rule
 (targets example_t.ml
          example_t.mli)
 (deps    example.atd)
 (action  (run atdgen -t %{deps})))

You can refer to example_t.ml and example_j.ml as usual (by default, they will be automatically linked into the library being built in the same directory). You will need to write rules for each .atd file individually until Dune supports wildcard rules.

Note that any options atdgen supports can be included in the run atdgen section (-open, -deriving-conv, etc.).

Dealing with untypable JSON

Sometimes we have to deal with JSON data that cannot be described using type definitions. In such case, we can represent the data as its JSON abstract syntax tree (AST), which lets the user inspect it at runtime.

Let’s consider a list of JSON objects for which we don’t know the type definitions, but somehow some other system knows how to deal with such data. Here is such data:

[
  {
    "label": "flower",
    "value": {
      "petals": [12, 45, 83.5555],
      "water": "a340bcf02e"
    }
  },
  {
    "label": "flower",
    "value": {
      "petals": "undefined",
      "fold": null,
      "water": 0
    }
  },
  { "labels": ["fork", "scissors"],
    "value": [ 8, 8 ]
  }
]

Hopefully this means something for someone. We are going to assume that each object has a value field of an unknown type, and may have a field label or a field labels of type string:

(* File untypable.atd *)

type obj_list = obj list

type obj = {
  ?label: string option;
  ?labels: string list option;
  value: abstract  (* requires ATD >= 2.6.0 *)
}

Until ATD 2.5, abstract could not be used as freely and would not stand for raw JSON by default. One had to write a dedicated type definition as shown below:

(* File untypable.atd *)

(* deprecated since ATD 2.6 *)
type json <ocaml module="Yojson.Safe"> = abstract
  (* uses type Yojson.Safe.t,
     with the functions Yojson.Safe.write_json
     and Yojson.Safe.read_json *)

type obj_list = obj list

type obj = {
  ?label: string option;
  ?labels: string list option;
  value: json
}

It is possible to give a different name than json to the type of the JSON AST, but then the name of the type used in the original module must be provided in the annotation, i.e.:

(* deprecated since ATD 2.6 *)
type raw_json <ocaml module="Yojson.Safe" t="json"> = abstract
  (* uses type Yojson.Safe.t,
     with the functions Yojson.Safe.write_json
     and Yojson.Safe.read_json *)

type obj_list = obj list

type obj = {
  ?label: string option;
  ?labels: string list option;
  value: raw_json
}

Compile either example with:

$ atdgen -t untypable.atd
$ atdgen -j -j-std untypable.atd
$ ocamlfind ocamlc -a -o untypable.cma -package atdgen \
    untypable_t.mli untypable_t.ml untypable_j.mli untypable_j.ml

Test the example with your favorite OCaml toplevel (ocaml or utop):

# #use "topfind";;
# #require "atdgen";;
# #load "untypable.cma";;
# Atdgen_runtime.Util.Json.from_channel Untypable_j.read_obj_list stdin;;
[
  {
    "label": "flower",
    "value": {
      "petals": [12, 45, 83.5555],
      "water": "a340bcf02e"
    }
  },
  {
    "label": "flower",
    "value": {
      "petals": "undefined",
      "fold": null,
      "water": 0
    }
  },
  { "labels": ["fork", "scissors"],
    "value": [ 8, 8 ]
  }
]
- : Untypable_t.obj_list =
[{Untypable_t.label = Some "flower"; labels = None;
  value =
  `Assoc
    [("petals", `List [`Int 12; `Int 45; `Float 83.5555]);
      ("water", `String "a340bcf02e")]};
{Untypable_t.label = Some "flower"; labels = None;
  value =
  `Assoc [("petals", `String "undefined");
          ("fold", `Null);
          ("water", `Int 0)]};
{Untypable_t.label = None; labels = Some ["fork"; "scissors"];
  value = `List [`Int 8; `Int 8]}]

Atdgen reference

Description

Atdgen is a command-line program that takes as input type definitions in the ATD syntax and produces OCaml code suitable for data serialization and deserialization.

Two data formats are currently supported, these are JSON and biniou, a binary format with extensibility properties similar to JSON. Atdgen-json and Atdgen-biniou will refer to Atdgen used in one context or the other.

Atdgen was designed with efficiency and durability in mind. Software authors are encouraged to use Atdgen directly and to write tools that may reuse part of Atdgen’s source code.

Atdgen uses the following packages that were developed in conjunction with Atdgen:

  • atd: parser for the syntax of type definitions
  • biniou: parser and printer for biniou, a binary extensible data format
  • `yojson <https://github.com/ocaml-community/yojson>`__: parser and printer for JSON, a widespread text-based data format

Command-line usage

Command-line help

Call atdgen -help for the full list of available options.

Atdgen-json example

$ atdgen -t example.atd
$ atdgen -j -j-std example.atd

Input file example.atd:

type profile = {
  id : string;
  email : string;
  ~email_validated : bool;
  name : string;
  ?real_name : string option;
  ~about_me : string list;
  ?gender : gender option;
  ?date_of_birth : date option;
}

type gender = [ Female | Male ]

type date = {
  year : int;
  month : int;
  day : int;
}

is used to produce files example_t.mli, example_t.ml, example_j.mli and example_j.ml. This is example_j.mli:

(* Auto-generated from "example.atd" *)


type gender = Example_t.gender

type date = Example_t.date = { year: int; month: int; day: int }

type profile = Example_t.profile = {
  id: string;
  email: string;
  email_validated: bool;
  name: string;
  real_name: string option;
  about_me: string list;
  gender: gender option;
  date_of_birth: date option
}

val write_gender :
  Bi_outbuf.t -> gender -> unit
  (** Output a JSON value of type {!gender}. *)

val string_of_gender :
  ?len:int -> gender -> string
  (** Serialize a value of type {!gender}
      into a JSON string.
      @param len specifies the initial length
                 of the buffer used internally.
                 Default: 1024. *)

val read_gender :
  Yojson.Safe.lexer_state -> Lexing.lexbuf -> gender
  (** Input JSON data of type {!gender}. *)

val gender_of_string :
  string -> gender
  (** Deserialize JSON data of type {!gender}. *)

val write_date :
  Bi_outbuf.t -> date -> unit
  (** Output a JSON value of type {!date}. *)

val string_of_date :
  ?len:int -> date -> string
  (** Serialize a value of type {!date}
      into a JSON string.
      @param len specifies the initial length
                 of the buffer used internally.
                 Default: 1024. *)

val read_date :
  Yojson.Safe.lexer_state -> Lexing.lexbuf -> date
  (** Input JSON data of type {!date}. *)

val date_of_string :
  string -> date
  (** Deserialize JSON data of type {!date}. *)

val write_profile :
  Bi_outbuf.t -> profile -> unit
  (** Output a JSON value of type {!profile}. *)

val string_of_profile :
  ?len:int -> profile -> string
  (** Serialize a value of type {!profile}
      into a JSON string.
      @param len specifies the initial length
                 of the buffer used internally.
                 Default: 1024. *)

val read_profile :
  Yojson.Safe.lexer_state -> Lexing.lexbuf -> profile
  (** Input JSON data of type {!profile}. *)

val profile_of_string :
  string -> profile
  (** Deserialize JSON data of type {!profile}. *)

Module Example_t (files example_t.mli and example_t.ml) contains all OCaml type definitions that can be used independently from Biniou or JSON.

For convenience, these definitions are also made available from the Example_j module whose interface is shown above. Any type name, record field name or variant constructor can be referred to using either module. For example, the OCaml expressions ((x : Example_t.date) : Example_j.date) and x.Example_t.year = x.Example_j.year are both valid.

Atdgen-biniou example

$ atdgen -t example.atd
$ atdgen -b example.atd

Input file example.atd:

type profile = {
  id : string;
  email : string;
  ~email_validated : bool;
  name : string;
  ?real_name : string option;
  ~about_me : string list;
  ?gender : gender option;
  ?date_of_birth : date option;
}

type gender = [ Female | Male ]

type date = {
  year : int;
  month : int;
  day : int;
}

is used to produce files example_t.mli, example_t.ml, example_b.mli and example_b.ml.

This is example_b.mli:

(* Auto-generated from "example.atd" *)


type gender = Example_t.gender

type date = Example_t.date = { year: int; month: int; day: int }

type profile = Example_t.profile = {
  id: string;
  email: string;
  email_validated: bool;
  name: string;
  real_name: string option;
  about_me: string list;
  gender: gender option;
  date_of_birth: date option
}

(* Writers for type gender *)

val gender_tag : Bi_io.node_tag
  (** Tag used by the writers for type {!gender}.
      Readers may support more than just this tag. *)

val write_untagged_gender :
  Bi_outbuf.t -> gender -> unit
  (** Output an untagged biniou value of type {!gender}. *)

val write_gender :
  Bi_outbuf.t -> gender -> unit
  (** Output a biniou value of type {!gender}. *)

val string_of_gender :
  ?len:int -> gender -> string
  (** Serialize a value of type {!gender} into
      a biniou string. *)

(* Readers for type gender *)

val get_gender_reader :
  Bi_io.node_tag -> (Bi_inbuf.t -> gender)
  (** Return a function that reads an untagged
      biniou value of type {!gender}. *)

val read_gender :
  Bi_inbuf.t -> gender
  (** Input a tagged biniou value of type {!gender}. *)

val gender_of_string :
  ?pos:int -> string -> gender
  (** Deserialize a biniou value of type {!gender}.
      @param pos specifies the position where
                 reading starts. Default: 0. *)

(* Writers for type date *)

val date_tag : Bi_io.node_tag
  (** Tag used by the writers for type {!date}.
      Readers may support more than just this tag. *)

val write_untagged_date :
  Bi_outbuf.t -> date -> unit
  (** Output an untagged biniou value of type {!date}. *)

val write_date :
  Bi_outbuf.t -> date -> unit
  (** Output a biniou value of type {!date}. *)

val string_of_date :
  ?len:int -> date -> string
  (** Serialize a value of type {!date} into
      a biniou string. *)

(* Readers for type date *)

val get_date_reader :
  Bi_io.node_tag -> (Bi_inbuf.t -> date)
  (** Return a function that reads an untagged
      biniou value of type {!date}. *)

val read_date :
  Bi_inbuf.t -> date
  (** Input a tagged biniou value of type {!date}. *)

val date_of_string :
  ?pos:int -> string -> date
  (** Deserialize a biniou value of type {!date}.
      @param pos specifies the position where
                 reading starts. Default: 0. *)

(* Writers for type profile *)

val profile_tag : Bi_io.node_tag
  (** Tag used by the writers for type {!profile}.
      Readers may support more than just this tag. *)

val write_untagged_profile :
  Bi_outbuf.t -> profile -> unit
  (** Output an untagged biniou value of type {!profile}. *)

val write_profile :
  Bi_outbuf.t -> profile -> unit
  (** Output a biniou value of type {!profile}. *)

val string_of_profile :
  ?len:int -> profile -> string
  (** Serialize a value of type {!profile} into
      a biniou string. *)

(* Readers for type profile *)

val get_profile_reader :
  Bi_io.node_tag -> (Bi_inbuf.t -> profile)
  (** Return a function that reads an untagged
      biniou value of type {!profile}. *)

val read_profile :
  Bi_inbuf.t -> profile
  (** Input a tagged biniou value of type {!profile}. *)

val profile_of_string :
  ?pos:int -> string -> profile
  (** Deserialize a biniou value of type {!profile}.
      @param pos specifies the position where
                 reading starts. Default: 0. *)

Module Example_t (files example_t.mli and example_t.ml) contains all OCaml type definitions that can be used independently from Biniou or JSON.

For convenience, these definitions are also made available from the Example_b module whose interface is shown above. Any type name, record field name or variant constructor can be referred to using either module. For example, the OCaml expressions ((x : Example_t.date) : Example_b.date) and x.Example_t.year = x.Example_b.year are both valid.

Validator example

$ atdgen -t example.atd
$ atdgen -v example.atd

Input file example.atd:

type month = int <ocaml valid="fun x -> x >= 1 && x <= 12">
type day = int <ocaml valid="fun x -> x >= 1 && x <= 31">

type date = {
  year : int;
  month : month;
  day : day;
}
  <ocaml validator="Date_util.validate_date">

is used to produce files example_t.mli, example_t.ml, example_v.mli and example_v.ml. This is example_v.ml, showing how the user-specified validators are used:

(* Auto-generated from "example.atd" *)


type gender = Example_t.gender

type date = Example_t.date = { year: int; month: int; day: int }

type profile = Example_t.profile = {
  id: string;
  email: string;
  email_validated: bool;
  name: string;
  real_name: string option;
  about_me: string list;
  gender: gender option;
  date_of_birth: date option
}

val validate_gender :
  Atdgen_runtime.Util.Validation.path -> gender -> Atdgen_runtime.Util.Validation.error option
  (** Validate a value of type {!gender}. *)

val create_date :
  year: int ->
  month: int ->
  day: int ->
  unit -> date
  (** Create a record of type {!date}. *)

val validate_date :
  Atdgen_runtime.Util.Validation.path -> date -> Atdgen_runtime.Util.Validation.error option
  (** Validate a value of type {!date}. *)

val create_profile :
  id: string ->
  email: string ->
  ?email_validated: bool ->
  name: string ->
  ?real_name: string ->
  ?about_me: string list ->
  ?gender: gender ->
  ?date_of_birth: date ->
  unit -> profile
  (** Create a record of type {!profile}. *)

val validate_profile :
  Atdgen_runtime.Util.Validation.path -> profile -> Atdgen_runtime.Util.Validation.error option
  (** Validate a value of type {!profile}. *)

Default type mapping

The following table summarizes the default mapping between ATD types and OCaml, biniou and JSON data types. For each language more representations are available and are detailed in the next section of this manual.

ATD OCaml JSON Biniou
unit unit null unit
bool bool boolean bool
int int -?(0|[1-9][0-9]*) svint
float float number float64
string string string string
'a option 'a option "None" or ["Some", ...] numeric variants (tag 0)
'a nullable 'a option null or representation of 'a numeric variants (tag 0)
'a list 'a list array array
'a shared no wrapping not implemented no longer supported
'a wrap defined by annotation, converted from 'a representation of 'a representation of 'a
variants polymorphic variants variants regular variants
record record object record
('a * 'b) ('a * 'b) array tuple
('a) 'a array tuple

Notes:

  • Null JSON fields by default are treated as if the field was missing. They can be made meaningful with the keep_nulls flag.
  • JSON nulls are used to represent the unit value and is useful for instanciating parametrized types with “nothing”.
  • OCaml floats are written to JSON numbers with either a decimal point or an exponent such that they are distinguishable from ints, even though the JSON standard does not require a distinction between the two.
  • The optional values of record fields denoted in ATD by a question mark are unwrapped or omitted in both biniou and JSON.
  • JSON option values and JSON variants are represented in standard JSON (atdgen -j -j-std) by a single string e.g. "None" or a pair in which the first element is the name (constructor) e.g. ["Some", 1234]. Yojson also provides a specific syntax for variants using edgy brackets: <"None">, <"Some": 1234>.
  • Biniou field names and variant names other than the option types use the hash of the ATD field or variant name and cannot currently be overridden by annotations.
  • JSON tuples in standard JSON (atdgen -j -j-std) use the array notation e.g. ["ABC", 123]. Yojson also provides a specific syntax for tuples using parentheses, e.g. ("ABC", 123).
  • Types defined as abstract are defined in another module.

ATD Annotations

Section json

Field keep_nulls

Position: after record

Values: none, true or false

Semantics: this flag, if present or set to true, indicates that fields whose JSON value is null should not be treated as if they were missing. In this case, null is parsed as a normal value, possibly of a nullable type.

Example: patch semantics

(* Type of the objects stored in our database *)
type t = {
  ?x : int option;
  ?y : int option;
  ?z : int option;
}
(* Type of the requests to modify some of the fields of an object. *)
type t_patch = {
  ?x : int nullable option; (* OCaml type: int option option *)
  ?y : int nullable option;
  ?z : int nullable option;
} <ocaml field_prefix="patch_"> <json keep_nulls>

Let’s consider the following json patch that means “set x to 1, clear y and keep z as it is”:

{
  "x": 1,
  "y": null
}

It will be parsed by the generated function t_patch_of_string into the following OCaml value:

{
  patch_x = Some (Some 1);
  patch_y = Some None;
  patch_z = None;
}

Then presumably some code would be written to apply the patch to an object of type t. Such code is not generated by atdgen at this time.

Available: from atd 1.12

Field name

Position: after field name or variant name

Values: any string making a valid JSON string value

Semantics: specifies an alternate object field name or variant name to be used by the JSON representation.

Example:

type color = [
    Black <json name="black">
  | White <json name="white">
  | Grey <json name="grey">
]

type profile = {
  id <json name="ID"> : int;
  username : string;
  background_color : color;
}

A valid JSON object of the profile type above is:

{
  "ID": 12345678,
  "username": "kimforever",
  "background_color": "black"
}
Field repr
Association lists

Position: after (string * _) list type

Values: object

Semantics: uses JSON’s object notation to represent association lists.

Example:

type counts = (string * int) list <json repr="object">

A valid JSON object of the counts type above is:

{
  "bob": 3,
  "john": 1408,
  "mary": 450987,
  "peter": 93087
}

Without the annotation <json repr="object">, the data above would be represented as:

[
  [ "bob", 3 ],
  [ "john", 1408 ],
  [ "mary", 450987 ],
  [ "peter", 93087 ]
]
Floats

Position: after float type

Values: int

Semantics: specifies a float value that must be rounded to the nearest integer and represented in JSON without a decimal point nor an exponent.

Example:

type unixtime = float <json repr="int">
Field tag_field

Superseded by <json adapter.ocaml="...">. Available since atdgen 1.5.0 and yojson 1.2.0 until atdgen 1.13.

This feature makes it possible to read JSON objects representing variants that use one field for the tag and another field for the untagged value of the specific type associated with that tag.

Position: on a record field name, for a field holding a variant type.

Value: name of another JSON field which holds the string representing the constructor for the variant.

Semantics: The type definition

type t = {
  value <json tag_field="kind">: [ A | B <json name="b"> of int ];
}

covers JSON objects that have an extra field kind which holds either "A" or "b". Valid JSON values of type t include { "kind": "A" } and { "kind": "b", "value": 123 }.

Field untyped

Superseded by <json open_enum> and <json adapter.ocaml="...">. Available since atdgen 1.10.0 and atd 1.2.0 until atdgen 1.13.

This flag enables parsing of arbitrary variants without prior knowledge of their type. It is useful for constructing flexible parsers for extensible serializations. json untyped is compatible with regular variants, json tag_field variants, default values, and implicit tag_field constructors.

Position: on a variant constructor with argument type string * json option (at most one per variant type)

Value: none, true or false

Semantics: The type definition

type v = [
  | A
  | B <json name="b"> of int
  | Unknown <json untyped> of (string * json option)
]

will parse and print "A", ["b", 0], "foo", and ["bar", [null]] in a regular variant context. In the tag_field type t context in the previous section, v will parse and print { "kind": "foo" } and { "kind": "bar", "value": [null] } as well as the examples previously given.

Field open_enum

Where an enum (finite set of strings) is expected, this flag allows unexpected strings to be kept under a catch-all constructor rather than producing an error.

Position: on a variant type comprising exactly one constructor with an argument. The type of that argument must be string. All other constructors must have no arguments.

Value: none

For example:

type language = [
  | English
  | Chinese
  | Other of string
] <json open_enum>

maps the json string "Chinese" to the OCaml value `Chinese and maps "French" to `Other "French".

Available since atdgen 2.0.

Field adapter.ocaml

Json adapters are a mechanism for rearranging json data on-the-fly, so as to make them compatible with ATD. The programmer must provide an OCaml module that provides converters between the original json representation and the ATD-compatible representation. The signature of the user-provided module must be equal to Atdgen_runtime.Json_adapter.S, which is:

sig
  (** Convert from original json to ATD-compatible json *)
  val normalize : Yojson.Safe.t -> Yojson.Safe.t

  (** Convert from ATD-compatible json to original json *)
  val restore : Yojson.Safe.t -> Yojson.Safe.t
end

The type Yojson.Safe.t is the type of parsed JSON as provided by the yojson library.

Position: on a variant type or on a record type.

Value: an OCaml module identifier. Note that Atdgen_runtime.Json_adapter provides a few modules and functors that are ready to use. Users are however encouraged to write their own to suit their needs.

Sample ATD definitions:

type document = [
  | Image of image
  | Text of text
] <json adapter.ocaml="Atdgen_runtime.Json_adapter.Type_field">

type image = {
  url: string;
}

type text = {
  title: string;
  body: string;
}

ATD-compliant json values:

  • ["Image", {"url": "https://example.com/ocean123.jpg"}]
  • ["Text", {"title": "Cheeses Around the World", "body": "..."}]

Corresponding json values given by some API:

  • {"type": "Image", "url": "https://example.com/ocean123.jpg"}
  • {"type": "Text", "title": "Cheeses Around the World", "body": "..."}

The json adapter Type_field that ships with the atdgen runtime takes care of converting between these two forms. For information on how to write your own adapter, please consult the documentation for the yojson library.

Fields adapter.to_ocaml and adapter.from_ocaml

This is an alternative form of specifying adapter.ocaml. It permits to specify arbitrary code and doesn’t require the “adapter” module to be defined in advance.

For example, the above usage of adapter.ocaml can be rewritten as following:

type document = [
  | Image of image
  | Text of text
]
<json
  adapter.to_ocaml="Atdgen_runtime.Json_adapter.normalize_type_field \"type\""
  adapter.from_ocaml="Atdgen_runtime.Json_adapter.restore_type_field \"type\""
>

type image = {
  url: string;
}

type text = {
  title: string;
  body: string;
}

Section biniou

Field repr
Integers

Position: after int type

Values: svint (default), uvint, int8, int16, int32, int64

Semantics: specifies an alternate type for representing integers. The default type is svint. The other integers types provided by biniou are supported by Atdgen-biniou. They have to map to the corresponding OCaml types in accordance with the following table:

Biniou type Supported OCaml type OCaml value range
svint int min_intmax_int
uvint int 0 … max_int, min_int … -1
int8 char '\000'\255
int16 int 0 … 65535
int32 int32 Int32.min_intInt32.max_int
int64 int64 Int64.min_intInt64.max_int

In addition to the mapping above, if the OCaml type is int, any biniou integer type can be read into OCaml data regardless of the declared biniou type.

Example:

type t = {
  id : int
    <ocaml repr="int64">
    <biniou repr="int64">;
  data : string list;
}
Floating-point numbers

Position: after float type

Values: float64 (default), float32

Semantics: float32 allows for a shorter serialized representation of floats, using 4 bytes instead of 8, with reduced precision. OCaml floats always use 8 bytes, though.

Example:

type t = {
  lat : float <biniou repr="float32">;
  lon : float <biniou repr="float32">;
}
Arrays and tables

Position: applies to lists of records

Values: array (default), table

Semantics: table uses biniou’s table format instead of a regular array for serializing OCaml data into biniou. Both formats are supported for reading into OCaml data regardless of the annotation. The table format allows

Example:

type item = {
  id : int;
  data : string list;
}

type items = item list <biniou repr="table">

Section ocaml

Field attr

Position: on a type definition, i.e. on the left-handside just before the equal sign =

Semantics: specifies custom ppx attributes for the type definition. Overrides any default attributes set globally via the command line option -type-attr.

Values: the contents of a ppx annotation without the enclosing [@@ and ]

Example:

type foo <ocaml attr="deriving show,eq"> = int list

translates to

type foo = int list [@@deriving show,eq]
Field predef

Position: left-hand side of a type definition, after the type name

Values: none, true or false

Semantics: this flag indicates that the corresponding OCaml type definition must be omitted.

Example:

(* Some third-party OCaml code *)
type message = {
  from : string;
  subject : string;
  body : string;
}
(*
   Our own ATD file used for making message_of_string and
   string_of_message functions.
*)
type message <ocaml predef> = {
  from : string;
  subject : string;
  body : string;
}
Field mutable

Position: after a record field name

Values: none, true or false

Semantics: this flag indicates that the corresponding OCaml record field is mutable.

Example:

type counter = {
  total <ocaml mutable> : int;
  errors <ocaml mutable> : int;
}

translates to the following OCaml definition:

type counter = {
  mutable total : int;
  mutable errors : int;
}
Field default

Position: after a record field name marked with a \~{} symbol or at the beginning of a tuple field.

Values: any valid OCaml expression

Semantics: specifies an explicit default value for a field of an OCaml record or tuple, allowing that field to be omitted. Default strings must be escaped.

Example:

type color = [ Black | White | Rgb of (int * int * int) ]

type ford_t = {
  year : int;
  ~color <ocaml default="`Black"> : color;
  ~name <ocaml default="\"Ford Model T\""> : string;
}

type point = (int * int * <ocaml default="0"> : int)
Field from

Position: left-hand side of a type definition, after the type name

Values: OCaml module name without the _t, _b, _j or _v suffix. This can be also seen as the name of the original ATD file, without the .atd extension and capitalized like an OCaml module name.

Semantics: specifies the base name of the OCaml modules where the type and values coming with that type are defined.

It is useful for ATD types defined as abstract and for types annotated as predefined using the annotation <ocaml predef>. In both cases, the missing definitions must be provided by modules composed of the base name and the standard suffix assumed by Atdgen which is _t, _b, _j or _v.

Example: First input file part1.atd:

type point = { x : int; y : int }

Second input file part2.atd depending on the first one:

type point <ocaml from="Part1"> = abstract
type points = point list

To use a different type name than defined in the Part1 module, add a t field declaration to the annotation which refers to the original type name:

type point_xy <ocaml from="Part1" t="point"> = abstract
type points = point_xy list
Field module
Using a custom wrapper

Using the built-in wrap constructor, it is possible to add a layer of abstraction on top of the concrete structure used for serialization.

Position: after a wrap type constructor

Values: OCaml module name

A common use case is to parse strings used as unique identifiers and wrap the result into an abstract type. Our OCaml module Uid needs to provide a type t, and two functions wrap and unwrap as follows:

type t
val wrap : string -> t
val unwrap : t -> string

Given that Uid OCaml module, we can write the following ATD definition:

type uid = string wrap <ocaml module="Uid">

Other languages than OCaml using the same ATD type definitions may or may not add their own abstract layer. Without an annotation, the wrap construct has no effect on the value being wrapped, i.e. wrap and unwrap default to the identity function.

It is also possible to define t, wrap, and unwrap inline:

type uid = string wrap <ocaml t="Uid.t"
                              wrap="Uid.wrap"
                              unwrap="Uid.unwrap">

This can be useful for very simple validation:

type uid = string wrap
  <ocaml wrap="fun s ->
                 if String.length s <> 16 then
                   failwith \"Invalid user ID\";
                 s"
  >
Importing an external type definition

In most cases since Atdgen 1.2.0 module annotations are deprecated in favor of from annotations previously described.

Position: left-hand side of a type definition, after the type name

Values: OCaml module name

Semantics: specifies the OCaml module where the type and values coming with that type are defined. It is useful for ATD types defined as abstract and for types annotated as predefined using the annotation <ocaml predef>. In both cases, the missing definitions can be provided either by globally opening an OCaml module with an OCaml directive or by specifying locally the name of the module to use.

The latter approach is recommended because it allows to create type and value aliases in the OCaml module being generated. It results in a complete module signature regardless of the external nature of some items.

Example: Input file example.atd:

type document <ocaml module="Doc"> = abstract

type color <ocaml predef module="Color"> =
  [ Black | White ] <ocaml repr="classic">

type point <ocaml predef module="Point"> = {
  x : float;
  y : float;
}

gives the following OCaml type definitions (file example.mli):

type document = Doc.document

type color = Color.color =  Black | White

type point = Point.point = { x: float; y: float }

Now for instance Example.Black and Color.Black can be used interchangeably in other modules.

Field t
Using a custom wrapper

Specifies the OCaml type of an abstract wrap construct, possibly overriding the default M.t if M is the module where the wrap and unwrap functions are found.

Position: after a wrap type constructor

Values: OCaml type name

Example:

type uid = string wrap <ocaml module="Uid" t="Uid.uid">

is equivalent to:

type uid = string wrap <ocaml t="Uid.uid" wrap="Uid.wrap" unwrap="Uid.unwrap">
Importing an external type definition

Position: left-hand side of a type definition, after the type name. Must be used in conjunction with a module field.

Values: OCaml type name as found in an external module.

Semantics: This option allows to specify the name of an OCaml type defined in an external module.

It is useful when the type needs to be renamed because its original name is already in use or not enough informative. Typically we may want to give the name foo to a type originally defined in OCaml as Foo.t.

Example:

type foo <ocaml_biniou module="Foo" t="t"> = abstract
type bar <ocaml_biniou module="Bar" t="t"> = abstract
type t <ocaml_biniou module="Baz"> = abstract

allows local type names to be unique and gives the following OCaml type definitions:

type foo = Foo.t
type bar = Bar.t
type t = Baz.t
Fields wrap and unwrap

See “Using a custom wrapper” under section ocaml, fields module and t.

Field field_prefix

Position: record type expression

Values: any string making a valid prefix for OCaml record field names

Semantics: specifies a prefix to be prepended to each field of the OCaml definition of the record. Overridden by alternate field names defined on a per-field basis.

Example:

type point2 = {
  x : int;
  y : int;
} <ocaml field_prefix="p2_">

gives the following OCaml type definition:

type point2 = {
  p2_x : int;
  p2_y : int;
}
Field name

Position: after record field name or variant name

Values: any string making a valid OCaml record field name or variant name

Semantics: specifies an alternate record field name or variant names to be used in OCaml.

Example:

type color = [
    Black <ocaml name="Grey0">
  | White <ocaml name="Grey100">
  | Grey <ocaml name="Grey50">
]

type profile = {
  id <ocaml name="profile_id"> : int;
  username : string;
}

gives the following OCaml type definitions:

type color = [
    `Grey0
  | `Grey100
  | `Grey50
]

type profile = {
  profile_id : int;
  username : string;
}
Field repr
Integers

Position: after int type

Values: char, int32, int64, float

Semantics: specifies an alternate type for representing integers. The default type is int, but char, int32, int64 or float can be used instead.

The three types char, int32 and int64 are supported by both Atdgen-biniou and Atdgen-json but Atdgen-biniou currently requires that they map to the corresponding fixed-width types provided by the biniou format.

The type float is only supported in conjunction with JSON and is useful when an OCaml float is used to represent an integral value, such as a time in seconds returned by Unix.time(). When converted into JSON, floats are rounded to the nearest integer.

Example:

type t = {
  id : int
    <ocaml repr="int64">
    <biniou repr="int64">;
  data : string list;
}
Lists and arrays

Position: after a list type

Values: array

Semantics: maps to OCaml’s array type instead of list.

Example:

type t = {
  id : int;
  data : string list
    <ocaml repr="array">;
}
Sum types

Position: after a sum type (denoted by square brackets)

Values: classic

Semantics: maps to OCaml’s classic variants instead of polymorphic variants.

Example:

type fruit = [ Apple | Orange ] <ocaml repr="classic">

translates to the following OCaml type definition:

type fruit = Apple | Orange
Shared values (obsolete)

Position: after a shared type

This feature is obsolete and was last supported by atdgen 1.3.1.

Field valid

Since atdgen 1.6.0.

Position: after any type expression except type variables

Values: OCaml function that takes one argument of the given type and returns a bool

Semantics: atdgen -v produces for each type named t a function validate_t:

val validate_t : Atdgen_runtime.Util.Validation.path -> t -> Atdgen_runtime.Util.Validation.error option

Such a function returns None if and only if the value and all of its subnodes pass all the validators specified by annotations of the form <ocaml validator="..."> or <ocaml valid="..."> (at most one per node).

Example:

type positive = int <ocaml validator="fun x -> x > 0">

type point = {
  x : positive;
  y : positive;
  z : int;
}
  <ocaml valid="Point.validate">
  (* Some validating function from a user-defined module Point *)

The generated validate_point function calls the validator for the containing object first (Point.validate) and continues on its fields x then y until an error is returned.

match validate_point [] { x = 1; y = 0; z = 1 } with
| None -> ()
| Some e ->
    Printf.eprintf "Error: %s\n%!"
      (Atdgen_runtime.Util.Validation.string_of_error e)

The above code prints the following error message:

Error: Validation error; path = <root>.y

In order to customize the error message and print the faulty value, use validator instead of valid, as described next.

Field validator

This is a variant of the valid annotation that allows full control over the error message that gets generated in case of an error.

Position: after any type expression except type variables

Values: OCaml function that takes the path in current JSON structure and the object to validate, and returns an optional error.

Semantics: atdgen -v produces for each type named t a function validate_t:

val validate_t : Atdgen_runtime.Util.Validation.path -> t -> Atdgen_runtime.Util.Validation.error option

Such a function returns None if and only if the value and all of its subnodes pass all the validators specified by annotations of the form <ocaml validator="..."> or <ocaml valid="..."> (at most one per node).

Example:

type positive = int <ocaml validator="
  fun path x ->
    if x > 0 then None
    else
      Some (
        Atdgen_runtime.Util.Validation.error
          ~msg: (\"Not a positive integer: \" ^ string_of_int x)
          path
      )
">

type point = {
  x : positive;
  y : positive;
  z : int;
}
  <ocaml validator="Point.validate">
  (* Some validating function from a user-defined module Point *)

The following user code

match Toto_v.validate_point [] { x = 1; y = 0; z = 1 } with
| None -> ()
| Some e ->
    Printf.eprintf "Error: %s\n%!"
      (Atdgen_runtime.Util.Validation.string_of_error e)

results in printing:

Error: Validation error: Not a positive integer: 0; path = <root>.y

Section ocaml_biniou

Section ocaml_biniou takes precedence over section ocaml in Biniou mode (-b) for the following fields:

  • predef (see section ocaml, field predef)
  • module (see section ocaml, field module)
  • t (see section ocaml.t)

Section ocaml_json (obsolete)

Section ocaml_json takes precedence over section ocaml in JSON mode (-json or -j) for the following fields:

  • predef (see section ocaml, field predef)
  • module (see section ocaml, field module)
  • t (see section ocaml, field t)

Please note that atdgen -json is now deprecated in favor of atdgen -j (json) and atdgen -t (types). The latter is in charge of producing type definitions independently from JSON and will ignore <ocaml_json ...> annotations, making them almost useless. The equivalent <ocaml ...> annotations are almost always preferable.

Example:

This example shows how to parse a field into a generic tree of type Yojson.Safe.t rather than a value of a specialized OCaml type.

type dyn <ocaml_json module="Yojson.Safe" t="json"> = abstract

type t = { foo: int; bar: dyn }

translates to the following OCaml type definitions:

type dyn = Yojson.Safe.t

type t = { foo : int; bar : dyn }

Sample OCaml value of type t:

{
  foo = 12345;
  bar =
    `List [
      `Int 12;
      `String "abc";
      `Assoc [
        "x", `Float 3.14;
        "y", `Float 0.0;
        "color", `List [ `Float 0.3; `Float 0.0; `Float 1.0 ]
      ]
    ]
}

Corresponding JSON data as obtained with string_of_t:

{"foo":12345,"bar":[12,"abc",{"x":3.14,"y":0.0,"color":[0.3,0.0,1.0]}]}

Section doc

Unlike comments, doc annotations are meant to be propagated into the generated source code. This is useful for making generated interface files readable without having to consult the original ATD file.

Generated source code comments can comply to a standard format and take advantage of documentation generators such as javadoc or ocamldoc.

Field text

Position:

  • after the type name on the left-hand side of a type definition
  • after the type expression on the right hand of a type definition (but not after any type expression)
  • after record field names
  • after variant names

Values: UTF-8-encoded text using a minimalistic markup language

Semantics: The markup language is defined as follows:

  • Blank lines separate paragraphs.
  • {{ }} can be used to enclose inline verbatim text.
  • {{{ }}} can be used to enclose verbatim text where whitespace is preserved.
  • The backslash character is used to escape special character sequences. In regular paragraph mode the special sequences are \, {{ and {{{. In inline verbatim text, special sequences are \ and }}. In verbatim text, special sequences are \ and }}}.

Example: The following is an example demonstrating the use of doc annotations generated using:

$ atdgen -t ocamldoc_example.atd

Input file ocamldoc_example.atd:

<doc text="This is the title">

type point = {
  x <doc text="The first coordinate">: float;
  y <doc text="The second coordinate">: float;
}
  <doc text="
The type of a point. A value {{p}} can be created as follows:
{{{
let p = { x = 1.2; y = 5.0 }
}}}
">

type color = [
 | Black <doc text="Same as {{RGB (0,0,0)}}">
 | White <doc text="Same as {{RGB (255, 255, 255)}}">
 | RGB
     <doc text="Red, green, blue components">
     of (int * int * int)
]

translates using atdgen -t ocamldoc_example.atd into the following OCaml interface file ocamldoc_example_t.mli with ocamldoc-compliant comments:

(* Auto-generated from "ocamldoc_example.atd" *)


(** This is the title *)

(**
  The type of a point. A value [p] can be created as follows:

{v
let p = \{ x = 1.2; y = 5.0 \}
v}
*)
type point = {
  x: float (** The first coordinate *);
  y: float (** The second coordinate *)
}

type color = [
    `Black (** Same as [RGB (0,0,0)] *)
  | `White (** Same as [RGB (255, 255, 255)] *)
  | `RGB of (int * int * int) (** Red, green, blue components *)
]

Atdgen runtime library

A library named atdgen-runtime is installed by the standard installation process. Only a fraction of it is officially supported and documented.

Modules intended for all users are:

  • Util
  • Json_adapter

The other modules exported by the library are used directly by generated code. Tool developers may use them but we don’t guarantee strong compatibility across releases.