PySchema

PySchema is a library for Python class declaration with typed fields that can be introspected and have data contracts associated with them. This allows for better data integrity checks when serializing/deserializing data and safe interaction with external tools that require typed data.

The foremost design principle when creating the library was to keep the definitions very concise and easy to read. Inspiration was taken from Django’s ORM and the main use cases in mind has been database interaction (Postgres) and Apache Avro schema/datum generation.

It has been tested on Python 2.6 and Python 2.7

API Reference

Usage

The Record base class is the easiest way to define schemas. Typed fields in the schema are defined using subclasses of Field. Standard field types are defined in pyschema.types and are also aliased in the pyschema package.

from pyschema import Record, dumps, loads
from pyschema.types import *

Declaration

To define a schema, use a Python class declaration inheriting from pyschema.Record.

class MyRecord(Record):
    foo = Text()
    bar = Integer()

Class instantiation

Like a typical Python class instantiation. All fields are keyword arguments to the constructor of the record.

r = MyRecord(foo="hej", bar=3)

Member access

r.foo
'hej'

Default string representation/repr

print r
MyRecord(foo='hej', bar=3)

Simple json serialization

Creates a json compatible string representing the object. A special $schema field is added to the json to allow parsing of the record without prior knowledge of the which schema to use. The name of this special field can be set to something else using the pyschema.core.set_schema_name_field

s = dumps(r)
print s
{"$schema": "MyRecord", "foo": "hej", "bar": 3}

####... and deserialization

o = loads(s)
print o.bar
3

Built-in types

PySchema comes with a standard set of field types that can be used to represent the most commonly used data types

  • Text
  • Integer
  • Float
  • Bytes - for binary data, the equivalent of Python < 3 str or Python 3 bytes
  • Boolean - True or False
  • Date - datetime.date objects
  • DateTime - datetime.datetime objects
  • Enum - only allows a preset of text values (specified as an arguemnt to the constructor)
  • List
  • Map
  • SubRecord

Complex types

Some types function as containers for other types.

Lists allow storage of ordered sequences of a single type of data, specified as an argument to the field constructor

class RecordWithList(Record):
    foo = List(Integer())

RecordWithList(foo=[1, 2, 3])
RecordWithList(foo=[1, 2, 3])

Storage for dictionaries mapping from strings to values of a single type, specified as an argument to the field constructor

class RecordWithMap(Record):
    foo = Map(Boolean())

RecordWithMap(foo={u"word": True})
RecordWithMap(foo={u'word': True})

SubRecords allow for nesting of records, i.e. storing records of some sort as fields in other records. SubRecord takes an argument being the schema (i.e. Record class) of the intended stored object. Recursive nesting can also be used by supplying pyschema.SELF as the schema type to SubRecord, in which case the field accepts records of the parent record type.

class NestedRecord(Record):
    foo = SubRecord(MyRecord)  # MyRecord is defined above...

NestedRecord(foo=MyRecord(foo="foo", bar=5))
NestedRecord(foo=MyRecord(foo='foo', bar=5))
class NestedSelfRecord(Record):
    foo = SubRecord(SELF)
    bar = Text()

NestedSelfRecord(foo=NestedSelfRecord(foo=None, bar="Second"), bar="First")
NestedSelfRecord(foo=NestedSelfRecord(foo=None, bar='Second'), bar='First')

Complex types are field types just like any other, so they can be combined to create complex data structures

class Part(Record):
    value = Integer()
    good = Boolean()
    attributes = List(Text())

class AdvancedRecord(Record):
    name = Text()
    parts = Map(SubRecord(Part))

AdvancedRecord(
    name=u"tool_1",
    parts={
        u"moo": Part(
            value=u"buzz",
            good=False,
            attributes=["something", "other"]
        )
    }
)
AdvancedRecord(name=u'tool_1', parts={u'moo': Part(value=u'buzz', good=False, attributes=['something', 'other'])})

Defaults

All fields are optional in the constructor, left-out fields are None by default, except for the Map and List types where they default to their respective empty containers {} and [].

MyRecord(bar=10)

class OtherRecord(Record):
    bar = Map(Float())
    baz = List(Integer())
OtherRecord()
OtherRecord(bar={}, baz=[])

Fails at serialization time when types don’t match

broken_record = MyRecord(foo=5) # object creation works with any types (to allow for temporary unallowed values)
print broken_record  # repr format also still works
MyRecord(foo=5, bar=None)
print dumps(broken_record)  # raises an Exception because 5 isn't a text format
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)

<ipython-input-17-92114c8b8749> in <module>()
----> 1 print dumps(broken_record)  # raises an Exception because 5 isn't a text format


/Users/freider/Code/spotify/pyschema/pyschema/core.py in dumps(obj, attach_schema_name)
    494
    495 def dumps(obj, attach_schema_name=True):
--> 496     json_dct = to_json_compatible(obj)
    497     if attach_schema_name:
    498         json_dct[SCHEMA_FIELD_NAME] = obj._schema_name


/Users/freider/Code/spotify/pyschema/pyschema/core.py in to_json_compatible(record)
    384         val = getattr(record, fname)
    385         if val is not None:
--> 386             d[fname] = f.dump(val)
    387     return d
    388


/Users/freider/Code/spotify/pyschema/pyschema/types.pyc in dump(self, obj)
     34             except:
     35                 raise ValueError(
---> 36                     "%r is not a valid UTF-8 string" % obj
     37                 )
     38


ValueError: 5 is not a valid UTF-8 string

Extending PySchema

Create new custom Field types

import datetime

class Date(Field):
    def dump(self, obj):
        return obj.strftime("%Y-%m-%d")

    def load(self, text):
        return datetime.date(*(int(part) for part in text.split('-')))
class MyOtherRecord(Record):
    date = Date()
s = dumps(MyOtherRecord(date=datetime.date(2013, 10, 7)))
print "Serialized:", s
print "Reloaded:", repr(loads(s).date)
Serialized: {"date": "2013-10-07", "$schema": "MyOtherRecord"}
Reloaded: datetime.date(2013, 10, 7)

Add mixins on existing field types to simplify adding functionality while maintaining OO structure

Text.postgres_type = "TEXT"
Integer.postgres_type = "INTEGER"

@List.mixin
class ListPostgresMixin:
    @property
    def postgres_type(self):
        return self.field_type.postgres_type + " ARRAY"
def create_table_from_record(schema):
    parts = []
    for name, field_type in schema._fields.iteritems():
        parts.append("%s %s" % (name, field_type.postgres_type))
    return "CREATE TABLE %s (" % (schema._schema_name,) + ", ".join(parts) + ")"
class MyTable(Record):
    list_name = Text()
    numbers = List(Integer())

create_table_from_record(MyTable)
'CREATE TABLE MyTable (list_name TEXT, numbers INTEGER ARRAY)'

The following will trigger an error since we haven’t mixed in the postgres_type field for the Map field type in this example.

class Impossibru(Record):
    numbers = Map(Integer())

create_table_from_record(Impossibru)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)

<ipython-input-24-65f7b1968e1a> in <module>()
      2     numbers = Map(Integer())
      3
----> 4 create_table_from_record(Impossibru)


<ipython-input-22-61f11a5f0ce7> in create_table_from_record(schema)
      2     parts = []
      3     for name, field_type in schema._fields.iteritems():
----> 4         parts.append("%s %s" % (name, field_type.postgres_type))
      5     return "CREATE TABLE %s (" % (schema._schema_name,) + ", ".join(parts) + ")"


AttributeError: 'Map' object has no attribute 'postgres_type'

Under the hood

In this section, a brief explanation of the underlying architecture of the package is presented.

Declaration

PySchema utilizes a Schema metaclass for the Record class that hooks into the class declaration logic of the python interpreter.

When a subclass of Record is declared, the metaclass will go through the class properties and create some helper variables needed for schema introspection and general setup. To be able to keep ordering of fields, a counter is increased every time a Field is declared and this is used a the sorting key in the ordered schema.

The metaclass is responsible for setting up the following magic variables on the schema class: * _fields - contains an OrderedDict of (name, field) mappings, where name is the field name and field is the Field instance, i.e. the type definition instance for the field. E.g. ("foo", Integer(size=4)) * _schema_name - the name of the schema. Typically the same as the class name.

Instantiation

When a Record is instantiated, a new object is created where each field is filled with its default value.

Indices and tables