Tangram::Ref - map references to persistent objects

   use Tangram;

   # or
   use Tangram::Core;
   use Tangram::Ref;

   $schema = Tangram::Schema->new(
       classes => { Company => { fields => {
         ref => [ qw( ceo vice_ceo ) ]

   # or

   $schema = Tangram::Schema->new(
       classes => { Company => { fields => {
         ref =>
         {
            ceo =>
            {
               col => 'ceo',
               null => 0
            },

            vice_ceo =>
            {
               ...
            }

This class is responsible for mapping fields that contain a reference to a Perl object. The persistent fields are grouped in a hash under the CWref key in the field hash.

The target object must belong to a persistent class.

Tangram uses a column on the source object to store the id of the target object.

The persistent fields may be specified either as a hash or as an array of field names.

In the hash form, each entry consists in a field name and an associated option hash. The option hash may contain the following fields:

* aggreg

* col

* null

* deep_update

Optional field CWaggreg specifies that the referenced element (if any) must be removed (erased) from persistent storage along with the source object. The default is not to aggregate.

CWcol sets the name of the column that contains the target object's id. This field is optional, it default to the persistent field name. You need to specify a column name if the field name is not an allowed SQL column identifier.

CWnull, if present and set to true, directs deploy() to generate SQL code that allows NULL values for that column.

Optional field CWdeep_update specificies that the target object has to be updated automatically when CWupdate is called on the source object. Automatic update ensures consisitency between the Perl representation and the DBMS state, but degrades update performance so use it with caution. The default is not to do automatic updates.

warning: there is currently something of a bug with all deep_update collections. If you (for instance) insert an object, and a property of that object that is deep_update, then sometimes things end up getting inserted twice - especially in mapping tables. This is currently under investigation.

You may also pass the names of the persistent fields in an array, in which case Tangram uses the field names as column names and allows NULL values.

Ref directly inherits from Tangram::Scalar, and indirectly from Type.

When Tangram stores a reference to another object, it stores the target object's OID in the source object's table, just like what happens with other scalar types like String and Number.

At least that's what basically happens, but there can be complications.

Complication #1: the target object is not persistent yet. Thus it doesn't have an OID yet. Tangram will attempt to store the target object; if this succeeds, the target object gets an OID too and Tangram can proceed.

Complication #2: Tangram may detect that the target object is already being saved; this happens in presence of cycles. Let's take an example:

        $homer = Person->new();
        $marge = Person->new();
        $homer->{partner} = $marge;
        $marge->{partner} = $homer;

        $storage->insert( $homer );

What happens here? Tangram detects that CW$homer refers to CW$marge, and that CW$marge is not yet persistent. So it ignores CW$homer for a while, and proceeds to storing CW$marge.

In the process, Tangram sees that CW$marge refers to CW$homer - which is not persistent yet! its insert() is suspended. Fortunately Tangram realizes that, and doesn't attempt to store CW$homer again. Instead it stores a NULL in the 'partner' column of Marge's row.

Tangram also schedules an UPDATE statement that will be executed just before returning from the call to insert(). That statement will patch Marge's 'partner' column with the proper OID.

Tangram never loads the target object in the process of retrieving the source object. Doing so could have disastrous consequences. For example, consider a family tree, where each Person has a reference to Mom and Dad. Pulling any single Person would eventually retrieve everybody up to Adam and Eve! If Tangram did work that way, that is...

Instead, Tangram ties all the fields that are declared as outgoing references to a package (Tangram::RefOnDemand). The underlying object keeps track of the source object's OID, the Storage object, and the name of the persistent field.

When - if - a reference field is accessed, the target object is pulled from the database (if it's not already present in memory), and the field is untied and its value is replaced with a reference to the target object. As a result, the target object is loaded on demand, but in an almost transparent fashion (why almost? because you can always use tied() on the field and detect that strange things are taking place).