A versioned domain model framework.

## Project description

Versioned Domain Model (vdm) is a package which allows you to ‘version’ your domain model in the same way that source code version control systems such as subversion allow you version your code. In particular, versioned domain model versions a complete model and not just individual domain objects (for more on this distinction see below).

At present the package is provided as an extension to SQLAlchemy (tested against v0.4-v0.8).

The library is pretty stable and has been used by the authors in production systems since v0.2 (May 2008).

## Authors

Rufus Pollock (Open Knowledge Foundation) - http://rufuspollock.org/ http://www.okfn.org/

## A Full Versioned Domain Model

To permit ‘atomic’ changes involving multiple objects at once as well as to facilitate domain object traversal it is necessary to introduce an explicit ‘Revision’ object to represent a single changeset to the domain model.

One also needs to introduce the concept of ‘State’. This allows us to make (some) domain objects stateful, in particular those which are to be versioned (State is necessary to support delete/undelete functionality as well as to implement versioned many-to-many relationships).

For each original domain object that comes versioned we end up with 2 domain objects:

• The ‘continuity’: the original domain object.
• The ‘version/revision’: the versions/revisions of that domain object.

Often a user will never need to be concerned (explicitly) with the version/revision object as they will just interact with the original domain object, which will, where necessary, ‘proxy’ requests down to the ‘version/revision’.

To give a flavour of all of this here is a pseudo-code example:

# We need a session of some kind to track which objects have been changed
# In SQLAlchemy can use its Session object
session = get_session_in_some_way()

# Our Revision object
rev1 = Revision(author='me')
# Associate revision with session
# Any given session will have a single associated revision
session.revision = rev1

# Book and Author are domain objects which has been made versioned using this library
# Note the typo!
b1 = Book(name='warandpeace', title='War and Peacee')
b2 = Book(name='annakarenina', title='Anna')
# Note the duplicate!
b3 = Book(name='warandpeace')
a1 = Author(name='tolstoy')

# this is just shorthand for ending this revision and saving all changes
# this may vary depending on the implementation
rev1.commit()
timestamp1 = rev1.timestamp

# some time later
rev2 = Revision(author='me')
session.revision = rev2

b1 = Book.get(name='warandpeace')
# correct typo
b1.title = 'War and Peace'
a1 = Author.get(name='tolstoy')
b1.authors.append(a1)
# duplicate item so delete
b3.delete()
rev2.commit()

# some time even later
rev1 = Revision.get(timestamp=timestamp1)
b1 = Book.get(name='warandpeace')
b1 = b1.get_as_of(rev1)
assert b1.title == 'War and Peacee'
assert b1.authors == []
# etc


### Code in Action

To see some real code in action take a look at, for SQLAlchemy:

vdm/sqlalchemy/demo.py
vdm/sqlalchemy/demo_test.py


## General Conceptual Documentation

A great starting point is Fowler’s Patterns for things that change with time:

In particular Temporal Object:

Two possible approaches:

1. (simpler) Versioned domain objects are versioned independently (like a wiki). This is less of a versioned ‘domain model’ and more of plain versioned domain objects.
2. (more complex) Have explicit ‘Revision’ object and multiple objects can be changed simultaneously in each revision (atomicity). This is a proper versioned domain model.

Remark: using the first approach it is:

• Impossible to support versioning of many-to-many links between versioned domain objects.
• Impossible to change multiple objects ‘at once’ – that is as part of one atomic change
• Difficult to support domain model traversal, that is the ability to navigate around the domain model at a particular ‘revision’/point-in-time.
• More discussions of limitations can be found in this thread [1].

The versioned domain model (vdm) package focuses on supporting the second case (this obviously includes the first one as a subcase) – hence the name.

### Use Cases

SA = Implemented in SQLAlchemy

1. (SA) CRUD for a simple versioned object (no references other than HasA)

2. (SA) Versioning of Many-2-Many and many-2-one relationships where one or both of the related objects are versioned.

1. (SA) Undelete for the above.
2. (SA) Purge for the above.
3. (SA) Support for changing multiple objects in a single commit.
4. (SA) Consistent object traversal both at HEAD and “in the past”

7. (SA) Diffing support on versioning objects and listing of changes for a given Revision.

1. Concurrency checking:
1. Simultaneous edits of different parts of the domain model

2. Simultaneous edits of same parts of domain model (conflict resolution or locking)

1. Alice and Bob both get object X
2. Bob updates object X and commits (A’s X is now out of date)
3. Alice updates object X and commits
4. Conflict!!

This can be resolved in the following ways:

1. Locking
2. Merging

Rather than summarize all situations just see Fowler on concurrency

1. Support for pending updates (so updates must be approved before being visible)
1. A non-approved user makes a change
2. This change is marked as pending
3. This change is notified to a moderator
4. A moderator either allows or disallows the change

## Project details

Uploaded source