1 package #hide from PAUSE
2 DBIx::Class::Storage::DBIHacks;
5 # This module contains code that should never have seen the light of day,
6 # does not belong in the Storage, or is otherwise unfit for public
7 # display. The arrival of SQLA2 should immediately obsolete 90% of this
13 use base 'DBIx::Class::Storage';
16 use List::Util 'first';
17 use Scalar::Util 'blessed';
18 use Sub::Name 'subname';
22 # This code will remove non-selecting/non-restricting joins from
23 # {from} specs, aiding the RDBMS query optimizer
25 sub _prune_unused_joins {
27 my ($from, $select, $where, $attrs) = @_;
29 return $from unless $self->_use_join_optimizer;
31 if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY') {
32 return $from; # only standard {from} specs are supported
35 my $aliastypes = $self->_resolve_aliastypes_from_select_args(@_);
37 # a grouped set will not be affected by amount of rows. Thus any
38 # {multiplying} joins can go
39 delete $aliastypes->{multiplying} if $attrs->{group_by};
41 my @newfrom = $from->[0]; # FROM head is always present
44 for (values %$aliastypes) {
45 # add all requested aliases
46 $need_joins{$_} = 1 for keys %$_;
48 # add all their parents (as per joinpath which is an AoH { table => alias })
49 $need_joins{$_} = 1 for map { values %$_ } map { @$_ } values %$_;
51 for my $j (@{$from}[1..$#$from]) {
52 push @newfrom, $j if (
53 (! $j->[0]{-alias}) # legacy crap
55 $need_joins{$j->[0]{-alias}}
63 # This is the code producing joined subqueries like:
64 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
66 sub _adjust_select_args_for_complex_prefetch {
67 my ($self, $from, $select, $where, $attrs) = @_;
69 $self->throw_exception ('Nothing to prefetch... how did we get here?!')
70 if not @{$attrs->{_prefetch_selector_range}};
72 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
73 if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY');
76 # generate inner/outer attribute lists, remove stuff that doesn't apply
77 my $outer_attrs = { %$attrs };
78 delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/;
80 my $inner_attrs = { %$attrs, _is_internal_subuery => 1 };
81 delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range select as/;
83 # generate the inner/outer select lists
84 # for inside we consider only stuff *not* brought in by the prefetch
85 # on the outside we substitute any function for its alias
86 my $outer_select = [ @$select ];
87 my $inner_select = [];
89 my ($p_start, $p_end) = @{$outer_attrs->{_prefetch_selector_range}};
90 for my $i (0 .. $p_start - 1, $p_end + 1 .. $#$outer_select) {
91 my $sel = $outer_select->[$i];
93 if (ref $sel eq 'HASH' ) {
94 $sel->{-as} ||= $attrs->{as}[$i];
95 $outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") );
98 push @$inner_select, $sel;
100 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
103 # construct the inner $from and lock it in a subquery
104 # we need to prune first, because this will determine if we need a group_by below
105 # the fake group_by is so that the pruner throws away all non-selecting, non-restricting
106 # multijoins (since we def. do not care about those inside the subquery)
108 my $inner_subq = do {
110 # must use it here regardless of user requests
111 local $self->{_use_join_optimizer} = 1;
113 my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, {
114 group_by => ['dummy'], %$inner_attrs,
117 my $inner_aliastypes =
118 $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs );
120 # we need to simulate collapse in the subq if a multiplying join is pulled
121 # by being a non-selecting restrictor
123 ! $inner_attrs->{group_by}
126 $inner_aliastypes->{restricting}{$_}
128 ! $inner_aliastypes->{selecting}{$_}
129 } ( keys %{$inner_aliastypes->{multiplying}||{}} )
131 my $unprocessed_order_chunks;
132 ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
133 $inner_from, $inner_select, $inner_attrs->{order_by}
136 $self->throw_exception (
137 'A required group_by clause could not be constructed automatically due to a complex '
138 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
140 ) if $unprocessed_order_chunks;
143 # we already optimized $inner_from above
144 local $self->{_use_join_optimizer} = 0;
146 # generate the subquery
147 $self->_select_args_to_query (
155 # Generate the outer from - this is relatively easy (really just replace
156 # the join slot with the subquery), with a major caveat - we can not
157 # join anything that is non-selecting (not part of the prefetch), but at
158 # the same time is a multi-type relationship, as it will explode the result.
160 # There are two possibilities here
161 # - either the join is non-restricting, in which case we simply throw it away
162 # - it is part of the restrictions, in which case we need to collapse the outer
163 # result by tackling yet another group_by to the outside of the query
167 # so first generate the outer_from, up to the substitution point
169 while (my $j = shift @$from) {
170 $j = [ $j ] unless ref $j eq 'ARRAY'; # promote the head-from to an AoH
172 if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap
176 -alias => $attrs->{alias},
177 -rsrc => $j->[0]{-rsrc},
178 $attrs->{alias} => $inner_subq,
182 last; # we'll take care of what's left in $from below
185 push @outer_from, $j;
189 # scan the *remaining* from spec against different attributes, and see which joins are needed
191 my $outer_aliastypes =
192 $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs );
195 my ($outer_select_chain, $outer_restrict_chain) = map { +{
196 map { $_ => 1 } map { values %$_} map { @$_ } values %{ $outer_aliastypes->{$_} || {} }
197 } } qw/selecting restricting/;
199 # see what's left - throw away if not selecting/restricting
200 # also throw in a group_by if a non-selecting multiplier,
201 # to guard against cross-join explosions
202 my $need_outer_group_by;
203 while (my $j = shift @$from) {
204 my $alias = $j->[0]{-alias};
207 $outer_select_chain->{$alias}
211 elsif ($outer_restrict_chain->{$alias}) {
212 push @outer_from, $j;
213 $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
217 # demote the outer_from head
218 $outer_from[0] = $outer_from[0][0];
220 if ($need_outer_group_by and ! $outer_attrs->{group_by}) {
222 my $unprocessed_order_chunks;
223 ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
224 \@outer_from, $outer_select, $outer_attrs->{order_by}
227 $self->throw_exception (
228 'A required group_by clause could not be constructed automatically due to a complex '
229 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
231 ) if $unprocessed_order_chunks;
235 # This is totally horrific - the $where ends up in both the inner and outer query
236 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
237 # then if where conditions apply to the *right* side of the prefetch, you may have
238 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
239 # the outer select to exclude joins you didin't want in the first place
241 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
242 return (\@outer_from, $outer_select, $where, $outer_attrs);
246 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
248 # Due to a lack of SQLA2 we fall back to crude scans of all the
249 # select/where/order/group attributes, in order to determine what
250 # aliases are neded to fulfill the query. This information is used
251 # throughout the code to prune unnecessary JOINs from the queries
252 # in an attempt to reduce the execution time.
253 # Although the method is pretty horrific, the worst thing that can
254 # happen is for it to fail due to some scalar SQL, which in turn will
255 # result in a vocal exception.
256 sub _resolve_aliastypes_from_select_args {
257 my ( $self, $from, $select, $where, $attrs ) = @_;
259 $self->throw_exception ('Unable to analyze custom {from}')
260 if ref $from ne 'ARRAY';
262 # what we will return
265 # see what aliases are there to work with
269 $j = $j->[0] if ref $j eq 'ARRAY';
270 my $al = $j->{-alias}
273 $alias_list->{$al} = $j;
274 $aliases_by_type->{multiplying}{$al} ||= $j->{-join_path}||[] if (
275 # not array == {from} head == can't be multiplying
276 ( ref($_) eq 'ARRAY' and ! $j->{-is_single} )
278 # a parent of ours is already a multiplier
279 ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } )
283 # get a column to source/alias map (including unqualified ones)
284 my $colinfo = $self->_resolve_column_info ($from);
286 # set up a botched SQLA
287 my $sql_maker = $self->sql_maker;
289 # these are throw away results, do not pollute the bind stack
290 local $sql_maker->{select_bind};
291 local $sql_maker->{where_bind};
292 local $sql_maker->{group_bind};
293 local $sql_maker->{having_bind};
295 # we can't scan properly without any quoting (\b doesn't cut it
296 # everywhere), so unless there is proper quoting set - use our
297 # own weird impossible character.
298 # Also in the case of no quoting, we need to explicitly disable
299 # name_sep, otherwise sorry nasty legacy syntax like
300 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
301 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
302 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
304 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
305 $sql_maker->{quote_char} = ["\x00", "\xFF"];
306 # if we don't unset it we screw up retarded but unfortunately working
307 # 'MAX(foo.bar)' => { '>', 3 }
308 $sql_maker->{name_sep} = '';
311 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
313 # generate sql chunks
316 $sql_maker->_recurse_where ($where),
317 $sql_maker->_parse_rs_attrs ({
318 map { $_ => $attrs->{$_} } (qw/group_by having/)
322 $sql_maker->_recurse_fields ($select),
323 ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ),
327 # throw away empty chunks
328 $_ = [ map { $_ || () } @$_ ] for values %$to_scan;
330 # first loop through all fully qualified columns and get the corresponding
331 # alias (should work even if they are in scalarrefs)
332 for my $alias (keys %$alias_list) {
334 $lquote $alias $rquote $sep
339 for my $type (keys %$to_scan) {
340 for my $piece (@{$to_scan->{$type}}) {
341 $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[]
342 if ($piece =~ $al_re);
347 # now loop through unqualified column names, and try to locate them within
349 for my $col (keys %$colinfo) {
350 next if $col =~ / \. /x; # if column is qualified it was caught by the above
352 my $col_re = qr/ $lquote $col $rquote /x;
354 for my $type (keys %$to_scan) {
355 for my $piece (@{$to_scan->{$type}}) {
356 if ($piece =~ $col_re) {
357 my $alias = $colinfo->{$col}{-source_alias};
358 $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[];
364 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
365 for my $j (values %$alias_list) {
366 my $alias = $j->{-alias} or next;
367 $aliases_by_type->{restricting}{$alias} ||= $j->{-join_path}||[] if (
368 (not $j->{-join_type})
370 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
374 return $aliases_by_type;
377 # This is the engine behind { distinct => 1 }
378 sub _group_over_selection {
379 my ($self, $from, $select, $order_by) = @_;
381 my $rs_column_list = $self->_resolve_column_info ($from);
383 my (@group_by, %group_index);
385 # the logic is: if it is a { func => val } we assume an aggregate,
386 # otherwise if \'...' or \[...] we assume the user knows what is
387 # going on thus group over it
389 if (! ref($_) or ref ($_) ne 'HASH' ) {
392 if ($rs_column_list->{$_} and $_ !~ /\./ ) {
393 # add a fully qualified version as well
394 $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++;
399 # add any order_by parts that are not already present in the group_by
400 # we need to be careful not to add any named functions/aggregates
401 # i.e. order_by => [ ... { count => 'foo' } ... ]
403 for ($self->_extract_order_criteria($order_by)) {
404 # only consider real columns (for functions the user got to do an explicit group_by)
410 my $colinfo = $rs_column_list->{$chunk} or do {
415 $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./;
416 push @group_by, $chunk unless $group_index{$chunk}++;
420 ? (\@group_by, (@leftovers ? \@leftovers : undef) )
425 sub _resolve_ident_sources {
426 my ($self, $ident) = @_;
428 my $alias2source = {};
431 # the reason this is so contrived is that $ident may be a {from}
432 # structure, specifying multiple tables to join
433 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
434 # this is compat mode for insert/update/delete which do not deal with aliases
435 $alias2source->{me} = $ident;
438 elsif (ref $ident eq 'ARRAY') {
442 if (ref $_ eq 'HASH') {
444 $rs_alias = $tabinfo->{-alias};
446 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
450 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
451 if ($tabinfo->{-rsrc});
455 return ($alias2source, $rs_alias);
458 # Takes $ident, \@column_names
460 # returns { $column_name => \%column_info, ... }
461 # also note: this adds -result_source => $rsrc to the column info
463 # If no columns_names are supplied returns info about *all* columns
465 sub _resolve_column_info {
466 my ($self, $ident, $colnames) = @_;
467 my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident);
469 my (%seen_cols, @auto_colnames);
471 # compile a global list of column names, to be able to properly
472 # disambiguate unqualified column names (if at all possible)
473 for my $alias (keys %$alias2src) {
474 my $rsrc = $alias2src->{$alias};
475 for my $colname ($rsrc->columns) {
476 push @{$seen_cols{$colname}}, $alias;
477 push @auto_colnames, "$alias.$colname" unless $colnames;
483 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
486 my (%return, $colinfos);
487 foreach my $col (@$colnames) {
488 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
490 # if the column was seen exactly once - we know which rsrc it came from
491 $source_alias ||= $seen_cols{$colname}[0]
492 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
494 next unless $source_alias;
496 my $rsrc = $alias2src->{$source_alias}
501 ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
503 $self->throw_exception(
504 "No such column '$colname' on source " . $rsrc->source_name
507 -result_source => $rsrc,
508 -source_alias => $source_alias,
509 -fq_colname => $col eq $colname ? "$source_alias.$col" : $col,
510 -colname => $colname,
513 $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname;
519 # The DBIC relationship chaining implementation is pretty simple - every
520 # new related_relationship is pushed onto the {from} stack, and the {select}
521 # window simply slides further in. This means that when we count somewhere
522 # in the middle, we got to make sure that everything in the join chain is an
523 # actual inner join, otherwise the count will come back with unpredictable
524 # results (a resultset may be generated with _some_ rows regardless of if
525 # the relation which the $rs currently selects has rows or not). E.g.
526 # $artist_rs->cds->count - normally generates:
527 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
528 # which actually returns the number of artists * (number of cds || 1)
530 # So what we do here is crawl {from}, determine if the current alias is at
531 # the top of the stack, and if not - make sure the chain is inner-joined down
534 sub _inner_join_to_node {
535 my ($self, $from, $alias) = @_;
537 # subqueries and other oddness are naturally not supported
543 ref $from->[0] ne 'HASH'
547 $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do
550 # find the current $alias in the $from structure
553 for my $j (@{$from}[1 .. $#$from]) {
554 if ($j->[0]{-alias} eq $alias) {
555 $switch_branch = $j->[0]{-join_path};
560 # something else went quite wrong
561 return $from unless $switch_branch;
563 # So it looks like we will have to switch some stuff around.
564 # local() is useless here as we will be leaving the scope
565 # anyway, and deep cloning is just too fucking expensive
566 # So replace the first hashref in the node arrayref manually
567 my @new_from = ($from->[0]);
568 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
570 for my $j (@{$from}[1 .. $#$from]) {
571 my $jalias = $j->[0]{-alias};
573 if ($sw_idx->{$jalias}) {
574 my %attrs = %{$j->[0]};
575 delete $attrs{-join_type};
589 # yet another atrocity: attempt to extract all columns from a
590 # where condition by hooking _quote
591 sub _extract_condition_columns {
592 my ($self, $cond, $sql_maker) = @_;
594 return [] unless $cond;
596 $sql_maker ||= $self->{_sql_ident_capturer} ||= do {
597 # FIXME - replace with a Moo trait
598 my $orig_sm_class = ref $self->sql_maker;
599 my $smic_class = "${orig_sm_class}::_IdentCapture_";
601 unless ($smic_class->isa('SQL::Abstract')) {
604 *{"${smic_class}::_quote"} = subname "${smic_class}::_quote" => sub {
605 my ($self, $ident) = @_;
606 if (ref $ident eq 'SCALAR') {
608 my $storage_quotes = $self->sql_quote_char || '"';
611 (ref $storage_quotes eq 'ARRAY' ? @$storage_quotes : ($storage_quotes) x 2 )
619 $self->{_captured_idents}{$1||$2}++;
623 $self->{_captured_idents}{$ident}++;
628 *{"${smic_class}::_get_captured_idents"} = subname "${smic_class}::_get_captures" => sub {
629 (delete shift->{_captured_idents}) || {};
632 $self->inject_base ($smic_class, $orig_sm_class);
639 $sql_maker->_recurse_where($cond);
641 return [ sort keys %{$sql_maker->_get_captured_idents} ];
644 sub _extract_order_criteria {
645 my ($self, $order_by, $sql_maker) = @_;
648 my ($sql_maker, $order_by) = @_;
650 return scalar $sql_maker->_order_by_chunks ($order_by)
654 for ($sql_maker->_order_by_chunks ($order_by) ) {
655 my $chunk = ref $_ ? $_ : [ $_ ];
656 $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix;
657 push @chunks, $chunk;
664 return $parser->($sql_maker, $order_by);
667 $sql_maker = $self->sql_maker;
668 local $sql_maker->{quote_char};
669 return $parser->($sql_maker, $order_by);
673 sub _order_by_is_stable {
674 my ($self, $ident, $order_by, $where) = @_;
676 my $colinfo = $self->_resolve_column_info($ident, [
677 (map { $_->[0] } $self->_extract_order_criteria($order_by)),
678 $where ? @{$self->_extract_fixed_condition_columns($where)} :(),
681 return undef unless keys %$colinfo;
684 $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo;
686 for (values %$cols_per_src) {
687 my $src = (values %$_)[0]->{-result_source};
688 return 1 if $src->_identifying_column_set($_);
694 # returns an arrayref of column names which *definitely* have som
695 # sort of non-nullable equality requested in the given condition
696 # specification. This is used to figure out if a resultset is
697 # constrained to a column which is part of a unique constraint,
698 # which in turn allows us to better predict how ordering will behave
701 # this is a rudimentary, incomplete, and error-prone extractor
702 # however this is OK - it is conservative, and if we can not find
703 # something that is in fact there - the stack will recover gracefully
704 # Also - DQ and the mst it rode in on will save us all RSN!!!
705 sub _extract_fixed_condition_columns {
706 my ($self, $where, $nested) = @_;
708 return unless ref $where eq 'HASH';
711 for my $lhs (keys %$where) {
712 if ($lhs =~ /^\-and$/i) {
713 push @cols, ref $where->{$lhs} eq 'ARRAY'
714 ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} )
715 : $self->_extract_fixed_condition_columns($where->{$lhs}, 1)
718 elsif ($lhs !~ /^\-/) {
719 my $val = $where->{$lhs};
721 push @cols, $lhs if (defined $val and (
724 (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='})
728 return $nested ? @cols : \@cols;