Python Support - atdpy

This documentation is incomplete. Your help would be appreciated! In particular, some how-to guides would be great.


Hello World

Install atdpy with opam:

opam install atdpy

Create a file hello.atd containing this:

type message = {
  subject: string;
  body: string;

Call atdpy to produce

$ atdpy hello.atd

There’s now a file that contains a class looking like this:


class Message:
  """Original type: message = { ... }"""

  subject: str
  body: str

  def from_json(cls, x: Any) -> 'Message':

  def to_json(self) -> Any:

  def from_json_string(cls, x: str) -> 'Message':

  def to_json_string(self, **kw: Any) -> str:

Let’s write a Python program that uses this code:

import hello

msg = hello.Message("Hello", "Dear friend, I hope you are well.")

Running it will print the JSON message:

$ python3
{"subject": "Hello", "body": "Dear friend, I hope you are well."}

Such JSON data can be parsed. Let’s write a program that consumes JSON data from standard input:

import hello, sys, json

data = json.load(sys.stdin)
msg = hello.Message.from_json(data)
print(f"subject: {msg.subject}")


$ echo '{"subject": "big news", "body": ""}' | python3
subject: big news

It works! But what happens if the JSON data lacks a "subject" field? Let’s see:

$ echo '{"subj": "big news", "body": ""}' | python3
Traceback (most recent call last):
ValueError: missing field 'subject' in JSON object of type 'Message'

And what if our program also thought that the correct field name was subj rather than subject? Here’s which tries to access a subj field:

import hello, sys, json

data = json.load(sys.stdin)
msg = hello.Message.from_json(data)
print(f"subject: {msg.subj}")

Let’s run the program through mypy:

$ mypy error: "Message" has no attribute "subj"
Found 1 error in 1 file (checked 1 source file)

Mypy detected that our program makes incorrect assumptions about the message format without running it. On the correct program, we get a reassuring message:

$ mypy
Success: no issues found in 1 source file

ATD Records, JSON objects, Python classes

An ATD file contains types that describe the structure of JSON data. JSON objects map to Python classes and objects. They’re called records in the ATD language. Let’s define a simple record type in the file hello_plus.atd:

type message = {
  subject: string;
  ~body: string;

Note the ~ in front of the body field. It means that this field has a default value. Whenever the JSON field is missing from a JSON object, a default value is assumed. The implicit default value for a string is "".

Let’s add a signature field whose default value isn’t the empty string:

type message = {
  subject: string;
  ~body: string;
  ~signature <python default="'anonymous'">: string;

Finally, we’ll add an optional url field that doesn’t take a default value at all:

type message = {
  subject: string;
  ~body: string;
  ~signature <python default="'anonymous'">: string;
  ?url: string option;

Let’s generate the Python code for this.

$ atdpy hello_plus.atd

Let’s update our reader program to this:

import hello_plus, sys, json

data = json.load(sys.stdin)
msg = hello_plus.Message.from_json(data)

We can test it, showing us the final value of each field:

$ echo '{"subject":"hi"}' | python3
Message(subject='hi', body='', signature='anonymous', url=None)

How-to guides

Defining default field values


Renaming field names


Deep dives



Type mapping

ATD type

Python type

JSON example
















int list


[1, 2, 3]

(int * int)

Tuple[int, int]

[-1, 1]

int nullable


42 or null

int option


["Some", 42] or "None"




record type


{"id": 17}

[A | B of int]

Union[A, B]

"A" or ["B", 5]



Supported ATD annotations

Default field values

Record fields following a ~ assume a default value. The default value can be implicit as mandated by the ATD language specification (false for bool, zero for int, etc.) or it can be a user-provided value.

A user-provided default uses an annotation of the form <python default="VALUE"> where VALUE evaluates to a Python expression e.g.

type foo = {
  ~answer <python default="42">: int;

Default values are always honored when reading JSON data from Python.

Field and constructor renaming

Alternate JSON object field names can be specified using an annotation of the form <json name="NAME"> where NAME is the desired field name to be used in the JSON representation. For example, the following specifies the JSON name of the id field is ID:

type foo = {
  id <json name="ID">: string

Similarly, the constructor names of sum types can also be given alternate names in the JSON representation. Here’s an example:

type bar = [
| Alpha <json name="alpha">
| Beta <json name="beta"> of int

Note that field names and constructor names in the generated Python code are assigned automatically so as to avoid conflicts with Python keywords or reserved identifiers.

Alternate representations for association lists

List of pairs can be represented by JSON objects or by Python dictionaries if the correct annotations are provided:

  • (string * bar) list <json repr="object"> will use JSON objects to represent a list of pairs of Python type List[str, Bar]. Using the annotation <json repr="array"> is equivalent to the default.

  • (foo * bar) list <python repr="dict"> will use a Python dictionary of type Dict[Foo, Bar] to represent the association list. Using the annotation <python repr="list"> is equivalent to the default.

Additional imports

At the beginning of the ATD file, placing annotations like this one allow inserting arbitrary Python code or comments:

<python text="import deco">

This is the recommended mechanism for inserting imports. In contrast, it should be used only as last resort for inserting functions or classes.

In the future, atdpy may generate more than one kind of files. An annotation of the form <python text="..."> will insert that text into all the generated files. In order to insert code only in the .py file that handles JSON, it is recommended to use a more specific annotation of the form <python json_py.text="...">:

<python json_py.text="import deco">

Custom class decorators

Extra class decorators can be specified in addition to @dataclass. The following ATD definition will add 3 decorators:

type thing <python decorator="deco.deco1"
                   decorator="dataclass(order=True)"> = {
  foo: int;
  bar: string;

The generated Python class will start like this:

class Thing:

If extra class decorators are specifed on a sum type, the python classes generated for the constructors of the sum type will also have the extra class decorators.