X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBIHacks.pm;h=ba1faa486776270fa0c66da9075e5837a0226ddf;hb=14e26c5f3516e39d9191ca9b2a9d460f8f495654;hp=a54477e071b5251b90056b5a59699e8d59292d1d;hpb=09e14fdcc53797129436ace4fc16f64e6f750a75;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/Storage/DBIHacks.pm b/lib/DBIx/Class/Storage/DBIHacks.pm index a54477e..ba1faa4 100644 --- a/lib/DBIx/Class/Storage/DBIHacks.pm +++ b/lib/DBIx/Class/Storage/DBIHacks.pm @@ -14,6 +14,43 @@ use base 'DBIx::Class::Storage'; use mro 'c3'; use Carp::Clan qw/^DBIx::Class/; +use List::Util 'first'; +use Scalar::Util 'blessed'; +use namespace::clean; + +# +# This code will remove non-selecting/non-restricting joins from +# {from} specs, aiding the RDBMS query optimizer +# +sub _prune_unused_joins { + my $self = shift; + my ($from, $select, $where, $attrs) = @_; + + return $from unless $self->_use_join_optimizer; + + if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY') { + return $from; # only standard {from} specs are supported + } + + my $aliastypes = $self->_resolve_aliastypes_from_select_args(@_); + + # a grouped set will not be affected by amount of rows. Thus any + # {multiplying} joins can go + delete $aliastypes->{multiplying} if $attrs->{group_by}; + + my @newfrom = $from->[0]; # FROM head is always present + + my %need_joins = (map { %{$_||{}} } (values %$aliastypes) ); + for my $j (@{$from}[1..$#$from]) { + push @newfrom, $j if ( + (! $j->[0]{-alias}) # legacy crap + || + $need_joins{$j->[0]{-alias}} + ); + } + + return \@newfrom; +} # # This is the code producing joined subqueries like: @@ -46,7 +83,6 @@ sub _adjust_select_args_for_complex_prefetch { ]; } - # generate the inner/outer select lists # for inside we consider only stuff *not* brought in by the prefetch # on the outside we substitute any function for its alias @@ -61,124 +97,66 @@ sub _adjust_select_args_for_complex_prefetch { } push @$inner_select, $sel; - } - # normalize a copy of $from, so it will be easier to work with further - # down (i.e. promote the initial hashref to an AoH) - $from = [ @$from ]; - $from->[0] = [ $from->[0] ]; - my %original_join_info = map { $_->[0]{-alias} => $_->[0] } (@$from); - - - # decide which parts of the join will remain in either part of - # the outer/inner query - - # First we compose a list of which aliases are used in restrictions - # (i.e. conditions/order/grouping/etc). Since we do not have - # introspectable SQLA, we fall back to ugly scanning of raw SQL for - # WHERE, and for pieces of ORDER BY in order to determine which aliases - # need to appear in the resulting sql. - # It may not be very efficient, but it's a reasonable stop-gap - # Also unqualified column names will not be considered, but more often - # than not this is actually ok - # - # In the same loop we enumerate part of the selection aliases, as - # it requires the same sqla hack for the time being - my ($restrict_aliases, $select_aliases, $prefetch_aliases); - { - # produce stuff unquoted, so it can be scanned - my $sql_maker = $self->sql_maker; - local $sql_maker->{quote_char}; - my $sep = $self->_sql_maker_opts->{name_sep} || '.'; - $sep = "\Q$sep\E"; - - my $non_prefetch_select_sql = $sql_maker->_recurse_fields ($inner_select); - my $prefetch_select_sql = $sql_maker->_recurse_fields ($outer_attrs->{_prefetch_select}); - my $where_sql = $sql_maker->where ($where); - my $group_by_sql = $sql_maker->_order_by({ - map { $_ => $inner_attrs->{$_} } qw/group_by having/ - }); - my @non_prefetch_order_by_chunks = (map - { ref $_ ? $_->[0] : $_ } - $sql_maker->_order_by_chunks ($inner_attrs->{order_by}) - ); - - - for my $alias (keys %original_join_info) { - my $seen_re = qr/\b $alias $sep/x; - - for my $piece ($where_sql, $group_by_sql, @non_prefetch_order_by_chunks ) { - if ($piece =~ $seen_re) { - $restrict_aliases->{$alias} = 1; - } - } - - if ($non_prefetch_select_sql =~ $seen_re) { - $select_aliases->{$alias} = 1; - } - - if ($prefetch_select_sql =~ $seen_re) { - $prefetch_aliases->{$alias} = 1; - } - - } + push @{$inner_attrs->{as}}, $attrs->{as}[$i]; } - # Add any non-left joins to the restriction list (such joins are indeed restrictions) - for my $j (values %original_join_info) { - my $alias = $j->{-alias} or next; - $restrict_aliases->{$alias} = 1 if ( - (not $j->{-join_type}) - or - ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) - ); - } + # construct the inner $from and lock it in a subquery + # we need to prune first, because this will determine if we need a group_by below + # the fake group_by is so that the pruner throws away all non-selecting, non-restricting + # multijoins (since we def. do not care about those inside the subquery) - # mark all join parents as mentioned - # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too ) - for my $collection ($restrict_aliases, $select_aliases) { - for my $alias (keys %$collection) { - $collection->{$_} = 1 - for (@{ $original_join_info{$alias}{-join_path} || [] }); - } - } + my $subq_joinspec = do { - # construct the inner $from for the subquery - my %inner_joins = (map { %{$_ || {}} } ($restrict_aliases, $select_aliases) ); - my @inner_from; - for my $j (@$from) { - push @inner_from, $j if $inner_joins{$j->[0]{-alias}}; - } + # must use it here regardless of user requests + local $self->{_use_join_optimizer} = 1; - # if a multi-type join was needed in the subquery ("multi" is indicated by - # presence in {collapse}) - add a group_by to simulate the collapse in the subq - unless ($inner_attrs->{group_by}) { - for my $alias (keys %inner_joins) { + my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, { + group_by => ['dummy'], %$inner_attrs, + }); - # the dot comes from some weirdness in collapse - # remove after the rewrite - if ($attrs->{collapse}{".$alias"}) { - $inner_attrs->{group_by} ||= $inner_select; - last; - } + my $inner_aliastypes = + $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs ); + + # if a multi-type non-selecting (only restricting) join was needed in the subquery + # add a group_by to simulate the collapse in the subq + if ( + ! $inner_attrs->{group_by} + and + first { + $inner_aliastypes->{restricting}{$_} + and + ! $inner_aliastypes->{selecting}{$_} + } ( keys %{$inner_aliastypes->{multiplying}||{}} ) + ) { + my $unprocessed_order_chunks; + ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( + $inner_from, $inner_select, $inner_attrs->{order_by} + ); + + $self->throw_exception ( + 'A required group_by clause could not be constructed automatically due to a complex ' + . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' + . 'group_by by hand' + ) if $unprocessed_order_chunks; } - } - # demote the inner_from head - $inner_from[0] = $inner_from[0][0]; + # we already optimized $inner_from above + local $self->{_use_join_optimizer} = 0; - # generate the subquery - my $subq = $self->_select_args_to_query ( - \@inner_from, - $inner_select, - $where, - $inner_attrs, - ); + # generate the subquery + my $subq = $self->_select_args_to_query ( + $inner_from, + $inner_select, + $where, + $inner_attrs, + ); - my $subq_joinspec = { - -alias => $attrs->{alias}, - -source_handle => $inner_from[0]{-source_handle}, - $attrs->{alias} => $subq, + +{ + -alias => $attrs->{alias}, + -source_handle => $inner_from->[0]{-source_handle}, + $attrs->{alias} => $subq, + }; }; # Generate the outer from - this is relatively easy (really just replace @@ -191,9 +169,13 @@ sub _adjust_select_args_for_complex_prefetch { # - it is part of the restrictions, in which case we need to collapse the outer # result by tackling yet another group_by to the outside of the query + $from = [ @$from ]; + # so first generate the outer_from, up to the substitution point my @outer_from; while (my $j = shift @$from) { + $j = [ $j ] unless ref $j eq 'ARRAY'; # promote the head-from to an AoH + if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap push @outer_from, [ $subq_joinspec, @@ -206,6 +188,11 @@ sub _adjust_select_args_for_complex_prefetch { } } + # scan the *remaining* from spec against different attributes, and see which joins are needed + # in what role + my $outer_aliastypes = + $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs ); + # see what's left - throw away if not selecting/restricting # also throw in a group_by if restricting to guard against # cross-join explosions @@ -213,27 +200,12 @@ sub _adjust_select_args_for_complex_prefetch { while (my $j = shift @$from) { my $alias = $j->[0]{-alias}; - if ($select_aliases->{$alias} || $prefetch_aliases->{$alias}) { + if ($outer_aliastypes->{selecting}{$alias}) { push @outer_from, $j; } - elsif ($restrict_aliases->{$alias}) { + elsif ($outer_aliastypes->{restricting}{$alias}) { push @outer_from, $j; - - # FIXME - this should be obviated by SQLA2, as I'll be able to - # have restrict_inner and restrict_outer... or something to that - # effect... I think... - - # FIXME2 - I can't find a clean way to determine if a particular join - # is a multi - instead I am just treating everything as a potential - # explosive join (ribasushi) - # - # if (my $handle = $j->[0]{-source_handle}) { - # my $rsrc = $handle->resolve; - # ... need to bail out of the following if this is not a multi, - # as it will be much easier on the db ... - - $outer_attrs->{group_by} ||= $outer_select; - # } + $outer_attrs->{group_by} ||= $outer_select unless $j->[0]{-is_single}; } } @@ -250,6 +222,181 @@ sub _adjust_select_args_for_complex_prefetch { return (\@outer_from, $outer_select, $where, $outer_attrs); } +# +# I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! +# +# Due to a lack of SQLA2 we fall back to crude scans of all the +# select/where/order/group attributes, in order to determine what +# aliases are neded to fulfill the query. This information is used +# throughout the code to prune unnecessary JOINs from the queries +# in an attempt to reduce the execution time. +# Although the method is pretty horrific, the worst thing that can +# happen is for it to fail due to some scalar SQL, which in turn will +# result in a vocal exception. +sub _resolve_aliastypes_from_select_args { + my ( $self, $from, $select, $where, $attrs ) = @_; + + $self->throw_exception ('Unable to analyze custom {from}') + if ref $from ne 'ARRAY'; + + # what we will return + my $aliases_by_type; + + # see what aliases are there to work with + my $alias_list; + for (@$from) { + my $j = $_; + $j = $j->[0] if ref $j eq 'ARRAY'; + my $al = $j->{-alias} + or next; + + $alias_list->{$al} = $j; + $aliases_by_type->{multiplying}{$al} = 1 + if ref($_) eq 'ARRAY' and ! $j->{-is_single}; # not array == {from} head == can't be multiplying + } + + # get a column to source/alias map (including unqualified ones) + my $colinfo = $self->_resolve_column_info ($from); + + # set up a botched SQLA + my $sql_maker = $self->sql_maker; + + # these are throw away results, do not pollute the bind stack + local $sql_maker->{having_bind}; + local $sql_maker->{select_bind}; + + # we can't scan properly without any quoting (\b doesn't cut it + # everywhere), so unless there is proper quoting set - use our + # own weird impossible character. + # Also in the case of no quoting, we need to explicitly disable + # name_sep, otherwise sorry nasty legacy syntax like + # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( + local $sql_maker->{quote_char} = $sql_maker->{quote_char}; + local $sql_maker->{name_sep} = $sql_maker->{name_sep}; + + unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { + $sql_maker->{quote_char} = ["\x00", "\xFF"]; + # if we don't unset it we screw up retarded but unfortunately working + # 'MAX(foo.bar)' => { '>', 3 } + $sql_maker->{name_sep} = ''; + } + + my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); + + # generate sql chunks + my $to_scan = { + restricting => [ + $sql_maker->_recurse_where ($where), + $sql_maker->_parse_rs_attrs ({ + map { $_ => $attrs->{$_} } (qw/group_by having/) + }), + ], + selecting => [ + $sql_maker->_recurse_fields ($select), + ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ), + ], + }; + + # throw away empty chunks + $_ = [ map { $_ || () } @$_ ] for values %$to_scan; + + # first loop through all fully qualified columns and get the corresponding + # alias (should work even if they are in scalarrefs) + for my $alias (keys %$alias_list) { + my $al_re = qr/ + $lquote $alias $rquote $sep + | + \b $alias \. + /x; + + for my $type (keys %$to_scan) { + for my $piece (@{$to_scan->{$type}}) { + $aliases_by_type->{$type}{$alias} = 1 if ($piece =~ $al_re); + } + } + } + + # now loop through unqualified column names, and try to locate them within + # the chunks + for my $col (keys %$colinfo) { + next if $col =~ / \. /x; # if column is qualified it was caught by the above + + my $col_re = qr/ $lquote $col $rquote /x; + + for my $type (keys %$to_scan) { + for my $piece (@{$to_scan->{$type}}) { + $aliases_by_type->{$type}{$colinfo->{$col}{-source_alias}} = 1 if ($piece =~ $col_re); + } + } + } + + # Add any non-left joins to the restriction list (such joins are indeed restrictions) + for my $j (values %$alias_list) { + my $alias = $j->{-alias} or next; + $aliases_by_type->{restricting}{$alias} = 1 if ( + (not $j->{-join_type}) + or + ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) + ); + } + + # mark all restricting/selecting join parents as such + # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too ) + for my $type (qw/restricting selecting/) { + for my $alias (keys %{$aliases_by_type->{$type}||{}}) { + $aliases_by_type->{$type}{$_} = 1 + for (map { values %$_ } @{ $alias_list->{$alias}{-join_path} || [] }); + } + } + + return $aliases_by_type; +} + +# This is the engine behind { distinct => 1 } +sub _group_over_selection { + my ($self, $from, $select, $order_by) = @_; + + my $rs_column_list = $self->_resolve_column_info ($from); + + my (@group_by, %group_index); + + for (@$select) { + if (! ref($_) or ref ($_) ne 'HASH' ) { + push @group_by, $_; + $group_index{$_}++; + if ($rs_column_list->{$_} and $_ !~ /\./ ) { + # add a fully qualified version as well + $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++; + } + } + } + + # add any order_by parts that are not already present in the group_by + # we need to be careful not to add any named functions/aggregates + # i.e. order_by => [ ... { count => 'foo' } ... ] + my @leftovers; + for ($self->_extract_order_criteria($order_by)) { + # only consider real columns (for functions the user got to do an explicit group_by) + if (@$_ != 1) { + push @leftovers, $_; + next; + } + my $chunk = $_->[0]; + my $colinfo = $rs_column_list->{$chunk} or do { + push @leftovers, $_; + next; + }; + + $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./; + push @group_by, $chunk unless $group_index{$chunk}++; + } + + return wantarray + ? (\@group_by, (@leftovers ? \@leftovers : undef) ) + : \@group_by + ; +} + sub _resolve_ident_sources { my ($self, $ident) = @_; @@ -258,7 +405,7 @@ sub _resolve_ident_sources { # the reason this is so contrived is that $ident may be a {from} # structure, specifying multiple tables to join - if ( Scalar::Util::blessed($ident) && $ident->isa("DBIx::Class::ResultSource") ) { + if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { # this is compat mode for insert/update/delete which do not deal with aliases $alias2source->{me} = $ident; $rs_alias = 'me'; @@ -294,10 +441,7 @@ sub _resolve_column_info { my ($self, $ident, $colnames) = @_; my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); - my $sep = $self->_sql_maker_opts->{name_sep} || '.'; - my $qsep = quotemeta $sep; - - my (%return, %seen_cols, @auto_colnames); + my (%seen_cols, @auto_colnames); # compile a global list of column names, to be able to properly # disambiguate unqualified column names (if at all possible) @@ -305,7 +449,7 @@ sub _resolve_column_info { my $rsrc = $alias2src->{$alias}; for my $colname ($rsrc->columns) { push @{$seen_cols{$colname}}, $alias; - push @auto_colnames, "$alias$sep$colname" unless $colnames; + push @auto_colnames, "$alias.$colname" unless $colnames; } } @@ -314,25 +458,23 @@ sub _resolve_column_info { grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), ]; - COLUMN: + my (%return, $colinfos); foreach my $col (@$colnames) { - my ($alias, $colname) = $col =~ m/^ (?: ([^$qsep]+) $qsep)? (.+) $/x; + my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; - unless ($alias) { - # see if the column was seen exactly once (so we know which rsrc it came from) - if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1) { - $alias = $seen_cols{$colname}[0]; - } - else { - next COLUMN; - } - } + # if the column was seen exactly once - we know which rsrc it came from + $source_alias ||= $seen_cols{$colname}[0] + if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); - my $rsrc = $alias2src->{$alias}; - $return{$col} = $rsrc && { - %{$rsrc->column_info($colname)}, + next unless $source_alias; + + my $rsrc = $alias2src->{$source_alias} + or next; + + $return{$col} = { + %{ ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} }, -result_source => $rsrc, - -source_alias => $alias, + -source_alias => $source_alias, }; } @@ -354,7 +496,7 @@ sub _resolve_column_info { # the top of the stack, and if not - make sure the chain is inner-joined down # to the root. # -sub _straight_join_to_node { +sub _inner_join_to_node { my ($self, $from, $alias) = @_; # subqueries and other oddness are naturally not supported @@ -388,7 +530,7 @@ sub _straight_join_to_node { # anyway, and deep cloning is just too fucking expensive # So replace the first hashref in the node arrayref manually my @new_from = ($from->[0]); - my $sw_idx = { map { $_ => 1 } @$switch_branch }; + my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path for my $j (@{$from}[1 .. $#$from]) { my $jalias = $j->[0]{-alias}; @@ -441,20 +583,29 @@ sub _strip_cond_qualifiers { for (my $i = 0; $i < @cond; $i++) { my $entry = $cond[$i]; my $hash; - if (ref $entry eq 'HASH') { + my $ref = ref $entry; + if ($ref eq 'HASH' or $ref eq 'ARRAY') { $hash = $self->_strip_cond_qualifiers($entry); } - else { + elsif (! $ref) { $entry =~ /([^.]+)$/; $hash->{$1} = $cond[++$i]; } + else { + $self->throw_exception ("_strip_cond_qualifiers() is unable to handle a condition reftype $ref"); + } push @{$cond->{-and}}, $hash; } } else { foreach my $key (keys %$where) { - $key =~ /([^.]+)$/; - $cond->{$1} = $where->{$key}; + if ($key eq '-or' && ref $where->{$key} eq 'ARRAY') { + $cond->{$key} = $self->_strip_cond_qualifiers($where->{$key}); + } + else { + $key =~ /([^.]+)$/; + $cond->{$1} = $where->{$key}; + } } } } @@ -465,5 +616,33 @@ sub _strip_cond_qualifiers { return $cond; } +sub _extract_order_criteria { + my ($self, $order_by, $sql_maker) = @_; + + my $parser = sub { + my ($sql_maker, $order_by) = @_; + + return scalar $sql_maker->_order_by_chunks ($order_by) + unless wantarray; + + my @chunks; + for ($sql_maker->_order_by_chunks ($order_by) ) { + my $chunk = ref $_ ? $_ : [ $_ ]; + $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix; + push @chunks, $chunk; + } + + return @chunks; + }; + + if ($sql_maker) { + return $parser->($sql_maker, $order_by); + } + else { + $sql_maker = $self->sql_maker; + local $sql_maker->{quote_char}; + return $parser->($sql_maker, $order_by); + } +} 1;