Limone is a library for generating content types from a Colander schema. A content type is, in this context, a class that implements the structure and constraints specified by the schema. This allows a developer to easily generate model objects which enforce the constraints of the schema, performing validation during initialization and attribute assignment. Objects are serializable and deserializable via Colander's serialization. Because types are generated at runtime, Limone also suggests the development of applications where the structure of the objects used to store your application's data can be derived from configuration or user input.
Content types can be generated declaratively from schema definitions using decorators. Let's take a look at the following Colander schema as an example, taken from the Colander documentation:
import colander
class Friend(colander.TupleSchema):
rank = colander.SchemaNode(colander.Int(),
validator=colander.Range(0, 9999))
name = colander.SchemaNode(colander.String())
class Phone(colander.MappingSchema):
location = colander.SchemaNode(colander.String(),
validator=colander.OneOf(['home', 'work']))
number = colander.SchemaNode(colander.String())
class Friends(colander.SequenceSchema):
friend = Friend()
class Phones(colander.SequenceSchema):
phone = Phone()
class Person(colander.MappingSchema):
name = colander.SchemaNode(colander.String())
age = colander.SchemaNode(colander.Int(),
validator=colander.Range(0, 200))
friends = Friends()
phones = Phones()
The simplest way to generate a Person content type is to add the limone.content_schema decorator:
import colander
import limone
... <elided for brevity>
@limone.content_schema
class Person(colander.MappingSchema):
name = etc...
Instances of Person can then be created in the usual way:
jack = Person(
name='Jack',
age=52,
friends=[
(1, 'Fred'),
(2, 'Barney')
],
phones=[
{'location': 'home',
'number': '555-1212'},
])
Assigning a value to an attribute triggers Colander schema validation. For example, when a value of 300 is assigned to age:
jack.age = 300
A colander.Invalid exception is raised:
colander.Invalid: {'age': u'300 is greater than maximum value 200'}
When instantiating a content type, values for all required attributes must be provided:
fred = Person()
Raises:
colander.Invalid: {'age': u'Required', 'name': u'Required'}
In some cases you might want to define a class separately from its schema. For this you can use the limone.content_type decorator. Let's say that instead of turning the Person schema into a content type directly, we have an HRPerson class which extends a hypothetical HRRecord class that we want to use for our content type:
@limone.content_type(Person)
class HRPerson(HRRecord):
pass
fred = HRPerson(name='Fred', age=54)
NOTE The decorated class must have a no-arg constructor.
The above examples use a declarative style for creating content types. Using the make_content_type function, we can also generate new content types imperatively. Assuming HRPerson has been defined as a class, the example above could have been written:
content_type = limone.make_content_type(Person, 'Person', bases=(HRPerson,)) fred = content_type(name='Joe', age=54)
The full signature for the make_content_type function is:
make_content_type(schema, name, module=None, bases=(object,))
- schema is the Colander schema to use to generate the class.
- The value of the name parameter will be assigned to the __name__ attribute of the generated class. If added to a registry, the name will also be used as the key for looking up the content type later. (See Using the Limone Registry.)
- module, if specified, will be used to set the __module__ attribute of the generated class.
- bases can be specified as a tuple of types that are the superclasses for the generated classes. NOTE The first base class must have a no-arg constructor.
Instances limone.Registry can be used to keep track of available content types. An instance of limone.Registry is required to make content types available via an import hook. (See Using the Import Hook.)
Content types are added to the registry using the register_content_type method:
registry = limone.Registry() registry.regsister_content_type(Person)
The get_content_type method is used to retrieve a content type by name:
content_type = registry.get_content_type('Person')
joe = content_type(name='Joe', age=54)
A tuple of all of the registered content types can be retrieved using the get_content_types method:
for content_type in registry.get_content_types():
print content_type.__name__, content_type
Prints:
Person <class 'Person'>
A registry instance can also find content types by scanning a package looking for content types to add to the registry. This is possible if you have used either the content_type or content_schema decorator somewhere in your package. The scan method is used to search for content types defined with those decorators and add them to the registry:
import limone import myapp.models registry = limone.Registry() registry.scan(myapp.models)
In the above two declarative examples, because types were being generated at module scope, they can be imported using the standard Python import mechanism. For content types that are generated imperatively, however, there may not be a global name that can be used to import the type. This would definitely be the case in an application that generated content types from schemas that were generated at runtime through configuration or user input. This can lead to difficulties--pickling, for example, does not work if the class can't be found by Python's import mechanism. Using the imperative example from earlier, let's see what happens when we try to pickle and then unpickle an instance of the Person content type:
import pickle content_type = make_content_type(PersonSchema, 'Person', bases=(HRPerson,)) fred = content_type(name='Fred', age=54) fred2 = pickle.loads(pickle.dumps(fred)) assert fred is not fred2 assert fred.serialize() == fred2.serialize()
We get this exception:
pickle.PicklingError: Can't pickle <class 'Person'>: it's not found as __main__.Person
What we can do, though, is hook Python's import mechanism so that Python can look up the content type in our Limone instance. This requires that the content type be registered with an instance of limone.Registry:
import pickle registry = limone.Registry() registry.register_content_type(Person) registry.hook_import() content_type = make_content_type(PersonSchema, 'Person', bases=(HRPerson,)) fred = content_type(name='Fred', age=54) fred2 = pickle.loads(pickle.dumps(fred)) assert fred is not fred2 assert fred.serialize() == fred2.serialize() registry.unhook_import()
The pickle and unpickle operations are now successful because pickle is able to look up the type using Python's import mechanism.
The signature for hook_import is:
hook_import(module='__limone__')
The hook_import method inserts an object into sys.meta_path that can look up content types in the registry. The module parameter is used to set the __module__ attribute on generated content types. This will also be used by the import hook to identify the types that it is able to import. Using the default value for module, with the import hook in place, we see that we can import imperatively generated content types in the standard Pythonic way:
from __limone__ import Person fred = Person(name='Fred', age=54)
The default value for module should not be used if you expect that an application will use more than one limone.Registry instance inside of a single process. In this case, a different value of module should be used for each instance so that each instance only tries to find its own content types.
The unhook_import method cleans up a previously made import hook, returning sys.meta_path to its previous state.
Instances of a content type can be converted to their Colander appstruct representations:
jack = Person(
name='Jack',
age=52,
friends=[
(1, 'Fred'),
(2, 'Barney')
],
phones=[
{'location': 'home',
'number': '555-1212'},
])
from pprint import pprint
pprint(jack.appstuct())
Produces this output:
{'age': 52,
'friends': [(1, u'Fred'), (2, u'Barney')],
'name': u'Jack',
'phones': [{'location': u'home', 'number': u'555-1212'}]}
A new instance can be created from an appstruct:
jack = Person.from_appstruct(
{'age': 52,
'friends': [(1, u'Fred'), (2, u'Barney')],
'name': u'Jack',
'phones': [{'location': u'home', 'number': u'555-1212'}]})
A partial appstruct may be used to update an instance:
jack.update_from_appstruct({'age': 53})
Instances of a content type can be serialized using Colander's serialization:
jack = Person(
name='Jack',
age=52,
friends=[
(1, 'Fred'),
(2, 'Barney')
],
phones=[
{'location': 'home',
'number': '555-1212'},
])
from pprint import pprint
pprint(jack.serialize())
Produces this output:
{'age': '52',
'friends': [('1', u'Fred'), ('2', u'Barney')],
'name': u'Jack',
'phones': [{'location': u'home', 'number': u'555-1212'}]}
Note that Colander's serialization is a kind of intermediate format. All scalar values are serialized to strings, but sequences, tuples and mappings are returned as lists, tuples and dicts, respectively. This intermediate form is easily fed into other serializers, like json, to produce a serialized byte sequence.
Instances can be instantiated via Colander's deserialization:
jack = Person.deserialize(
{'age': '52',
'friends': [('1', u'Fred'), ('2', u'Barney')],
'name': u'Jack',
'phones': [{'location': u'home', 'number': u'555-1212'}]})
Deserialization can also be used to update an existing instance:
jack.deserialize_update({'age': '53'})