Memory-efficient data serialization library

Design goals

1. Memory efficiency:
   • Low memory overhead for the objects
   • Generator-based serialization & deserialization
2. Serialization schema:
   • Define data schema with classes and property-like decorators
   • Use (multi-) inheritance to extend, modify and combine schemas
3. User interface:
   • Detailed error report for data that don't match the defined schema

Vocabulary

To minimize possible confusions of the description below, we define the following vocabulary:

• serializable class: a class inheriting the Serializable class
• serializable object: an instance of a serializable class
• serializable property: a SerializableAttribute, SerializableChildObject or SerializableTextContent property of a serializable class
• key: key in the serialized format, that is, tag in XML, key in JSON/YAML

Basic usage

The key class is the Serializable class. Inherit this class to create a schema for serializable objects. Use SerializableAttribute, SerializableChildObject and SerializableTextContent to add schema on the corresponding contents to this object.

A simple example:

from serializer import *

class Animal(Serializable):
    type = SerializableAttribute(required=True)
    description = SerializableTextContent()

class Zoo(Serializable):
    animal = SerializableChildObject(Animal, required=True, multiple=True)

Use deserialize_xml to deserialize a XML file with a serializable class (deserialize_json and deserialize_yaml will be added later). This function takes three arguments, the first being a file-like object to deserialize from, the second being the root key, and the third being a factory function which can create serializable objects (could be the serializable class itself).

from StringIO import StringIO
zoo = deserialize_xml(StringIO('''
    <zoo>
        <animal type="cat">The cat, often referred to as the domestic cat to
        distinguish from other felids and felines, is a small, typically
        furry, carnivorous mammal.</animal>
        <animal type="dog">The domestic dog is a member of the genus Canis,
        which forms part of the wolf-like canids, and is the most widely
        abundant terrestrial carnivore.</animal>
    </zoo>
    '''), 'zoo', Zoo)

After deserialization, the serializable object zoo has everything stored in it. All serializable properties can be accessed just like normal properties.

for animal in zoo.animal:
    print animal.type
# output
# > cat
# > dog

In addition to deserializing from a file, serializable objects can also be created and modified in the Python program.

cow = Animal(type='cow')
cow.description=("Cattle-colloquially cows-are the most common type of large "
    "domesticated ungulates.")
zoo.animal.append(cow)

Pitfalls: uninitialized properties

Finally, use serialize_xml to serialize to a XML file (serialze_json and serialize_yaml will be added later). This function also takes three arguments, the first being a file-like object to deserialize to , the second being the root key, and the third being a serializable object. An optional argument is pretty, which enables pretty printing, and is set to False by default.

outstream = StringIO()
serialize_xml(outstream, 'zoo', zoo, pretty=True)
print outstream.getvalue()
# output
# > <zoo>
# >     <animal type="cat">The cat, often referred to as the domestic cat to
# >     distinguish from other felids and felines, is a small, typically
# >     furry, carnivorous mammal.</animal>
# >     <animal type="dog">The domestic dog is a member of the genus Canis,
# >     which forms part of the wolf-like canids, and is the most widely
# >     abundant terrestrial carnivore.</animal>
# >     <animal type="cow">Cattle-colloquially cows-are the most common type
# >     of large domesticated ungulates.</animal>
# > </zoo>

Advanced usage I: Combine schemas

It's possible to inherit serializable classes to extend or combine schemas. Note that it's also possible to override some serializable properties with regular properties or methods, and they simply follow Python's MRO.

class AnimalCategory(Animal):
    @property
    def description(self):  # remove the text content from the schema
        raise NotImplementedError

    subtype = SerializableChildObject(Animal, required=True, multiple=True)

class BetterZoo(Zoo):
    category = SerializableChildObject(AnimalCategory, required=True,
        multiple=True)

zoo2 = deserialize_xml(StringIO('<zoo><category type="mammal">'
    '<subtype type="human">Humans (taxonomically, Homo sapiens) are the only '
    'extant members of the subtribe Hominina.</subtype>'
    '<subtype type="whale">Whales are a widely distributed and diverse group '
    'of fully aquatic placental marine mammals.</subtype>'
    '</category><animal type="fish">Fish are gill-bearing aquatic craniate '
    'animals that lack limbs with digits.</animal></zoo>'),
    'zoo', BetterZoo)

for category in zoo2.category:
    for subtype in category.subtype:
        print subtype.type
for animal in zoo2.animal:
    print animal.type
# output
# > human
# > whale
# > fish

Advanced usage II: Advanced serializable properties

Sometimes we want the key and property name of a serializable property to be different. For example, the key might be a reserved keyword in Python. Use key in SerializableAttribute, SerializableChildObject and SerializableTextContent to specify a different key.

SerializableAttribute, SerializableChildObject and SerializableTextContent can also be used just like Python's built-in property decorator. Besides, XML, JSON and YAML only support a limited number of basic data types, but we often want to have some more specific type of data. So, SerializableAttribute and SerializableTextContent also have serializer and serializer methods which can be used to convert between custom data types and basic data types.

class DetailedAnimal(Animal):
    features = SerializableAttribute(required=False, key='feature')

    @features.deserializer
    def features(s):
        return s.split()

    @features.serializer
    def features(v):
        return ' '.join(v)

animal = deserialize_xml(StringIO('<animal type="bird" feature="fly sing">'
    'Birds, also known as Aves, are a group of endothermic vertebrates, '
    'characterised by feathers, toothless beaked jaws, the laying of '
    'hard-shelled eggs, a high metabolic rate, a four-chambered heart, and a '
    'strong yet lightweight skeleton.</animal>'), 'animal', DetailedAnimal)
for feature in animal.features:
    print feature

# output
# > fly
# > sing

It's also possible to ask the serializer to ignore a serializable property. Return IGNORE in the serializer method to do so.

Advanced usage III: Fully customized serialization & deserialization

Pitfalls

Uninitialized Properties

A SerializableAttribute, SerializableChildObject, and SerializableTextContent property are uninitialized when a serializable object is created, unless:

1. a default value is set when defining the property
2. the property is set with argument passed to __init__
3. the property is deserialized from a file

If a property is uninitialized, accessing it will raise a SerializableAttributeError. To avoid complex logic checking if a serializable property is initialized, always define default value, or set it in the inherited __init__.