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, $ignore_multiplication) = @_;
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(@_);
38 delete $aliastypes->{joining};
40 # a grouped set will not be affected by amount of rows. Thus any
41 # {multiplying} joins can go
42 delete $aliastypes->{multiplying}
43 if $ignore_multiplication or $attrs->{group_by};
45 my @newfrom = $from->[0]; # FROM head is always present
49 for (values %$aliastypes) {
50 # add all requested aliases
51 $need_joins{$_} = 1 for keys %$_;
53 # add all their parents (as per joinpath which is an AoH { table => alias })
54 $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_;
57 for my $j (@{$from}[1..$#$from]) {
58 push @newfrom, $j if (
59 (! $j->[0]{-alias}) # legacy crap
61 $need_joins{$j->[0]{-alias}}
69 # This is the code producing joined subqueries like:
70 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
72 sub _adjust_select_args_for_complex_prefetch {
73 my ($self, $from, $select, $where, $attrs) = @_;
75 $self->throw_exception ('Nothing to prefetch... how did we get here?!')
76 if not @{$attrs->{_prefetch_selector_range}||[]};
78 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
79 if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY');
81 # generate inner/outer attribute lists, remove stuff that doesn't apply
82 my $outer_attrs = { %$attrs };
83 delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/;
85 my $inner_attrs = { %$attrs };
86 delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range select as/;
88 # if the user did not request it, there is no point using it inside
89 delete $inner_attrs->{order_by} if delete $inner_attrs->{_order_is_artificial};
91 # generate the inner/outer select lists
92 # for inside we consider only stuff *not* brought in by the prefetch
93 # on the outside we substitute any function for its alias
94 my $outer_select = [ @$select ];
97 my ($root_node, $root_node_offset);
99 for my $i (0 .. $#$from) {
100 my $node = $from->[$i];
101 my $h = (ref $node eq 'HASH') ? $node
102 : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
106 if ( ($h->{-alias}||'') eq $attrs->{alias} and $h->{-rsrc} ) {
108 $root_node_offset = $i;
113 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
116 # use the heavy duty resolver to take care of aliased/nonaliased naming
117 my $colinfo = $self->_resolve_column_info($from);
118 my $selected_root_columns;
120 my ($p_start, $p_end) = @{$outer_attrs->{_prefetch_selector_range}};
121 for my $i (0 .. $p_start - 1, $p_end + 1 .. $#$outer_select) {
122 my $sel = $outer_select->[$i];
124 if (ref $sel eq 'HASH' ) {
125 $sel->{-as} ||= $attrs->{as}[$i];
126 $outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") );
128 elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
129 $selected_root_columns->{$ci->{-colname}} = 1;
132 push @$inner_select, $sel;
134 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
137 # We will need to fetch all native columns in the inner subquery, which may
138 # be a part of an *outer* join condition, or an order_by (which needs to be
140 # We can not just fetch everything because a potential has_many restricting
141 # join collapse *will not work* on heavy data types.
142 my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args(
143 [grep { ref($_) eq 'ARRAY' or ref($_) eq 'HASH' } @{$from}[$root_node_offset .. $#$from]],
149 for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) {
150 my $ci = $colinfo->{$_} or next;
152 $ci->{-source_alias} eq $attrs->{alias}
154 ! $selected_root_columns->{$ci->{-colname}}++
156 # adding it to both to keep limits not supporting dark selectors happy
157 push @$inner_select, $ci->{-fq_colname};
158 push @{$inner_attrs->{as}}, $ci->{-fq_colname};
162 # construct the inner $from and lock it in a subquery
163 # we need to prune first, because this will determine if we need a group_by below
164 # throw away all non-selecting, non-restricting multijoins
165 # (since we def. do not care about multiplication those inside the subquery)
166 my $inner_subq = do {
168 # must use it here regardless of user requests
169 local $self->{_use_join_optimizer} = 1;
171 # throw away multijoins since we def. do not care about those inside the subquery
172 my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, $inner_attrs, 'ignore_multiplication');
174 my $inner_aliastypes =
175 $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs );
177 # we need to simulate collapse in the subq if a multiplying join is pulled
178 # by being a non-selecting restrictor
180 ! $inner_attrs->{group_by}
183 $inner_aliastypes->{restricting}{$_}
185 ! $inner_aliastypes->{selecting}{$_}
186 } ( keys %{$inner_aliastypes->{multiplying}||{}} )
188 my $unprocessed_order_chunks;
189 ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
190 $inner_from, $inner_select, $inner_attrs->{order_by}
193 $self->throw_exception (
194 'A required group_by clause could not be constructed automatically due to a complex '
195 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
197 ) if $unprocessed_order_chunks;
200 # we already optimized $inner_from above
201 # and already local()ized
202 $self->{_use_join_optimizer} = 0;
204 # generate the subquery
205 $self->_select_args_to_query (
213 # Generate the outer from - this is relatively easy (really just replace
214 # the join slot with the subquery), with a major caveat - we can not
215 # join anything that is non-selecting (not part of the prefetch), but at
216 # the same time is a multi-type relationship, as it will explode the result.
218 # There are two possibilities here
219 # - either the join is non-restricting, in which case we simply throw it away
220 # - it is part of the restrictions, in which case we need to collapse the outer
221 # result by tackling yet another group_by to the outside of the query
223 # work on a shallow copy
228 # we may not be the head
229 if ($root_node_offset) {
230 # first generate the outer_from, up to the substitution point
231 @outer_from = splice @$from, 0, $root_node_offset;
235 -alias => $attrs->{alias},
236 -rsrc => $root_node->{-rsrc},
237 $attrs->{alias} => $inner_subq,
239 @{$from->[0]}[1 .. $#{$from->[0]}],
244 -alias => $attrs->{alias},
245 -rsrc => $root_node->{-rsrc},
246 $attrs->{alias} => $inner_subq,
250 shift @$from; # it's replaced in @outer_from already
252 # scan the *remaining* from spec against different attributes, and see which joins are needed
254 my $outer_aliastypes =
255 $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs );
258 my ($outer_select_chain, $outer_restrict_chain) = map { +{
259 map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} }
260 } } qw/selecting restricting/;
262 # see what's left - throw away if not selecting/restricting
263 # also throw in a group_by if a non-selecting multiplier,
264 # to guard against cross-join explosions
265 my $need_outer_group_by;
266 while (my $j = shift @$from) {
267 my $alias = $j->[0]{-alias};
270 $outer_select_chain->{$alias}
274 elsif ($outer_restrict_chain->{$alias}) {
275 push @outer_from, $j;
276 $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
280 if ($need_outer_group_by and ! $outer_attrs->{group_by}) {
282 my $unprocessed_order_chunks;
283 ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
284 \@outer_from, $outer_select, $outer_attrs->{order_by}
287 $self->throw_exception (
288 'A required group_by clause could not be constructed automatically due to a complex '
289 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
291 ) if $unprocessed_order_chunks;
295 # This is totally horrific - the $where ends up in both the inner and outer query
296 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
297 # then if where conditions apply to the *right* side of the prefetch, you may have
298 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
299 # the outer select to exclude joins you didin't want in the first place
301 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
302 return (\@outer_from, $outer_select, $where, $outer_attrs);
306 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
308 # Due to a lack of SQLA2 we fall back to crude scans of all the
309 # select/where/order/group attributes, in order to determine what
310 # aliases are neded to fulfill the query. This information is used
311 # throughout the code to prune unnecessary JOINs from the queries
312 # in an attempt to reduce the execution time.
313 # Although the method is pretty horrific, the worst thing that can
314 # happen is for it to fail due to some scalar SQL, which in turn will
315 # result in a vocal exception.
316 sub _resolve_aliastypes_from_select_args {
317 my ( $self, $from, $select, $where, $attrs ) = @_;
319 $self->throw_exception ('Unable to analyze custom {from}')
320 if ref $from ne 'ARRAY';
322 # what we will return
325 # see what aliases are there to work with
329 $j = $j->[0] if ref $j eq 'ARRAY';
330 my $al = $j->{-alias}
333 $alias_list->{$al} = $j;
334 $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } if (
335 # not array == {from} head == can't be multiplying
336 ( ref($_) eq 'ARRAY' and ! $j->{-is_single} )
338 # a parent of ours is already a multiplier
339 ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } )
343 # get a column to source/alias map (including unqualified ones)
344 my $colinfo = $self->_resolve_column_info ($from);
346 # set up a botched SQLA
347 my $sql_maker = $self->sql_maker;
349 # these are throw away results, do not pollute the bind stack
350 local $sql_maker->{select_bind};
351 local $sql_maker->{where_bind};
352 local $sql_maker->{group_bind};
353 local $sql_maker->{having_bind};
354 local $sql_maker->{from_bind};
356 # we can't scan properly without any quoting (\b doesn't cut it
357 # everywhere), so unless there is proper quoting set - use our
358 # own weird impossible character.
359 # Also in the case of no quoting, we need to explicitly disable
360 # name_sep, otherwise sorry nasty legacy syntax like
361 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
362 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
363 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
365 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
366 $sql_maker->{quote_char} = ["\x00", "\xFF"];
367 # if we don't unset it we screw up retarded but unfortunately working
368 # 'MAX(foo.bar)' => { '>', 3 }
369 $sql_maker->{name_sep} = '';
372 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
374 # generate sql chunks
377 $sql_maker->_recurse_where ($where),
378 $sql_maker->_parse_rs_attrs ({
379 map { $_ => $attrs->{$_} } (qw/group_by having/)
383 $sql_maker->_recurse_from (
384 ref $from->[0] eq 'ARRAY' ? $from->[0][0] : $from->[0],
385 @{$from}[1 .. $#$from],
389 $sql_maker->_recurse_fields ($select),
390 ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ),
394 # throw away empty chunks
395 $_ = [ map { $_ || () } @$_ ] for values %$to_scan;
397 # first loop through all fully qualified columns and get the corresponding
398 # alias (should work even if they are in scalarrefs)
399 for my $alias (keys %$alias_list) {
401 $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
403 \b $alias \. ([^\s\)\($rquote]+)?
406 for my $type (keys %$to_scan) {
407 for my $piece (@{$to_scan->{$type}}) {
408 if (my @matches = $piece =~ /$al_re/g) {
409 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
410 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = 1
411 for grep { defined $_ } @matches;
417 # now loop through unqualified column names, and try to locate them within
419 for my $col (keys %$colinfo) {
420 next if $col =~ / \. /x; # if column is qualified it was caught by the above
422 my $col_re = qr/ $lquote ($col) $rquote /x;
424 for my $type (keys %$to_scan) {
425 for my $piece (@{$to_scan->{$type}}) {
426 if (my @matches = $piece =~ /$col_re/g) {
427 my $alias = $colinfo->{$col}{-source_alias};
428 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
429 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = 1
430 for grep { defined $_ } @matches;
436 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
437 for my $j (values %$alias_list) {
438 my $alias = $j->{-alias} or next;
439 $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if (
440 (not $j->{-join_type})
442 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
446 return $aliases_by_type;
449 # This is the engine behind { distinct => 1 }
450 sub _group_over_selection {
451 my ($self, $from, $select, $order_by) = @_;
453 my $rs_column_list = $self->_resolve_column_info ($from);
455 my (@group_by, %group_index);
457 # the logic is: if it is a { func => val } we assume an aggregate,
458 # otherwise if \'...' or \[...] we assume the user knows what is
459 # going on thus group over it
461 if (! ref($_) or ref ($_) ne 'HASH' ) {
464 if ($rs_column_list->{$_} and $_ !~ /\./ ) {
465 # add a fully qualified version as well
466 $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++;
471 # add any order_by parts that are not already present in the group_by
472 # we need to be careful not to add any named functions/aggregates
473 # i.e. order_by => [ ... { count => 'foo' } ... ]
475 for ($self->_extract_order_criteria($order_by)) {
476 # only consider real columns (for functions the user got to do an explicit group_by)
482 my $colinfo = $rs_column_list->{$chunk} or do {
487 $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./;
488 push @group_by, $chunk unless $group_index{$chunk}++;
492 ? (\@group_by, (@leftovers ? \@leftovers : undef) )
497 sub _resolve_ident_sources {
498 my ($self, $ident) = @_;
500 my $alias2source = {};
503 # the reason this is so contrived is that $ident may be a {from}
504 # structure, specifying multiple tables to join
505 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
506 # this is compat mode for insert/update/delete which do not deal with aliases
507 $alias2source->{me} = $ident;
510 elsif (ref $ident eq 'ARRAY') {
514 if (ref $_ eq 'HASH') {
516 $rs_alias = $tabinfo->{-alias};
518 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
522 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
523 if ($tabinfo->{-rsrc});
527 return ($alias2source, $rs_alias);
530 # Takes $ident, \@column_names
532 # returns { $column_name => \%column_info, ... }
533 # also note: this adds -result_source => $rsrc to the column info
535 # If no columns_names are supplied returns info about *all* columns
537 sub _resolve_column_info {
538 my ($self, $ident, $colnames) = @_;
539 my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident);
541 my (%seen_cols, @auto_colnames);
543 # compile a global list of column names, to be able to properly
544 # disambiguate unqualified column names (if at all possible)
545 for my $alias (keys %$alias2src) {
546 my $rsrc = $alias2src->{$alias};
547 for my $colname ($rsrc->columns) {
548 push @{$seen_cols{$colname}}, $alias;
549 push @auto_colnames, "$alias.$colname" unless $colnames;
555 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
558 my (%return, $colinfos);
559 foreach my $col (@$colnames) {
560 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
562 # if the column was seen exactly once - we know which rsrc it came from
563 $source_alias ||= $seen_cols{$colname}[0]
564 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
566 next unless $source_alias;
568 my $rsrc = $alias2src->{$source_alias}
573 ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
575 $self->throw_exception(
576 "No such column '$colname' on source " . $rsrc->source_name
579 -result_source => $rsrc,
580 -source_alias => $source_alias,
581 -fq_colname => $col eq $colname ? "$source_alias.$col" : $col,
582 -colname => $colname,
585 $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname;
591 # The DBIC relationship chaining implementation is pretty simple - every
592 # new related_relationship is pushed onto the {from} stack, and the {select}
593 # window simply slides further in. This means that when we count somewhere
594 # in the middle, we got to make sure that everything in the join chain is an
595 # actual inner join, otherwise the count will come back with unpredictable
596 # results (a resultset may be generated with _some_ rows regardless of if
597 # the relation which the $rs currently selects has rows or not). E.g.
598 # $artist_rs->cds->count - normally generates:
599 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
600 # which actually returns the number of artists * (number of cds || 1)
602 # So what we do here is crawl {from}, determine if the current alias is at
603 # the top of the stack, and if not - make sure the chain is inner-joined down
606 sub _inner_join_to_node {
607 my ($self, $from, $alias) = @_;
609 # subqueries and other oddness are naturally not supported
615 ref $from->[0] ne 'HASH'
619 $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do
622 # find the current $alias in the $from structure
625 for my $j (@{$from}[1 .. $#$from]) {
626 if ($j->[0]{-alias} eq $alias) {
627 $switch_branch = $j->[0]{-join_path};
632 # something else went quite wrong
633 return $from unless $switch_branch;
635 # So it looks like we will have to switch some stuff around.
636 # local() is useless here as we will be leaving the scope
637 # anyway, and deep cloning is just too fucking expensive
638 # So replace the first hashref in the node arrayref manually
639 my @new_from = ($from->[0]);
640 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
642 for my $j (@{$from}[1 .. $#$from]) {
643 my $jalias = $j->[0]{-alias};
645 if ($sw_idx->{$jalias}) {
646 my %attrs = %{$j->[0]};
647 delete $attrs{-join_type};
661 sub _extract_order_criteria {
662 my ($self, $order_by, $sql_maker) = @_;
665 my ($sql_maker, $order_by) = @_;
667 return scalar $sql_maker->_order_by_chunks ($order_by)
671 for ($sql_maker->_order_by_chunks ($order_by) ) {
672 my $chunk = ref $_ ? $_ : [ $_ ];
673 $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix;
674 push @chunks, $chunk;
681 return $parser->($sql_maker, $order_by);
684 $sql_maker = $self->sql_maker;
685 local $sql_maker->{quote_char};
686 return $parser->($sql_maker, $order_by);
690 sub _order_by_is_stable {
691 my ($self, $ident, $order_by, $where) = @_;
693 my $colinfo = $self->_resolve_column_info($ident, [
694 (map { $_->[0] } $self->_extract_order_criteria($order_by)),
695 $where ? @{$self->_extract_fixed_condition_columns($where)} :(),
698 return undef unless keys %$colinfo;
701 $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo;
703 for (values %$cols_per_src) {
704 my $src = (values %$_)[0]->{-result_source};
705 return 1 if $src->_identifying_column_set($_);
711 # returns an arrayref of column names which *definitely* have som
712 # sort of non-nullable equality requested in the given condition
713 # specification. This is used to figure out if a resultset is
714 # constrained to a column which is part of a unique constraint,
715 # which in turn allows us to better predict how ordering will behave
718 # this is a rudimentary, incomplete, and error-prone extractor
719 # however this is OK - it is conservative, and if we can not find
720 # something that is in fact there - the stack will recover gracefully
721 # Also - DQ and the mst it rode in on will save us all RSN!!!
722 sub _extract_fixed_condition_columns {
723 my ($self, $where, $nested) = @_;
725 return unless ref $where eq 'HASH';
728 for my $lhs (keys %$where) {
729 if ($lhs =~ /^\-and$/i) {
730 push @cols, ref $where->{$lhs} eq 'ARRAY'
731 ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} )
732 : $self->_extract_fixed_condition_columns($where->{$lhs}, 1)
735 elsif ($lhs !~ /^\-/) {
736 my $val = $where->{$lhs};
738 push @cols, $lhs if (defined $val and (
741 (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='})
745 return $nested ? @cols : \@cols;