package Class::DBI::__::Base;

require 5.006;

use Class::Trigger 0.07;
use base qw(Class::Accessor Class::Data::Inheritable Ima::DBI);

package Class::DBI;

use version; $VERSION = qv('3.0.17');

use strict;
use warnings;

use base "Class::DBI::__::Base";

use Class::DBI::ColumnGrouper;
use Class::DBI::Query;
use Carp ();
use List::Util;
use Clone ();
use UNIVERSAL::moniker;

use vars qw($Weaken_Is_Available);

	$Weaken_Is_Available = 1;
	eval {
		require Scalar::Util;
		import Scalar::Util qw(weaken);
	if ($@) {
		$Weaken_Is_Available = 0;

use overload
	'""'     => sub { shift->stringify_self },
	bool     => sub { not shift->_undefined_primary },
	fallback => 1;

sub stringify_self {
	my $self = shift;
	return (ref $self || $self) unless $self;    # empty PK
	my @cols = $self->columns('Stringify');
	@cols = $self->primary_columns unless @cols;
	return join "/", $self->get(@cols);

sub _undefined_primary {
	my $self = shift;
	return grep !defined, $self->_attrs($self->primary_columns);

# Deprecations

__PACKAGE__->mk_classdata('__hasa_rels' => {});

	my %deprecated = (
   # accessor_name => 'accessor_name_for', # 3.0.7
   # mutator_name  => 'accessor_name_for', # 3.0.7

	no strict 'refs';
	while (my ($old, $new) = each %deprecated) {
		*$old = sub {
			my @caller = caller;
				"Use of '$old' is deprecated at $caller[1] line $caller[2]. Use '$new' instead\n";
			goto &$new;

# Our Class Data
__PACKAGE__->mk_classdata('__grouper'   => Class::DBI::ColumnGrouper->new());
__PACKAGE__->mk_classdata('__data_type' => {});
__PACKAGE__->mk_classdata('iterator_class' => 'Class::DBI::Iterator');
__PACKAGE__->mk_classdata('purge_object_index_every' => 1000);
__PACKAGE__->add_searcher(search => "Class::DBI::Search::Basic",);

	has_a      => "Class::DBI::Relationship::HasA",
	has_many   => "Class::DBI::Relationship::HasMany",
	might_have => "Class::DBI::Relationship::MightHave",
__PACKAGE__->mk_classdata('__meta_info' => {});

# SQL we'll need
__PACKAGE__->set_sql(MakeNewObj => <<'');

__PACKAGE__->set_sql(update => <<"");
SET    %s

__PACKAGE__->set_sql(Nextval => <<'');

__PACKAGE__->set_sql(SearchSQL => <<'');
FROM   %s

__PACKAGE__->set_sql(RetrieveAll => <<'');

__PACKAGE__->set_sql(Retrieve => <<'');

__PACKAGE__->set_sql(Flesh => <<'');

__PACKAGE__->set_sql(single => <<'');

__PACKAGE__->set_sql(DeleteMe => <<"");


# Override transform_sql from Ima::DBI to provide some extra
# transformations
sub transform_sql {
	my ($self, $sql, @args) = @_;
	my $tclass = $self->sql_transformer_class;
	my $T = $tclass->new($self, $sql, @args);
	return $self->SUPER::transform_sql($T->sql => $T->args);


sub _carp {
	my ($self, $msg) = @_;
	Carp::carp($msg || $self);

sub _croak {
	my ($self, $msg) = @_;
	Carp::croak($msg || $self);

sub _db_error {
	my ($self, %info) = @_;
	my $msg = delete $info{msg};
	return $self->_croak($msg, %info);


sub connection {
	my $class = shift;
	$class->set_db(Main => @_);

	my %Per_DB_Attr_Defaults = (
		pg     => { AutoCommit => 0 },
		oracle => { AutoCommit => 0 },

	sub _default_attributes {
		my $class = shift;
		return (
			FetchHashKeyName   => 'NAME_lc',
			ShowErrorStatement => 1,
			AutoCommit         => 1,
			ChopBlanks         => 1,
			%{ $Per_DB_Attr_Defaults{ lc $class->__driver } || {} },

sub set_db {
	my ($class, $db_name, $data_source, $user, $password, $attr) = @_;

	# 'dbi:Pg:dbname=foo' we want 'Pg'. I think this is enough.
	my ($driver) = $data_source =~ /^dbi:(\w+)/i;
	$class->SUPER::set_db('Main', $data_source, $user, $password, $attr);

sub table {
	my ($proto, $table, $alias) = @_;
	my $class = ref $proto || $proto;
	$class->_table($table)      if $table;
	$class->table_alias($alias) if $alias;
	return $class->_table || $class->_table($class->table_alias);

sub table_alias {
	my ($proto, $alias) = @_;
	my $class = ref $proto || $proto;
	$class->_table_alias($alias) if $alias;
	return $class->_table_alias || $class->_table_alias($class->moniker);

sub columns {
	my $proto = shift;
	my $class = ref $proto || $proto;
	my $group = shift || "All";
	return $class->_set_columns($group => @_) if @_;
	return $class->all_columns    if $group eq "All";
	return $class->primary_column if $group eq "Primary";
	return $class->_essential     if $group eq "Essential";
	return $class->__grouper->group_cols($group);

sub _column_class { 'Class::DBI::Column' }

sub _set_columns {
	my ($class, $group, @columns) = @_;

	my @cols = map ref $_ ? $_ : $class->_column_class->new($_), @columns;

	# Careful to take copy
			->add_group($group => @cols));
	return @columns;

sub all_columns { shift->__grouper->all_columns }

sub id {
	my $self  = shift;
	my $class = ref($self)
		or return $self->_croak("Can't call id() as a class method");

	# we don't use get() here because all objects should have
	# exisitng values for PK columns, or else loop endlessly
	my @pk_values = $self->_attrs($self->primary_columns);
	UNIVERSAL::can($_ => 'id') and $_ = $_->id for @pk_values;
	return @pk_values if wantarray;
		"id called in scalar context for class with multiple primary key columns")
		if @pk_values > 1;
	return $pk_values[0];

sub primary_column {
	my $self            = shift;
	my @primary_columns = $self->__grouper->primary;
	return @primary_columns if wantarray;
			. " has multiple primary columns, but fetching in scalar context")
		if @primary_columns > 1;
	return $primary_columns[0];
*primary_columns = \&primary_column;

sub _essential { shift->__grouper->essential }

sub find_column {
	my ($class, $want) = @_;
	return $class->__grouper->find_column($want);

sub _find_columns {
	my $class = shift;
	my $cg    = $class->__grouper;
	return map $cg->find_column($_), @_;

sub has_real_column {    # is really in the database
	my ($class, $want) = @_;
	return ($class->find_column($want) || return)->in_database;

sub data_type {
	my $class    = shift;
	my %datatype = @_;
	while (my ($col, $type) = each %datatype) {
		$class->_add_data_type($col, $type);

sub _add_data_type {
	my ($class, $col, $type) = @_;
	my $datatype = $class->__data_type;
	$datatype->{$col} = $type;

# Make a set of accessors for each of a list of columns. We construct
# the method name by calling accessor_name_for() and mutator_name_for()
# with the normalized column name.

# mutator name will be the same as accessor name unless you override it.

# If both the accessor and mutator are to have the same method name,
# (which will always be true unless you override mutator_name_for), a
# read-write method is constructed for it. If they differ we create both
# a read-only accessor and a write-only mutator.

sub _mk_column_accessors {
	my $class = shift;
	foreach my $col (@_) {

		my $default_accessor = $col->accessor;

		my $acc = $class->accessor_name_for($col);
		my $mut = $class->mutator_name_for($col);

		my %method = ();

		if (
			($acc    eq $mut)                # if they are the same
			or ($mut eq $default_accessor)
			) {                              # or only the accessor was customized
			%method = ('_' => $acc);         # make the accessor the mutator too
			} else {
			%method = (
				_ro_ => $acc,
				_wo_ => $mut,

		foreach my $type (keys %method) {
			my $name     = $method{$type};
			my $acc_type = "make${type}accessor";
			my $accessor = $class->$acc_type($col->name_lc);
			$class->_make_method($_, $accessor) for ($name, "_${name}_accessor");

sub _make_method {
	my ($class, $name, $method) = @_;
	return if defined &{"$class\::$name"};
	$class->_carp("Column '$name' in $class clashes with built-in method")
		if Class::DBI->can($name)
		and not($name eq "id" and join(" ", $class->primary_columns) eq "id");
	no strict 'refs';
	*{"$class\::$name"} = $method;
	$class->_make_method(lc $name => $method);

sub accessor_name_for {
	my ($class, $column) = @_;
  if ($class->can('accessor_name')) { 
		warn "Use of 'accessor_name' is deprecated. Use 'accessor_name_for' instead\n";
		return $class->accessor_name($column) 
	return $column->accessor;

sub mutator_name_for {
	my ($class, $column) = @_;
  if ($class->can('mutator_name')) { 
		warn "Use of 'mutator_name' is deprecated. Use 'mutator_name_for' instead\n";
		return $class->mutator_name($column) 
	return $column->mutator;

sub autoupdate {
	my $proto = shift;
	ref $proto ? $proto->_obj_autoupdate(@_) : $proto->_class_autoupdate(@_);

sub _obj_autoupdate {
	my ($self, $set) = @_;
	my $class = ref $self;
	$self->{__AutoCommit} = $set if defined $set;
	defined $self->{__AutoCommit}
		? $self->{__AutoCommit}
		: $class->_class_autoupdate;

sub _class_autoupdate {
	my ($class, $set) = @_;
	$class->__AutoCommit($set) if defined $set;
	return $class->__AutoCommit;

sub make_read_only {
	my $proto = shift;
	$proto->add_trigger("before_$_" => sub { _croak "$proto is read only" })
		foreach qw/create delete update/;
	return $proto;

sub find_or_create {
	my $class    = shift;
	my $hash     = ref $_[0] eq "HASH" ? shift: {@_};
	my ($exists) = $class->search($hash);
	return defined($exists) ? $exists : $class->insert($hash);

sub insert {
	my $class = shift;
	return $class->_croak("insert needs a hashref") unless ref $_[0] eq 'HASH';
	my $info = { %{ +shift } };    # make sure we take a copy

	my $data;
	while (my ($k, $v) = each %$info) {
		my $col = $class->find_column($k)
			|| (List::Util::first { $_->mutator  eq $k } $class->columns)
			|| (List::Util::first { $_->accessor eq $k } $class->columns)
			|| $class->_croak("$k is not a column of $class");
		$data->{$col} = $v;

	return $class->_insert($data);

*create = \&insert;

# Low Level Data Access

sub _attrs {
	my ($self, @atts) = @_;
	return @{$self}{@atts};
*_attr = \&_attrs;

sub _attribute_store {
	my $self   = shift;
	my $vals   = @_ == 1 ? shift: {@_};
	my (@cols) = keys %$vals;
	@{$self}{@cols} = @{$vals}{@cols};

# If you override this method, you must use the same mechanism to log changes
# for future updates, as other parts of Class::DBI depend on it.
sub _attribute_set {
	my $self = shift;
	my $vals = @_ == 1 ? shift: {@_};

	# We increment instead of setting to 1 because it might be useful to
	# someone to know how many times a value has changed between updates.
	for my $col (keys %$vals) { $self->{__Changed}{$col}++; }

sub _attribute_delete {
	my ($self, @attributes) = @_;
	delete @{$self}{@attributes};

sub _attribute_exists {
	my ($self, $attribute) = @_;
	exists $self->{$attribute};

# Live Object Index (using weak refs if available)

my %Live_Objects;
my $Init_Count = 0;

sub _init {
	my $class = shift;
	my $data  = shift || {};
	my $key   = $class->_live_object_key($data);
	return $Live_Objects{$key} || $class->_fresh_init($key => $data);

sub _fresh_init {
	my ($class, $key, $data) = @_;
	my $obj = bless {}, $class;

	# don't store it unless all keys are present
	if ($key && $Weaken_Is_Available) {
		weaken($Live_Objects{$key} = $obj);

		# time to clean up your room?
			if ++$Init_Count % $class->purge_object_index_every == 0;
	return $obj;

sub _live_object_key {
	my ($me, $data) = @_;
	my $class   = ref($me) || $me;
	my @primary = $class->primary_columns;

	# no key unless all PK columns are defined
	return "" unless @primary == grep defined $data->{$_}, @primary;

	# create single unique key for this object
	return join "\030", $class, map $_ . "\032" . $data->{$_}, sort @primary;

sub purge_dead_from_object_index {
	delete @Live_Objects{ grep !defined $Live_Objects{$_}, keys %Live_Objects };

sub remove_from_object_index {
	my $self    = shift;
	my $obj_key = $self->_live_object_key({ $self->_as_hash });
	delete $Live_Objects{$obj_key};

sub clear_object_index {
	%Live_Objects = ();


sub _prepopulate_id {
	my $self            = shift;
	my @primary_columns = $self->primary_columns;
	return $self->_croak(
		sprintf "Can't create %s object with null primary key columns (%s)",
		ref $self, $self->_undefined_primary)
		if @primary_columns > 1;
	$self->_attribute_store($primary_columns[0] => $self->_next_in_sequence)
		if $self->sequence;

sub _insert {
	my ($proto, $data) = @_;
	my $class = ref $proto || $proto;

	my $self = $class->_init($data);

	$self->_prepopulate_id if $self->_undefined_primary;

	# Reinstate data
	my ($real, $temp) = ({}, {});
	foreach my $col (grep $self->_attribute_exists($_), $self->all_columns) {
		($class->has_real_column($col) ? $real : $temp)->{$col} =

	my @primary_columns = $class->primary_columns;
		$primary_columns[0] => $real->{ $primary_columns[0] })
		if @primary_columns == 1;

	delete $self->{__Changed};

	my %primary_columns;
	@primary_columns{@primary_columns} = ();
	my @discard_columns = grep !exists $primary_columns{$_}, keys %$real;
	$self->call_trigger('create', discard_columns => \@discard_columns);   # XXX

	# Empty everything back out again!
	return $self;

sub _next_in_sequence {
	my $self = shift;
	return $self->sql_Nextval($self->sequence)->select_val;

sub _auto_increment_value {
	my $self = shift;
	my $dbh  = $self->db_Main;

	# Try to do this in a standard method. Fall back to MySQL/SQLite
	# specific versions. TODO remove these when last_insert_id is more
	# widespread.
	# Note: I don't believe the last_insert_id can be zero. We need to
	# switch to defined() checks if it can.
	my $id = $dbh->last_insert_id(undef, undef, $self->table, undef)    # std
		|| $dbh->{mysql_insertid}                                         # mysql
		|| eval { $dbh->func('last_insert_rowid') }
		or $self->_croak("Can't get last insert id");
	return $id;

sub _insert_row {
	my $self = shift;
	my $data = shift;
	eval {
		my @columns = keys %$data;
		my $sth     = $self->sql_MakeNewObj(
			join(', ', @columns),
			join(', ', map $self->_column_placeholder($_), @columns),
		$self->_bind_param($sth, \@columns);
		$sth->execute(values %$data);
		my @primary_columns = $self->primary_columns;
		$data->{ $primary_columns[0] } = $self->_auto_increment_value
			if @primary_columns == 1
			&& !defined $data->{ $primary_columns[0] };
	if ($@) {
		my $class = ref $self;
		return $self->_db_error(
			msg    => "Can't insert new $class: $@",
			err    => $@,
			method => 'insert'
	return 1;

sub _bind_param {
	my ($class, $sth, $keys) = @_;
	my $datatype = $class->__data_type or return;
	for my $i (0 .. $#$keys) {
		if (my $type = $datatype->{ $keys->[$i] }) {
			$sth->bind_param($i + 1, undef, $type);

sub retrieve {
	my $class           = shift;
	my @primary_columns = $class->primary_columns
		or return $class->_croak(
		"Can't retrieve unless primary columns are defined");
	my %key_value;
	if (@_ == 1 && @primary_columns == 1) {
		my $id = shift;
		return unless defined $id;
		return $class->_croak("Can't retrieve a reference") if ref($id);
		$key_value{ $primary_columns[0] } = $id;
	} else {
		%key_value = @_;
			"$class->retrieve(@_) parameters don't include values for all primary key columns (@primary_columns)"
			if keys %key_value < @primary_columns;
	my @rows = $class->search(%key_value);
	$class->_carp("$class->retrieve(@_) selected " . @rows . " rows")
		if @rows > 1;
	return $rows[0];

# Get the data, as a hash, but setting certain values to whatever
# we pass. Used by copy() and move().
# This can take either a primary key, or a hashref of all the columns
# to change.
sub _data_hash {
	my $self            = shift;
	my %data            = $self->_as_hash;
	my @primary_columns = $self->primary_columns;
	delete @data{@primary_columns};
	if (@_) {
		my $arg = shift;
		unless (ref $arg) {
			$self->_croak("Need hash-ref to edit copied column values")
				unless @primary_columns == 1;
			$arg = { $primary_columns[0] => $arg };
		@data{ keys %$arg } = values %$arg;
	return \%data;

sub _as_hash {
	my $self    = shift;
	my @columns = $self->all_columns;
	my %data;
	@data{@columns} = $self->get(@columns);
	return %data;

sub copy {
	my $self = shift;
	return $self->insert($self->_data_hash(@_));


sub construct {
	my ($proto, $data) = @_;
	my $class = ref $proto || $proto;
	my $self = $class->_init($data);
	return $self;

sub move {
	my ($class, $old_obj, @data) = @_;
	$class->_carp("move() is deprecated. If you really need it, "
			. "you should tell me quickly so I can abandon my plan to remove it.");
	return $old_obj->_croak("Can't move to an unrelated class")
		unless $class->isa(ref $old_obj)
		or $old_obj->isa($class);
	return $class->insert($old_obj->_data_hash(@data));

sub delete {
	my $self = shift;
	return $self->_search_delete(@_) if not ref $self;

	eval { $self->sql_DeleteMe->execute($self->id) };
	if ($@) {
		return $self->_db_error(
			msg    => "Can't delete $self: $@",
			err    => $@,
			method => 'delete'
	undef %$self;
	bless $self, 'Class::DBI::Object::Has::Been::Deleted';
	return 1;

sub _search_delete {
	my ($class, @args) = @_;
		"Delete as class method is deprecated. Use search and delete_all instead."
	my $it = $class->search_like(@args);
	while (my $obj = $it->next) { $obj->delete }
	return 1;

# Return the placeholder to be used in UPDATE and INSERT queries.
# Overriding this is deprecated in favour of
#   __PACKAGE__->find_column('entered')->placeholder('IF(1, CURDATE(), ?));

sub _column_placeholder {
	my ($self, $column) = @_;
	return $self->find_column($column)->placeholder;

sub update {
	my $self  = shift;
	my $class = ref($self)
		or return $self->_croak("Can't call update as a class method");

	return -1 unless my @changed_cols = $self->is_changed;
	my @primary_columns = $self->primary_columns;
	my $sth             = $self->sql_update($self->_update_line);
	$class->_bind_param($sth, \@changed_cols);
	my $rows = eval { $sth->execute($self->_update_vals, $self->id); };
	if ($@) {
		return $self->_db_error(
			msg    => "Can't update $self: $@",
			err    => $@,
			method => 'update'

	# enable this once new fixed DBD::SQLite is released:
	if (0 and $rows != 1) {    # should always only update one row
		$self->_croak("Can't update $self: row not found") if $rows == 0;
		$self->_croak("Can't update $self: updated more than one row");

	$self->call_trigger('after_update', discard_columns => \@changed_cols);

	# delete columns that changed (in case adding to DB modifies them again)
	delete $self->{__Changed};
	return 1;

sub _update_line {
	my $self = shift;
	join(', ', map "$_ = " . $self->_column_placeholder($_), $self->is_changed);

sub _update_vals {
	my $self = shift;

	my ($self) = shift;
	if (my @changed = $self->is_changed) {
		my $class = ref $self;
		$self->_carp("$class $self destroyed without saving changes to "
				. join(', ', @changed));

sub discard_changes {
	my $self = shift;
	return $self->_croak("Can't discard_changes while autoupdate is on")
		if $self->autoupdate;
	delete $self->{__Changed};
	return 1;

# We override the get() method from Class::Accessor to fetch the data for
# the column (and associated) columns from the database, using the _flesh()
# method. We also allow get to be called with a list of keys, instead of
# just one.

sub get {
	my $self = shift;
	return $self->_croak("Can't fetch data as class method") unless ref $self;

	my @cols = $self->_find_columns(@_);
	return $self->_croak("Can't get() nothing!") unless @cols;

	if (my @fetch_cols = grep !$self->_attribute_exists($_), @cols) {

	return $self->_attrs(@cols);

sub _flesh {
	my ($self, @groups) = @_;
	my @real = grep $_ ne "TEMP", @groups;
	if (my @want = grep !$self->_attribute_exists($_),
		$self->__grouper->columns_in(@real)) {
		my %row;
		@row{@want} = $self->sql_Flesh(join ", ", @want)->select_row($self->id);
	return 1;

# We also override set() from Class::Accessor so we can keep track of
# changes, and either write to the database now (if autoupdate is on),
# or when update() is called.
sub set {
	my $self          = shift;
	my $column_values = {@_};


	while (my ($column, $value) = each %$column_values) {
		my $col = $self->find_column($column) or die "No such column: $column\n";
		$self->_attribute_set($col => $value);

		# $self->SUPER::set($column, $value);

		eval { $self->call_trigger("after_set_$column") };    # eg inflate
		if ($@) {
			return $self->_croak("after_set_$column trigger error: $@", err => $@);

	$self->update if $self->autoupdate;
	return 1;

sub is_changed {
	my $self = shift;
	grep $self->has_real_column($_), keys %{ $self->{__Changed} };

sub any_changed { keys %{ shift->{__Changed} } }

# By default do nothing. Subclasses should override if required.
# Given a hash ref of column names and proposed new values,
# edit the values in the hash if required.
# For insert $self is the class name (not an object ref).
sub normalize_column_values {
	my ($self, $column_values) = @_;

# Given a hash ref of column names and proposed new values
# validate that the whole set of new values in the hash
# is valid for the object in relation to its current values
# For insert $self is the class name (not an object ref).
sub validate_column_values {
	my ($self, $column_values) = @_;
	my @errors;
	foreach my $column (keys %$column_values) {
		eval {
			$self->call_trigger("before_set_$column", $column_values->{$column},
		push @errors, $column => $@ if $@;
	return unless @errors;
		"validate_column_values error: " . join(" ", @errors),
		method => 'validate_column_values',
		data   => {@errors}

# We override set_sql() from Ima::DBI so it has a default database connection.
sub set_sql {
	my ($class, $name, $sql, $db, @others) = @_;
	$db ||= 'Main';
	$class->SUPER::set_sql($name, $sql, $db, @others);
	$class->_generate_search_sql($name) if $sql =~ /select/i;
	return 1;

sub _generate_search_sql {
	my ($class, $name) = @_;
	my $method = "search_$name";
	defined &{"$class\::$method"}
		and return $class->_carp("$method() already exists");
	my $sql_method = "sql_$name";
	no strict 'refs';
	*{"$class\::$method"} = sub {
		my ($class, @args) = @_;
		return $class->sth_to_objects($name, \@args);

sub dbi_commit   { my $proto = shift; $proto->SUPER::commit(@_); }
sub dbi_rollback { my $proto = shift; $proto->SUPER::rollback(@_); }

# Constraints / Triggers

sub constrain_column {
	my $class = shift;
	my $col   = $class->find_column(+shift)
		or return $class->_croak("constraint_column needs a valid column");
	my $how = shift
		or return $class->_croak("constrain_column needs a constraint");
	if (ref $how eq "ARRAY") {
		my %hash = map { $_ => 1 } @$how;
		$class->add_constraint(list => $col => sub { exists $hash{ +shift } });
	} elsif (ref $how eq "Regexp") {
		$class->add_constraint(regexp => $col => sub { shift =~ $how });
	} elsif (ref $how eq "CODE") {
			code => $col => sub { local $_ = $_[0]; $how->($_) });
	} else {
		my $try_method = sprintf '_constrain_by_%s', $how->moniker;
		if (my $dispatch = $class->can($try_method)) {
			$class->$dispatch($col => ($how, @_));
		} else {
			$class->_croak("Don't know how to constrain $col with $how");

sub add_constraint {
	my $class = shift;
	$class->_invalid_object_method('add_constraint()') if ref $class;
	my $name = shift or return $class->_croak("Constraint needs a name");
	my $column = $class->find_column(+shift)
		or return $class->_croak("Constraint $name needs a valid column");
	my $code = shift
		or return $class->_croak("Constraint $name needs a code reference");
	return $class->_croak("Constraint $name '$code' is not a code reference")
		unless ref($code) eq "CODE";

		"before_set_$column" => sub {
			my ($self, $value, $column_values) = @_;
			$code->($value, $self, $column, $column_values)
				or return $self->_croak(
				"$class $column fails '$name' constraint with '$value'",
				method         => "before_set_$column",
				exception_type => 'constraint_failure',
				data           => {
					column          => $column,
					value           => $value,
					constraint_name => $name,

sub add_trigger {
	my ($self, $name, @args) = @_;
	return $self->_croak("on_setting trigger no longer exists")
		if $name eq "on_setting";
		"$name trigger deprecated: use before_$name or after_$name instead")
		if ($name eq "create" or $name eq "delete");
	$self->SUPER::add_trigger($name => @args);

# Inflation

sub add_relationship_type {
	my ($self, %rels) = @_;
	while (my ($name, $class) = each %rels) {
		no strict 'refs';
		*{"$self\::$name"} = sub {
			my $proto = shift;
			$class->set_up($name => $proto => @_);

sub _extend_meta {
	my ($class, $type, $subtype, $val) = @_;
	my %hash = %{ Clone::clone($class->__meta_info || {}) };
	$hash{$type}->{$subtype} = $val;

sub meta_info {
	my ($class, $type, $subtype) = @_;
	my $meta = $class->__meta_info;
	return $meta          unless $type;
	return $meta->{$type} unless $subtype;
	return $meta->{$type}->{$subtype};

sub _simple_bless {
	my ($class, $pri) = @_;
	return $class->_init({ $class->primary_column => $pri });

sub _deflated_column {
	my ($self, $col, $val) = @_;
	$val ||= $self->_attrs($col) if ref $self;
	return $val unless ref $val;
	my $meta = $self->meta_info(has_a => $col) or return $val;
	my ($a_class, %meths) = ($meta->foreign_class, %{ $meta->args });
	if (my $deflate = $meths{'deflate'}) {
		$val = $val->$deflate(ref $deflate eq 'CODE' ? $self : ());
		return $val unless ref $val;
	return $self->_croak("Can't deflate $col: $val is not a $a_class")
		unless UNIVERSAL::isa($val, $a_class);
	return $val->id if UNIVERSAL::isa($val => 'Class::DBI');
	return "$val";


sub retrieve_all { shift->sth_to_objects('RetrieveAll') }

sub retrieve_from_sql {
	my ($class, $sql, @vals) = @_;
	$sql =~ s/^\s*(WHERE)\s*//i;
	return $class->sth_to_objects($class->sql_Retrieve($sql), \@vals);

sub add_searcher {
	my ($self, %rels) = @_;
	while (my ($name, $class) = each %rels) {
		$self->_croak("$class is not a valid Searcher")
			unless $class->can('run_search');
		no strict 'refs';
		*{"$self\::$name"} = sub {

# This should really be its own Search subclass. But the _do_search
# version has been publicised as the way to do this. We need to
# deprecate this eventually.

sub search_like { shift->_do_search(LIKE => @_) }

sub _do_search {
	my ($class, $type, @args) = @_;
	my $search = Class::DBI::Search::Basic->new($class, @args);


sub add_constructor {
	my ($class, $method, $fragment) = @_;
	return $class->_croak("constructors needs a name") unless $method;
	no strict 'refs';
	my $meth = "$class\::$method";
	return $class->_carp("$method already exists in $class")
		if *$meth{CODE};
	*$meth = sub {
		my $self = shift;
		$self->sth_to_objects($self->sql_Retrieve($fragment), \@_);

sub sth_to_objects {
	my ($class, $sth, $args) = @_;
	$class->_croak("sth_to_objects needs a statement handle") unless $sth;
	unless (UNIVERSAL::isa($sth => "DBI::st")) {
		my $meth = "sql_$sth";
		$sth = $class->$meth();
	my (%data, @rows);
	eval {
		$sth->execute(@$args) unless $sth->{Active};
		$sth->bind_columns(\(@data{ @{ $sth->{NAME_lc} } }));
		push @rows, {%data} while $sth->fetch;
	return $class->_croak("$class can't $sth->{Statement}: $@", err => $@)
		if $@;
	return $class->_ids_to_objects(\@rows);
*_sth_to_objects = \&sth_to_objects;

sub _my_iterator {
	my $self  = shift;
	my $class = $self->iterator_class;
	return $class;

sub _ids_to_objects {
	my ($class, $data) = @_;
	return $#$data + 1 unless defined wantarray;
	return map $class->construct($_), @$data if wantarray;
	return $class->_my_iterator->new($class => $data);


sub _single_row_select {
	my ($self, $sth, @args) = @_;
	Carp::confess("_single_row_select is deprecated in favour of select_row");
	return $sth->select_row(@args);

sub _single_value_select {
	my ($self, $sth, @args) = @_;
	$self->_carp("_single_value_select is deprecated in favour of select_val");
	return $sth->select_val(@args);

sub count_all { shift->sql_single("COUNT(*)")->select_val }

sub maximum_value_of {
	my ($class, $col) = @_;

sub minimum_value_of {
	my ($class, $col) = @_;

sub _unique_entries {
	my ($class, %tmp) = shift;
	return grep !$tmp{$_}++, @_;

sub _invalid_object_method {
	my ($self, $method) = @_;
		"$method should be called as a class method not an object method");

# misc stuff

sub _extend_class_data {
	my ($class, $struct, $key, $value) = @_;
	my %hash = %{ $class->$struct() || {} };
	$hash{$key} = $value;

my %required_classes; # { required_class => class_that_last_required_it, ... }

sub _require_class {
	my ($self, $load_class) = @_;
	$required_classes{$load_class} ||= my $for_class = ref($self) || $self;

	# return quickly if class already exists
	no strict 'refs';
	return if exists ${"$load_class\::"}{ISA};
	(my $load_module = $load_class) =~ s!::!/!g;
	return if eval { require "$" };

	# Only ignore "Can't locate" errors for the specific module we're loading
	return if $@ =~ /^Can't locate \Q$load_module\E\.pm /;

	# Other fatal errors (syntax etc) must be reported (as per
	chomp $@;

	# This error message prefix is especially handy when dealing with
	# classes that are being loaded by other classes recursively.
	# The final message shows the path, e.g.:
	# Foo can't load Bar: Bar can't load Baz: syntax error at line ...
	$self->_croak("$for_class can't load $load_class: $@");

sub _check_classes {    # may automatically call from CHECK block in future
	while (my ($load_class, $by_class) = each %required_classes) {
		next if $load_class->isa("Class::DBI");
			"Class $load_class used by $by_class has not been loaded");



=head1 NAME

Class::DBI - Simple Database Abstraction


  package Music::DBI;
  use base 'Class::DBI';
  Music::DBI->connection('dbi:mysql:dbname', 'username', 'password');

  package Music::Artist;
  use base 'Music::DBI';
  Music::Artist->columns(All => qw/artistid name/);
  Music::Artist->has_many(cds => 'Music::CD');

  package Music::CD;
  use base 'Music::DBI';
  Music::CD->columns(All => qw/cdid artist title year reldate/);
  Music::CD->has_many(tracks => 'Music::Track');
  Music::CD->has_a(artist => 'Music::Artist');
  Music::CD->has_a(reldate => 'Time::Piece',
    inflate => sub { Time::Piece->strptime(shift, "%Y-%m-%d") },
    deflate => 'ymd',

  Music::CD->might_have(liner_notes => LinerNotes => qw/notes/);

  package Music::Track;
  use base 'Music::DBI';
  Music::Track->columns(All => qw/trackid cd position title/); 

  #-- Meanwhile, in a nearby piece of code! --#

  my $artist = Music::Artist->insert({ artistid => 1, name => 'U2' });

  my $cd = $artist->add_to_cds({ 
    cdid   => 1,
    title  => 'October',
    year   => 1980,

  # Oops, got it wrong.

  # etc.

  foreach my $track ($cd->tracks) {
    print $track->position, $track->title

  $cd->delete; # also deletes the tracks

  my $cd  = Music::CD->retrieve(1);
  my @cds = Music::CD->retrieve_all;
  my @cds = Music::CD->search(year => 1980);
  my @cds = Music::CD->search_like(title => 'October%');


Class::DBI provides a convenient abstraction layer to a database.

It not only provides a simple database to object mapping layer, but can
be used to implement several higher order database functions (triggers,
referential integrity, cascading delete etc.), at the application level,
rather than at the database.

This is particularly useful when using a database which doesn't support
these (such as MySQL), or when you would like your code to be portable
across multiple databases which might implement these things in different

In short, Class::DBI aims to make it simple to introduce 'best
practice' when dealing with data stored in a relational database.

=head2 How to set it up

=over 4

=item I<Set up a database.>

You must have an existing database set up, have installed and
the necessary DBD:: driver module for that database.  See L<DBI> and
the documentation of your particular database and driver for details.

=item I<Set up a table for your objects to be stored in.>

Class::DBI works on a simple one class/one table model.  It is your
responsibility to have your database tables already set up. Automating that
process is outside the scope of Class::DBI.

Using our CD example, you might declare a table something like this:

    artist INTEGER, # references 'artist'
    title  VARCHAR(255),
    year   CHAR(4),

=item I<Set up an application base class>

It's usually wise to set up a "top level" class for your entire
application to inherit from, rather than have each class inherit
directly from Class::DBI.  This gives you a convenient point to
place system-wide overrides and enhancements to Class::DBI's behavior.

  package Music::DBI;
  use base 'Class::DBI';

=item I<Give it a database connection>

Class::DBI needs to know how to access the database.  It does this
through a DBI connection which you set up by calling the connection()

  Music::DBI->connection('dbi:mysql:dbname', 'user', 'password');

By setting the connection up in your application base class all the
table classes that inherit from it will share the same connection.

=item I<Set up each Class>

  package Music::CD;
  use base 'Music::DBI';

Each class will inherit from your application base class, so you don't
need to repeat the information on how to connect to the database.

=item I<Declare the name of your table>

Inform Class::DBI what table you are using for this class:


=item I<Declare your columns.>

This is done using the columns() method. In the simplest form, you tell
it the name of all your columns (with the single primary key first):

  Music::CD->columns(All => qw/cdid artist title year/);

If the primary key of your table spans multiple columns then
declare them using a separate call to columns() like this:

  Music::CD->columns(Primary => qw/pk1 pk2/);
  Music::CD->columns(Others => qw/foo bar baz/);

For more information about how you can more efficiently use subsets of
your columns, see L</"LAZY POPULATION">

=item I<Done.>

That's it! You now have a class with methods to L<"insert">,
L<"retrieve">, L<"search"> for, L<"update"> and L<"delete"> objects
from your table, as well as accessors and mutators for each of the
columns in that object (row).


Let's look at all that in more detail:


=head2 connection 

  __PACKAGE__->connection($data_source, $user, $password, \%attr);

This sets up a database connection with the given information. 

This uses L<Ima::DBI> to set up an inheritable connection (named Main). It is 
therefore usual to only set up a connection() in your application base class 
and let the 'table' classes inherit from it.

  package Music::DBI;
  use base 'Class::DBI';

  Music::DBI->connection('dbi:foo:dbname', 'user', 'password');

  package My::Other::Table;
  use base 'Music::DBI';

Class::DBI helps you along a bit to set up the database connection.
connection() provides its own default attributes depending on the driver
name in the data_source parameter. The connection() method provides defaults
for these attributes:

  FetchHashKeyName   => 'NAME_lc',
  ShowErrorStatement => 1,
  ChopBlanks         => 1,
  AutoCommit         => 1,

(Except for Oracle and Pg, where AutoCommit defaults 0, placing the
database in transactional mode).

The defaults can always be extended (or overridden if you know what
you're doing) by supplying your own \%attr parameter. For example:


The RootClass of L<DBIx::ContextualFetch> in also inherited from L<Ima::DBI>,
and you should be very careful not to change this unless you know what
you're doing!

=head3 Dynamic Database Connections / db_Main

It is sometimes desirable to generate your database connection information
dynamically, for example, to allow multiple databases with the same
schema to not have to duplicate an entire class hierarchy.

The preferred method for doing this is to supply your own db_Main()
method rather than calling L<"connection">. This method should return a
valid database handle, and should ensure it sets the standard attributes
described above, preferably by combining $class->_default_attributes()
with your own. Note, this handle *must* have its RootClass set to
L<DBIx::ContextualFetch>, so it is usually not possible to just supply a
$dbh obtained elsewhere.

Note that connection information is class data, and that changing it
at run time may have unexpected behaviour for instances of the class
already in existence.

=head2 table


  $table = Class->table;
  $table = $obj->table;

An accessor to get/set the name of the database table in which this
class is stored.  It -must- be set.

Table information is inherited by subclasses, but can be overridden.

=head2 table_alias

  package Shop::Order;

When Class::DBI constructs SQL, it aliases your table name to a name
representing your class. However, if your class's name is an SQL reserved
word (such as 'Order') this will cause SQL errors. In such cases you
should supply your own alias for your table name (which can, of course,
be the same as the actual table name).

This can also be passed as a second argument to 'table':

  __PACKAGE__->table('orders', 'orders');

As with table, this is inherited but can be overridden.

=head2 sequence / auto_increment


  $sequence_name = Class->sequence;
  $sequence_name = $obj->sequence;

If you are using a database which supports sequences and you want to use
a sequence to automatically supply values for the primary key of a table,
then you should declare this using the sequence() method:

  __PACKAGE__->columns(Primary => 'id');

Class::DBI will use the sequence to generate a primary key value when
objects are inserted without one.

*NOTE* This method does not work for Oracle. However, L<Class::DBI::Oracle>
(which can be downloaded separately from CPAN) provides a suitable
replacement sequence() method.

If you are using a database with AUTO_INCREMENT (e.g. MySQL) then you do
not need this, and any call to insert() without a primary key specified
will fill this in automagically.

Sequence and auto-increment mechanisms only apply to tables that have
a single column primary key. For tables with multi-column primary keys
you need to supply the key values manually.


The following are methods provided for convenience to insert, retrieve
and delete stored objects.  It's not entirely one-size fits all and you
might find it necessary to override them.

=head2 insert

  my $obj = Class->insert(\%data);

This is a constructor to insert new data into the database and create an
object representing the newly inserted row.

%data consists of the initial information to place in your object and
the database.  The keys of %data match up with the columns of your
objects and the values are the initial settings of those fields.

  my $cd = Music::CD->insert({ 
    cdid   => 1,
    artist => $artist,
    title  => 'October',
    year   => 1980,

If the table has a single primary key column and that column value
is not defined in %data, insert() will assume it is to be generated.
If a sequence() has been specified for this Class, it will use that.
Otherwise, it will assume the primary key can be generated by
AUTO_INCREMENT and attempt to use that.

The C<before_create> trigger is invoked directly after storing the
supplied values into the new object and before inserting the record
into the database. The object stored in $self may not have all the
functionality of the final object after_creation, particularly if the
database is going to be providing the primary key value.

For tables with multi-column primary keys you need to supply all
the key values, either in the arguments to the insert() method, or
by setting the values in a C<before_create> trigger.

If the class has declared relationships with foreign classes via
has_a(), you can pass an object to insert() for the value of that key.
Class::DBI will Do The Right Thing.

After the new record has been inserted into the database the data
for non-primary key columns is discarded from the object. If those
columns are accessed again they'll simply be fetched as needed.
This ensures that the data in the application is consistent with
what the database I<actually> stored.

The C<after_create> trigger is invoked after the database insert
has executed.

=head2 find_or_create

  my $cd = Music::CD->find_or_create({ artist => 'U2', title => 'Boy' });

This checks if a CD can be found to match the information passed, and
if not inserts it. 

=head2 delete

  Music::CD->search(year => 1980, title => 'Greatest %')->delete_all;

Deletes this object from the database and from memory. If you have set up
any relationships using C<has_many> or C<might_have>, this will delete
the foreign elements also, recursively (cascading delete).  $obj is no
longer usable after this call.

Multiple objects can be deleted by calling delete_all on the Iterator
returned from a search. Each object found will be deleted in turn,
so cascading delete and other triggers will be honoured.

The C<before_delete> trigger is when an object instance is about to be
deleted. It is invoked before any cascaded deletes.  The C<after_delete>
trigger is invoked after the record has been deleted from the database
and just before the contents in memory are discarded.


Class::DBI provides a few very simple search methods. 

It is not the goal of Class::DBI to replace the need for using SQL. Users
are expected to write their own searches for more complex cases.

L<Class::DBI::AbstractSearch>, available on CPAN, provides a much more
complex search interface than Class::DBI provides itself.

=head2 retrieve

  $obj = Class->retrieve( $id );
  $obj = Class->retrieve( %key_values );

Given key values it will retrieve the object with that key from the
database.  For tables with a single column primary key a single
parameter can be used, otherwise a hash of key-name key-value pairs
must be given.

  my $cd = Music::CD->retrieve(1) or die "No such cd";

=head2 retrieve_all

  my @objs = Class->retrieve_all;
  my $iterator = Class->retrieve_all;

Retrieves objects for all rows in the database. This is probably a
bad idea if your table is big, unless you use the iterator version.

=head2 search

  @objs = Class->search(column1 => $value, column2 => $value ...);

This is a simple search for all objects where the columns specified are
equal to the values specified e.g.:

  @cds = Music::CD->search(year => 1990);
  @cds = Music::CD->search(title => "Greatest Hits", year => 1990);

You may also specify the sort order of the results by adding a final
hash of arguments with the key 'order_by':

  @cds = Music::CD->search(year => 1990, { order_by=>'artist' });

This is passed through 'as is', enabling order_by clauses such
as 'year DESC, title'.

=head2 search_like

  @objs = Class->search_like(column1 => $like_pattern, ....);

This is a simple search for all objects where the columns specified are
like the values specified.  $like_pattern is a pattern given in SQL LIKE
predicate syntax.  '%' means "any zero or more characters", '_' means
"any single character". 

  @cds = Music::CD->search_like(title => 'October%');
  @cds = Music::CD->search_like(title => 'Hits%', artist => 'Various%');

You can also use 'order_by' with these, as with search().


  my $it = Music::CD->search_like(title => 'October%');
  while (my $cd = $it->next) {
    print $cd->title;

Any of the above searches (as well as those defined by has_many) can also
be used as an iterator.  Rather than creating a list of objects matching
your criteria, this will return a Class::DBI::Iterator instance, which
can return the objects required one at a time.

Currently the iterator initially fetches all the matching row data into
memory, and defers only the creation of the objects from that data until
the iterator is asked for the next object. So using an iterator will
only save significant memory if your objects will inflate substantially
when used.

In the case of has_many relationships with a mapping method, the mapping
method is not called until each time you call 'next'. This means that
if your mapping is not a one-to-one, the results will probably not be
what you expect.

=head2 Subclassing the Iterator


You can also subclass the default iterator class to override its
functionality.  This is done via class data, and so is inherited into
your subclasses.


  my $obj = Class->construct(\%data);

This is used to turn data from the database into objects, and should
thus only be used when writing constructors. It is very handy for
cheaply setting up lots of objects from data for without going back to
the database.

For example, instead of doing one SELECT to get a bunch of IDs and then
feeding those individually to retrieve() (and thus doing more SELECT
calls), you can do one SELECT to get the essential data of many objects
and feed that data to construct():

   return map $class->construct($_), $sth->fetchall_hash;

The construct() method creates a new empty object, loads in the column
values, and then invokes the C<select> trigger.


=head2 copy

  $new_obj = $obj->copy;
  $new_obj = $obj->copy($new_id);
  $new_obj = $obj->copy({ title => 'new_title', rating => 18 });

This creates a copy of the given $obj, removes the primary key,
sets any supplied column values and calls insert() to make a new
record in the database.

For tables with a single column primary key, copy() can be called
with no parameters and the new object will be assigned a key
automatically.  Or a single parameter can be supplied and will be
used as the new key.

For tables with a multi-column primary key, copy() must be called with
parameters which supply new values for all primary key columns, unless
a C<before_create> trigger will supply them. The insert() method will
fail if any primary key columns are not defined.

  my $blrunner_dc = $blrunner->copy("Bladerunner: Director's Cut");
  my $blrunner_unrated = $blrunner->copy({
    Title => "Bladerunner: Director's Cut",
    Rating => 'Unrated',

=head2 move

  my $new_obj = Sub::Class->move($old_obj);
  my $new_obj = Sub::Class->move($old_obj, $new_id);
  my $new_obj = Sub::Class->move($old_obj, \%changes);

For transferring objects from one class to another. Similar to copy(), an
instance of Sub::Class is inserted using the data in $old_obj (Sub::Class
is a subclass of $old_obj's subclass). Like copy(), you can supply
$new_id as the primary key of $new_obj (otherwise the usual sequence or
autoincrement is used), or a hashref of multiple new values.


  __PACKAGE__->add_trigger(trigger_point_name => \&code_to_execute);

  # e.g.

  __PACKAGE__->add_trigger(after_create  => \&call_after_create);

It is possible to set up triggers that will be called at various
points in the life of an object. Valid trigger points are:

  before_create       (also used for deflation)
  before_set_$column  (also used by add_constraint)
  after_set_$column   (also used for inflation and by has_a)
  before_update       (also used for deflation and by might_have)
  select              (also used for inflation and by construct and _flesh)

You can create any number of triggers for each point, but you cannot
specify the order in which they will be run. 

All triggers are passed the object they are being fired for, except
when C<before_set_$column> is fired during L<"insert">, in which case
the class is passed in place of the object, which does not yet exist.
You may change object values if required.

Some triggers are also passed extra parameters as name-value
pairs. The individual triggers are further documented with the methods
that trigger them.


  __PACKAGE__->add_constraint('name', column => \&check_sub);

  # e.g.

  __PACKAGE__->add_constraint('over18', age => \&check_age);

  # Simple version
  sub check_age { 
    my ($value) = @_;
    return $value >= 18;

  # Cross-field checking - must have SSN if age < 18
  sub check_age { 
    my ($value, $self, $column_name, $changing) = @_;
    return 1 if $value >= 18;     # We're old enough. 
    return 1 if $changing->{SSN}; # We're also being given an SSN
    return 0 if !ref($self);      # This is an insert, so we can't have an SSN
    return 1 if $self->ssn;       # We already have one in the database
    return 0;                     # We can't find an SSN anywhere

It is also possible to set up constraints on the values that can be set
on a column. The constraint on a column is triggered whenever an object
is created and whenever the value in that column is being changed.

The constraint code is called with four parameters:

  - The new value to be assigned
  - The object it will be assigned to
  (or class name when initially creating an object)
  - The name of the column
  (useful if many constraints share the same code)
  - A hash ref of all new column values being assigned
  (useful for cross-field validation)

The constraints are applied to all the columns being set before the
object data is changed. Attempting to create or modify an object
where one or more constraint fail results in an exception and the object
remains unchanged.

The exception thrown has its data set to a hashref of the column being
changed and the value being changed to. 

Note 1: Constraints are implemented using before_set_$column triggers.
This will only prevent you from setting these values through a
the provided insert() or set() methods. It will always be possible to
bypass this if you try hard enough.

Note 2: When an object is created constraints are currently only
checked for column names included in the parameters to insert().
This is probably a bug and is likely to change in future.

=head2 constrain_column

  Film->constrain_column(year => qr/^\d{4}$/);
  Film->constrain_column(rating => [qw/U Uc PG 12 15 18/]);
  Film->constrain_column(title => sub { length() <= 20 });

Simple anonymous constraints can also be added to a column using the
constrain_column() method.  By default this takes either a regex which
must match, a reference to a list of possible values, or a subref which
will have $_ aliased to the value being set, and should return a
true or false value.

However, this behaviour can be extended (or replaced) by providing a
constraint handler for the type of argument passed to constrain_column.
This behavior should be provided in a method named "_constrain_by_$type",
where $type is the moniker of the argument. For example, the
year example above could be provided by _constrain_by_array().


Before an object is assigned data from the application (via insert or
a set accessor) the normalize_column_values() method is called with
a reference to a hash containing the column names and the new values
which are to be assigned (after any validation and constraint checking,
as described below).

Currently Class::DBI does not offer any per-column mechanism here.
The default method is empty.  You can override it in your own classes
to normalize (edit) the data in any way you need. For example the values
in the hash for certain columns could be made lowercase.

The method is called as an instance method when the values of an existing
object are being changed, and as a class method when a new object is
being created.


Before an object is assigned data from the application (via insert or
a set accessor) the validate_column_values() method is called with a
reference to a hash containing the column names and the new values which
are to be assigned.

The method is called as an instance method when the values of an existing
object are being changed, and as a class method when a new object is
being inserted.

The default method calls the before_set_$column trigger for each column
name in the hash. Each trigger is called inside an eval.  Any failures
result in an exception after all have been checked.  The exception data
is a reference to a hash which holds the column name and error text for
each trigger error.

When using this mechanism for form data validation, for example,
this exception data can be stored in an exception object, via a
custom _croak() method, and then caught and used to redisplay the
form with error messages next to each field which failed validation.


All errors that are generated, or caught and propagated, by Class::DBI
are handled by calling the _croak() method (as an instance method
if possible, or else as a class method).

The _croak() method is passed an error message and in some cases
some extra information as described below. The default behaviour
is simply to call Carp::croak($message).

Applications that require custom behaviour should override the
_croak() method in their application base class (or table classes
for table-specific behaviour). For example:

  use Error;

  sub _croak {
    my ($self, $message, %info) = @_;
    # convert errors into exception objects
    # except for duplicate insert errors which we'll ignore
    Error->throw(-text => $message, %info)
      unless $message =~ /^Can't insert .* duplicate/;

The _croak() method is expected to trigger an exception and not
return. If it does return then it should use C<return;> so that an
undef or empty list is returned as required depending on the calling
context. You should only return other values if you are prepared to
deal with the (unsupported) consequences. 

For exceptions that are caught and propagated by Class::DBI, $message
includes the text of $@ and the original $@ value is available in $info{err}.
That allows you to correctly propagate exception objects that may have
been thrown 'below' Class::DBI (using L<Exception::Class::DBI> for example). 

Exceptions generated by some methods may provide additional data in
$info{data} and, if so, also store the method name in $info{method}.
For example, the validate_column_values() method stores details of
failed validations in $info{data}. See individual method documentation
for what additional data they may store, if any.


All warnings are handled by calling the _carp() method (as
an instance method if possible, or else as a class method).
The default behaviour is simply to call Carp::carp().


=head2 accessors

Class::DBI inherits from L<Class::Accessor> and thus provides individual
accessor methods for every column in your subclass.  It also overrides
the get() and set() methods provided by Accessor to automagically handle
database reading and writing. (Note that as it doesn't make sense to
store a list of values in a column, set() takes a hash of column =>
value pairs, rather than the single key => values of Class::Accessor).

=head2 the fundamental set() and get() methods

  $value = $obj->get($column_name);
  @values = $obj->get(@column_names);

  $obj->set($column_name => $value);
  $obj->set($col1 => $value1, $col2 => $value2 ... );

These methods are the fundamental entry points for getting and setting
column values.  The extra accessor methods automatically generated for
each column of your table are simple wrappers that call these get()
and set() methods.

The set() method calls normalize_column_values() then
validate_column_values() before storing the values.  The
C<before_set_$column> trigger is invoked by validate_column_values(),
checking any constraints that may have been set up.

The C<after_set_$column> trigger is invoked after the new value has
been stored.

It is possible for an object to not have all its column data in memory
(due to lazy inflation).  If the get() method is called for such a column
then it will select the corresponding group of columns and then invoke
the C<select> trigger.

=head1 Changing Your Column Accessor Method Names

=head2 accessor_name_for / mutator_name_for

It is possible to change the name of the accessor method created for a
column either declaratively or programmatically. 

If, for example, you have a column with a name that clashes with a
method otherwise created by Class::DBI, such as 'meta_info', you could
create that Column explicitly with a different accessor (and/or
mutator) when setting up your columns:

	my $meta_col = Class::DBI::Column->new(meta_info => {
		accessor => 'metadata',

  __PACKAGE__->columns(All => qw/id name/, $meta_col);

If you want to change the name of all your accessors, or all that match
a certain pattern, you need to provide an accessor_name_for($col) method, 
which will convert a column name to a method name.

e.g: if your local database naming convention was to prepend the word
'customer' to each column in the 'customer' table, so that you had the
columns 'customerid', 'customername' and 'customerage', but you wanted
your methods to just be $customer->name and $customer->age rather than
$customer->customername etc., you could create a

  sub accessor_name_for {
    my ($class, $column) = @_;
    $column =~ s/^customer//;
    return $column;

Similarly, if you wanted to have distinct accessor and mutator methods,
you could provide a mutator_name_for($col) method which would return
the name of the method to change the value:

  sub mutator_name_for {
    my ($class, $column) = @_;
    return "set_" . $column->accessor;

If you override the mutator name, then the accessor method will be
enforced as read-only, and the mutator as write-only.

=head2 update vs auto update

There are two modes for the accessors to work in: manual update and
autoupdate. When in autoupdate mode, every time one calls an accessor
to make a change an UPDATE will immediately be sent to the database.
Otherwise, if autoupdate is off, no changes will be written until update()
is explicitly called.

This is an example of manual updating:

  # The calls to NumExplodingSheep() and Rating() will only make the
  # changes in memory, not in the database.  Once update() is called
  # it writes to the database in one swell foop.

And of autoupdating:

  # Turn autoupdating on for this object.

  # Each accessor call causes the new value to immediately be written.

Manual updating is probably more efficient than autoupdating and
it provides the extra safety of a discard_changes() option to clear out all
unsaved changes.  Autoupdating can be more convenient for the programmer.
Autoupdating is I<off> by default.

If changes are neither updated nor rolled back when the object is
destroyed (falls out of scope or the program ends) then Class::DBI's
DESTROY method will print a warning about unsaved changes.

=head2 autoupdate

  $update_style = Class->autoupdate;

  $update_style = $obj->autoupdate;

This is an accessor to the current style of auto-updating.  When called
with no arguments it returns the current auto-updating state, true for on,
false for off.  When given an argument it turns auto-updating on and off:
a true value turns it on, a false one off.

When called as a class method it will control the updating style for
every instance of the class.  When called on an individual object it
will control updating for just that object, overriding the choice for
the class.

  __PACKAGE__->autoupdate(1);     # Autoupdate is now on for the class.

  $obj = Class->retrieve('Aliens Cut My Hair');
  $obj->autoupdate(0);      # Shut off autoupdating for this object.

The update setting for an object is not stored in the database.

=head2 update


If L<"autoupdate"> is not enabled then changes you make to your object are
not reflected in the database until you call update().  It is harmless
to call update() if there are no changes to be saved.  (If autoupdate
is on there'll never be anything to save.)

Note: If you have transactions turned on for your database (but see
L<"TRANSACTIONS"> below) you will also need to call dbi_commit(), as
update() merely issues the UPDATE to the database).

After the database update has been executed, the data for columns
that have been updated are deleted from the object. If those columns
are accessed again they'll simply be fetched as needed. This ensures
that the data in the application is consistent with what the database
I<actually> stored.

When update() is called the C<before_update>($self) trigger is
always invoked immediately.

If any columns have been updated then the C<after_update> trigger
is invoked after the database update has executed and is passed:
  ($self, discard_columns => \@discard_columns)

The trigger code can modify the discard_columns array to affect
which columns are discarded.

For example:

  Class->add_trigger(after_update => sub {
    my ($self, %args) = @_;
    my $discard_columns = $args{discard_columns};
    # discard the md5_hash column if any field starting with 'foo'
    # has been updated - because the md5_hash will have been changed
    # by a trigger.
    push @$discard_columns, 'md5_hash' if grep { /^foo/ } @$discard_columns;

Take care to not delete a primary key column unless you know what
you're doing.

The update() method returns the number of rows updated.  If the object
had not changed and thus did not need to issue an UPDATE statement,
the update() call will have a return value of -1.

If the record in the database has been deleted, or its primary key value
changed, then the update will not affect any records and so the update()
method will return 0.

=head2 discard_changes


Removes any changes you've made to this object since the last update.
Currently this simply discards the column values from the object.

If you're using autoupdate this method will throw an exception.

=head2 is_changed

  my $changed = $obj->is_changed;
  my @changed_keys = $obj->is_changed;

Indicates if the given $obj has changes since the last update. Returns
a list of keys which have changed. (If autoupdate is on, this method
will return an empty list, unless called inside a before_update or
after_set_$column trigger)

=head2 id

  $id = $obj->id;
  @id = $obj->id;

Returns a unique identifier for this object based on the values in the
database. It's the equivalent of $obj->get($self->columns('Primary')),
with inflated values reduced to their ids.

A warning will be generated if this method is used in scalar context on
a table with a multi-column primary key.


On some occasions, such as when you're writing triggers or constraint
routines, you'll want to manipulate data in a Class::DBI object without
using the usual get() and set() accessors, which may themselves call
triggers, fetch information from the database, etc.

Rather than interacting directly with the data hash stored in a Class::DBI
object (the exact implementation of which may change in future releases)
you could use Class::DBI's low-level accessors. These appear 'private'
to make you think carefully about using them - they should not be a
common means of dealing with the object.

The data within the object is modelled as a set of key-value pairs,
where the keys are normalized column names (returned by find_column()),
and the values are the data from the database row represented by the
object. Access is via these functions:

=over 4

=item _attrs

  @values = $object->_attrs(@cols);

Returns the values for one or more keys.

=item _attribute_store

  $object->_attribute_store( { $col0 => $val0, $col1 => $val1 } );
  $object->_attribute_store($col0, $val0, $col1, $val1);

Stores values in the object.  They key-value pairs may be passed in
either as a simple list or as a hash reference.  This only updates
values in the object itself; changes will not be propagated to the

=item _attribute_set

  $object->_attribute_set( { $col0 => $val0, $col1 => $val1 } );
  $object->_attribute_set($col0, $val0, $col1, $val1);

Updates values in the object via _attribute_store(), but also logs
the changes so that they are propagated to the database with the next
update.  (Unlike set(), however, _attribute_set() will not trigger an
update if autoupdate is turned on.)

=item _attribute_delete

  @values = $object->_attribute_delete(@cols);

Deletes values from the object, and returns the deleted values.

=item _attribute_exists

  $bool = $object->_attribute_exists($col);

Returns a true value if the object contains a value for the specified
column, and a false value otherwise.


By default, Class::DBI uses simple hash references to store object
data, but all access is via these routines, so if you want to
implement a different data model, just override these functions.


Class::DBI and its subclasses overload the perl builtin I<stringify>
and I<bool> operators. This is a significant convenience.

The perl builtin I<bool> operator is overloaded so that a Class::DBI
object reference is true so long as all its key columns have defined
values.  (This means an object with an id() of zero is not considered

When a Class::DBI object reference is used in a string context it will,
by default, return the value of the primary key. (Composite primary key
values will be separated by a slash).

You can also specify the column(s) to be used for stringification via
the special 'Stringify' column group. So, for example, if you're using
an auto-incremented primary key, you could use this to provide a more
meaningful display string:

  Widget->columns(Stringify => qw/name/);

If you need to do anything more complex, you can provide an stringify_self()
method which stringification will call:

  sub stringify_self { 
    my $self = shift;
    return join ":", $self->id, $self->name;

This overloading behaviour can be useful for columns that have has_a()
relationships.  For example, consider a table that has price and currency

  package Widget;
  use base 'My::Class::DBI';
  Widget->columns(All => qw/widgetid name price currency_code/);

  $obj = Widget->retrieve($id);
  print $obj->price . " " . $obj->currency_code;

The would print something like "C<42.07 USD>".  If the currency_code
field is later changed to be a foreign key to a new currency table then
$obj->currency_code will return an object reference instead of a plain
string. Without overloading the stringify operator the example would now
print something like "C<42.07 Widget=HASH(0x1275}>" and the fix would
be to change the code to add a call to id():

  print $obj->price . " " . $obj->currency_code->id;

However, with overloaded stringification, the original code continues
to work as before, with no code changes needed.

This makes it much simpler and safer to add relationships to existing
applications, or remove them later.


Databases are all about relationships. Thus Class::DBI provides a way
for you to set up descriptions of your relationhips.

Class::DBI provides three such relationships: 'has_a', 'has_many', and
'might_have'. Others are available from CPAN.

=head2 has_a

  Music::CD->has_a(column => 'Foreign::Class');

  Music::CD->has_a(artist => 'Music::Artist');
  print $cd->artist->name;

'has_a' is most commonly used to supply lookup information for a foreign
key. If a column is declared as storing the primary key of another
table, then calling the method for that column does not return the id,
but instead the relevant object from that foreign class.

It is also possible to use has_a to inflate the column value to a non
Class::DBI based. A common usage would be to inflate a date field to a
date/time object:

  Music::CD->has_a(reldate => 'Date::Simple');
  print $cd->reldate->format("%d %b, %Y");

  Music::CD->has_a(reldate => 'Time::Piece',
    inflate => sub { Time::Piece->strptime(shift, "%Y-%m-%d") },
    deflate => 'ymd',
  print $cd->reldate->strftime("%d %b, %Y");

If the foreign class is another Class::DBI representation retrieve is
called on that class with the column value. Any other object will be
instantiated either by calling new($value) or using the given 'inflate'
method. If the inflate method name is a subref, it will be executed,
and will be passed the value and the Class::DBI object as arguments.

When the object is being written to the database the object will be
deflated either by calling the 'deflate' method (if given), or by
attempting to stringify the object. If the deflate method is a subref,
it will be passed the Class::DBI object as an argument.

*NOTE* You should not attempt to make your primary key column inflate
using has_a() as bad things will happen. If you have two tables which
share a primary key, consider using might_have() instead.

=head2 has_many

  Class->has_many(method_to_create => "Foreign::Class");

  Music::CD->has_many(tracks => 'Music::Track');

  my @tracks = $cd->tracks;

  my $track6 = $cd->add_to_tracks({ 
    position => 6,
    title    => 'Tomorrow',

This method declares that another table is referencing us (i.e. storing
our primary key in its table).

It creates a named accessor method in our class which returns a list of
all the matching Foreign::Class objects.

In addition it creates another method which allows a new associated object
to be constructed, taking care of the linking automatically. This method
is the same as the accessor method with "add_to_" prepended.

The add_to_tracks example above is exactly equivalent to:

  my $track6 = Music::Track->insert({
    cd       => $cd,
    position => 6,
    title    => 'Tomorrow',

When setting up the relationship the foreign class's has_a() declarations
are examined to discover which of its columns reference our class. (Note
that because this happens at compile time, if the foreign class is defined
in the same file, the class with the has_a() must be defined earlier than
the class with the has_many(). If the classes are in different files,
Class::DBI should usually be able to do the right things, as long as all
classes inherit Class::DBI before 'use'ing any other classes.)

If the foreign class has no has_a() declarations linking to this class,
it is assumed that the foreign key in that class is named after the
moniker() of this class.

If this is not true you can pass an additional third argument to
the has_many() declaration stating which column of the foreign class
is the foreign key to this class.

=head3 Limiting

  Music::Artist->has_many(cds => 'Music::CD');
  my @cds = $artist->cds(year => 1980);

When calling the method created by has_many, you can also supply any
additional key/value pairs for restricting the search. The above example
will only return the CDs with a year of 1980.

=head3 Ordering

  Music::CD->has_many(tracks => 'Music::Track', { order_by => 'playorder' });

has_many takes an optional final hashref of options. If an 'order_by'
option is set, its value will be set in an ORDER BY clause in the SQL
issued. This is passed through 'as is', enabling order_by clauses such
as 'length DESC, position'.

=head3 Mapping

  Music::CD->has_many(styles => [ 'Music::StyleRef' => 'style' ]);

If the second argument to has_many is turned into a listref of the
Classname and an additional method, then that method will be called in
turn on each of the objects being returned.

The above is exactly equivalent to:

  Music::CD->has_many(_style_refs => 'Music::StyleRef');

  sub styles { 
    my $self = shift;
    return map $_->style, $self->_style_refs;

For an example of where this is useful see L<"MANY TO MANY RELATIONSHIPS">

=head3 Cascading Delete
  Music::Artist->has_many(cds => 'Music::CD', { cascade => 'Fail' });

It is also possible to control what happens to the 'child' objects when
the 'parent' object is deleted. By default this is set to 'Delete' - so,
for example, when you delete an artist, you also delete all their CDs,
leaving no orphaned records. However you could also set this to 'None',
which would leave all those orphaned records (although this generally
isn't a good idea), or 'Fail', which will throw an exception when you
try to delete an artist that still has any CDs.

You can also write your own Cascade strategies by supplying a Class
Name here.

For example you could write a Class::DBI::Cascade::Plugin::Nullify
which would set all related foreign keys to be NULL, and plug it into
your relationship:

  Music::Artist->has_many(cds => 'Music::CD', { 
    cascade => 'Class::DBI::Cascade::Plugin::Nullify' 

=head2 might_have

  Music::CD->might_have(method_name => Class => (@fields_to_import));

  Music::CD->might_have(liner_notes => LinerNotes => qw/notes/);

  my $liner_notes_object = $cd->liner_notes;
  my $notes = $cd->notes; # equivalent to $cd->liner_notes->notes;

might_have() is similar to has_many() for relationships that can have
at most one associated objects. For example, if you have a CD database
to which you want to add liner notes information, you might not want
to add a 'liner_notes' column to your main CD table even though there
is no multiplicity of relationship involved (each CD has at most one
'liner notes' field). So, you create another table with the same primary
key as this one, with which you can cross-reference.

But you don't want to have to keep writing methods to turn the the
'list' of liner_notes objects you'd get back from has_many into the
single object you'd need. So, might_have() does this work for you. It
creates an accessor to fetch the single object back if it exists, and
it also allows you import any of its methods into your namespace. So,
in the example above, the LinerNotes class can be mostly invisible -
you can just call $cd->notes and it will call the notes method on the
correct LinerNotes object transparently for you.

Making sure you don't have namespace clashes is up to you, as is correctly
creating the objects, but this may be made simpler in later versions.
(Particularly if someone asks for this!)

=head2 Notes

has_a(), might_have() and has_many() check that the relevant class has
already been loaded. If it hasn't then they try to load the module of
the same name using require.  If the require fails because it can't
find the module then it will assume it's not a simple require (i.e.,
Foreign::Class isn't in Foreign/ and that you will take care
of it and ignore the warning. Any other error, such as a syntax error,
triggers an exception.

NOTE: The two classes in a relationship do not have to be in the same
database, on the same machine, or even in the same type of database! It
is quite acceptable for a table in a MySQL database to be connected to
a different table in an Oracle database, and for cascading delete etc
to work across these. This should assist greatly if you need to migrate
a database gradually.


Class::DBI does not currently support Many to Many relationships, per se.
However, by combining the relationships that already exist it is possible
to set these up.

Consider the case of Films and Actors, with a linking Role table with a
multi-column Primary Key. First of all set up the Role class:

  Role->columns(Primary => qw/film actor/);
  Role->has_a(film => 'Film');
  Role->has_a(actor => 'Actor');

Then, set up the Film and Actor classes to use this linking table:

  Film->columns(All => qw/id title rating/);
  Film->has_many(stars => [ Role => 'actor' ]);

  Actor->columns(All => qw/id name/);
  Actor->has_many(films => [ Role => 'film' ]);

In each case the 'mapping method' variation of has_many() is used to 
call the lookup method on the Role object returned. As these methods are
the 'has_a' relationships on the Role, these will return the actual
Actor and Film objects, providing a cheap many-to-many relationship.

In the case of Film, this is equivalent to the more long-winded:

  Film->has_many(roles => "Role");

  sub actors { 
    my $self = shift;
    return map $_->actor, $self->roles 

As this is almost exactly what is created internally, add_to_stars and
add_to_films will generally do the right thing as they are actually
doing the equivalent of add_to_roles:

  $film->add_to_actors({ actor => $actor });

Similarly a cascading delete will also do the right thing as it will
only delete the relationship from the linking table.

If the Role table were to contain extra information, such as the name
of the character played, then you would usually need to skip these
short-cuts and set up each of the relationships, and associated helper
methods, manually.


=head2 add_relationship_type

The relationships described above are implemented through
Class::DBI::Relationship subclasses.  These are then plugged into
Class::DBI through an add_relationship_type() call:

    has_a      => "Class::DBI::Relationship::HasA",
    has_many   => "Class::DBI::Relationship::HasMany",
    might_have => "Class::DBI::Relationship::MightHave",

If is thus possible to add new relationship types, or modify the behaviour
of the existing types.  See L<Class::DBI::Relationship> for more information
on what is required.


There are several main approaches to setting up your own SQL queries:

For queries which could be used to create a list of matching objects
you can create a constructor method associated with this SQL and let
Class::DBI do the work for you, or just inline the entire query.

For more complex queries you need to fall back on the underlying Ima::DBI
query mechanism. (Caveat: since Ima::DBI uses sprintf-style interpolation,
you need to be careful to double any "wildcard" % signs in your queries).

=head2 add_constructor

  __PACKAGE__->add_constructor(method_name => 'SQL_where_clause');

The SQL can be of arbitrary complexity and will be turned into:

  SELECT (essential columns)
    FROM (table name)
   WHERE <your SQL>

This will then create a method of the name you specify, which returns
a list of objects as with any built in query.

For example:

  Music::CD->add_constructor(new_music => 'year > 2000');
  my @recent = Music::CD->new_music;

You can also supply placeholders in your SQL, which must then be
specified at query time:

  Music::CD->add_constructor(new_music => 'year > ?');
  my @recent = Music::CD->new_music(2000);

=head2 retrieve_from_sql

On occasions where you want to execute arbitrary SQL, but don't want
to go to the trouble of setting up a constructor method, you can inline
the entire WHERE clause, and just get the objects back directly:

  my @cds = Music::CD->retrieve_from_sql(qq{
    artist = 'Ozzy Osbourne' AND
    title like "%Crazy"      AND
    year <= 1986
    ORDER BY year
    LIMIT 2,3

=head2 Ima::DBI queries

When you can't use 'add_constructor', e.g. when using aggregate functions,
you can fall back on the fact that Class::DBI inherits from Ima::DBI
and prefers to use its style of dealing with statements, via set_sql().

The Class::DBI set_sql() method defaults to using prepare_cached()
unless the $cache parameter is defined and false (see L<Ima::DBI> docs for
more information).

To assist with writing SQL that is inheritable into subclasses, several
additional substitutions are available here: __TABLE__, __ESSENTIAL__
and __IDENTIFIER__.  These represent the table name associated with the
class, its essential columns, and the primary key of the current object,
in the case of an instance method on it.

For example, the SQL for the internal 'update' method is implemented as:

  __PACKAGE__->set_sql('update', <<"");
    SET    %s

The 'longhand' version of the new_music constructor shown above would
similarly be:

  Music::CD->set_sql(new_music => qq{
      FROM __TABLE__
     WHERE year > ?

For such 'SELECT' queries L<Ima::DBI>'s set_sql() method is extended to
create a helper shortcut method, named by prefixing the name of the
SQL fragment with 'search_'. Thus, the above call to set_sql() will
automatically set up the method Music::CD->search_new_music(), which
will execute this search and return the relevant objects or Iterator.
(If there are placeholders in the query, you must pass the relevant
arguments when calling your search method.)

This does the equivalent of:

  sub search_new_music {
    my ($class, @args) = @_;
    my $sth = $class->sql_new_music;
    return $class->sth_to_objects($sth);

The $sth which is used to return the objects here is a normal DBI-style
statement handle, so if the results can't be turned into objects easily,
it is still possible to call $sth->fetchrow_array etc and return whatever
data you choose.

Of course, any query can be added via set_sql, including joins.  So,
to add a query that returns the 10 Artists with the most CDs, you could
write (with MySQL):

  Music::Artist->set_sql(most_cds => qq{
      FROM artist, cd
     WHERE = cd.artist
     ORDER BY cds DESC
     LIMIT 10

  my @artists = Music::Artist->search_most_cds();

If you also need to access the 'cds' value returned from this query,
the best approach is to declare 'cds' to be a TEMP column. (See
L<"Non-Persistent Fields"> below).

=head2 Class::DBI::AbstractSearch

  my @music = Music::CD->search_where(
    artist => [ 'Ozzy', 'Kelly' ],
    status => { '!=', 'outdated' },

The L<Class::DBI::AbstractSearch> module, available from CPAN, is a
plugin for Class::DBI that allows you to write arbitrarily complex
searches using perl data structures, rather than SQL.

=head2 Single Value SELECTs

=head3 select_val

Selects which only return a single value can couple Class::DBI's
sql_single() SQL, with the $sth->select_val() call which we get from

  __PACKAGE__->set_sql(count_all => "SELECT COUNT(*) FROM __TABLE__");
  # .. then ..
  my $count = $class->sql_count_all->select_val;

This can also take placeholders and/or do column interpolation if required:

  __PACKAGE__->set_sql(count_above => q{
  # .. then ..
  my $count = $class->sql_count_above('year')->select_val(2001);

=head3 sql_single

Internally Class::DBI defines a very simple SQL fragment called 'single':

  "SELECT %s FROM __TABLE__".  

This is used to implement the above Class->count_all():


This interpolates the COUNT(*) into the %s of the SQL, and then executes
the query, returning a single value.

Any SQL set up via set_sql() can of course be supplied here, and
select_val can take arguments for any placeholders there.

Internally several helper methods are defined using this approach:

=over 4

=item - count_all

=item - maximum_value_of($column)

=item - minimum_value_of($column)



In the tradition of Perl, Class::DBI is lazy about how it loads your
objects.  Often, you find yourself using only a small number of the
available columns and it would be a waste of memory to load all of them
just to get at two, especially if you're dealing with large numbers of
objects simultaneously.

You should therefore group together your columns by typical usage, as
fetching one value from a group can also pre-fetch all the others in
that group for you, for more efficient access.

So for example, if we usually fetch the artist and title, but don't use
the 'year' so much, then we could say the following:

  Music::CD->columns(Primary   => qw/cdid/);
  Music::CD->columns(Essential => qw/artist title/);
  Music::CD->columns(Others    => qw/year runlength/);

Now when you fetch back a CD it will come pre-loaded with the 'cdid',
'artist' and 'title' fields. Fetching the 'year' will mean another visit
to the database, but will bring back the 'runlength' whilst it's there.

This can potentially increase performance.

If you don't like this behavior, then just add all your columns to the
Essential group, and Class::DBI will load everything at once. If you
have a single column primary key you can do this all in one shot with
one single column declaration:

  Music::CD->columns(Essential => qw/cdid artist title year runlength/);

=head2 columns

  my @all_columns  = $class->columns;
  my @columns      = $class->columns($group);

  my @primary      = $class->primary_columns;
  my $primary      = $class->primary_column;
  my @essential    = $class->_essential;

There are four 'reserved' groups: 'All', 'Essential', 'Primary' and

B<'All'> are all columns used by the class. If not set it will be
created from all the other groups.

B<'Primary'> is the primary key columns for this class. It I<must>
be set before objects can be used.

If 'All' is given but not 'Primary' it will assume the first column in
'All' is the primary key.

B<'Essential'> are the minimal set of columns needed to load and use the
object. Only the columns in this group will be loaded when an object
is retrieve()'d. It is typically used to save memory on a class that
has a lot of columns but where only use a few of them are commonly
used. It will automatically be set to B<'Primary'> if not explicitly set.
The 'Primary' column is always part of the 'Essential' group.

For simplicity primary_columns(), primary_column(), and _essential()
methods are provided to return these. The primary_column() method should
only be used for tables that have a single primary key column.

=head2 Non-Persistent Fields

  Music::CD->columns(TEMP => qw/nonpersistent/);

If you wish to have fields that act like columns in every other way, but
that don't actually exist in the database (and thus will not persist),
you can declare them as part of a column group of 'TEMP'.

=head2 find_column


The columns of a class are stored as Class::DBI::Column objects. This
method will return you the object for the given column, if it exists.
This is most useful either in a boolean context to discover if the column
exists, or to 'normalize' a user-entered column name to an actual Column.

The interface of the Column object itself is still under development,
so you shouldn't really rely on anything internal to it.


Class::DBI suffers from the usual problems when dealing with transactions.
In particular, you should be very wary when committing your changes that
you may actually be in a wider scope than expected and that your caller
may not be expecting you to commit.

However, as long as you are aware of this, and try to keep the scope
of your transactions small, ideally always within the scope of a single
method, you should be able to work with transactions with few problems.

=head2 dbi_commit / dbi_rollback


These are thin aliases through to the DBI's commit() and rollback()
commands to commit or rollback all changes to this object.

=head2 Localised Transactions

A nice idiom for turning on a transaction locally (with AutoCommit turned
on globally) (courtesy of Dominic Mitchell) is:

  sub do_transaction {
    my $class = shift;
    my ( $code ) = @_;
    # Turn off AutoCommit for this scope.
    # A commit will occur at the exit of this block automatically,
    # when the local AutoCommit goes out of scope.
    local $class->db_Main->{ AutoCommit };

    # Execute the required code inside the transaction.
    eval { $code->() };
    if ( $@ ) {
      my $commit_error = $@;
      eval { $class->dbi_rollback }; # might also die!
      die $commit_error;

  And then you just call:

  Music::DBI->do_transaction( sub {
    my $artist = Music::Artist->insert({ name => 'Pink Floyd' });
    my $cd = $artist->add_to_cds({ 
      title => 'Dark Side Of The Moon', 
      year => 1974,

Now either both will get added, or the entire transaction will be
rolled back.


Class::DBI supports uniqueness of objects in memory. In a given perl
interpreter there will only be one instance of any given object at
one time. Many variables may reference that object, but there can be
only one.

Here's an example to illustrate:

  my $artist1 = Music::Artist->insert({ artistid => 7, name => 'Polysics' });
  my $artist2 = Music::Artist->retrieve(7);
  my $artist3 = Music::Artist->search( name => 'Polysics' )->first;

Now $artist1, $artist2, and $artist3 all point to the same object. If you
update a property on one of them, all of them will reflect the update.

This is implemented using a simple object lookup index for all live
objects in memory. It is not a traditional cache - when your objects
go out of scope, they will be destroyed normally, and a future retrieve
will instantiate an entirely new object.

The ability to perform this magic for you replies on your perl having
access to the Scalar::Util::weaken function. Although this is part of
the core perl distribution, some vendors do not compile support for it.
To find out if your perl has support for it, you can run this on the
command line:

  perl -e 'use Scalar::Util qw(weaken)'

If you get an error message about weak references not being implemented,
Class::DBI will not maintain this lookup index, but give you a separate
instances for each retrieve.

A few new tools are offered for adjusting the behavior of the object
index. These are still somewhat experimental and may change in a
future release.

=head2 remove_from_object_index


This is an object method for removing a single object from the live
objects index. You can use this if you want to have multiple distinct
copies of the same object in memory.

=head2 clear_object_index


You can call this method on any class or instance of Class::DBI, but
the effect is universal: it removes all objects from the index.

=head2 purge_object_index_every


Weak references are not removed from the index when an object goes
out of scope. This means that over time the index will grow in memory.
This is really only an issue for long-running environments like mod_perl,
but every so often dead references are cleaned out to prevent this. By
default, this happens every 1000 object loads, but you can change that
default for your class by setting the 'purge_object_index_every' value.

(Eventually this may handled in the DESTROY method instead.)

As a final note, keep in mind that you can still have multiple distinct
copies of an object in memory if you have multiple perl interpreters
running. CGI, mod_perl, and many other common usage situations run
multiple interpreters, meaning that each one of them may have an instance
of an object representing the same data. However, this is no worse
than it was before, and is entirely normal for database applications in
multi-process environments.


The preferred method of interacting with Class::DBI is for you to write
a subclass for your database connection, with each table-class inheriting
in turn from it. 

As well as encapsulating the connection information in one place,
this also allows you to override default behaviour or add additional
functionality across all of your classes.

As the innards of Class::DBI are still in flux, you must exercise extreme
caution in overriding private methods of Class::DBI (those starting with
an underscore), unless they are explicitly mentioned in this documentation
as being safe to override. If you find yourself needing to do this,
then I would suggest that you ask on the mailing list about it, and
we'll see if we can either come up with a better approach, or provide
a new means to do whatever you need to do.

=head1 CAVEATS

=head2 Multi-Column Foreign Keys are not supported

You can't currently add a relationship keyed on multiple columns.
You could, however, write a Relationship plugin to do this, and the
world would be eternally grateful...

=head2 Don't change or inflate the value of your primary columns

Altering your primary key column currently causes Bad Things to happen.
I should really protect against this.


Theoretically Class::DBI should work with almost any standard RDBMS. Of
course, in the real world, we know that that's not true. It is known
to work with MySQL, PostgreSQL, Oracle and SQLite, each of which have
their own additional subclass on CPAN that you should explore if you're
using them:

  L<Class::DBI::mysql>, L<Class::DBI::Pg>, L<Class::DBI::Oracle>,

For the most part it's been reported to work with Sybase, although there
are some issues with multi-case column/table names. Beyond that lies
The Great Unknown(tm). If you have access to other databases, please
give this a test run, and let me know the results.

L<Ima::DBI> (and hence Class::DBI) requires a database that supports
table aliasing and a DBI driver that supports placeholders. This means
it won't work with older releases of L<DBD::AnyData> (and any releases
of its predecessor L<DBD::RAM>), and L<DBD::Sybase> + FreeTDS may or
may not work depending on your FreeTDS version.


Tony Bowden 


Michael G Schwern 

=head1 THANKS TO

Tim Bunce, Tatsuhiko Miyagawa, Perrin Harkins, Alexander Karelas, Barry
Hoggard, Bart Lateur, Boris Mouzykantskii, Brad Bowman, Brian Parker,
Casey West, Charles Bailey, Christopher L. Everett Damian Conway, Dan
Thill, Dave Cash, David Jack Olrik, Dominic Mitchell, Drew Taylor,
Drew Wilson, Jay Strauss, Jesse Sheidlower, Jonathan Swartz, Marty
Pauley, Michael Styer, Mike Lambert, Paul Makepeace, Phil Crow, Richard
Piacentini, Simon Cozens, Simon Wilcox, Thomas Klausner, Tom Renfro,
Uri Gutman, William McKee, the Class::DBI mailing list, the POOP group,
and all the others who've helped, but that I've forgetten to mention.


Class::DBI now uses a three-level versioning system. This release, for
example, is version 3.0.17

The general approach to releases will be that users who like a degree of
stability can hold off on upgrades until the major sub-version increases
(e.g. 3.1.0). Those who like living more on the cutting edge can keep up
to date with minor sub-version releases. 

Functionality which was introduced during a minor sub-version release may
disappear without warning in a later minor sub-version release. I'll try
to avoid doing this, and will aim to have a deprecation cycle of at least
a few minor sub-versions, but you should keep a close eye on the CHANGES
file, and have good tests in place. (This is good advice generally,
of course.) Anything that is in a major sub-version release will go
through a deprecation cycle of at least one further major sub-version
before it is removed (and usually longer).

=head2 Getting changes accepted

There is an active Class::DBI community, however I am not part of it.
I am not on the mailing list, and I don't follow the wiki. I also do
not follow Perl Monks or CPAN reviews or annoCPAN or whatever the tool
du jour happens to be. 

If you find a problem with Class::DBI, by all means discuss it in any of
these places, but don't expect anything to happen unless you actually
tell me about it.

The preferred method for doing this is via the CPAN RT interface, which
you can access at or by emailing

If you email me personally about Class::DBI issues, then I will
probably bounce them on to there, unless you specifically ask me not to.
Otherwise I can't keep track of what all needs fixed. (This of course
means that if you ask me not to send your mail to RT, there's a much
higher chance that nothing will every happen about your problem).

=head2 Bug Reports

If you're reporting a bug then it has a much higher chance of getting
fixed quicker if you can include a failing test case. This should be
a completely stand-alone test that could be added to the Class::DBI
distribution. That is, it should use L<Test::Simple> or L<Test::More>,
fail with the current code, but pass when I fix the problem. If it
needs to have a working database to show the problem, then this should
preferably use SQLite, and come with all the code to set this up. The
nice people on the mailing list will probably help you out if you need
assistance putting this together.

You don't need to include code for actually fixing the problem, but of
course it's often nice if you can. I may choose to fix it in a different
way, however, so it's often better to ask first whether I'd like a
patch, particularly before spending a lot of time hacking.

=head2 Patches

If you are sending patches, then please send either the entire code
that is being changed or the output of 'diff -Bub'.  Please also note
what version the patch is against. I tend to apply all patches manually,
so I'm more interested in being able to see what you're doing than in
being able to apply the patch cleanly. Code formatting isn't an issue,
as I automagically run perltidy against the source after any changes,
so please format for clarity.

Patches have a much better chance of being applied if they are small.
People often think that it's better for me to get one patch with a bunch
of fixes. It's not. I'd much rather get 100 small patches that can be
applied one by one. A change that I can make and release in five minutes
is always better than one that needs a couple of hours to ponder and work

I often reject patches that I don't like. Please don't take it personally.
I also like time to think about the wider implications of changes. Often
a I<lot> of time. Feel free to remind me about things that I may have
forgotten about, but as long as they're on I will get around
to them eventually.

=head2 Feature Requests

Wish-list requests are fine, although you should probably discuss them
on the mailing list (or equivalent) with others first. There's quite
often a plugin somewhere that already does what you want.

In general I am much more open to discussion on how best to provide the
flexibility for you to make your Cool New Feature(tm) a plugin rather
than adding it to Class::DBI itself.

For the most part the core of Class::DBI already has most of the
functionality that I believe it will ever need (and some more besides,
that will probably be split off at some point). Most other things are much
better off as plugins, with a separate life on CPAN or elsewhere (and with
me nowhere near the critical path). Most of the ongoing work on Class::DBI
is about making life easier for people to write extensions - whether
they're local to your own codebase or released for wider consumption.

=head1 SUPPORT

Support for Class::DBI is mostly via the mailing list.

To join the list, or read the archives, visit

There is also a Class::DBI wiki at

The wiki contains much information that should probably be in these docs
but isn't yet. (See above if you want to help to rectify this.)

As mentioned above, I don't follow the list or the wiki, so if you want
to contact me individually, then you'll have to track me down personally.

There are lots of 3rd party subclasses and plugins available.
For a list of the ones on CPAN see:

An article on Class::DBI was published on a while ago. It's
slightly out of date , but it's a good introduction:

The wiki has numerous references to other articles, presentations etc. provides a document comparing a variety
of different approaches to database persistence, such as Class::DBI,
Alazabo, Tangram, SPOPS etc.

=head1 LICENSE

This library is free software; you can redistribute it and/or modify
it under the same terms as Perl itself.

=head1 SEE ALSO

Class::DBI is built on top of L<Ima::DBI>, L<DBIx::ContextualFetch>,
L<Class::Accessor> and L<Class::Data::Inheritable>. The innards and
much of the interface are easier to understand if you have an idea of
how they all work as well.