X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=blobdiff_plain;f=lib%2FDBIx%2FClass%2FStorage%2FDBIHacks.pm;h=3efd4884a39813da053863b9b494c13038ee5dfe;hb=908aa1bb761ec1da5c061fe9f687598e3f1934bc;hp=d4a9df844c709ee05ea0bbc27254d3389b2015b8;hpb=d2fcb9b3612958ecc794acae77520f3da06d3c5b;p=dbsrgits%2FDBIx-Class.git diff --git a/lib/DBIx/Class/Storage/DBIHacks.pm b/lib/DBIx/Class/Storage/DBIHacks.pm index d4a9df8..3efd488 100644 --- a/lib/DBIx/Class/Storage/DBIHacks.pm +++ b/lib/DBIx/Class/Storage/DBIHacks.pm @@ -4,7 +4,7 @@ package #hide from PAUSE # # This module contains code that should never have seen the light of day, # does not belong in the Storage, or is otherwise unfit for public -# display. The arrival of SQLA2 should immediately oboslete 90% of this +# display. The arrival of SQLA2 should immediately obsolete 90% of this # use strict; @@ -13,9 +13,9 @@ use warnings; use base 'DBIx::Class::Storage'; use mro 'c3'; -use Carp::Clan qw/^DBIx::Class/; use List::Util 'first'; use Scalar::Util 'blessed'; +use Sub::Name 'subname'; use namespace::clean; # @@ -23,10 +23,11 @@ use namespace::clean; # {from} specs, aiding the RDBMS query optimizer # sub _prune_unused_joins { - my ($self) = shift; - + 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 } @@ -37,10 +38,16 @@ sub _prune_unused_joins { # {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 %$aliastypes) ); + my %need_joins; + for (values %$aliastypes) { + # add all requested aliases + $need_joins{$_} = 1 for keys %$_; + + # add all their parents (as per joinpath which is an AoH { table => alias }) + $need_joins{$_} = 1 for map { values %$_ } map { @$_ } values %$_; + } for my $j (@{$from}[1..$#$from]) { push @newfrom, $j if ( (! $j->[0]{-alias}) # legacy crap @@ -54,13 +61,13 @@ sub _prune_unused_joins { # # This is the code producing joined subqueries like: -# SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... +# SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... # sub _adjust_select_args_for_complex_prefetch { my ($self, $from, $select, $where, $attrs) = @_; $self->throw_exception ('Nothing to prefetch... how did we get here?!') - if not @{$attrs->{_prefetch_select}}; + if not @{$attrs->{_prefetch_selector_range}}; $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'); @@ -70,25 +77,17 @@ sub _adjust_select_args_for_complex_prefetch { my $outer_attrs = { %$attrs }; delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/; - my $inner_attrs = { %$attrs }; - delete $inner_attrs->{$_} for qw/for collapse _prefetch_select _collapse_order_by select as/; - - - # bring over all non-collapse-induced order_by into the inner query (if any) - # the outer one will have to keep them all - delete $inner_attrs->{order_by}; - if (my $ord_cnt = @{$outer_attrs->{order_by}} - @{$outer_attrs->{_collapse_order_by}} ) { - $inner_attrs->{order_by} = [ - @{$outer_attrs->{order_by}}[ 0 .. $ord_cnt - 1] - ]; - } + my $inner_attrs = { %$attrs, _is_internal_subuery => 1 }; + delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range select as/; # 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 = []; - for my $i (0 .. ( @$outer_select - @{$outer_attrs->{_prefetch_select}} - 1) ) { + + my ($p_start, $p_end) = @{$outer_attrs->{_prefetch_selector_range}}; + for my $i (0 .. $p_start - 1, $p_end + 1 .. $#$outer_select) { my $sel = $outer_select->[$i]; if (ref $sel eq 'HASH' ) { @@ -101,27 +100,56 @@ sub _adjust_select_args_for_complex_prefetch { push @{$inner_attrs->{as}}, $attrs->{as}[$i]; } - # construct the inner $from for the 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 - my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, $inner_attrs); - - # if a multi-type join was needed in the subquery - add a group_by to simulate the - # collapse in the subq - $inner_attrs->{group_by} ||= $inner_select - if first { ! $_->[0]{-is_single} } (@{$inner_from}[1 .. $#$inner_from]); - - # generate the subquery - my $subq = $self->_select_args_to_query ( - $inner_from, - $inner_select, - $where, - $inner_attrs, - ); + # 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) + + my $inner_subq = do { + + # must use it here regardless of user requests + local $self->{_use_join_optimizer} = 1; + + my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, { + group_by => ['dummy'], %$inner_attrs, + }); + + my $inner_aliastypes = + $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs ); + + # we need to simulate collapse in the subq if a multiplying join is pulled + # by being a non-selecting restrictor + 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; + } + + # we already optimized $inner_from above + local $self->{_use_join_optimizer} = 0; - my $subq_joinspec = { - -alias => $attrs->{alias}, - -source_handle => $inner_from->[0]{-source_handle}, - $attrs->{alias} => $subq, + # generate the subquery + $self->_select_args_to_query ( + $inner_from, + $inner_select, + $where, + $inner_attrs, + ); }; # Generate the outer from - this is relatively easy (really just replace @@ -134,17 +162,21 @@ 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 - # 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] ]; # 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, + { + -alias => $attrs->{alias}, + -rsrc => $j->[0]{-rsrc}, + $attrs->{alias} => $inner_subq, + }, @{$j}[1 .. $#$j], ]; last; # we'll take care of what's left in $from below @@ -154,30 +186,52 @@ sub _adjust_select_args_for_complex_prefetch { } } - # scan the from spec against different attributes, and see which joins are needed + # 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, $outer_attrs ); + # unroll parents + my ($outer_select_chain, $outer_restrict_chain) = map { +{ + map { $_ => 1 } map { values %$_} map { @$_ } values %{ $outer_aliastypes->{$_} || {} } + } } qw/selecting restricting/; + # see what's left - throw away if not selecting/restricting - # also throw in a group_by if restricting to guard against - # cross-join explosions - # + # 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 $alias = $j->[0]{-alias}; - if ($outer_aliastypes->{selecting}{$alias}) { - push @outer_from, $j; + if ( + $outer_select_chain->{$alias} + ) { + push @outer_from, $j } - elsif ($outer_aliastypes->{restricting}{$alias}) { + elsif ($outer_restrict_chain->{$alias}) { push @outer_from, $j; - $outer_attrs->{group_by} ||= $outer_select unless $j->[0]{-is_single}; + $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; } } # demote the outer_from head $outer_from[0] = $outer_from[0][0]; + if ($need_outer_group_by and ! $outer_attrs->{group_by}) { + + 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; + + } + # 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 @@ -217,8 +271,13 @@ sub _resolve_aliastypes_from_select_args { or next; $alias_list->{$al} = $j; - $aliases_by_type->{multiplying}{$al} = 1 - unless $j->{-is_single}; + $aliases_by_type->{multiplying}{$al} ||= $j->{-join_path}||[] if ( + # not array == {from} head == can't be multiplying + ( ref($_) eq 'ARRAY' and ! $j->{-is_single} ) + or + # a parent of ours is already a multiplier + ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } ) + ); } # get a column to source/alias map (including unqualified ones) @@ -227,7 +286,11 @@ sub _resolve_aliastypes_from_select_args { # set up a botched SQLA my $sql_maker = $self->sql_maker; - local $sql_maker->{having_bind}; # these are throw away results + # these are throw away results, do not pollute the bind stack + local $sql_maker->{select_bind}; + local $sql_maker->{where_bind}; + local $sql_maker->{group_bind}; + local $sql_maker->{having_bind}; # we can't scan properly without any quoting (\b doesn't cut it # everywhere), so unless there is proper quoting set - use our @@ -239,7 +302,9 @@ sub _resolve_aliastypes_from_select_args { 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"; + $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} = ''; } @@ -254,8 +319,8 @@ sub _resolve_aliastypes_from_select_args { }), ], selecting => [ - $self->_extract_order_columns ($attrs->{order_by}, $sql_maker), $sql_maker->_recurse_fields ($select), + ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ), ], }; @@ -273,7 +338,8 @@ sub _resolve_aliastypes_from_select_args { for my $type (keys %$to_scan) { for my $piece (@{$to_scan->{$type}}) { - $aliases_by_type->{$type}{$alias} = 1 if ($piece =~ $al_re); + $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[] + if ($piece =~ $al_re); } } } @@ -287,7 +353,10 @@ sub _resolve_aliastypes_from_select_args { for my $type (keys %$to_scan) { for my $piece (@{$to_scan->{$type}}) { - $aliases_by_type->{$type}{$colinfo->{$col}{-source_alias}} = 1 if ($piece =~ $col_re); + if ($piece =~ $col_re) { + my $alias = $colinfo->{$col}{-source_alias}; + $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[]; + } } } } @@ -295,23 +364,62 @@ sub _resolve_aliastypes_from_select_args { # 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 ( + $aliases_by_type->{restricting}{$alias} ||= $j->{-join_path}||[] if ( (not $j->{-join_type}) or ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) ); } - # mark all join parents as mentioned - # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too ) - for my $type (keys %$aliases_by_type) { - 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); + + # 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 (@$select) { + if (! ref($_) 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->{$_}{-source_alias}.$_"}++; + } } } - return $aliases_by_type; + # add any order_by parts that are not already present in the group_by + # we need to be careful not to add any named functions/aggregates + # i.e. order_by => [ ... { count => 'foo' } ... ] + my @leftovers; + for ($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 { @@ -339,8 +447,8 @@ sub _resolve_ident_sources { $tabinfo = $_->[0]; } - $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-source_handle}->resolve - if ($tabinfo->{-source_handle}); + $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} + if ($tabinfo->{-rsrc}); } } @@ -358,7 +466,7 @@ sub _resolve_column_info { my ($self, $ident, $colnames) = @_; my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); - 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) @@ -375,26 +483,34 @@ sub _resolve_column_info { grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), ]; - COLUMN: + my (%return, $colinfos); foreach my $col (@$colnames) { - my ($alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; + my ($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} + || + $self->throw_exception( + "No such column '$colname' on source " . $rsrc->source_name + ); + }, -result_source => $rsrc, - -source_alias => $alias, + -source_alias => $source_alias, + -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, + -colname => $colname, }; + + $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; } return \%return; @@ -447,7 +563,7 @@ sub _inner_join_to_node { # So it looks like we will have to switch some stuff around. # local() is useless here as we will be leaving the scope # anyway, and deep cloning is just too fucking expensive - # So replace the first hashref in the node arrayref manually + # So replace the first hashref in the node arrayref manually my @new_from = ($from->[0]); my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path @@ -470,67 +586,62 @@ sub _inner_join_to_node { return \@new_from; } -# Most databases do not allow aliasing of tables in UPDATE/DELETE. Thus -# a condition containing 'me' or other table prefixes will not work -# at all. What this code tries to do (badly) is introspect the condition -# and remove all column qualifiers. If it bails out early (returns undef) -# the calling code should try another approach (e.g. a subquery) -sub _strip_cond_qualifiers { - my ($self, $where) = @_; - - my $cond = {}; - - # No-op. No condition, we're updating/deleting everything - return $cond unless $where; - - if (ref $where eq 'ARRAY') { - $cond = [ - map { - my %hash; - foreach my $key (keys %{$_}) { - $key =~ /([^.]+)$/; - $hash{$1} = $_->{$key}; - } - \%hash; - } @$where - ]; - } - elsif (ref $where eq 'HASH') { - if ( (keys %$where) == 1 && ( (keys %{$where})[0] eq '-and' )) { - $cond->{-and} = []; - my @cond = @{$where->{-and}}; - for (my $i = 0; $i < @cond; $i++) { - my $entry = $cond[$i]; - my $hash; - my $ref = ref $entry; - if ($ref eq 'HASH' or $ref eq 'ARRAY') { - $hash = $self->_strip_cond_qualifiers($entry); - } - elsif (! $ref) { - $entry =~ /([^.]+)$/; - $hash->{$1} = $cond[++$i]; +# yet another atrocity: attempt to extract all columns from a +# where condition by hooking _quote +sub _extract_condition_columns { + my ($self, $cond, $sql_maker) = @_; + + return [] unless $cond; + + $sql_maker ||= $self->{_sql_ident_capturer} ||= do { + # FIXME - replace with a Moo trait + my $orig_sm_class = ref $self->sql_maker; + my $smic_class = "${orig_sm_class}::_IdentCapture_"; + + unless ($smic_class->isa('SQL::Abstract')) { + + no strict 'refs'; + *{"${smic_class}::_quote"} = subname "${smic_class}::_quote" => sub { + my ($self, $ident) = @_; + if (ref $ident eq 'SCALAR') { + $ident = $$ident; + my $storage_quotes = $self->sql_quote_char || '"'; + my ($ql, $qr) = map + { quotemeta $_ } + (ref $storage_quotes eq 'ARRAY' ? @$storage_quotes : ($storage_quotes) x 2 ) + ; + + while ($ident =~ / + $ql (\w+) $qr + | + ([\w\.]+) + /xg) { + $self->{_captured_idents}{$1||$2}++; + } } else { - $self->throw_exception ("_strip_cond_qualifiers() is unable to handle a condition reftype $ref"); + $self->{_captured_idents}{$ident}++; } - push @{$cond->{-and}}, $hash; - } - } - else { - foreach my $key (keys %$where) { - $key =~ /([^.]+)$/; - $cond->{$1} = $where->{$key}; - } + return $ident; + }; + + *{"${smic_class}::_get_captured_idents"} = subname "${smic_class}::_get_captures" => sub { + (delete shift->{_captured_idents}) || {}; + }; + + $self->inject_base ($smic_class, $orig_sm_class); + } - } - else { - return undef; - } - return $cond; + $smic_class->new(); + }; + + $sql_maker->_recurse_where($cond); + + return [ sort keys %{$sql_maker->_get_captured_idents} ]; } -sub _extract_order_columns { +sub _extract_order_criteria { my ($self, $order_by, $sql_maker) = @_; my $parser = sub { @@ -540,8 +651,9 @@ sub _extract_order_columns { unless wantarray; my @chunks; - for my $chunk (map { ref $_ ? @$_ : $_ } ($sql_maker->_order_by_chunks ($order_by) ) ) { - $chunk =~ s/\s+ (?: ASC|DESC ) \s* $//ix; + for ($sql_maker->_order_by_chunks ($order_by) ) { + my $chunk = ref $_ ? $_ : [ $_ ]; + $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix; push @chunks, $chunk; } @@ -558,4 +670,62 @@ sub _extract_order_columns { } } +sub _order_by_is_stable { + my ($self, $ident, $order_by, $where) = @_; + + my $colinfo = $self->_resolve_column_info($ident, [ + (map { $_->[0] } $self->_extract_order_criteria($order_by)), + $where ? @{$self->_extract_fixed_condition_columns($where)} :(), + ]); + + return undef unless keys %$colinfo; + + my $cols_per_src; + $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo; + + for (values %$cols_per_src) { + my $src = (values %$_)[0]->{-result_source}; + return 1 if $src->_identifying_column_set($_); + } + + return undef; +} + +# returns an arrayref of column names which *definitely* have som +# 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, +# which in turn allows us to better predict how ordering will behave +# etc. +# +# this is a rudimentary, incomplete, and error-prone extractor +# however this is OK - it is conservative, and if we can not find +# 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) = @_; + + return unless ref $where eq 'HASH'; + + my @cols; + 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->{$lhs}, 1) + ; + } + elsif ($lhs !~ /^\-/) { + my $val = $where->{$lhs}; + + push @cols, $lhs if (defined $val and ( + ! ref $val + or + (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='}) + )); + } + } + return $nested ? @cols : \@cols; +} + 1;