X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBIHacks.pm;h=8a3cb0279ff273c6e23c686cfab15aaca214a439;hb=93d7452f38b38b66d6d8282425a928873725f43e;hp=b3b017769a8c20d430b8567744439fdb95063fa1;hpb=90f10b5a4f971772f25d6304430124abfa4a890c;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/Storage/DBIHacks.pm b/lib/DBIx/Class/Storage/DBIHacks.pm index b3b0177..8a3cb02 100644 --- a/lib/DBIx/Class/Storage/DBIHacks.pm +++ b/lib/DBIx/Class/Storage/DBIHacks.pm @@ -23,30 +23,43 @@ use namespace::clean; # {from} specs, aiding the RDBMS query optimizer # sub _prune_unused_joins { - my $self = shift; - my ($from, $select, $where, $attrs) = @_; - - return $from unless $self->_use_join_optimizer; + my ($self, $attrs) = @_; + + # only standard {from} specs are supported, and we could be disabled in general + return ($attrs->{from}, {}) unless ( + ref $attrs->{from} eq 'ARRAY' + and + @{$attrs->{from}} > 1 + and + ref $attrs->{from}[0] eq 'HASH' + and + ref $attrs->{from}[1] eq 'ARRAY' + and + $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 $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs); - my $aliastypes = $self->_resolve_aliastypes_from_select_args(@_); + my $new_aliastypes = { %$orig_aliastypes }; - # don't care - delete $aliastypes->{joining}; + # we will be recreating this entirely + my @reclassify = 'joining'; # a grouped set will not be affected by amount of rows. Thus any - # {multiplying} joins can go - delete $aliastypes->{multiplying} + # purely multiplicator classifications can go + # (will be reintroduced below if needed by something else) + push @reclassify, qw(multiplying premultiplied) if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by}; - my @newfrom = $from->[0]; # FROM head is always present + # nuke what will be recalculated + delete @{$new_aliastypes}{@reclassify}; - my %need_joins; + my @newfrom = $attrs->{from}[0]; # FROM head is always present - for (values %$aliastypes) { + # recalculate what we need once the multipliers are potentially gone + # ignore premultiplies, since they do not add any value to anything + my %need_joins; + for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) { # add all requested aliases $need_joins{$_} = 1 for keys %$_; @@ -54,15 +67,24 @@ sub _prune_unused_joins { $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; } - for my $j (@{$from}[1..$#$from]) { + for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) { push @newfrom, $j if ( - (! $j->[0]{-alias}) # legacy crap + (! defined $j->[0]{-alias}) # legacy crap || $need_joins{$j->[0]{-alias}} ); } - return \@newfrom; + # we have a new set of joiners - for everything we nuked pull the classification + # off the original stack + for my $ctype (@reclassify) { + $new_aliastypes->{$ctype} = { map + { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () } + keys %{$orig_aliastypes->{$ctype}} + } + } + + return ( \@newfrom, $new_aliastypes ); } # @@ -70,19 +92,26 @@ sub _prune_unused_joins { # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... # sub _adjust_select_args_for_complex_prefetch { - my ($self, $from, $select, $where, $attrs) = @_; - - $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') - if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY'); + my ($self, $attrs) = @_; + + $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless ( + ref $attrs->{from} eq 'ARRAY' + and + @{$attrs->{from}} > 1 + and + ref $attrs->{from}[0] eq 'HASH' + and + ref $attrs->{from}[1] eq 'ARRAY' + ); my $root_alias = $attrs->{alias}; # generate inner/outer attribute lists, remove stuff that doesn't apply my $outer_attrs = { %$attrs }; - delete $outer_attrs->{$_} for qw/where bind rows offset group_by _grouped_by_distinct having/; + delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)}; my $inner_attrs = { %$attrs }; - delete $inner_attrs->{$_} for qw/from for collapse select as _related_results_construction/; + delete @{$inner_attrs}{qw(for collapse select as _related_results_construction)}; # there is no point of ordering the insides if there is no limit delete $inner_attrs->{order_by} if ( @@ -94,13 +123,12 @@ 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 - my $outer_select = [ @$select ]; - my $inner_select; + $outer_attrs->{select} = [ @{$attrs->{select}} ]; my ($root_node, $root_node_offset); - for my $i (0 .. $#$from) { - my $node = $from->[$i]; + for my $i (0 .. $#{$inner_attrs->{from}}) { + my $node = $inner_attrs->{from}[$i]; my $h = (ref $node eq 'HASH') ? $node : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] : next @@ -117,11 +145,11 @@ sub _adjust_select_args_for_complex_prefetch { unless $root_node; # use the heavy duty resolver to take care of aliased/nonaliased naming - my $colinfo = $self->_resolve_column_info($from); + my $colinfo = $self->_resolve_column_info($inner_attrs->{from}); my $selected_root_columns; - for my $i (0 .. $#$outer_select) { - my $sel = $outer_select->[$i]; + for my $i (0 .. $#{$outer_attrs->{select}}) { + my $sel = $outer_attrs->{select}->[$i]; next if ( $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias @@ -129,28 +157,27 @@ sub _adjust_select_args_for_complex_prefetch { if (ref $sel eq 'HASH' ) { $sel->{-as} ||= $attrs->{as}[$i]; - $outer_select->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); + $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); } elsif (! ref $sel and my $ci = $colinfo->{$sel}) { $selected_root_columns->{$ci->{-colname}} = 1; } - push @$inner_select, $sel; + push @{$inner_attrs->{select}}, $sel; push @{$inner_attrs->{as}}, $attrs->{as}[$i]; } # We will need to fetch all native columns in the inner subquery, which may # be a part of an *outer* join condition, or an order_by (which needs to be - # preserved outside) + # preserved outside), or wheres. In other words everything but the inner + # selector # We can not just fetch everything because a potential has_many restricting # join collapse *will not work* on heavy data types. - my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args( - [grep { ref($_) eq 'ARRAY' or ref($_) eq 'HASH' } @{$from}[$root_node_offset .. $#$from]], - [], - $where, - $inner_attrs - ); + my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({ + %$inner_attrs, + select => [], + }); for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { my $ci = $colinfo->{$_} or next; @@ -160,159 +187,64 @@ sub _adjust_select_args_for_complex_prefetch { ! $selected_root_columns->{$ci->{-colname}}++ ) { # adding it to both to keep limits not supporting dark selectors happy - push @$inner_select, $ci->{-fq_colname}; + push @{$inner_attrs->{select}}, $ci->{-fq_colname}; push @{$inner_attrs->{as}}, $ci->{-fq_colname}; } } - # construct the inner $from and lock it in a subquery + # 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 # throw away all non-selecting, non-restricting multijoins - # (since we def. do not care about multiplication those inside the subquery) + # (since we def. do not care about multiplication of the contents of the subquery) my $inner_subq = do { - # must use it here regardless of user requests + # must use it here regardless of user requests (vastly gentler on optimizer) local $self->{_use_join_optimizer} = 1; # throw away multijoins since we def. do not care about those inside the subquery - my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, { + ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({ %$inner_attrs, _force_prune_multiplying_joins => 1 }); - my $inner_aliastypes = - $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs ); - - # uh-oh a multiplier (which is not us) left in, this is a problem + # uh-oh a multiplier (which is not us) left in, this is a problem for limits + # we will need to add a group_by to collapse the resultset for proper counts if ( - $inner_aliastypes->{multiplying} + grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} } and - !$inner_aliastypes->{grouping} # if there are groups - assume user knows wtf they are up to - and - my @multipliers = grep { $_ ne $root_alias } keys %{$inner_aliastypes->{multiplying}} + # if there are user-supplied groups - assume user knows wtf they are up to + ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} ) ) { - # if none of the multipliers came from an order_by (guaranteed to have been combined - # with a limit) - easy - just slap a group_by to simulate a collape and be on our way - if ( - ! $inner_aliastypes->{ordering} - or - ! first { $inner_aliastypes->{ordering}{$_} } @multipliers - ) { + my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} }; - my $unprocessed_order_chunks; - ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( - $inner_from, $inner_select, $inner_attrs->{order_by} + # *possibly* supplement the main selection with pks if not already + # there, as they will have to be a part of the group_by to collapse + # things properly + my $inner_select_with_extras; + my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns + or $self->throw_exception( sprintf + 'Unable to perform complex limited prefetch off %s without declared primary key', + $root_node->{-rsrc}->source_name, ); - - $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; + for my $col (@pks) { + push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col + unless $cur_sel->{$col}++; } - else { - # We need to order by external columns and group at the same time - # so we can calculate the proper limit - # This doesn't really make sense in SQL, however from DBICs point - # of view is rather valid (order the leftmost objects by whatever - # criteria and get the offset/rows many). There is a way around - # this however in SQL - we simply tae the direction of each piece - # of the foreign order and convert them to MIN(X) for ASC or MAX(X) - # for DESC, and group_by the root columns. The end result should be - # exactly what we expect - - # FIXME REMOVE LATER - (just a sanity check) - if (defined ( my $impostor = first - { $_ ne $root_alias } - keys %{ $inner_aliastypes->{selecting} } - ) ) { - $self->throw_exception(sprintf - 'Unexpected inner selection during complex prefetch (%s)...', - join ', ', keys %{ $inner_aliastypes->{joining}{$impostor}{-seen_columns} || {} } - ); - } - - # supplement the main selection with pks if not already there, - # as they will have to be a part of the group_by to colapse - # things properly - my $cur_sel = { map { $_ => 1 } @$inner_select }; - my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns - or $self->throw_exception( sprintf - 'Unable to perform complex limited prefetch off %s without declared primary key', - $root_node->{-rsrc}->source_name, - ); - for my $col (@pks) { - push @$inner_select, $col - unless $cur_sel->{$col}++; - } - - # wrap any part of the order_by that "responds" to an ordering alias - # into a MIN/MAX - # FIXME - this code is a joke, will need to be completely rewritten in - # the DQ branch. But I need to push a POC here, otherwise the - # pesky tests won't pass - my $sql_maker = $self->sql_maker; - my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); - my $own_re = qr/ $lquote \Q$root_alias\E $rquote $sep | \b \Q$root_alias\E $sep /x; - my @order_chunks = map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}); - my @new_order = map { \$_ } @order_chunks; - my $inner_columns_info = $self->_resolve_column_info($inner_from); - - # loop through and replace stuff that is not "ours" with a min/max func - # everything is a literal at this point, since we are likely properly - # quoted and stuff - for my $i (0 .. $#new_order) { - my $chunk = $order_chunks[$i][0]; - - # skip ourselves - next if $chunk =~ $own_re; - - ($chunk, my $is_desc) = $sql_maker->_split_order_chunk($chunk); - - # maybe our own unqualified column - my $ord_bit = ( - $lquote and $sep and $chunk =~ /^ $lquote ([^$sep]+) $rquote $/x - ) ? $1 : $chunk; - - next if ( - $ord_bit - and - $inner_columns_info->{$ord_bit} - and - $inner_columns_info->{$ord_bit}{-source_alias} eq $root_alias - ); - - $new_order[$i] = \[ - sprintf( - '%s(%s)%s', - ($is_desc ? 'MAX' : 'MIN'), - $chunk, - ($is_desc ? ' DESC' : ''), - ), - @ {$order_chunks[$i]} [ 1 .. $#{$order_chunks[$i]} ] - ]; - } - - $inner_attrs->{order_by} = \@new_order; - - # do not care about leftovers here - it will be all the functions - # we just created - ($inner_attrs->{group_by}) = $self->_group_over_selection ( - $inner_from, $inner_select, $inner_attrs->{order_by} - ); - } + ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({ + %$inner_attrs, + $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (), + _aliastypes => $inner_aliastypes, + }); } - # we already optimized $inner_from above + # we already optimized $inner_attrs->{from} above # and already local()ized $self->{_use_join_optimizer} = 0; # generate the subquery $self->_select_args_to_query ( - $inner_from, - $inner_select, - $where, + @{$inner_attrs}{qw(from select where)}, $inner_attrs, ); }; @@ -328,22 +260,25 @@ sub _adjust_select_args_for_complex_prefetch { # result by tackling yet another group_by to the outside of the query # work on a shallow copy - $from = [ @$from ]; + my @orig_from = @{$attrs->{from}}; - my @outer_from; + + $outer_attrs->{from} = \ my @outer_from; # we may not be the head if ($root_node_offset) { # first generate the outer_from, up to the substitution point - @outer_from = splice @$from, 0, $root_node_offset; + @outer_from = splice @orig_from, 0, $root_node_offset; + # substitute the subq at the right spot push @outer_from, [ { -alias => $root_alias, -rsrc => $root_node->{-rsrc}, $root_alias => $inner_subq, }, - @{$from->[0]}[1 .. $#{$from->[0]}], + # preserve attrs from what is now the head of the from after the splice + @{$orig_from[0]}[1 .. $#{$orig_from[0]}], ]; } else { @@ -354,12 +289,12 @@ sub _adjust_select_args_for_complex_prefetch { }; } - shift @$from; # it's replaced in @outer_from already + shift @orig_from; # what we just replaced above # 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 ); + my $outer_aliastypes = $outer_attrs->{_aliastypes} = + $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from }); # unroll parents my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ @@ -367,10 +302,8 @@ sub _adjust_select_args_for_complex_prefetch { } } qw/selecting restricting grouping ordering/; # see what's left - throw away if not selecting/restricting - # also throw in a group_by if a non-selecting multiplier, - # to guard against cross-join explosions - my $need_outer_group_by; - while (my $j = shift @$from) { + my $may_need_outer_group_by; + while (my $j = shift @orig_from) { my $alias = $j->[0]{-alias}; if ( @@ -380,32 +313,29 @@ sub _adjust_select_args_for_complex_prefetch { } elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { push @outer_from, $j; - $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; + $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; } } - if ( $need_outer_group_by and $attrs->{_grouped_by_distinct} ) { - my $unprocessed_order_chunks; - ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( - \@outer_from, $outer_select, $outer_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; - + # also throw in a synthetic group_by if a non-selecting multiplier, + # to guard against cross-join explosions + # the logic is somewhat fragile, but relies on the idea that if a user supplied + # a group by on their own - they know what they were doing + if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) { + ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({ + %$outer_attrs, + from => \@outer_from, + }); } - # This is totally horrific - the $where ends up in both the inner and outer query + # This is totally horrific - the {where} ends up in both the inner and outer query # Unfortunately not much can be done until SQLA2 introspection arrives, and even # then if where conditions apply to the *right* side of the prefetch, you may have # to both filter the inner select (e.g. to apply a limit) and then have to re-filter - # the outer select to exclude joins you didin't want in the first place + # the outer select to exclude joins you didn't want in the first place # # OTOH it can be seen as a plus: (notes that this query would make a DBA cry ;) - return (\@outer_from, $outer_select, $where, $outer_attrs); + return $outer_attrs; } # @@ -413,41 +343,47 @@ sub _adjust_select_args_for_complex_prefetch { # # 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 +# aliases are needed 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 ) = @_; + my ( $self, $attrs ) = @_; $self->throw_exception ('Unable to analyze custom {from}') - if ref $from ne 'ARRAY'; + if ref $attrs->{from} ne 'ARRAY'; # what we will return my $aliases_by_type; # see what aliases are there to work with + # and record who is a multiplier and who is premultiplied my $alias_list; - for (@$from) { - my $j = $_; + for my $node (@{$attrs->{from}}) { + + my $j = $node; $j = $j->[0] if ref $j eq 'ARRAY'; my $al = $j->{-alias} or next; $alias_list->{$al} = $j; - $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } if ( + + $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } # not array == {from} head == can't be multiplying - ( ref($_) eq 'ARRAY' and ! $j->{-is_single} ) - or - # a parent of ours is already a multiplier - ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } ) - ); + if ref($node) eq 'ARRAY' and ! $j->{-is_single}; + + $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] } + # parts of the path that are not us but are multiplying + if grep { $aliases_by_type->{multiplying}{$_} } + grep { $_ ne $al } + map { values %$_ } + @{ $j->{-join_path}||[] } } # get a column to source/alias map (including unambiguous unqualified ones) - my $colinfo = $self->_resolve_column_info ($from); + my $colinfo = $self->_resolve_column_info ($attrs->{from}); # set up a botched SQLA my $sql_maker = $self->sql_maker; @@ -480,7 +416,7 @@ sub _resolve_aliastypes_from_select_args { # generate sql chunks my $to_scan = { restricting => [ - $sql_maker->_recurse_where ($where), + $sql_maker->_recurse_where ($attrs->{where}), $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), ], grouping => [ @@ -488,12 +424,12 @@ sub _resolve_aliastypes_from_select_args { ], joining => [ $sql_maker->_recurse_from ( - ref $from->[0] eq 'ARRAY' ? $from->[0][0] : $from->[0], - @{$from}[1 .. $#$from], + ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0], + @{$attrs->{from}}[1 .. $#{$attrs->{from}}], ), ], selecting => [ - $sql_maker->_recurse_fields ($select), + $sql_maker->_recurse_fields ($attrs->{select}), ], ordering => [ map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker), @@ -569,52 +505,139 @@ sub _resolve_aliastypes_from_select_args { return $aliases_by_type; } -# This is the engine behind { distinct => 1 } +# This is the engine behind { distinct => 1 } and the general +# complex prefetch grouper sub _group_over_selection { - my ($self, $from, $select, $order_by) = @_; + my ($self, $attrs) = @_; - my $rs_column_list = $self->_resolve_column_info ($from); + my $colinfos = $self->_resolve_column_info ($attrs->{from}); my (@group_by, %group_index); # the logic is: if it is a { func => val } we assume an aggregate, # otherwise if \'...' or \[...] we assume the user knows what is # going on thus group over it - for (@$select) { + for (@{$attrs->{select}}) { if (! ref($_) or ref ($_) ne 'HASH' ) { push @group_by, $_; $group_index{$_}++; - if ($rs_column_list->{$_} and $_ !~ /\./ ) { + if ($colinfos->{$_} and $_ !~ /\./ ) { # add a fully qualified version as well - $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++; + $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; } } } + my @order_by = $self->_extract_order_criteria($attrs->{order_by}) + or return (\@group_by, $attrs->{order_by}); + # add any order_by parts that are not already present in the group_by - # we need to be careful not to add any named functions/aggregates - # i.e. order_by => [ ... { count => 'foo' } ... ] - my @leftovers; - for ($self->_extract_order_criteria($order_by)) { + # to maintain SQL cross-compatibility and general sanity + # + # also in case the original selection is *not* unique, or in case part + # of the ORDER BY refers to a multiplier - we will need to replace the + # skipped order_by elements with their MIN/MAX equivalents as to maintain + # the proper overall order without polluting the group criteria (and + # possibly changing the outcome entirely) + + my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes); + + my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by); + + for my $o_idx (0 .. $#order_by) { + + # if the chunk is already a min/max function - there is nothing left to touch + next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix; + # only consider real columns (for functions the user got to do an explicit group_by) - if (@$_ != 1) { - push @leftovers, $_; - next; + my $chunk_ci; + if ( + @{$order_by[$o_idx]} != 1 + or + # only declare an unknown *plain* identifier as "leftover" if we are called with + # aliastypes to examine. If there are none - we are still in _resolve_attrs, and + # can just assume the user knows what they want + ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} ) + ) { + push @$leftovers, $order_by[$o_idx][0]; } - 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}++; + next unless $chunk_ci; + + # no duplication of group criteria + next if $group_index{$chunk_ci->{-fq_colname}}; + + $aliastypes ||= ( + $attrs->{_aliastypes} + or + $self->_resolve_aliastypes_from_select_args({ + from => $attrs->{from}, + order_by => $attrs->{order_by}, + }) + ) if $group_already_unique; + + # check that we are not ordering by a multiplier (if a check is requested at all) + if ( + $group_already_unique + and + ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}} + and + ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}} + ) { + push @group_by, $chunk_ci->{-fq_colname}; + $group_index{$chunk_ci->{-fq_colname}}++ + } + else { + # We need to order by external columns without adding them to the group + # (eiehter a non-unique selection, or a multi-external) + # + # This doesn't really make sense in SQL, however from DBICs point + # of view is rather valid (e.g. order the leftmost objects by whatever + # criteria and get the offset/rows many). There is a way around + # this however in SQL - we simply tae the direction of each piece + # of the external order and convert them to MIN(X) for ASC or MAX(X) + # for DESC, and group_by the root columns. The end result should be + # exactly what we expect + + # FIXME - this code is a joke, will need to be completely rewritten in + # the DQ branch. But I need to push a POC here, otherwise the + # pesky tests won't pass + # wrap any part of the order_by that "responds" to an ordering alias + # into a MIN/MAX + $sql_maker ||= $self->sql_maker; + $order_chunks ||= [ + map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}) + ]; + + my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]); + + $new_order_by[$o_idx] = \[ + sprintf( '%s( %s )%s', + ($is_desc ? 'MAX' : 'MIN'), + $chunk, + ($is_desc ? ' DESC' : ''), + ), + @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ] + ]; + } } - return wantarray - ? (\@group_by, (@leftovers ? \@leftovers : undef) ) - : \@group_by - ; + $self->throw_exception ( sprintf + 'A required group_by clause could not be constructed automatically due to a complex ' + . 'order_by criteria (%s). Either order_by columns only (no functions) or construct a suitable ' + . 'group_by by hand', + join ', ', map { "'$_'" } @$leftovers, + ) if $leftovers; + + # recreate the untouched order parts + if (@new_order_by) { + $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks ); + } + + return ( + \@group_by, + (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged + ); } sub _resolve_ident_sources { @@ -826,15 +849,25 @@ sub _extract_order_criteria { sub _order_by_is_stable { my ($self, $ident, $order_by, $where) = @_; - my $colinfo = $self->_resolve_column_info($ident, [ + my @cols = ( (map { $_->[0] } $self->_extract_order_criteria($order_by)), $where ? @{$self->_extract_fixed_condition_columns($where)} :(), - ]); + ) or return undef; + + my $colinfo = $self->_resolve_column_info($ident, \@cols); + + return keys %$colinfo + ? $self->_columns_comprise_identifying_set( $colinfo, \@cols ) + : undef + ; +} - return undef unless keys %$colinfo; +sub _columns_comprise_identifying_set { + my ($self, $colinfo, $columns) = @_; my $cols_per_src; - $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo; + $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_ + for grep { defined $_ } @{$colinfo}{@$columns}; for (values %$cols_per_src) { my $src = (values %$_)[0]->{-result_source}; @@ -915,7 +948,7 @@ sub _main_source_order_by_portion_is_stable { die 'How did we get here...'; } -# returns an arrayref of column names which *definitely* have som +# returns an arrayref of column names which *definitely* have some # sort of non-nullable equality requested in the given condition # specification. This is used to figure out if a resultset is # constrained to a column which is part of a unique constraint, @@ -927,7 +960,7 @@ sub _main_source_order_by_portion_is_stable { # something that is in fact there - the stack will recover gracefully # Also - DQ and the mst it rode in on will save us all RSN!!! sub _extract_fixed_condition_columns { - my ($self, $where, $nested) = @_; + my ($self, $where) = @_; return unless ref $where eq 'HASH'; @@ -935,8 +968,8 @@ sub _extract_fixed_condition_columns { for my $lhs (keys %$where) { if ($lhs =~ /^\-and$/i) { push @cols, ref $where->{$lhs} eq 'ARRAY' - ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} ) - : $self->_extract_fixed_condition_columns($where->{$lhs}, 1) + ? ( map { @{ $self->_extract_fixed_condition_columns($_) } } @{$where->{$lhs}} ) + : @{ $self->_extract_fixed_condition_columns($where->{$lhs}) } ; } elsif ($lhs !~ /^\-/) { @@ -949,7 +982,7 @@ sub _extract_fixed_condition_columns { )); } } - return $nested ? @cols : \@cols; + return \@cols; } 1;