X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FResultSet.pm;h=a2e3a4cadefee218ea56f860fb07ac18896ab2ea;hb=1e4f9fb3b8bd1f54518bc2942554099356fa6524;hp=70025dbaf187c871349d3862b9b76caa6ee4bbea;hpb=65ad59ab00bf973e5fd7a375927fd831e6b6e6dc;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/ResultSet.pm b/lib/DBIx/Class/ResultSet.pm index 70025db..a2e3a4c 100644 --- a/lib/DBIx/Class/ResultSet.pm +++ b/lib/DBIx/Class/ResultSet.pm @@ -5,7 +5,7 @@ use warnings; use base qw/DBIx::Class/; use DBIx::Class::Carp; use DBIx::Class::ResultSetColumn; -use Scalar::Util qw/blessed weaken/; +use Scalar::Util qw/blessed weaken reftype/; use Try::Tiny; use Data::Compare (); # no imports!!! guard against insane architecture @@ -25,6 +25,10 @@ use overload 'bool' => "_bool", fallback => 1; +# this is real - CDBICompat overrides it with insanity +# yes, prototype won't matter, but that's for now ;) +sub _bool () { 1 } + __PACKAGE__->mk_group_accessors('simple' => qw/_result_class result_source/); =head1 NAME @@ -137,11 +141,15 @@ another. =head3 Resolving conditions and attributes -When a resultset is chained from another resultset, conditions and -attributes with the same keys need resolving. +When a resultset is chained from another resultset (ie: +Csearch(\%extra_cond, \%attrs)>), conditions +and attributes with the same keys need resolving. + +If any of L, L, L are present, they reset the +original selection, and start the selection "clean". -L, L, L, L attributes are merged -into the existing ones from the original resultset. +The L, L, L, L, L attributes +are merged into the existing ones from the original resultset. The L and L attributes, and any search conditions, are merged with an SQL C to the existing condition from the original @@ -389,11 +397,11 @@ sub search_rs { my $cache; my %safe = (alias => 1, cache => 1); if ( ! List::Util::first { !$safe{$_} } keys %$call_attrs and ( - ! defined $_[0] + ! defined $call_cond or - ref $_[0] eq 'HASH' && ! keys %{$_[0]} + ref $call_cond eq 'HASH' && ! keys %$call_cond or - ref $_[0] eq 'ARRAY' && ! @{$_[0]} + ref $call_cond eq 'ARRAY' && ! @$call_cond )) { $cache = $self->get_cache; } @@ -1069,8 +1077,8 @@ sub single { $attrs->{where}, $attrs )]; return undef unless @$data; - $self->{stashed_rows} = [ $data ]; - $self->_construct_objects->[0]; + $self->{_stashed_rows} = [ $data ]; + $self->_construct_results->[0]; } @@ -1239,39 +1247,47 @@ sub next { return ($self->all)[0]; } - return shift(@{$self->{stashed_objects}}) if @{ $self->{stashed_objects}||[] }; + return shift(@{$self->{_stashed_results}}) if @{ $self->{_stashed_results}||[] }; - $self->{stashed_objects} = $self->_construct_objects + $self->{_stashed_results} = $self->_construct_results or return undef; - return shift @{$self->{stashed_objects}}; + return shift @{$self->{_stashed_results}}; } -# Constructs as many objects as it can in one pass while respecting +# Constructs as many results as it can in one pass while respecting # cursor laziness. Several modes of operation: # -# * Always builds everything present in @{$self->{stashed_rows}} +# * Always builds everything present in @{$self->{_stashed_rows}} # * If called with $fetch_all true - pulls everything off the cursor and -# builds all objects in one pass +# builds all result structures (or objects) in one pass # * If $self->_resolved_attrs->{collapse} is true, checks the order_by # and if the resultset is ordered properly by the left side: # * Fetches stuff off the cursor until the "master object" changes, -# and saves the last extra row (if any) in @{$self->{stashed_rows}} +# and saves the last extra row (if any) in @{$self->{_stashed_rows}} # OR # * Just fetches, and collapses/constructs everything as if $fetch_all # was requested (there is no other way to collapse except for an # eager cursor) # * If no collapse is requested - just get the next row, construct and # return -sub _construct_objects { +sub _construct_results { my ($self, $fetch_all) = @_; my $rsrc = $self->result_source; my $attrs = $self->_resolved_attrs; - if (!$fetch_all and ! $attrs->{order_by} and $attrs->{collapse}) { + if ( + ! $fetch_all + and + ! $attrs->{order_by} + and + $attrs->{collapse} + and + my @pcols = $rsrc->primary_columns + ) { # default order for collapsing unless the user asked for something - $attrs->{order_by} = [ map { join '.', $attrs->{alias}, $_} $rsrc->primary_columns ]; + $attrs->{order_by} = [ map { join '.', $attrs->{alias}, $_} @pcols ]; $attrs->{_ordered_for_collapse} = 1; $attrs->{_order_is_artificial} = 1; } @@ -1279,20 +1295,19 @@ sub _construct_objects { my $cursor = $self->cursor; # this will be used as both initial raw-row collector AND as a RV of - # _construct_objects. Not regrowing the array twice matters a lot... - # a suprising amount actually - my $rows = (delete $self->{stashed_rows}) || []; + # _construct_results. Not regrowing the array twice matters a lot... + # a surprising amount actually + my $rows = delete $self->{_stashed_rows}; + + my $did_fetch_all = $fetch_all; + if ($fetch_all) { # FIXME SUBOPTIMAL - we can do better, cursor->next/all (well diff. methods) should return a ref - $rows = [ @$rows, $cursor->all ]; + $rows = [ ($rows ? @$rows : ()), $cursor->all ]; } - elsif (!$attrs->{collapse}) { - # FIXME SUBOPTIMAL - we can do better, cursor->next/all (well diff. methods) should return a ref - push @$rows, do { my @r = $cursor->next; @r ? \@r : () } - unless @$rows; - } - else { - $attrs->{_ordered_for_collapse} ||= (!$attrs->{order_by}) ? undef : do { + elsif( $attrs->{collapse} ) { + + $attrs->{_ordered_for_collapse} = (!$attrs->{order_by}) ? 0 : do { my $st = $rsrc->schema->storage; my @ord_cols = map { $_->[0] } @@ -1319,74 +1334,131 @@ sub _construct_objects { { $colinfos->{$_}{-colname} => $colinfos->{$_} } @ord_cols })) ? 1 : 0; - }; + } unless defined $attrs->{_ordered_for_collapse}; + + if (! $attrs->{_ordered_for_collapse}) { + $did_fetch_all = 1; - if ($attrs->{_ordered_for_collapse}) { - push @$rows, do { my @r = $cursor->next; @r ? \@r : () }; + # instead of looping over ->next, use ->all in stealth mode + # *without* calling a ->reset afterwards + # FIXME ENCAPSULATION - encapsulation breach, cursor method additions pending + if (! $cursor->{_done}) { + $rows = [ ($rows ? @$rows : ()), $cursor->all ]; + $cursor->{_done} = 1; + } } - # instead of looping over ->next, use ->all in stealth mode - # *without* calling a ->reset afterwards - # FIXME - encapsulation breach, got to be a better way - elsif (! $cursor->{_done}) { - push @$rows, $cursor->all; - $cursor->{_done} = 1; - $fetch_all = 1; + } + + if (! $did_fetch_all and ! @{$rows||[]} ) { + # FIXME SUBOPTIMAL - we can do better, cursor->next/all (well diff. methods) should return a ref + if (scalar (my @r = $cursor->next) ) { + $rows = [ \@r ]; } } - return undef unless @$rows; + return undef unless @{$rows||[]}; - my $res_class = $self->result_class; - my $inflator = $res_class->can ('inflate_result') - or $self->throw_exception("Inflator $res_class does not provide an inflate_result() method"); + my @extra_collapser_args; + if ($attrs->{collapse} and ! $did_fetch_all ) { + + @extra_collapser_args = ( + # FIXME SUBOPTIMAL - we can do better, cursor->next/all (well diff. methods) should return a ref + sub { my @r = $cursor->next or return; \@r }, # how the collapser gets more rows + ($self->{_stashed_rows} = []), # where does it stuff excess + ); + } + + # hotspot - skip the setter + my $res_class = $self->_result_class; + + my $inflator_cref = $self->{_result_inflator}{cref} ||= do { + $res_class->can ('inflate_result') + or $self->throw_exception("Inflator $res_class does not provide an inflate_result() method"); + }; my $infmap = $attrs->{as}; - if (!$attrs->{collapse} and $attrs->{_single_object_inflation}) { - # construct a much simpler array->hash folder for the one-table cases right here + $self->{_result_inflator}{is_core_row} = ( ( + $inflator_cref + == + ( \&DBIx::Class::Row::inflate_result || die "No ::Row::inflate_result() - can't happen" ) + ) ? 1 : 0 ) unless defined $self->{_result_inflator}{is_core_row}; + + $self->{_result_inflator}{is_hri} = ( ( + ! $self->{_result_inflator}{is_core_row} + and + $inflator_cref == ( + require DBIx::Class::ResultClass::HashRefInflator + && + DBIx::Class::ResultClass::HashRefInflator->can('inflate_result') + ) + ) ? 1 : 0 ) unless defined $self->{_result_inflator}{is_hri}; + + + if (! $attrs->{_related_results_construction}) { + # construct a much simpler array->hash folder for the one-table cases right here + if ($self->{_result_inflator}{is_hri}) { + for my $r (@$rows) { + $r = { map { $infmap->[$_] => $r->[$_] } 0..$#$infmap }; + } + } # FIXME SUBOPTIMAL this is a very very very hot spot # while rather optimal we can *still* do much better, by - # building a smarter [Row|HRI]::inflate_result(), and + # building a smarter Row::inflate_result(), and # switch to feeding it data via a much leaner interface # # crude unscientific benchmarking indicated the shortcut eval is not worth it for # this particular resultset size - if (@$rows < 60) { - my @as_idx = 0..$#$infmap; + elsif (@$rows < 60) { for my $r (@$rows) { - $r = $inflator->($res_class, $rsrc, { map { $infmap->[$_] => $r->[$_] } @as_idx } ); + $r = $inflator_cref->($res_class, $rsrc, { map { $infmap->[$_] => $r->[$_] } (0..$#$infmap) } ); } } else { eval sprintf ( - '$_ = $inflator->($res_class, $rsrc, { %s }) for @$rows', + '$_ = $inflator_cref->($res_class, $rsrc, { %s }) for @$rows', join (', ', map { "\$infmap->[$_] => \$_->[$_]" } 0..$#$infmap ) ); } } - else { - $self->{_row_parser} ||= eval sprintf 'sub { %s }', $rsrc->_mk_row_parser({ + # Special-case multi-object HRI (we always prune, and there is no $inflator_cref pass) + elsif ($self->{_result_inflator}{is_hri}) { + ( $self->{_row_parser}{hri} ||= $rsrc->_mk_row_parser({ + eval => 1, inflate_map => $infmap, selection => $attrs->{select}, collapse => $attrs->{collapse}, premultiplied => $attrs->{_main_source_premultiplied}, - }) or die $@; + hri_style => 1, + prune_null_branches => 1, + }) )->($rows, @extra_collapser_args); + } + # Regular multi-object + else { + my $parser_type = $self->{_result_inflator}{is_core_row} ? 'classic_pruning' : 'classic_nonpruning'; - # modify $rows in-place, shrinking/extending as necessary - $self->{_row_parser}->($rows, $fetch_all ? () : ( - # FIXME SUBOPTIMAL - we can do better, cursor->next/all (well diff. methods) should return a ref - sub { my @r = $cursor->next or return; \@r }, # how the collapser gets more rows - ($self->{stashed_rows} = []), # where does it stuff excess - )); + ( $self->{_row_parser}{$parser_type} ||= $rsrc->_mk_row_parser({ + eval => 1, + inflate_map => $infmap, + selection => $attrs->{select}, + collapse => $attrs->{collapse}, + premultiplied => $attrs->{_main_source_premultiplied}, + prune_null_branches => $self->{_result_inflator}{is_core_row}, + }) )->($rows, @extra_collapser_args); - $_ = $inflator->($res_class, $rsrc, @$_) for @$rows; + $_ = $inflator_cref->($res_class, $rsrc, @$_) for @$rows; } - # CDBI compat stuff - if ($attrs->{record_filter}) { - $_ = $attrs->{record_filter}->($_) for @$rows; - } + # The @$rows check seems odd at first - why wouldn't we want to warn + # regardless? The issue is things like find() etc, where the user + # *knows* only one result will come back. In these cases the ->all + # is not a pessimization, but rather something we actually want + carp_unique( + 'Unable to properly collapse has_many results in iterator mode due ' + . 'to order criteria - performed an eager cursor slurp underneath. ' + . 'Consider using ->all() instead' + ) if ( ! $fetch_all and @$rows > 1 ); return $rows; } @@ -1428,14 +1500,22 @@ in the original source class will not run. sub result_class { my ($self, $result_class) = @_; if ($result_class) { - unless (ref $result_class) { # don't fire this for an object - $self->ensure_class_loaded($result_class); + + # don't fire this for an object + $self->ensure_class_loaded($result_class) + unless ref($result_class); + + if ($self->get_cache) { + carp_unique('Changing the result_class of a ResultSet instance with cached results is a noop - the cache contents will not be altered'); + } + # FIXME ENCAPSULATION - encapsulation breach, cursor method additions pending + elsif ($self->{cursor} && $self->{cursor}{_pos}) { + $self->throw_exception('Changing the result_class of a ResultSet instance with an active cursor is not supported'); } + $self->_result_class($result_class); - # THIS LINE WOULD BE A BUG - this accessor specifically exists to - # permit the user to set result class on one result set only; it only - # chains if provided to search() - #$self->{attrs}{result_class} = $result_class if ref $self; + + delete $self->{_result_inflator}; } $self->_result_class; } @@ -1531,7 +1611,7 @@ sub _count_rs { my $tmp_attrs = { %$attrs }; # take off any limits, record_filter is cdbi, and no point of ordering nor locking a count - delete @{$tmp_attrs}{qw/rows offset order_by record_filter for/}; + delete @{$tmp_attrs}{qw/rows offset order_by _related_results_construction record_filter for/}; # overwrite the selector (supplied by the storage) $tmp_attrs->{select} = $rsrc->storage->_count_select ($rsrc, $attrs); @@ -1553,7 +1633,7 @@ sub _count_subq_rs { my $sub_attrs = { %$attrs }; # extra selectors do not go in the subquery and there is no point of ordering it, nor locking it - delete @{$sub_attrs}{qw/collapse columns as select _prefetch_selector_range order_by for/}; + delete @{$sub_attrs}{qw/collapse columns as select _related_results_construction order_by for/}; # if we multi-prefetch we group_by something unique, as this is what we would # get out of the rs via ->next/->all. We *DO WANT* to clobber old group_by regardless @@ -1640,9 +1720,6 @@ sub _count_subq_rs { ->get_column ('count'); } -sub _bool { - return 1; -} =head2 count_literal @@ -1684,7 +1761,7 @@ sub all { $self->throw_exception("all() doesn't take any arguments, you probably wanted ->search(...)->all()"); } - delete @{$self}{qw/stashed_rows stashed_objects/}; + delete @{$self}{qw/_stashed_rows _stashed_results/}; if (my $c = $self->get_cache) { return @$c; @@ -1692,7 +1769,7 @@ sub all { $self->cursor->reset; - my $objs = $self->_construct_objects('fetch_all') || []; + my $objs = $self->_construct_results('fetch_all') || []; $self->set_cache($objs) if $self->{attrs}{cache}; @@ -1718,7 +1795,7 @@ another query. sub reset { my ($self) = @_; - delete @{$self}{qw/stashed_rows stashed_objects/}; + delete @{$self}{qw/_stashed_rows _stashed_results/}; $self->{all_cache_position} = 0; $self->cursor->reset; return $self; @@ -1759,7 +1836,7 @@ sub _rs_update_delete { my $attrs = { %{$self->_resolved_attrs} }; my $join_classifications; - my $existing_group_by = delete $attrs->{group_by}; + my ($existing_group_by) = delete @{$attrs}{qw(group_by _grouped_by_distinct)}; # do we need a subquery for any reason? my $needs_subq = ( @@ -1820,7 +1897,7 @@ sub _rs_update_delete { ); # make a new $rs selecting only the PKs (that's all we really need for the subq) - delete $attrs->{$_} for qw/collapse select _prefetch_selector_range as/; + delete $attrs->{$_} for qw/select as collapse _related_results_construction/; $attrs->{columns} = [ map { "$attrs->{alias}.$_" } @$idcols ]; $attrs->{group_by} = \ ''; # FIXME - this is an evil hack, it causes the optimiser to kick in and throw away the LEFT joins my $subrs = (ref $self)->new($rsrc, $attrs); @@ -2319,15 +2396,29 @@ sub new_result { my ($merged_cond, $cols_from_relations) = $self->_merge_with_rscond($values); - my %new = ( + my $new = $self->result_class->new({ %$merged_cond, - @$cols_from_relations + ( @$cols_from_relations ? (-cols_from_relations => $cols_from_relations) - : (), + : () + ), -result_source => $self->result_source, # DO NOT REMOVE THIS, REQUIRED - ); + }); - return $self->result_class->new(\%new); + if ( + reftype($new) eq 'HASH' + and + ! keys %$new + and + blessed($new) + ) { + carp_unique (sprintf ( + "%s->new returned a blessed empty hashref - a strong indicator something is wrong with its inheritance chain", + $self->result_class, + )); + } + + $new; } # _merge_with_rscond @@ -2996,7 +3087,6 @@ Returns a related resultset for the supplied relationship name. sub related_resultset { my ($self, $rel) = @_; - $self->{related_resultsets} ||= {}; return $self->{related_resultsets}{$rel} ||= do { my $rsrc = $self->result_source; my $rel_info = $rsrc->relationship_info($rel); @@ -3023,13 +3113,13 @@ sub related_resultset { #XXX - temp fix for result_class bug. There likely is a more elegant fix -groditi delete @{$attrs}{qw(result_class alias)}; - my $new_cache; + my $related_cache; if (my $cache = $self->get_cache) { - if ($cache->[0] && $cache->[0]->related_resultset($rel)->get_cache) { - $new_cache = [ map { @{$_->related_resultset($rel)->get_cache||[]} } - @$cache ]; - } + $related_cache = [ map + { @{$_->related_resultset($rel)->get_cache||[]} } + @$cache + ]; } my $rel_source = $rsrc->related_source($rel); @@ -3052,7 +3142,7 @@ sub related_resultset { where => $attrs->{where}, }); }; - $new->set_cache($new_cache) if $new_cache; + $new->set_cache($related_cache) if $related_cache; $new; }; } @@ -3192,7 +3282,7 @@ sub _chain_relationship { # ->_resolve_join as otherwise they get lost - captainL my $join = $self->_merge_joinpref_attr( $attrs->{join}, $attrs->{prefetch} ); - delete @{$attrs}{qw/join prefetch collapse group_by distinct select as columns +select +as +columns/}; + delete @{$attrs}{qw/join prefetch collapse group_by distinct _grouped_by_distinct select as columns +select +as +columns/}; my $seen = { %{ (delete $attrs->{seen_join}) || {} } }; @@ -3277,40 +3367,6 @@ sub _chain_relationship { return {%$attrs, from => $from, seen_join => $seen}; } -# FIXME - this needs to go live in Schema with the tree walker... or -# something -my $inflatemap_checker; -$inflatemap_checker = sub { - my ($rsrc, $relpaths) = @_; - - my $rels; - - for (@$relpaths) { - $_ =~ /^ ( [^\.]+ ) \. (.+) $/x - or next; - - push @{$rels->{$1}}, $2; - } - - for my $rel (keys %$rels) { - my $rel_rsrc = try { - $rsrc->related_source ($rel) - } catch { - $rsrc->throw_exception(sprintf( - "Inflation into non-existent relationship '%s' of '%s' requested, " - . "check the inflation specification (columns/as) ending in '...%s.%s'", - $rel, - $rsrc->source_name, - $rel, - ( sort { length($a) <=> length ($b) } @{$rels->{$rel}} )[0], - ))}; - - $inflatemap_checker->($rel_rsrc, $rels->{$rel}); - } - - return; -}; - sub _resolved_attrs { my $self = shift; return $self->{_attrs} if $self->{_attrs}; @@ -3382,14 +3438,6 @@ sub _resolved_attrs { } } - # validate the user-supplied 'as' chain - # folks get too confused by the (logical) exception message, need to - # go to some lengths to clarify the text - # - # FIXME - this needs to go live in Schema with the tree walker... or - # something - $inflatemap_checker->($source, \@as); - $attrs->{select} = \@sel; $attrs->{as} = \@as; @@ -3444,6 +3492,7 @@ sub _resolved_attrs { carp_unique ("Useless use of distinct on a grouped resultset ('distinct' is ignored when a 'group_by' is present)"); } else { + $attrs->{_grouped_by_distinct} = 1; # distinct affects only the main selection part, not what prefetch may # add below. $attrs->{group_by} = $source->storage->_group_over_selection ( @@ -3489,18 +3538,14 @@ sub _resolved_attrs { my @prefetch = $source->_resolve_prefetch( $prefetch, $alias, $join_map ); - # we need to somehow mark which columns came from prefetch - if (@prefetch) { - my $sel_end = $#{$attrs->{select}}; - $attrs->{_prefetch_selector_range} = [ $sel_end + 1, $sel_end + @prefetch ]; - } - push @{ $attrs->{select} }, (map { $_->[0] } @prefetch); push @{ $attrs->{as} }, (map { $_->[1] } @prefetch); } - if ( ! List::Util::first { $_ =~ /\./ } @{$attrs->{as}} ) { - $attrs->{_single_object_inflation} = 1; + if ( List::Util::first { $_ =~ /\./ } @{$attrs->{as}} ) { + $attrs->{_related_results_construction} = 1; + } + else { $attrs->{collapse} = 0; } @@ -3774,7 +3819,7 @@ sub STORABLE_freeze { # A cursor in progress can't be serialized (and would make little sense anyway) # the parser can be regenerated (and can't be serialized) - delete @{$to_serialize}{qw/cursor _row_parser/}; + delete @{$to_serialize}{qw/cursor _row_parser _result_inflator/}; # nor is it sensical to store a not-yet-fired-count pager if ($to_serialize->{pager} and ref $to_serialize->{pager}{total_entries} eq 'CODE') { @@ -3811,6 +3856,10 @@ sub throw_exception { } } +1; + +__END__ + # XXX: FIXME: Attributes docs need clearing up =head1 ATTRIBUTES @@ -3860,7 +3909,7 @@ syntax as outlined above. =over 4 -=item Value: \@columns +=item Value: \@columns | \%columns | $column =back @@ -3962,14 +4011,6 @@ an explicit list. =back -=head2 +as - -=over 4 - -Indicates additional column names for those added via L. See L. - -=back - =head2 as =over 4 @@ -4012,6 +4053,14 @@ use C instead: You can create your own accessors if required - see L for details. +=head2 +as + +=over 4 + +Indicates additional column names for those added via L. See L. + +=back + =head2 join =over 4 @@ -4075,7 +4124,7 @@ similarly for a third time). For e.g. will return a set of all artists that have both a cd with title 'Down to Earth' and a cd with title 'Popular'. -If you want to fetch related objects from other tables as well, see C +If you want to fetch related objects from other tables as well, see L below. NOTE: An internal join-chain pruner will discard certain joins while @@ -4086,185 +4135,133 @@ below. For more help on using joins with search, see L. -=head2 prefetch +=head2 collapse =over 4 -=item Value: ($rel_name | \@rel_names | \%rel_names) +=item Value: (0 | 1) =back -Contains one or more relationships that should be fetched along with -the main query (when they are accessed afterwards the data will -already be available, without extra queries to the database). This is -useful for when you know you will need the related objects, because it -saves at least one query: - - my $rs = $schema->resultset('Tag')->search( - undef, - { - prefetch => { - cd => 'artist' - } - } - ); - -The initial search results in SQL like the following: - - SELECT tag.*, cd.*, artist.* FROM tag - JOIN cd ON tag.cd = cd.cdid - JOIN artist ON cd.artist = artist.artistid - -L has no need to go back to the database when we access the -C or C relationships, which saves us two SQL statements in this -case. - -Simple prefetches will be joined automatically, so there is no need -for a C attribute in the above search. - -L can be used with the any of the relationship types and -multiple prefetches can be specified together. Below is a more complex -example that prefetches a CD's artist, its liner notes (if present), -the cover image, the tracks on that cd, and the guests on those -tracks. - - # Assuming: - My::Schema::CD->belongs_to( artist => 'My::Schema::Artist' ); - My::Schema::CD->might_have( liner_note => 'My::Schema::LinerNotes' ); - My::Schema::CD->has_one( cover_image => 'My::Schema::Artwork' ); - My::Schema::CD->has_many( tracks => 'My::Schema::Track' ); - - My::Schema::Artist->belongs_to( record_label => 'My::Schema::RecordLabel' ); - - My::Schema::Track->has_many( guests => 'My::Schema::Guest' ); +When set to a true value, indicates that any rows fetched from joined has_many +relationships are to be aggregated into the corresponding "parent" object. For +example, the resultset: + my $rs = $schema->resultset('CD')->search({}, { + '+columns' => [ qw/ tracks.title tracks.position / ], + join => 'tracks', + collapse => 1, + }); - my $rs = $schema->resultset('CD')->search( - undef, - { - prefetch => [ - { artist => 'record_label'}, # belongs_to => belongs_to - 'liner_note', # might_have - 'cover_image', # has_one - { tracks => 'guests' }, # has_many => has_many - ] - } - ); - -This will produce SQL like the following: - - SELECT cd.*, artist.*, record_label.*, liner_note.*, cover_image.*, - tracks.*, guests.* - FROM cd me - JOIN artist artist - ON artist.artistid = me.artistid - JOIN record_label record_label - ON record_label.labelid = artist.labelid - LEFT JOIN track tracks - ON tracks.cdid = me.cdid - LEFT JOIN guest guests - ON guests.trackid = track.trackid - LEFT JOIN liner_notes liner_note - ON liner_note.cdid = me.cdid - JOIN cd_artwork cover_image - ON cover_image.cdid = me.cdid - ORDER BY tracks.cd - -Now the C, C, C, C, -C, and C of the CD will all be available through the -relationship accessors without the need for additional queries to the -database. - -However, there is one caveat to be observed: it can be dangerous to -prefetch more than one L -relationship on a given level. e.g.: - - my $rs = $schema->resultset('CD')->search( - undef, - { - prefetch => [ - 'tracks', # has_many - { cd_to_producer => 'producer' }, # has_many => belongs_to (i.e. m2m) - ] - } - ); - -The collapser currently can't identify duplicate tuples for multiple -L relationships and as a -result the second L -relation could contain redundant objects. - -=head3 Using L with L - -L implies a L with the equivalent argument, and is -properly merged with any existing L specification. So the -following: - - my $rs = $schema->resultset('CD')->search( - {'record_label.name' => 'Music Product Ltd.'}, - { - join => {artist => 'record_label'}, - prefetch => 'artist', - } - ); +While executing the following query: -... will work, searching on the record label's name, but only -prefetching the C. + SELECT me.*, tracks.title, tracks.position + FROM cd me + LEFT JOIN track tracks + ON tracks.cdid = me.cdid -=head3 Using L with L / L / L / L +Will return only as many objects as there are rows in the CD source, even +though the result of the query may span many rows. Each of these CD objects +will in turn have multiple "Track" objects hidden behind the has_many +generated accessor C. Without C<< collapse => 1 >>, the return values +of this resultset would be as many CD objects as there are tracks (a "Cartesian +product"), with each CD object containing exactly one of all fetched Track data. -L implies a L/L with the fields of the -prefetched relations. So given: +When a collapse is requested on a non-ordered resultset, an order by some +unique part of the main source (the left-most table) is inserted automatically. +This is done so that the resultset is allowed to be "lazy" - calling +L<< $rs->next|/next >> will fetch only as many rows as it needs to build the next +object with all of its related data. - my $rs = $schema->resultset('CD')->search( - undef, - { - select => ['cd.title'], - as => ['cd_title'], - prefetch => 'artist', - } - ); +If an L is already declared, and orders the resultset in a way that +makes collapsing as described above impossible (e.g. C<< ORDER BY +has_many_rel.column >> or C), DBIC will automatically +switch to "eager" mode and slurp the entire resultset before consturcting the +first object returned by L. -The L becomes: C<'cd.title', 'artist.*'> and the L -becomes: C<'cd_title', 'artist.*'>. +Setting this attribute on a resultset that does not join any has_many +relations is a no-op. -=head3 CAVEATS +For a more in-depth discussion, see L. -Prefetch does a lot of deep magic. As such, it may not behave exactly -as you might expect. +=head2 prefetch =over 4 -=item * - -Prefetch uses the L to populate the prefetched relationships. This -may or may not be what you want. +=item Value: ($rel_name | \@rel_names | \%rel_names) -=item * +=back -If you specify a condition on a prefetched relationship, ONLY those -rows that match the prefetched condition will be fetched into that relationship. -This means that adding prefetch to a search() B what is returned by -traversing a relationship. So, if you have C<< Artist->has_many(CDs) >> and you do +This attribute is a shorthand for specifying a L spec, adding all +columns from the joined related sources as L and setting +L to a true value. For example, the following two queries are +equivalent: - my $artist_rs = $schema->resultset('Artist')->search({ - 'cds.year' => 2008, - }, { - join => 'cds', + my $rs = $schema->resultset('Artist')->search({}, { + prefetch => { cds => ['genre', 'tracks' ] }, }); - my $count = $artist_rs->first->cds->count; +and - my $artist_rs_prefetch = $artist_rs->search( {}, { prefetch => 'cds' } ); + my $rs = $schema->resultset('Artist')->search({}, { + join => { cds => ['genre', 'tracks' ] }, + collapse => 1, + '+columns' => [ + (map + { +{ "cds.$_" => "cds.$_" } } + $schema->source('Artist')->related_source('cds')->columns + ), + (map + { +{ "cds.genre.$_" => "genre.$_" } } + $schema->source('Artist')->related_source('cds')->related_source('genre')->columns + ), + (map + { +{ "cds.tracks.$_" => "tracks.$_" } } + $schema->source('Artist')->related_source('cds')->related_source('tracks')->columns + ), + ], + }); - my $prefetch_count = $artist_rs_prefetch->first->cds->count; +Both producing the following SQL: + + SELECT me.artistid, me.name, me.rank, me.charfield, + cds.cdid, cds.artist, cds.title, cds.year, cds.genreid, cds.single_track, + genre.genreid, genre.name, + tracks.trackid, tracks.cd, tracks.position, tracks.title, tracks.last_updated_on, tracks.last_updated_at + FROM artist me + LEFT JOIN cd cds + ON cds.artist = me.artistid + LEFT JOIN genre genre + ON genre.genreid = cds.genreid + LEFT JOIN track tracks + ON tracks.cd = cds.cdid + ORDER BY me.artistid + +While L implies a L, it is ok to mix the two together, as +the arguments are properly merged and generally do the right thing. For +example, you may want to do the following: + + my $artists_and_cds_without_genre = $schema->resultset('Artist')->search( + { 'genre.genreid' => undef }, + { + join => { cds => 'genre' }, + prefetch => 'cds', + } + ); - cmp_ok( $count, '==', $prefetch_count, "Counts should be the same" ); +Which generates the following SQL: -that cmp_ok() may or may not pass depending on the datasets involved. This -behavior may or may not survive the 0.09 transition. + SELECT me.artistid, me.name, me.rank, me.charfield, + cds.cdid, cds.artist, cds.title, cds.year, cds.genreid, cds.single_track + FROM artist me + LEFT JOIN cd cds + ON cds.artist = me.artistid + LEFT JOIN genre genre + ON genre.genreid = cds.genreid + WHERE genre.genreid IS NULL + ORDER BY me.artistid -=back +For a more in-depth discussion, see L. =head2 alias @@ -4442,6 +4439,131 @@ Set to 'update' for a SELECT ... FOR UPDATE or 'shared' for a SELECT ... FOR SHARED. If \$scalar is passed, this is taken directly and embedded in the query. +=head1 PREFETCHING + +DBIx::Class supports arbitrary related data prefetching from multiple related +sources. Any combination of relationship types and column sets are supported. +If L is requested, there is an additional requirement of +selecting enough data to make every individual object uniquely identifiable. + +Here are some more involved examples, based on the following relationship map: + + # Assuming: + My::Schema::CD->belongs_to( artist => 'My::Schema::Artist' ); + My::Schema::CD->might_have( liner_note => 'My::Schema::LinerNotes' ); + My::Schema::CD->has_many( tracks => 'My::Schema::Track' ); + + My::Schema::Artist->belongs_to( record_label => 'My::Schema::RecordLabel' ); + + My::Schema::Track->has_many( guests => 'My::Schema::Guest' ); + + + + my $rs = $schema->resultset('Tag')->search( + undef, + { + prefetch => { + cd => 'artist' + } + } + ); + +The initial search results in SQL like the following: + + SELECT tag.*, cd.*, artist.* FROM tag + JOIN cd ON tag.cd = cd.cdid + JOIN artist ON cd.artist = artist.artistid + +L has no need to go back to the database when we access the +C or C relationships, which saves us two SQL statements in this +case. + +Simple prefetches will be joined automatically, so there is no need +for a C attribute in the above search. + +The L attribute can be used with any of the relationship types +and multiple prefetches can be specified together. Below is a more complex +example that prefetches a CD's artist, its liner notes (if present), +the cover image, the tracks on that CD, and the guests on those +tracks. + + my $rs = $schema->resultset('CD')->search( + undef, + { + prefetch => [ + { artist => 'record_label'}, # belongs_to => belongs_to + 'liner_note', # might_have + 'cover_image', # has_one + { tracks => 'guests' }, # has_many => has_many + ] + } + ); + +This will produce SQL like the following: + + SELECT cd.*, artist.*, record_label.*, liner_note.*, cover_image.*, + tracks.*, guests.* + FROM cd me + JOIN artist artist + ON artist.artistid = me.artistid + JOIN record_label record_label + ON record_label.labelid = artist.labelid + LEFT JOIN track tracks + ON tracks.cdid = me.cdid + LEFT JOIN guest guests + ON guests.trackid = track.trackid + LEFT JOIN liner_notes liner_note + ON liner_note.cdid = me.cdid + JOIN cd_artwork cover_image + ON cover_image.cdid = me.cdid + ORDER BY tracks.cd + +Now the C, C, C, C, +C, and C of the CD will all be available through the +relationship accessors without the need for additional queries to the +database. + +=head3 CAVEATS + +Prefetch does a lot of deep magic. As such, it may not behave exactly +as you might expect. + +=over 4 + +=item * + +Prefetch uses the L to populate the prefetched relationships. This +may or may not be what you want. + +=item * + +If you specify a condition on a prefetched relationship, ONLY those +rows that match the prefetched condition will be fetched into that relationship. +This means that adding prefetch to a search() B what is returned by +traversing a relationship. So, if you have C<< Artist->has_many(CDs) >> and you do + + my $artist_rs = $schema->resultset('Artist')->search({ + 'cds.year' => 2008, + }, { + join => 'cds', + }); + + my $count = $artist_rs->first->cds->count; + + my $artist_rs_prefetch = $artist_rs->search( {}, { prefetch => 'cds' } ); + + my $prefetch_count = $artist_rs_prefetch->first->cds->count; + + cmp_ok( $count, '==', $prefetch_count, "Counts should be the same" ); + +That cmp_ok() may or may not pass depending on the datasets involved. In other +words the C condition would apply to the entire dataset, just like +it would in regular SQL. If you want to add a condition only to the "right side" +of a C - consider declaring and using a L + +=back + =head1 DBIC BIND VALUES Because DBIC may need more information to bind values than just the column name @@ -4498,6 +4620,3 @@ See L and L in You may distribute this code under the same terms as Perl itself. -=cut - -1;