jsonobject is a python library for handling deeply nested JSON objects as well-schema'd python objects.

jsonobject is made by Dimagi, where we build, use, and contribute to OSS in our mission to reduce inequity in the world.

jsonobject is inspired by and largely API compatible with the Document/DocumentSchema portion of couchdbkit. Because jsonobject is not only simpler and standalone, but also faster, we also maintain a fork of couchdbkit, jsonobject-couchdbkit, that is backed by jsonobject and works seamlessly as a swap-in replacement for the main library.

It is used heavily in CommCare HQ (source), and the API is largely stable, but more advanced features may change in the future.

To install using pip, simply run

pip install jsonobject

The code below defines a simple user model, and its natural mapping to JSON.

from jsonobject import *

class User(JsonObject):
    username = StringProperty()
    name = StringProperty()
    active = BooleanProperty(default=False)
    date_joined = DateTimeProperty()
    tags = ListProperty(unicode)

Once it is defined, it can be used to wrap or produce deserialized JSON.

>>> user1 = User(
    name='John Doe',
    username='jdoe',
    date_joined=datetime.datetime.utcnow(),
    tags=['generic', 'anonymous']
)
>>> user1.to_json()
{
    'name': u'John Doe',
    'username': u'jdoe',
    'active': False,
    'date_joined': '2013-08-05T02:46:58Z',
    'tags': [u'generic', u'anonymous']
}

Notice that the datetime is converted to an ISO format string in JSON, but is a real datetime on the object:

>>> user1.date_joined
datetime.datetime(2013, 8, 5, 2, 46, 58)

A JsonObject subclass that has been defined as User above comes with a lot of built-in functionality. The basic operations are

1. Make a new object from deserialized JSON (e.g. the output of json.loads)
2. Construct a new object with given values
3. Modify an object
4. Dump to deserialized json (e.g. the input of json.dumps)

1 & 2 are accomplished with the constructor. There are two main ways to call the constructor:

User(
    name='John Doe',
    username='jdoe',
    date_joined=datetime.datetime.utcnow(),
    tags=['generic', 'anonymous']
)

as above (satisfies #2) and

User({
    'name': u'John Doe',
    'username': u'jdoe',
    'active': False,
    'date_joined': '2013-08-05T02:46:58Z',
    'tags': [u'generic', u'anonymous']
})

(satisfies #1). These two styles can also be mixed and matched:

User({
    'name': u'John Doe',
    'username': u'jdoe',
    'active': False,
    'tags': [u'generic', u'anonymous']
}, date_joined=datetime.datetime.utcnow())

Notice how datetimes are stored as strings in the deserialized JSON, but as datetime.datetimes in the nice python object—we will refer to these as the "json" representation and the "python" representation, or alternatively the "unwrapped" representation and the "wrapped" representation.

Gotcha. When calling the constructor, remember that the keyword argument style requires you to pass in the "python" representation (e.g. a datetime) while the json-wrapping style of passing in a dict requires you to give it in the "json" representation (e.g. a datetime-formatted string).

There are two main kinds of property types: scalar types (like string, bool, int, datetime, etc.) and container types (list, dict, set). They are dealt with separately below.

All scalar properties can take the value None in addition to the values particular to their type (strings, bools, etc). If set to the wrong type, properties raise a jsonobject.exceptions.BadValueError:

class Foo(jsonobject.JsonObject):
    b = jsonobject.BooleanProperty()

>>> Foo(b=0)
Traceback (most recent call last):
  [...]
jsonobject.exceptions.BadValueError: 0 not of type <type 'bool'>

Maps to a unicode. Usage:

class Foo(jsonobject.JsonObject):
    s = jsonobject.StringProperty()

If you set it to an ascii str it will implicitly convert to unicode:

>>> Foo(s='hi')  # converts to unicode
Foo(s=u'hi')

If you set it to a non-ascii str, it will fail with a UnicodeDecodeError:

>>> Foo(s='\xff')
Traceback (most recent call last):
  [...]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xff in position 0: ordinal not in range(128)

Maps to a bool.

Maps to an int or long.

Maps to a float.

Maps to a decimal.Decimal and stored as a JSON string. This type, unlike FloatProperty, stores the "human" representation of the digits. Usage:

class Foo(jsonobject.JsonObject):
    number = jsonobject.DecimalProperty()

If you set it to an int or float, it will implicitly convert to Decimal:

>>> Foo(number=1)
Foo(number=Decimal('1'))
>>> Foo(number=1.2)
Foo(number=Decimal('1.2'))

If you set it to a str or unicode, however, it raises an AssertionError:

>>> Foo(number='1.0')
Traceback (most recent call last):
  [...]
AssertionError

Todo: this should really raise a BadValueError.

If you pass in json in which the Decimal value is a str or unicode, but it is malformed, it throws the same errors as decimal.Decimal.

>>> Foo({'number': '1.0'})
Foo(number=Decimal('1.0'))
>>> Foo({'number': '1.0.0'})
Traceback (most recent call last):
  [...]
decimal.InvalidOperation: Invalid literal for Decimal: '1.0.0'

Maps to a datetime.date and stored as a JSON string of the format '%Y-%m-%d'. Usage:

class Foo(jsonobject.JsonObject):
    date = jsonobject.DateProperty()

Wrapping a badly formatted string raises a BadValueError:

>>> Foo({'date': 'foo'})
Traceback (most recent call last):
  [...]
jsonobject.exceptions.BadValueError: 'foo' is not a date-formatted string

Maps to a timezone-unaware datetime.datetime and stored as a JSON string of the format '%Y-%m-%dT%H:%M:%SZ'.

While it works perfectly with good inputs, it is extremely sloppy when it comes to dealing with inputs that don't match the exact specified format. Rather than matching stricty, it simply truncates the string to the first 19 characters and tries to parse that as '%Y-%m-%dT%H:%M:%S'. This ignores both microseconds and, even worse, the timezone. This is a holdover from couchdbkit.

In newer versions of jsonboject, you may optionally specify a DateTimeProperty as exact:

class Foo(jsonobject.JsonObject):
    date = jsonobject.DateTimeProperty(exact=True)

This provides a much cleaner conversion model that has the following properties:

1. It preserves microseconds
2. The incoming JSON representation must match '%Y-%m-%dT%H:%M:%S.%fZ' exactly. (This is similar to the default output, except for the mandatory 6 decimal places, i.e. milliseconds.)
3. Representations that don't match exactly will be rejected with a BadValueError.

Recommendation: If you are not locked into couchdbkit's earlier bad behavior, you should always use the exact=True flag on DateTimePropertys and TimePropertys (below).

Maps to a datetime.time, stored as a JSON string of the format '%H:%M:%S'.

To get access to milliseconds and strict behavior, use the exact=True setting which strictly accepts the format '%H:%M:%S.%f'. This is always recommended. For more information please read the previous section on DateTimeProperty.

Container types generally take a first argument, item_type, specifying the type of the contained objects.

Maps to a dict that has a schema specified by item_type, which must be itself a subclass of JsonObject. Usage:

class Bar(jsonobject.JsonObject):
    name = jsonobject.StringProperty()


class Foo(jsonobject.JsonObject):
    bar = jsonobject.ObjectProperty(Bar)

If not specified, it will be set to a new object with default values:

>>> Foo()
Foo(bar=Bar(name=None))

If you want it set to None you must do so explicitly.

Maps to a list with items of type item_type, which can be any of the following:

• an instance of a property class. This is the most flexible option, and all validation (required, etc.) will be done as as specified by the property instance.
• a property class, which will be instantiated with required=True
• one of their corresponding python types (i.e. int for IntegerProperty, etc.)
• a JsonObject subclass

Note that a property class (as well as the related python type syntax) will be instantiated with required=True, so ListProperty(IntegerProperty) and ListProperty(int) do not allow None, and ListProperty(IntegerProperty()) does allow None.

The serialization behavior of whatever item type is given is recursively applied to each member of the list.

If not specified, it will be set to an empty list.

Maps to a set and stored as a list (with only unique elements). Otherwise its behavior is very much like ListProperty's.

Maps to a dict with string keys and values specified by item_type. Otherwise its behavior is very much like ListProperty's.

If not specified, it will be set to an empty dict.

This flexibly wraps any valid JSON, including all scalar and container types, dynamically detecting the value's type and treating it with the corresponding property.

Certain parameters may be passed in to any property.

For example, required is one such parameter in the example below:

class User(JsonObject):
    username = StringProperty(required=True)

Here is a complete list of properties:

• default

  Specifies a default value for the property

• name

  The name of the property within the JSON representation*. This defaults to the name of the python property, but you can override it if you wish. This can be useful, for example, to get around conflicting with python keywords:

  >>> class Route(JsonObject):
  ...     from_ = StringProperty(name='from')
  ...     to = StringProperty()  # name='to' by default
  >>> Route(from_='me', to='you').to_json()
  {'from': u'me', 'to': u'you'}

  Notice how an underscore is present in the python property name ('from_'), but absent in the JSON property name ('from').

  \*If you're wondering how `StringProperty`'s `name` parameter could possibly default to `to` in the example above, when it doesn't have access to the `Route` class's properties at init time, you're completely right. The behavior described is implemented in `JsonObject`'s `__metaclass__`, which *does* have access to the `Route` class's properties.

• choices

  A list of allowed values for the property. (Unless otherwise specified, None is also an allowed value.)

• required

  Defaults to False. For scalar properties requires means that the value None may not be used. For container properties it means they may not be empty or take the value None.

• exclude_if_none

  Defaults to False. When set to true, this property will be excluded from the JSON output when its value is falsey. (Note that currently this is at odds with the parameter's name, since the condition is that it is falsey, not that it is None).

• validators

  A single validator function or list of validator functions. Each validator function should raise an exception on invalid input and do nothing otherwise.

• verbose_name

  This property does nothing and was added to match couchdbkit's API.

In order to do a direct comparison with couchdbkit, the test suite includes a large sample schema originally written with couchdbkit. It is easy to swap in jsonobject for couchdbkit and run the tests with each. Here are the results:

$ python -m unittest test.test_couchdbkit
....
----------------------------------------------------------------------
Ran 4 tests in 1.403s

OK
$ python -m unittest test.test_couchdbkit
....
----------------------------------------------------------------------
Ran 4 tests in 0.153s

OK

jsonobject versions follow semantic versioning. Version information can be found in CHANGES.md.

Information for developers and maintainers, such as how to run tests and release new versions, can be found in LIFECYCLE.md.