package Class::DBI::Plugin::Pager; use strict; use warnings; use Carp; use UNIVERSAL::require; use SQL::Abstract; use base qw( Data::Page Class::Data::Inheritable ); use vars qw( $VERSION ); $VERSION = '0.6_3'; # D::P inherits from Class::Accessor::Chained::Fast __PACKAGE__->mk_accessors( qw( _where abstract_attr per_page page _order_by _cdbi_app ) ); __PACKAGE__->mk_classdata( '_syntax' ); __PACKAGE__->mk_classdata( '_pager_class' ); =head1 NAME Class::DBI::Plugin::Pager - paged queries for CDBI =head1 DESCRIPTION Adds a pager method to your class that can query using SQL::Abstract where clauses, and limit the number of rows returned to a specific subset. =head1 SYNOPSIS package CD; use base 'Class::DBI'; use Class::DBI::Plugin::AbstractCount; # pager needs this use Class::DBI::Plugin::Pager; # or to use a different syntax # use Class::DBI::Plugin::Pager::RowsTo; __PACKAGE__->set_db(...); # in a nearby piece of code... use CD; # see SQL::Abstract for how to specify the query my $where = { ... }; my $order_by => [ qw( foo bar ) ]; # bit by bit: my $pager = CD->pager; $pager->per_page( 10 ); $pager->page( 3 ); $pager->where( $where ); $pager->order_by( $order_by ); $pager->set_syntax( 'RowsTo' ); my @cds = $pager->search_where; # or all at once my $pager = CD->pager( $where, $order_by, 10, 3 ); my @cds = $pager->search_where; # or my $pager = CD->pager; my @cds = $pager->search_where( $where, $order_by, 10, 3 ); # $pager isa Data::Page # @cds contains the CDs just for the current page =head1 METHODS =over =item import Loads the C method into the CDBI app. =cut sub import { my ( $class ) = @_; # the pager class or subclass __PACKAGE__->_pager_class( $class ); my $caller; # find the app foreach my $level ( 0 .. 10 ) { $caller = caller( $level ); last if UNIVERSAL::isa( $caller, 'Class::DBI' ) } croak( "can't find the CDBI app" ) unless $caller; no strict 'refs'; *{"$caller\::pager"} = \&pager; } =item pager( [$where, [$abstract_attr]], [$order_by], [$per_page], [$page], [$syntax] ) Also accepts named arguments: where => $where, abstract_attr => $attr, order_by => $order_by, per_page => $per_page, page => $page, syntax => $syntax Returns a pager object. This subclasses L. Note that for positional arguments, C<$abstract_attr> can only be passed if preceded by a C<$where> argument. C<$abstract_attr> can contain the C<$order_by> setting (just as in L). =over 4 =item configuration The named arguments all exist as get/set methods. =over 4 =item where A hashref specifying the query. See L. =item abstract_attr A hashref specifying extra options to be passed through to the L constructor. =item order_by Single column name or arrayref of column names for the ORDER BY clause. Defaults to the primary key(s) if not set. =item per_page Number of results per page. Defaults to 10, b if using the positional arguments style, and supplying the C attribute, C must also be supplied. =item page The pager will retrieve results just for this page. Defaults to 1. =item syntax Change the way the 'limit' clause is constructed. See C. Default is C. =back =back =cut sub pager { my $cdbi = shift; my $class = __PACKAGE__->_pager_class; my $self = bless {}, $class; $self->_cdbi_app( $cdbi ); # This has to come before _init, so the caller can choose to set the syntax # instead. But don't auto-set if we're a subclass. $self->auto_set_syntax if $class eq __PACKAGE__; $self->_init( @_ ); return $self; } # _init is also called by results, so preserve any existing settings if # new settings are not provided sub _init { my $self = shift; return unless @_; my ( $where, $abstract_attr, $order_by, $per_page, $page, $syntax, $named_args ); # I wish I'd never implemented positional arguments in the first place! # Does anyone use this? if ( @_ % 2 == 0 ) { # _might_ be named args my %args = @_; $where = $args{where}; $abstract_attr = $args{abstract_attr}; $order_by = $args{order_by}; $per_page = $args{per_page}; $page = $args{page}; $syntax = $args{syntax}; $named_args = $where || $abstract_attr || $order_by || $per_page || $page || $syntax; } #if ( ref( $_[0] ) or $_[0] =~ /^\d+$/ ) unless ( $named_args ) { $where = shift if ref $_[0]; # SQL::Abstract accepts a hashref or an arrayref $abstract_attr = shift if ref $_[0] eq 'HASH'; $order_by = shift if ( @_ and $_[0] !~ /^\d+$/ ); $per_page = shift if ( @_ and $_[0] =~ /^\d+$/ ); $page = shift if ( @_ and $_[0] =~ /^\d+$/ ); $syntax = shift; } # Emulate AbstractSearch's search_where ordering -VV 20041209 $order_by = delete $$abstract_attr{order_by} if ($abstract_attr and !$order_by); $per_page ||= $self->per_page; $page ||= $self->page; $self->per_page( $per_page || 10 ); # if $per_page; $self->set_syntax( $syntax ) if $syntax; $self->abstract_attr( $abstract_attr )if $abstract_attr; $self->where( $where ) if $where; $self->order_by( $order_by ) if $order_by; $self->page( $page || 1 ); # if $page; } =item where( [ $where ] ) Translates accessor names into column names (or is it vice versa?). The original C<$where> can be retrieved via C<_where()>. =cut sub where { my ( $self, $where ) = @_; $where ||= $self->_where; $self->_where( $where ); my %columns = $self->_real_col_names( keys %$where ); # keys are candidates, values are real names my %munged_where = map { $columns{ $_ } => $where->{ $_ } } keys %columns; return \%munged_where; } =item order_by( [ $order_by ] ) Does the same translation as C. =cut sub order_by { my ( $self, $order_by ) = @_; $order_by ||= $self->_order_by; $self->_order_by( $order_by ); my @candidates = ref( $order_by ) ? @$order_by : $order_by ? ( $order_by ) : (); my %columns = $self->_real_col_names( @candidates ); return [ map { "$_" } values %columns ]; } sub _real_col_names { my ( $self, @candidates ) = @_; my $cdbi = $self->_cdbi_app; my %cols = (); foreach my $candidate ( @candidates ) { # lifted from Class::DBI::Plugin::CountSearch my $column = $cdbi->find_column( $candidate ) || ( List::Util::first { $_->accessor eq $candidate } $cdbi->columns ) || $cdbi->_croak("'$candidate' is not a column of $cdbi"); $cols{ $candidate } = $column; } return %cols; } =item add_attr( %attributes ) Add to (or override) the L attributes: $pager->add_attr( cmp => 'like' ); =cut sub add_attr { my ( $self, %attr ) = @_; my $abs_attr = $self->abstract_attr; $abs_attr->{ $_ } = $attr{ $_ } for keys %attr; $self->abstract_attr( $abs_attr ); } =item search_where Retrieves results from the pager. Accepts the same arguments as the C method. =cut # like CDBI::AbstractSearch::search_where, with extra limitations sub search_where { my $self = shift; $self->_init( @_ ); $self->_setup_pager; my $cdbi = $self->_cdbi_app; my $order_by = $self->order_by || [ $cdbi->primary_columns ]; my $where = $self->where; my $syntax = $self->_syntax || $self->set_syntax; my $limit_phrase = $self->$syntax; my $sql = SQL::Abstract->new( %{ $self->abstract_attr || {} } ); $order_by = [ $order_by ] unless ref $order_by; my ( $phrase, @bind ) = $sql->where( $where, $order_by ); $phrase .= ' ' . $limit_phrase; $phrase =~ s/^\s*WHERE\s*//i; return $cdbi->retrieve_from_sql( $phrase, @bind ); } =item retrieve_all Convenience method, generates a WHERE clause that matches all rows from the table. Accepts the same arguments as the C or C methods, except that no WHERE clause should be specified. Note that the argument parsing routine called by the C method cannot cope with positional arguments that lack a WHERE clause, so either use named arguments, or the 'bit by bit' approach, or pass the arguments directly to C. =cut sub retrieve_all { my $self = shift; my $get_all = { 1 => 1 }; unless ( @_ ) { # already set pager up via method calls $self->where( $get_all ); return $self->search_where; } my @args = ( ref( $_[0] ) or $_[0] =~ /^\d+$/ ) ? ( $get_all, @_ ) : # send an array ( where => $get_all, @_ ); # send a hash return $self->search_where( @args ); } =item retrieve_all_sorted_by( $order ) Useful for L. =cut sub retrieve_all_sorted_by { my ( $self, $order ) = @_; return $self->retrieve_all( { order_by => $order } ); } sub _setup_pager { my ( $self ) = @_; my $where = $self->where || croak( 'must set a query before retrieving results' ); my $per_page = $self->per_page || croak( 'no. of entries per page not specified' ); my $cdbi = $self->_cdbi_app; my $count = $cdbi->count_search_where( $where, $self->abstract_attr ); #my $count = $self->_get_count; my $page = $self->page || 1; $self->total_entries( $count ); $self->entries_per_page( $per_page ); $self->current_page( $page ); croak( 'Fewer than one entry per page!' ) if $self->entries_per_page < 1; $self->current_page( $self->first_page ) unless defined $self->current_page; $self->current_page( $self->first_page ) if $self->current_page < $self->first_page; $self->current_page( $self->last_page ) if $self->current_page > $self->last_page; } =for retiring sub _get_count { my ( $self ) = @_; my $where = $self->where || croak( 'must set a query before retrieving results' ); my $cdbi = $self->_cdbi_app; # Class::DBI::Plugin::AbstractCount can handle SQL::Abstract attributes return $cdbi->count_search_where( $where, $self->abstract_attr ) if $cdbi->can( 'count_search_where' ); # Class::DBI::Plugin::CountSearch correctly handles aliased column accessors # (i.e. accessors for columns where the accessor has a different name from the # column). But it can't handle SQL::Abstract attributes. croak( 'no way to count total entries' ) unless $cdbi->can( 'count_search' ); croak( 'Class::DBI::Plugin::CountSearch does not handle SQL::Abstract attributes - ' . 'use Class::DBI::Plugin::AbstractCount instead' ) if $self->abstract_attr; my @cols = keys %$where; croak('no keys in where clause') unless @cols > 0; # unqualified count return $cdbi->count_search if ( @cols == 1 and $cols[0] eq '1' ); # count with search fields my @args; foreach my $col ( @cols ) { my $expr = $where->{$col}; croak( "code doesn't handle anything but 'like'!" ) unless keys %$expr == 1 and $expr->{like}; push @args, $col => $expr->{like}; } return $cdbi->count_search_like( @args ); } =cut # SQL::Abstract::_recurse_where eats the WHERE clause #sub where { # my ( $self, $where_ref ) = @_; # # return $self->_where unless $where_ref; # # my $where_copy; # # if ( ref( $where_ref ) eq 'HASH' ) { # $where_copy = { %$where_ref }; # } # elsif ( ref( $where_ref ) eq 'ARRAY' ) # { # $where_copy = [ @$where_ref ]; # } # else # { # die "WHERE clause [$where_ref] must be specified as an ARRAYREF or HASHREF"; # } # # # this will get eaten, but the caller's value is now protected # $self->_where( $where_copy ); #} =item set_syntax( [ $name || $class || $coderef ] ) Changes the syntax used to generate the C or other phrase that restricts the results set to the required page. The syntax is implemented as a method called on the pager, which can be queried to provide the C<$rows> and C<$offset> parameters (see the subclasses included in this distribution). =over 4 =item $class A class with a C method. =item $name Name of a class in the C namespace, which has a C method. =item $coderef Will be called as a method on the pager object, so receives the pager as its argument. =item (no args) Called without args, will default to C, which causes L to be used. =back =cut sub set_syntax { my ( $proto, $syntax ) = @_; # pick up default from subclass, or load from LimitOffset $syntax ||= $proto->can( 'make_limit' ); $syntax ||= 'LimitOffset'; if ( ref( $syntax ) eq 'CODE' ) { $proto->_syntax( $syntax ); return $syntax; } my $format_class = $syntax =~ '::' ? $syntax : "Class::DBI::Plugin::Pager::$syntax"; $format_class->require || croak "error loading $format_class: $UNIVERSAL::require::ERROR"; my $formatter = $format_class->can( 'make_limit' ) || croak "no make_limit method in $format_class"; $proto->_syntax( $formatter ); return $formatter; } =item auto_set_syntax This is called automatically when you call C, and attempts to set the syntax automatically. If you are using a subclass of the pager, this method will not be called. Will C if using Oracle or DB2, since there is no simple syntax for limiting the results set. DB2 has a C keyword, but that seems to apply to a cursor and I don't know if there is a cursor available to the pager. There should probably be others to add to the unsupported list. Supports the following drivers: DRIVER CDBI::P::Pager subclass my %supported = ( pg => 'LimitOffset', mysql => 'LimitOffset', # older versions need LimitXY sqlite => 'LimitOffset', # or LimitYX interbase => 'RowsTo', firebird => 'RowsTo', ); Older versions of MySQL should use the LimitXY syntax. You'll need to set it manually, either by C, or by passing C 'LimitXY'> to a method call, or call C directly. Any driver not in the supported or unsupported lists defaults to LimitOffset. Any additions to the supported and unsupported lists gratefully received. =cut sub auto_set_syntax { my ( $self ) = @_; # not an exhaustive list my %not_supported = ( oracle => 'Oracle', db2 => 'DB2', ); # additions welcome my %supported = ( pg => 'LimitOffset', mysql => 'LimitOffset', # older versions need LimitXY sqlite => 'LimitOffset', # or LimitYX interbase => 'RowsTo', firebird => 'RowsTo', ); my $cdbi = $self->_cdbi_app; my $driver = lc( $cdbi->__driver ); die __PACKAGE__ . " can't build limit clauses for $not_supported{ $driver }" if $not_supported{ $driver }; $self->set_syntax( $supported{ $driver } || 'LimitOffset' ); } 1; __END__ #=for notes # #Would this work? # #with $limit and $offset defined. # #my $last = $limit + $offset # #my $order_by_str = join( ', ', @$order_by ) # #$cdbi->set_sql( emulate_limit => <<''); # SELECT * FROM ( # SELECT TOP $limit * FROM ( # SELECT TOP $last __ESSENTIAL__ # FROM __TABLE__ # ORDER BY $order_by_str ASC # ) AS foo ORDER BY $order_by_str DESC # ) AS bar ORDER BY $order_by_str ASC # # #e.g. MS Access (thanks Emanuele Zeppieri) # #to add LIMIT/OFFSET to this query: # #SELECT my_column #FROM my_table #ORDER BY my_column ASC # #say with the values LIMIT=5 OFFSET=10, you have to resort to the TOP #clause and re-write it this way: # #SELECT * FROM ( # SELECT TOP 5 * FROM ( # SELECT TOP 15 my_column # FROM my_table # ORDER BY my_column ASC # ) AS foo ORDER BY my_column DESC #) AS bar ORDER BY my_column ASC # #=cut =back =head2 SUBCLASSING The 'limit' syntax can be set by using a subclass, e.g. use Class::DBI::Plugin::Pager::RowsTo; instead of setting at runtime. A subclass looks like this: package Class::DBI::Plugin::Pager::RowsTo; use base 'Class::DBI::Plugin::Pager'; sub make_limit { my ( $self ) = @_; my $offset = $self->skipped; my $rows = $self->entries_per_page; my $last = $rows + $offset; return "ROWS $offset TO $last"; } 1; You can omit the C and switch syntax by calling C<$pager-Eset_syntax( 'RowsTo' )>. Or you can leave in the C and still say C<$pager-Eset_syntax( 'RowsTo' )>, because in this case the class is Cd and the C in the base class doesn't get called. Or something. At any rate, It Works. The subclasses implement the following LIMIT syntaxes: =over =item Class::DBI::Plugin::Pager::LimitOffset LIMIT $rows OFFSET $offset This is the default if your driver is not in the list of known drivers. This should work for PostgreSQL, more recent MySQL, SQLite, and maybe some others. =item Class::DBI::Plugin::LimitXY LIMIT $offset, $rows Older versions of MySQL. =item Class::DBI::Plugin::LimitYX LIMIT $rows, $offset SQLite. =item Class::DBI::Plugin::RowsTo ROWS $offset TO $offset + $rows InterBase, also FireBird, maybe others? =back =head1 TODO I've only used this on an older version of MySQL. Reports of this thing working (or not) elsewhere would be useful. It should be possible to use C to build the complex queries required by some databases to emulate LIMIT (see notes in source). =head1 CAVEATS This class can't implement the subselect mechanism required by some databases to emulate the LIMIT phrase, because it only has access to the WHERE clause, not the whole SQL statement. At the moment. Integration with L is on the horizon. Each query issues two requests to the database - the first to count the entire result set, the second to retrieve the required subset of results. If your tables are small it may be quicker to use L. This class will use L to count the result set, if it is loaded, otherwise it will look for L. The latter is more limited in its search capabilities but does correctly handle Class::DBI accessor functions whose name differs from that of the column. The C clause means the database has to retrieve (internally) and sort the entire results set, before chopping out the requested subset. It's probably a good idea to have an index on the column(s) used to order the results. For huge tables, this approach to paging may be too inefficient. =head1 DEPENDENCIES L, L, L, L, L, L. =head1 SEE ALSO L does a similar job, but retrieves the entire results set into memory before chopping out the page you want. =head1 BUGS Please report all bugs via the CPAN Request Tracker at L. =head1 COPYRIGHT AND LICENSE Copyright 2004 by David Baird. This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself. =head1 AUTHOR David Baird, C