0.30  2004/12/03 Tom Insam

I've completely removed the Tracker and Deleted tables, along with the
setup_DB_infrastructure and destroy_DB_infrastructure calls. We don't store
deleted objects at all now, once deleted they're gone, and tracking object
classes is just done by iterating through every class till we find the
right one - this has been made cheap by the object cache.

creation_date and timestamp are now stored as DATETIME column types, in
mysql dbs anyway. timestamp has changed from a TIMESTAMP column, that for
some reason mysql wasn't updating properly, to a simple datetime col that
we set to 'now' in perl space at every store.

      2004/11/25 Tom Insam

has_a relationships are no longer handled magically in private methods - we
install db_inflate_<column> and db_deflate_<column> methods that know how to
handle them. This has the secondary advantage that the object state is never
collapsed to oids in a hash - the accessors will _always_ have the objects in.

might_have relationships are no longer special either - they act exactly like
has_a relationships, except that you don't need a column in the owning
table for them. This is a change from the previous behaviour, where there
was always a proxy in their place, that meremly might not be able to
inflate. The old behaviour was inconsistent and confusing.

Class::Persist::Proxy::Collection had almost nothing to do with
Class::Persist::Proxy - it has been moved to Class::Persist::Collection, is no
longer a subclass of Class::Persist::Proxy, and the confusing code that was
shared between the two has been split up and simplified.

Proxies now proxy an oid, for a class. Nothing else. When created, you have
to specify the oid and the class as params to the new() call - proxies can
never be just floating objects.

The big change is the singleton cache. Singleton is the wrong word here -
the concept is that requesting an object of a given ID will always give you
the same instance of that object. It should not ever be possible to get two
seperate instances of an object with the same ID. Because of this, various
operations, like asking for a proxy of an object, will actually return the
real copy of the object, if there's an instance of it in memory. The cache
is implemented in Class::Perist::Cache, and is a hashref of _weak_ references
to the objects, so it's not a source of circular references.

Because of the singleton cache, I've removed every fixed 'upwards' reference
in Class::Persist - object don't link to their parents or containing collections,
all 'strong' references are from the parent to the child. The owner() method
will fetch the right object out of the singleton cache of possible, or a
proxy otherwise. This means that Class::Persist should never contain circular
references unless there is a circular loop in the object tree, which is
wrong and will break right now anyway.

0.05  2004/11/04 Tom Insam

Added a Class::Perist subclass that stores various fields of the object
in a blob, in a similar fashion to Vx, so you can add data to an object
without having to change the DB spec.

0.02  Monday 23rd August, 2004

Improved the code example in SYNOPSIS
Frobbed search to take no query, and documented the order_by parameter.
Tweaked ->set to set multiple key/value pairs in one call.

0.01  Thursday 15th July, 2004

Initial CPAN release