Tangram::Schema - describe a system of persistent classes

   use Tangram;

   $schema = Tangram::Schema->new( $hashref );

   Tangram::Relational->connect( $schema, ... );

   # write SQL to FILE
   $schema->deploy( \*FILE );

   # write SQL to STDOUT
   $schema->deploy();

A Schema contains all the information about the persistent aspects of a system of classes. That information is used to perform the mapping between OO constructs and a relational database.

Schema objects are initialized from a nested data structure called a schema hash. The general structure of the schema hash is described here.

The resulting Schema object becomes the owner of the schema hash passed to new(). The hash may not be modified afterwards, and no assumptions can be made regarding its content.

   $schema = Tangram::Schema->new( $hash );

Returns a new Schema object.

The newly created Schema object becomes the owner of the hash, which can no longer be modified nor reused by client code.

The schema hash describes the persistent aspects of a system of classes. It is a multilevel data structure.

1

The first level of the hash contains information that is relevant to the system as a whole.

2

The second level contains information on a per-class basis.

3

The third level contains information about the individual fields in a class. That information depends on the type of the field and is not documented here; see field hash for a list of predefined persistent types.

The first level of the schema hash describes aspects that are global to a system of persistent classes. It has the following aspect:

   {
      classes =>
     [
        Identity =>
         {
            table => 'IdentityState',
         abstract => 1
         },

         NaturalPerson =>
         {
            bases => [ qw( Identity ) ],
         },

         LegalPerson =>
         {
            bases => [ qw( Identity ) ],
         },
      ],

      make_object => sub { ... },
      set_id => sub { ... }
      get_id => sub { ... }
      normalize => sub { ... },

      control => '...'

      sql => { ... },
   }

CWclasses is an array called the class registry. It contains a description of each persistent class.

CWmake_object contains a reference to a closure that returns a new object of a given class. This field is optional: by default, Tangram calls class method new().

CWset_id and CWget_id are used together to associate an object ID with a persistent object. By default, Tangram converts a reference to an object to a unique integer value by evaluating the expression CW0 + $obj. The result is used as a key in a hash contained in the Storage object. The values in that hash are the object IDs.

If any of your classes use overloading, this approach will not work and you will need to supply your own get/set_id methods.

CWcontrol is the name of a table that has a single row, containing the major and minor version numbers of the Tangram that created the storage, and the highest allocated object id. It defaults to 'Tangram'.

Optional field CWnormalize contains a subroutine that's called to transform classnames and fieldnames into table and column names. The function is called with two arguments; the name to be transformed, and a 'type' argument (currently one of 'tablename' or 'fieldname'). The return value should be the transformed string.

Note that it is expected that the normalize sub will return identical strings with identical arguments, that CWnormalize(normalize($string, $flag), $flag) eq normalize($string, $flag)

Optional field CWsql contains a hash that can be used to customize some of the SQL generated by Tangram. The available options are:

* default_null

CWdefault_null can be used to deal with those databases that don't support the explicit 'NULL' specifier in column definitions. Defaults to 'NULL'. Note that this does not get automatically appended to attributes that have a SQL type explicitly declared.

* id

Object ids encode the type of the object. Tangram assigns a class id to each persistent concrete class within a Schema. When an object is inserted, Tangram allocates a unique integer from a class-specific allocator, then appends the class id to it. Thus the object id for a NaturalPerson may look like CW270005, where CW0005 is the class id. Field CWid contains the SQL type that is used to map the rowid part of the object id. It defaults to 'INTEGER'.

* oid_sequence

If set, this is the name of a sequence to use as the default OID generator, should a particular class not define one. Sequences are an alternate way of generating unique identifiers for rows. They are more scalable when you have high concurrency, but most people won't notice that. It is also possible to define an OID sequence on a per-class level; see below for details. Sequences are emulated on pretend databases like MySQL.

* make_id

This is a closure that is expected to return an unique ID. It is called like this:

   $make_id->($class_id, $storage)

Where CW$class_id is the Class identifier for_the newly created object, and CW$storage is the Tangram::Storage object.

* cid

Field CWcid contains the SQL type that is used to map the class id part of the object id. It defaults to 'INTEGER'.

* cid_size

Field CWcid_size contains the number of decimal positions that the class id occupies within the object id. It defaults to '4'. This does not affect the database, only the in-memory representation of object IDs.

* oid

Historical spurious documentation bug. Documentation described the function that the CWid option performs.

* table_type

Field CWtable_type is a string that if set, will be appended to all CREATE TABLE commands with TYPE=x. For instance, to use transactions with a MySQL database with InnoDB enabled, set CWtable_type to CWInnoDB, and (re)deploy your database.

* dumper

This field sets the default mechanism by which arbitrary structures are serialised to columns, in the absence of a portion of the Tangram schema covering their mapping. The default value is CWStorable (see Storable), seeing as Storable is something like the native serialisation mechanism for Perl. Currently, this setting only applies to the idbif mapping type (see Tangram::IDBIF). It would make more sense for the default to be CWYAML, but unfortunately YAML doesn't support dumping the entire range of native Perl data types, which sucks immensely.

The other fields are related to the SQL types that Tangram uses to store meta-information.

The class registry is an array containing one entry per persistent class. The array contains a list of CWkey => value pairs. The key is the class name, the value is a reference to a hash called the class hash. It contains information on how to map the class.

The class hash can have the following fields:

* abstract

* bases

* fields

* table

* table_type

* id

* oid_sequence

* make_id

Field CWabstract contains a boolean that should be true if the class is abstract. If this field is not present, the class is considered to be concrete.

Field CWbases contains a reference to an array of base classes.

Field CWfields contains a reference to the field hash.

Field CWtable sets the name of the table that Tangram should use to store the state of objects pertaining to this class. This field is optional: it defaults to the class name. If the class name is not an acceptable SQL table identifier, you will need to set this field.

Field CWtable_type sets the type of the table, for instance, the storage back-end to the RDBMS or storage format; it specifies on a per-table basis what the CWtable_type attribute of the schema defines. You almost certainly don't want to set this on a per-table basis.

Field CWid contains an integer identifier for this class. That identifier must be unique within the same schema. If this field is not present, Tangram sets it to the last class id plus one.

Fields CWoid_sequence and CWmake_id are per-class versions of their schema-wide versions documented above. These should be inherited by their subclasses, but currently (as of 2.07_06) aren't. To be safe, until this documentation is fixed, define them in all subclasses.

Each persistent type is identified by a 'typetag', e.g. CWint, CWstring or CWarray.

All the persistent fields of a given type are grouped together inside the field hash, where the typetag is used as a key. The individual fields are specified in an array or a hash, whose layout is type-dependant. For example:

   fields =>
   {
      string   => [ qw( firstName name ) ],
      int      => [ qw( age ) ],
      ref      => { partner => { null => 1 } },
      array    => { children => 'NaturalPerson' },
   },

The typetag not only specifies the type of a field, but also the way in which it should be mapped to SQL constructs. Sometimes the same Perl type lends itself to more than one mapping, for example there are at least two plausible ways of mapping a Perl array (see Tangram::Array and Tangram::IntrArray).

Tangram's persistent type system is extensible, allowing you to mount your own types and make them persistent. All you have to do is to register your type and provide mapping code. See Tangram::Type.

Tangram comes with built-in support for the following types:

* string, int, real: see Tangram::Scalar

* reference : see Tangram::Ref

* array : see Tangram::Array, Tangram::IntrArray

* Set::Object : see Tangram::Set, Tangram::IntrSet

This method is deprecated. See Tangram::Relational.

This method is deprecated. See Tangram::Relational.