NAME
DBIx::Class::ResultSource::MultipleTableInheritance Use multiple tables
to define your classes
NOTICE
This only works with PostgreSQL at the moment. It has been tested with
PostgreSQL 9.0, 9.1 beta, and 9.1.
There is one additional caveat: the "parent" result classes that you
defined with this resultsource must have one primary column and it must
be named "id."
SYNOPSIS
{
package Cafe::Result::Coffee;
use strict;
use warnings;
use parent 'DBIx::Class::Core';
use aliased 'DBIx::Class::ResultSource::MultipleTableInheritance'
=> 'MTI';
__PACKAGE__->table_class(MTI);
__PACKAGE__->table('coffee');
__PACKAGE__->add_columns(
"id", { data_type => "integer" },
"flavor", {
data_type => "text",
default_value => "good" },
);
__PACKAGE__->set_primary_key("id");
1;
}
{
package Cafe::Result::Sumatra;
use parent 'Cafe::Result::Coffee';
__PACKAGE__->table('sumatra');
__PACKAGE__->add_columns( "aroma",
{ data_type => "text" }
);
1;
}
...
my $schema = Cafe->connect($dsn,$user,$pass);
my $cup = $schema->resultset('Sumatra');
print STDERR Dwarn $cup->result_source->columns;
"id"
"flavor"
"aroma"
..
Inherit from this package and you can make a resultset class from a
view, but that's more than a little bit misleading: the result is
transparently writable.
This is accomplished through the use of stored procedures that map
changes written to the view to changes to the underlying concrete
tables.
WHY?
In many applications, many classes are subclasses of others. Let's say
you have this schema:
# Conceptual domain model
class User {
has id,
has name,
has password
}
class Investor {
has id,
has name,
has password,
has dollars
}
That's redundant. Hold on a sec...
class User {
has id,
has name,
has password
}
class Investor extends User {
has dollars
}
Good idea, but how to put this into code?
One far-too common and absolutely horrendous solution is to have a
"checkbox" in your database: a nullable "investor" column, which entails
a nullable "dollars" column, in the user table.
create table "user" (
"id" integer not null primary key autoincrement,
"name" text not null,
"password" text not null,
"investor" tinyint(1),
"dollars" integer
);
Let's not discuss that further.
A second, better, solution is to break out the two tables into user and
investor:
create table "user" (
"id" integer not null primary key autoincrement,
"name" text not null,
"password" text not null
);
create table "investor" (
"id" integer not null references user("id"),
"dollars" integer
);
So that investor's PK is just an FK to the user. We can clearly see the
class hierarchy here, in which investor is a subclass of user. In
DBIx::Class applications, this second strategy looks like:
my $user_rs = $schema->resultset('User');
my $new_user = $user_rs->create(
name => $args->{name},
password => $args->{password},
);
...
my $new_investor = $schema->resultset('Investor')->create(
id => $new_user->id,
dollars => $args->{dollars},
);
One can cope well with the second strategy, and it seems to be the most
popular smart choice.
HOW?
There is a third strategy implemented here. Make the database do more of
the work: hide the nasty bits so we don't have to handle them unless we
really want to. It'll save us some typing and it'll make for more
expressive code. What if we could do this:
my $new_investor = $schema->resultset('Investor')->create(
name => $args->{name},
password => $args->{password},
dollars => $args->{dollars},
);
And have it Just Work? The user...
{
name => $args->{name},
password => $args->{password},
}
should be created behind the scenes, and the use of either user or
investor in your code should require no special handling. Deleting and
updating $new_investor should also delete or update the user row.
It does. User and investor are both views, their concrete tables
abstracted away behind a set of rules and triggers. You would expect the
above DBIC create statement to look like this in SQL:
INSERT INTO investor ("name","password","dollars") VALUES (...);
But using MTI, it is really this:
INSERT INTO _user_table ("username","password") VALUES (...);
INSERT INTO _investor_table ("id","dollars") VALUES (currval('_user_table_id_seq',...) );
For deletes, the triggers fire in reverse, to preserve referential
integrity (foreign key constraints). For instance:
my $investor = $schema->resultset('Investor')->find({id => $args->{id}});
$investor->delete;
Becomes:
DELETE FROM _investor_table WHERE ("id" = ?);
DELETE FROM _user_table WHERE ("id" = ?);
METHODS
new MTI find the parents, if any, of your resultset class and adds them
to the list of parent_sources for the table.
add_additional_parents
Continuing with coffee:
__PACKAGE__->result_source_instance->add_additional_parents(
qw/
MyApp::Schema::Result::Beverage
MyApp::Schema::Result::Liquid
/
);
This just lets you manually add additional parents beyond the ones
MTI finds.
add_additional_parent
__PACKAGE__->result_source_instance->add_additional_parent(
MyApp::Schema::Result::Beverage
);
You can also add just one.
attach_additional_sources
MTI takes the parents' sources and relationships, creates a new
DBIx::Class::Table object from them, and registers this as a new,
raw, source in the schema, e.g.,
use MyApp::Schema;
print STDERR map { "$_\n" } MyApp::Schema->sources;
# Coffee
# Beverage
# Liquid
# Sumatra
# Raw::Sumatra
Raw::Sumatra will be used to generate the view.
view_definition
This takes the raw table and generates the view (and stored
procedures) you will use.
AUTHOR
Matt S. Trout, <mst@shadowcatsystems.co.uk>
CONTRIBUTORS
Amiri Barksdale, <amiri@roosterpirates.com>
COPYRIGHT
Copyright (c) 2011 the
DBIx::Class::ResultSource::MultipleTableInheritance "AUTHOR" and
"CONTRIBUTORS" as listed above.
LICENSE
This library is free software; you can redistribute it and/or modify it
under the same terms as Perl itself.
SEE ALSO
DBIx::Class DBIx::Class::ResultSource