Skip to main content

Relational object persistance framework

Project description


Dobbin is an object database implemented on top of SQLAlchemy. It’s designed to mimick the behavior of the Zope object database (ZODB) while providing greater flexibility and control of the storage.

It supports strong typing with native SQL columns by utilizing the declarative field definitions from zope.schema. Weak typing is supported using the Python pickle protocol. Attributes are automatically persisted with the exception of those starting with the characters “_v_” (volatile attributes).

Tables to support the strongly typed attributes are created on-the-fly with a 1:1 correspondence to interfaces with no inheritance (base interface). As such, objects are modelled as a join between the interfaces they implement plus a table that maintains object metadata and weakly typed instance attributes.


This package was designed and implemented by Malthe Borch and Stefan Eletzhofer with parts contributed by Kapil Thangavelu and Laurence Rowe. It’s licensed as ZPL.

Developer documentation

Objects are mapped by their specification. Polymorphic attributes are declared as interface attributes; strong typing may be declared using schema fields; Attributes that are not declared in a schema or interface are considered volatile.

Unique identifiers (UUID)

A 16-byte unique identification number is used.


Polymorphic attributes are always stored using foreign key relations. This is handled transparently by the framework.

The target of a relation may be a basic type such as a string, integer, tuple or list, or it may be a mapped object.

The following fields allow polymorphic relations of any kind with the type declared on assignment.

  • zope.schema.Object

  • zope.interface.Attribute

Collections are instrumented objects and may be declared using the sequence fields:

  • zope.schema.List

  • zope.schema.Dict

  • zope.schema.Set

A note on dictionaries: Dictionaries are keyed by (unicode) string. Mapped instances may be used as keys in which case a string representation of the unique instance identifier is used. Dictionaries support polymorphic values with type set on assignment.

Walk-through of the framework

This section demonstrates the functionality of the package.


Dobbin uses SQLAlchemy’s object relational mapper to transparently store objects in the database. Objects are persisted in two levels:

Attributes may correspond directly to a table column in which case we say that the attribute is strongly typed. This is the most optimal way to store data.

We may also store attributes that are not mapped directly to a column; in this case, the value of the attribute is stored as a Python pickle. This allows weak typing, but also persistence of amorphic data, e.g. data which does not fit naturally in a relational database.

A universally unique id (UUID) is automatically assigned to all objects.

We begin with a new database session.

>>> import z3c.saconfig
>>> session = z3c.saconfig.Session()

Declarative configuration

We can map attributes to table columns using zope.schema. Instead of using SQL column definitions, we rely on the declarative properties of schema fields.

We start out with an interface decribing a recorded album.

>>> class IAlbum(interface.Interface):
...     artist = schema.TextLine(
...         title=u"Artist",
...         default=u"")
...     title = schema.TextLine(
...         title=u"Title",
...         default=u"")

We can now fabricate instances that implement this interface by using the create method. This is a shorthand for setting up a mapper and creating an instance by calling it.

>>> from z3c.dobbin.factory import create
>>> album = create(IAlbum)

Set attributes.

>>> album.artist = "The Beach Boys"
>>> album.title = u"Pet Sounds"

Interface inheritance is supported. For instance, a vinyl record is a particular type of album.

>>> class IVinyl(IAlbum):
...     rpm = schema.Int(
...         title=u"RPM")
>>> vinyl = create(IVinyl)
>>> vinyl.artist = "Diana Ross and The Supremes"
>>> vinyl.title = "Taking Care of Business"
>>> vinyl.rpm = 45

The attributes are instrumented by SQLAlchemy and map directly to a column in a table.

>>> IVinyl.__mapper__.artist
<sqlalchemy.orm.attributes.InstrumentedAttribute object at ...>

A compact disc is another kind of album.

>>> class ICompactDisc(IAlbum):
...     year = schema.Int(title=u"Year")

Let’s pick a more recent Diana Ross, to fit the format.

>>> cd = create(ICompactDisc)
>>> cd.artist = "Diana Ross"
>>> cd.title = "The Great American Songbook"
>>> cd.year = 2005

To verify that we’ve actually inserted objects to the database, we commit the transacation, thus flushing the current session.


We must actually query the database once before proceeding; this seems to be a bug in zope.sqlalchemy.

>>> results = session.query(album.__class__).all()

Proceed with the transaction.

>>> import transaction
>>> transaction.commit()

We get a reference to the database metadata object, to locate each underlying table.

>>> engine = session.bind
>>> metadata = engine.metadata

Tables are given a name based on the dotted path of the interface they describe. A utility method is provided to create a proper table name for an interface.

>>> from z3c.dobbin.mapper import encode

Verify tables for IVinyl, IAlbum and ICompactDisc.

>>> session.bind = metadata.bind
>>> session.execute(metadata.tables[encode(IVinyl)].select()).fetchall()
[(2, 45)]
>>> session.execute(metadata.tables[encode(IAlbum)].select()).fetchall()
[(1, u'Pet Sounds', u'The Beach Boys'),
 (2, u'Taking Care of Business', u'Diana Ross and The Supremes'),
 (3, u'The Great American Songbook', u'Diana Ross')]
>>> session.execute(metadata.tables[encode(ICompactDisc)].select()).fetchall()
[(3, 2005)]

Mapping concrete classes

Now we’ll create a mapper based on a concrete class. We’ll let the class implement the interface that describes the attributes we want to store, but also provides a custom method.

>>> class Vinyl(object):
...     interface.implements(IVinyl)
...     def __repr__(self):
...         return "<Vinyl %s: %s (@ %d RPM)>" % \
...                (self.artist, self.title, self.rpm)

Although the symbols we define in this test report that they’re available from the __builtin__ module, they really aren’t.

We’ll manually add these symbols.

>>> import __builtin__
>>> __builtin__.IVinyl = IVinyl
>>> __builtin__.IAlbum = IAlbum
>>> __builtin__.Vinyl = Vinyl

Create an instance using the create factory.

>>> vinyl = create(Vinyl)

Verify that we’ve instantiated and instance of our class.

>>> isinstance(vinyl, Vinyl)

Copy the attributes from the Diana Ross vinyl record.

>>> diana = session.query(IVinyl.__mapper__).filter_by(
...     artist=u"Diana Ross and The Supremes")[0]
>>> vinyl.artist = diana.artist
>>> vinyl.title = diana.title
>>> vinyl.rpm = diana.rpm

Verify that the methods on our Vinyl-class are available on the mapper.

>>> repr(vinyl)
'<Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>'

When mapping a class we may run into properties that should take the place of a column (a read-only value). As an example, consider this experimental record class where rotation speed is a function of the title and artist.

>>> class Experimental(Vinyl):
...     @property
...     def rpm(self):
...         return len(self.title+self.artist)
>>> experimental = create(Experimental)

XXX: There’s currently an issue with SQLAlchemy that hinders this behavior; it specifically won’t work if a default value is set on the column that we’re overriding.

>>> #
>>> experimental.artist = vinyl.artist
>>> experimental.title = vinyl.title

Let’s see how fast this record should be played back.

>>> experimental.rpm


Relations are columns that act as references to other objects. They’re declared using the zope.schema.Object field.

Note that we needn’t declare the relation target type in advance, although it may be useful in general to specialize the schema keyword parameter.

>>> class IFavorite(interface.Interface):
...     item = schema.Object(
...         title=u"Item",
...         schema=interface.Interface)
>>> __builtin__.IFavorite = IFavorite

Let’s make our Diana Ross record a favorite.

>>> favorite = create(IFavorite)
>>> favorite.item = vinyl
>>> favorite.item
<Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>

We’ll commit the transaction and lookup the object by its unique id.

>>> transaction.commit()
>>> from z3c.dobbin.soup import lookup
>>> favorite = lookup(favorite.uuid)

When we retrieve the related items, it’s automatically reconstructed to match the specification to which it was associated.

>>> favorite.item
<Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>

We can create relations to objects that are not mapped. Let’s model an accessory item.

>>> class IAccessory(interface.Interface):
...     name = schema.TextLine(title=u"Name of accessory")
>>> class Accessory(object):
...     interface.implements(IAccessory)
...     def __repr__(self):
...          return "<Accessory '%s'>" %

If we now instantiate an accessory and assign it as a favorite item, we’ll implicitly create a mapper from the class specification and insert it into the database.

>>> cleaner = Accessory()
>>> = u"Record cleaner"

Set up relation.

>>> favorite.item = cleaner

Let’s try and get back our record cleaner item.

>>> __builtin__.Accessory = Accessory
>>> favorite.item
<Accessory 'Record cleaner'>

Within the same transaction, the relation will return the original object, maintaining integrity.

>>> favorite.item is cleaner

The session keeps a copy of the pending object until the transaction is ended.

>>> cleaner in session._d_pending.values()

However, once we commit the transaction, the relation is no longer attached to the relation source, and the correct data will be persisted in the database.

>>> = u"CD cleaner"
>>> session.flush()
>>> session.update(favorite)
u'CD cleaner'

This behavior should work well in a request-response type environment, where the request will typically end with a commit.


We can instrument properties that behave like collections by using the sequence and mapping schema fields.

Let’s set up a record collection as an ordered list.

>>> class ICollection(interface.Interface):
...     records = schema.List(
...         title=u"Records",
...         value_type=schema.Object(schema=IAlbum)
...         )
>>> __builtin__.ICollection = ICollection
>>> collection = create(ICollection)
>>> collection.records

Add the Diana Ross record, and save the collection to the session.

>>> collection.records.append(diana)

We can get our collection back.

>>> collection = lookup(collection.uuid)

Let’s verify that we’ve stored the Diana Ross record.

>>> record = collection.records[0]
>>> record.artist, record.title
(u'Diana Ross and The Supremes', u'Taking Care of Business')
>>> session.flush()

When we create a new, transient object and append it to a list, it’s automatically saved on the session.

>>> collection = lookup(collection.uuid)
>>> kool = create(IVinyl)
>>> kool.artist = u"Kool & the Gang"
>>> kool.title = u"Music Is the Message"
>>> kool.rpm = 33
>>> collection.records.append(kool)
>>> [record.artist for record in collection.records]
[u'Diana Ross and The Supremes', u'Kool & the Gang']
>>> session.flush()
>>> session.update(collection)

We can remove items.

>>> collection.records.remove(kool)
>>> len(collection.records) == 1

And extend.

>>> collection.records.extend((kool,))
>>> len(collection.records) == 2

Items can appear twice in the list.

>>> collection.records.append(kool)
>>> len(collection.records) == 3

We can add concrete instances to collections.

>>> marvin = Vinyl()
>>> marvin.artist = u"Marvin Gaye"
>>> marvin.title = u"Let's get it on"
>>> marvin.rpm = 33
>>> collection.records.append(marvin)
>>> len(collection.records) == 4

And remove them, too.

>>> collection.records.remove(marvin)
>>> len(collection.records) == 3

The standard list methods are available.

>>> collection.records = [marvin, vinyl]
>>> collection.records.sort(key=lambda record: record.artist)
>>> collection.records
[<Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>,
 <Vinyl Marvin Gaye: Let's get it on (@ 33 RPM)>]
>>> collection.records.reverse()
>>> collection.records
[<Vinyl Marvin Gaye: Let's get it on (@ 33 RPM)>,
 <Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>]
>>> collection.records.index(vinyl)
>>> collection.records.pop()
<Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>
>>> collection.records.insert(0, vinyl)
>>> collection.records
[<Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>,
 <Vinyl Marvin Gaye: Let's get it on (@ 33 RPM)>]
>>> collection.records.count(vinyl)
>>> collection.records[1] = vinyl
>>> collection.records.count(vinyl)

For good measure, let’s create a new instance without adding any elements to its list.

>>> empty_collection = create(ICollection)

To demonstrate the mapping implementation, let’s set up a catalog for our record collection. We’ll index the records by their ASIN string.

>>> class ICatalog(interface.Interface):
...     index = schema.Dict(
...         title=u"Record index")
>>> catalog = create(ICatalog)

Add a record to the index.

>>> catalog.index[u"B00004WZ5Z"] = diana
>>> catalog.index[u"B00004WZ5Z"]
<Mapper (__builtin__.IVinyl) at ...>

Verify state after commit.

>>> transaction.commit()
>>> catalog.index[u"B00004WZ5Z"]
<Mapper (__builtin__.IVinyl) at ...>

Let’s check that the standard dict methods are supported.

>>> catalog.index.values()
[<Mapper (__builtin__.IVinyl) at ...>]
>>> tuple(catalog.index.itervalues())
(<Mapper (__builtin__.IVinyl) at ...>,)
>>> catalog.index.setdefault(u"B00004WZ5Z", None)
<Mapper (__builtin__.IVinyl) at ...>
>>> catalog.index.pop(u"B00004WZ5Z")
<Mapper (__builtin__.IVinyl) at ...>
>>> len(catalog.index)

Concrete instances are supported.

>>> vinyl = Vinyl()
>>> vinyl.artist = diana.artist
>>> vinyl.title = diana.title
>>> vinyl.rpm = diana.rpm
>>> catalog.index[u"B00004WZ5Z"] = vinyl
>>> len(catalog.index)
>>> catalog.index.popitem()
 <Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>)
>>> catalog.index = {u"B00004WZ5Z": vinyl}
>>> len(catalog.index)
>>> catalog.index.clear()
>>> len(catalog.index)

We may use a mapped object as index.

>>> catalog.index[diana] = diana
>>> catalog.index.keys()[0] == diana.uuid
>>> transaction.commit()
>>> catalog.index[diana]
<Mapper (__builtin__.IVinyl) at ...>
>>> class IDiscography(ICatalog):
...     records = schema.Dict(
...         title=u"Discographies by artist",
...         value_type=schema.List())

Amorphic objects

We can set and retrieve attributes that aren’t declared in an interface.

>>> record = create(interface.Interface)
>>> record.publisher = u"Columbia records"
>>> record.publisher
u'Columbia records'
>>> session.query(record.__class__).filter_by(
...     uuid=record.uuid)[0].publisher
u'Columbia records'

Using this kind of weak we can store (almost) any kind of structure. Values are kept as Python pickles.

>>> favorite = create(interface.Interface)

A transaction hook makes sure that assigned values are transient during a session.

>>> obj = object()
>>> favorite.item = obj
>>> favorite.item is obj

Integers, floats and unicode strings are straight-forward.

>>> favorite.item = 42; transaction.commit()
>>> favorite.item
>>> favorite.item = 42.01; transaction.commit()
>>> 42 < favorite.item <= 42.01
>>> favorite.item = u"My favorite number is 42."; transaction.commit()
>>> favorite.item
u'My favorite number is 42.'

Normal strings need explicit coercing to str.

>>> favorite.item = "My favorite number is 42."; transaction.commit()
>>> str(favorite.item)
'My favorite number is 42.'

Or sequences of items.

>>> favorite.item = (u"green", u"blue", u"red"); transaction.commit()
>>> favorite.item
(u'green', u'blue', u'red')


>>> favorite.item = {u"green": 0x00FF00, u"blue": 0x0000FF, u"red": 0xFF0000}
>>> transaction.commit()
>>> favorite.item
{u'blue': 255, u'green': 65280, u'red': 16711680}
>>> favorite.item[u"black"] = 0x000000
>>> sorted(favorite.item.items())
[(u'black', 0), (u'blue', 255), (u'green', 65280), (u'red', 16711680)]

We do need explicitly set the dirty bit of this instance.

>>> favorite.item = favorite.item
>>> transaction.commit()
>>> sorted(favorite.item.items())
[(u'black', 0), (u'blue', 255), (u'green', 65280), (u'red', 16711680)]

When we create relations to mutable objects, a hook is made into the transaction machinery to keep track of the pending state.

>>> some_list = [u"green", u"blue"]
>>> favorite.item = some_list
>>> some_list.append(u"red"); transaction.commit()
>>> favorite.item
[u'green', u'blue', u'red']

Amorphic structures.

>>> favorite.item = ((1, u"green"), (2, u"blue"), (3, u"red")); transaction.commit()
>>> favorite.item
((1, u'green'), (2, u'blue'), (3, u'red'))

Structures involving relations to other instances.

>>> favorite.item = vinyl; transaction.commit()
>>> favorite.item
<Vinyl Diana Ross and The Supremes: Taking Care of Business (@ 45 RPM)>

Self-referencing works because polymorphic attributes are lazy.

>>> favorite.item = favorite; transaction.commit()
>>> favorite.item
<z3c.dobbin.bootstrap.Soup object at ...>


The security model from Zope is applied to mappers.

>>> from import getCheckerForInstancesOf

Our Vinyl class does not have a security checker defined.

>>> from z3c.dobbin.mapper import getMapper
>>> mapper = getMapper(Vinyl)
>>> getCheckerForInstancesOf(mapper) is None

Let’s set a checker and regenerate the mapper.

>>> from import defineChecker, CheckerPublic
>>> defineChecker(Vinyl, CheckerPublic)
>>> from z3c.dobbin.mapper import createMapper
>>> mapper = createMapper(Vinyl)
>>> getCheckerForInstancesOf(mapper) is CheckerPublic

Known limitations

Certain names are disallowed, and will be ignored when constructing the mapper.

>>> class IKnownLimitations(interface.Interface):
...     __name__ = schema.TextLine()
>>> from z3c.dobbin.interfaces import IMapper
>>> mapper = IMapper(IKnownLimitations)
>>> mapper.__name__


Commit session.

>>> transaction.commit()

Change log



  • Added patch to support old-style mixin classes in the inheritance tree of a mapper class.

  • All attributes that are not declared as interface names are now persisted automatically using the Pickle-protocol. The exception to this rule is attributes starting with the characters “_v_” (volatile attributes).

  • Changed target to SQLAlchemy 0.5-series.


  • Use pickles to store polymorphic attributes; there’s no benefit in using native columns for amorphic data.

  • Dobbin now uses zope.sqlalchemy for transaction and session glue.


  • Use native UUID column type (available on PostgreSQL); compatibility with SQLite is preserved due to its weak typing.

  • Basic type factories are now registered as components.


  • Implemented rest of list methods.

  • Refactoring of table bootstrapping; internal tables now using a naming convention less likely to clash with existing tables.

  • Added support for schema.Dict (including polymorphic dictionary relation).

  • Implemented polymorphic relations for a subset of the basic types (int, str, unicode, tuple and list).


  • Tables are now only created once per minimal interface; this fixes issue on both SQLite and Postgres when we create mappers with an explicit polymorphic class.

  • First entry in change-log.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

z3c.dobbin-0.4.0.tar.gz (30.1 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page