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 DBIx::Class::_Util qw(UNRESOLVABLE_CONDITION serialize);
19 use SQL::Abstract qw(is_plain_value is_literal_value);
20 use DBIx::Class::Carp;
24 # This code will remove non-selecting/non-restricting joins from
25 # {from} specs, aiding the RDBMS query optimizer
27 sub _prune_unused_joins {
28 my ($self, $attrs) = @_;
30 # only standard {from} specs are supported, and we could be disabled in general
31 return ($attrs->{from}, {}) unless (
32 ref $attrs->{from} eq 'ARRAY'
36 ref $attrs->{from}[0] eq 'HASH'
38 ref $attrs->{from}[1] eq 'ARRAY'
40 $self->_use_join_optimizer
44 $attrs->{_precalculated_aliastypes}
46 $self->_resolve_aliastypes_from_select_args($attrs)
49 my $new_aliastypes = { %$orig_aliastypes };
51 # we will be recreating this entirely
52 my @reclassify = 'joining';
54 # a grouped set will not be affected by amount of rows. Thus any
55 # purely multiplicator classifications can go
56 # (will be reintroduced below if needed by something else)
57 push @reclassify, qw(multiplying premultiplied)
58 if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by};
60 # nuke what will be recalculated
61 delete @{$new_aliastypes}{@reclassify};
63 my @newfrom = $attrs->{from}[0]; # FROM head is always present
65 # recalculate what we need once the multipliers are potentially gone
66 # ignore premultiplies, since they do not add any value to anything
68 for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) {
69 # add all requested aliases
70 $need_joins{$_} = 1 for keys %$_;
72 # add all their parents (as per joinpath which is an AoH { table => alias })
73 $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_;
76 for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) {
77 push @newfrom, $j if (
78 (! defined $j->[0]{-alias}) # legacy crap
80 $need_joins{$j->[0]{-alias}}
84 # we have a new set of joiners - for everything we nuked pull the classification
85 # off the original stack
86 for my $ctype (@reclassify) {
87 $new_aliastypes->{$ctype} = { map
88 { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () }
89 keys %{$orig_aliastypes->{$ctype}}
93 return ( \@newfrom, $new_aliastypes );
97 # This is the code producing joined subqueries like:
98 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
100 sub _adjust_select_args_for_complex_prefetch {
101 my ($self, $attrs) = @_;
103 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless (
104 ref $attrs->{from} eq 'ARRAY'
106 @{$attrs->{from}} > 1
108 ref $attrs->{from}[0] eq 'HASH'
110 ref $attrs->{from}[1] eq 'ARRAY'
113 my $root_alias = $attrs->{alias};
115 # generate inner/outer attribute lists, remove stuff that doesn't apply
116 my $outer_attrs = { %$attrs };
117 delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)};
119 my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 };
120 delete @{$inner_attrs}{qw(for collapse select as)};
122 # there is no point of ordering the insides if there is no limit
123 delete $inner_attrs->{order_by} if (
124 delete $inner_attrs->{_order_is_artificial}
126 ! $inner_attrs->{rows}
129 # generate the inner/outer select lists
130 # for inside we consider only stuff *not* brought in by the prefetch
131 # on the outside we substitute any function for its alias
132 $outer_attrs->{select} = [ @{$attrs->{select}} ];
134 my ($root_node, $root_node_offset);
136 for my $i (0 .. $#{$inner_attrs->{from}}) {
137 my $node = $inner_attrs->{from}[$i];
138 my $h = (ref $node eq 'HASH') ? $node
139 : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
143 if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) {
145 $root_node_offset = $i;
150 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
153 # use the heavy duty resolver to take care of aliased/nonaliased naming
154 my $colinfo = $self->_resolve_column_info($inner_attrs->{from});
155 my $selected_root_columns;
157 for my $i (0 .. $#{$outer_attrs->{select}}) {
158 my $sel = $outer_attrs->{select}->[$i];
161 $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias
164 if (ref $sel eq 'HASH' ) {
165 $sel->{-as} ||= $attrs->{as}[$i];
166 $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") );
168 elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
169 $selected_root_columns->{$ci->{-colname}} = 1;
172 push @{$inner_attrs->{select}}, $sel;
174 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
177 my $inner_aliastypes = $self->_resolve_aliastypes_from_select_args($inner_attrs);
179 # In the inner subq we will need to fetch *only* native columns which may
180 # be a part of an *outer* join condition, or an order_by (which needs to be
181 # preserved outside), or wheres. In other words everything but the inner
183 # We can not just fetch everything because a potential has_many restricting
184 # join collapse *will not work* on heavy data types.
186 # essentially a map of all non-selecting seen columns
187 # the sort is there for a nicer select list
191 { keys %{$_->{-seen_columns}||{}} }
193 { values %{$inner_aliastypes->{$_}} }
195 { $_ ne 'selecting' }
196 keys %$inner_aliastypes
198 my $ci = $colinfo->{$_} or next;
200 $ci->{-source_alias} eq $root_alias
202 ! $selected_root_columns->{$ci->{-colname}}++
204 # adding it to both to keep limits not supporting dark selectors happy
205 push @{$inner_attrs->{select}}, $ci->{-fq_colname};
206 push @{$inner_attrs->{as}}, $ci->{-fq_colname};
210 # construct the inner {from} and lock it in a subquery
211 # we need to prune first, because this will determine if we need a group_by below
212 # throw away all non-selecting, non-restricting multijoins
213 # (since we def. do not care about multiplication of the contents of the subquery)
214 my $inner_subq = do {
216 # must use it here regardless of user requests (vastly gentler on optimizer)
217 local $self->{_use_join_optimizer} = 1;
219 # throw away multijoins since we def. do not care about those inside the subquery
220 # $inner_aliastypes *will* be redefined at this point
221 ($inner_attrs->{from}, $inner_aliastypes ) = $self->_prune_unused_joins ({
223 _force_prune_multiplying_joins => 1,
224 _precalculated_aliastypes => $inner_aliastypes,
227 # uh-oh a multiplier (which is not us) left in, this is a problem for limits
228 # we will need to add a group_by to collapse the resultset for proper counts
230 grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} }
232 # if there are user-supplied groups - assume user knows wtf they are up to
233 ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} )
236 my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} };
238 # *possibly* supplement the main selection with pks if not already
239 # there, as they will have to be a part of the group_by to collapse
241 my $inner_select_with_extras;
242 my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns
243 or $self->throw_exception( sprintf
244 'Unable to perform complex limited prefetch off %s without declared primary key',
245 $root_node->{-rsrc}->source_name,
248 push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col
249 unless $cur_sel->{$col}++;
252 ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({
254 $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (),
255 _aliastypes => $inner_aliastypes,
259 # we already optimized $inner_attrs->{from} above
260 # and already local()ized
261 $self->{_use_join_optimizer} = 0;
263 # generate the subquery
264 $self->_select_args_to_query (
265 @{$inner_attrs}{qw(from select where)},
270 # Generate the outer from - this is relatively easy (really just replace
271 # the join slot with the subquery), with a major caveat - we can not
272 # join anything that is non-selecting (not part of the prefetch), but at
273 # the same time is a multi-type relationship, as it will explode the result.
275 # There are two possibilities here
276 # - either the join is non-restricting, in which case we simply throw it away
277 # - it is part of the restrictions, in which case we need to collapse the outer
278 # result by tackling yet another group_by to the outside of the query
280 # work on a shallow copy
281 my @orig_from = @{$attrs->{from}};
284 $outer_attrs->{from} = \ my @outer_from;
286 # we may not be the head
287 if ($root_node_offset) {
288 # first generate the outer_from, up to the substitution point
289 @outer_from = splice @orig_from, 0, $root_node_offset;
291 # substitute the subq at the right spot
294 -alias => $root_alias,
295 -rsrc => $root_node->{-rsrc},
296 $root_alias => $inner_subq,
298 # preserve attrs from what is now the head of the from after the splice
299 @{$orig_from[0]}[1 .. $#{$orig_from[0]}],
304 -alias => $root_alias,
305 -rsrc => $root_node->{-rsrc},
306 $root_alias => $inner_subq,
310 shift @orig_from; # what we just replaced above
312 # scan the *remaining* from spec against different attributes, and see which joins are needed
314 my $outer_aliastypes = $outer_attrs->{_aliastypes} =
315 $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from });
318 my ($outer_select_chain, @outer_nonselecting_chains) = map { +{
319 map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} }
320 } } qw/selecting restricting grouping ordering/;
322 # see what's left - throw away if not selecting/restricting
323 my $may_need_outer_group_by;
324 while (my $j = shift @orig_from) {
325 my $alias = $j->[0]{-alias};
328 $outer_select_chain->{$alias}
332 elsif (first { $_->{$alias} } @outer_nonselecting_chains ) {
333 push @outer_from, $j;
334 $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
338 # also throw in a synthetic group_by if a non-selecting multiplier,
339 # to guard against cross-join explosions
340 # the logic is somewhat fragile, but relies on the idea that if a user supplied
341 # a group by on their own - they know what they were doing
342 if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) {
343 ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({
345 from => \@outer_from,
349 # This is totally horrific - the {where} ends up in both the inner and outer query
350 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
351 # then if where conditions apply to the *right* side of the prefetch, you may have
352 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
353 # the outer select to exclude joins you didn't want in the first place
355 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
360 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
362 # Due to a lack of SQLA2 we fall back to crude scans of all the
363 # select/where/order/group attributes, in order to determine what
364 # aliases are needed to fulfill the query. This information is used
365 # throughout the code to prune unnecessary JOINs from the queries
366 # in an attempt to reduce the execution time.
367 # Although the method is pretty horrific, the worst thing that can
368 # happen is for it to fail due to some scalar SQL, which in turn will
369 # result in a vocal exception.
370 sub _resolve_aliastypes_from_select_args {
371 my ( $self, $attrs ) = @_;
373 $self->throw_exception ('Unable to analyze custom {from}')
374 if ref $attrs->{from} ne 'ARRAY';
376 # what we will return
379 # see what aliases are there to work with
380 # and record who is a multiplier and who is premultiplied
382 for my $node (@{$attrs->{from}}) {
385 $j = $j->[0] if ref $j eq 'ARRAY';
386 my $al = $j->{-alias}
389 $alias_list->{$al} = $j;
391 $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] }
392 # not array == {from} head == can't be multiplying
393 if ref($node) eq 'ARRAY' and ! $j->{-is_single};
395 $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] }
396 # parts of the path that are not us but are multiplying
397 if grep { $aliases_by_type->{multiplying}{$_} }
400 @{ $j->{-join_path}||[] }
403 # get a column to source/alias map (including unambiguous unqualified ones)
404 my $colinfo = $self->_resolve_column_info ($attrs->{from});
406 # set up a botched SQLA
407 my $sql_maker = $self->sql_maker;
409 # these are throw away results, do not pollute the bind stack
410 local $sql_maker->{where_bind};
411 local $sql_maker->{group_bind};
412 local $sql_maker->{having_bind};
413 local $sql_maker->{from_bind};
415 # we can't scan properly without any quoting (\b doesn't cut it
416 # everywhere), so unless there is proper quoting set - use our
417 # own weird impossible character.
418 # Also in the case of no quoting, we need to explicitly disable
419 # name_sep, otherwise sorry nasty legacy syntax like
420 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
421 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
422 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
424 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
425 $sql_maker->{quote_char} = ["\x00", "\xFF"];
426 # if we don't unset it we screw up retarded but unfortunately working
427 # 'MAX(foo.bar)' => { '>', 3 }
428 $sql_maker->{name_sep} = '';
431 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
433 # generate sql chunks
436 ($sql_maker->_recurse_where ($attrs->{where}))[0],
437 $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }),
440 $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }),
443 $sql_maker->_recurse_from (
444 ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0],
445 @{$attrs->{from}}[1 .. $#{$attrs->{from}}],
449 # kill all selectors which look like a proper subquery
450 # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
451 # fail to run, so we are relatively safe
453 { $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi }
455 { ($sql_maker->_recurse_fields($_))[0] }
460 ( my $sql = (ref $_ ? $_->[0] : $_) ) =~ s/ \s+ (?: ASC | DESC ) \s* \z //xi;
463 $sql_maker->_order_by_chunks( $attrs->{order_by} ),
467 # we will be bulk-scanning anyway - pieces will not matter in that case,
468 # thus join everything up
469 # throw away empty-string chunks, and make sure no binds snuck in
470 # note that we operate over @{$to_scan->{$type}}, hence the
471 # semi-mindbending ... map ... for values ...
472 ( $_ = join ' ', map {
474 ( ! defined $_ ) ? ()
475 : ( length ref $_ ) ? (require Data::Dumper::Concise && $self->throw_exception(
476 "Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($_)
478 : ( $_ =~ /^\s*$/ ) ? ()
481 } @$_ ) for values %$to_scan;
483 # throw away empty to-scan's
485 length $to_scan->{$_}
487 delete $to_scan->{$_}
488 ) for keys %$to_scan;
492 # these will be used for matching in the loop below
493 my $all_aliases = join ' | ', map { quotemeta $_ } keys %$alias_list;
495 $lquote ( $all_aliases ) $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
497 \b ( $all_aliases ) \. ( [^\s\)\($rquote]+ )?
501 my $all_unq_columns = join ' | ',
505 # using a regex here shows up on profiles, boggle
506 { index( $_, '.') < 0 }
509 my $unq_col_re = $all_unq_columns
511 $lquote ( $all_unq_columns ) $rquote
513 (?: \A | \s ) ( $all_unq_columns ) (?: \s | \z )
519 # the actual scan, per type
520 for my $type (keys %$to_scan) {
523 # now loop through all fully qualified columns and get the corresponding
524 # alias (should work even if they are in scalarrefs)
526 # The regex captures in multiples of 4, with one of the two pairs being
527 # undef. There may be a *lot* of matches, hence the convoluted loop
528 my @matches = $to_scan->{$type} =~ /$fq_col_re/g;
530 while( $i < $#matches ) {
535 $aliases_by_type->{$type}{$matches[$i]} ||= { -parents => $alias_list->{$matches[$i]}{-join_path}||[] };
537 $aliases_by_type->{$type}{$matches[$i]}{-seen_columns}{"$matches[$i].$matches[$i+1]"} = "$matches[$i].$matches[$i+1]"
538 if defined $matches[$i+1];
547 # now loop through unqualified column names, and try to locate them within
548 # the chunks, if there are any unqualified columns in the 1st place
549 next unless $unq_col_re;
551 # The regex captures in multiples of 2, one of the two being undef
552 for ( $to_scan->{$type} =~ /$unq_col_re/g ) {
554 my $alias = $colinfo->{$_}{-source_alias} or next;
555 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
556 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
561 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
565 ! $aliases_by_type->{restricting}{ $_->{-alias} }
570 $_->{-join_type} !~ /^left (?: \s+ outer)? $/xi
573 $aliases_by_type->{restricting}{ $_->{-alias} } = { -parents => $_->{-join_path}||[] }
574 ) for values %$alias_list;
579 keys %{$aliases_by_type->{$_}}
581 delete $aliases_by_type->{$_}
582 ) for keys %$aliases_by_type;
588 # This is the engine behind { distinct => 1 } and the general
589 # complex prefetch grouper
590 sub _group_over_selection {
591 my ($self, $attrs) = @_;
593 my $colinfos = $self->_resolve_column_info ($attrs->{from});
595 my (@group_by, %group_index);
597 # the logic is: if it is a { func => val } we assume an aggregate,
598 # otherwise if \'...' or \[...] we assume the user knows what is
599 # going on thus group over it
600 for (@{$attrs->{select}}) {
601 if (! ref($_) or ref ($_) ne 'HASH' ) {
604 if ($colinfos->{$_} and $_ !~ /\./ ) {
605 # add a fully qualified version as well
606 $group_index{"$colinfos->{$_}{-source_alias}.$_"}++;
611 my @order_by = $self->_extract_order_criteria($attrs->{order_by})
612 or return (\@group_by, $attrs->{order_by});
614 # add any order_by parts that are not already present in the group_by
615 # to maintain SQL cross-compatibility and general sanity
617 # also in case the original selection is *not* unique, or in case part
618 # of the ORDER BY refers to a multiplier - we will need to replace the
619 # skipped order_by elements with their MIN/MAX equivalents as to maintain
620 # the proper overall order without polluting the group criteria (and
621 # possibly changing the outcome entirely)
623 my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes);
625 my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by);
627 for my $o_idx (0 .. $#order_by) {
629 # if the chunk is already a min/max function - there is nothing left to touch
630 next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix;
632 # only consider real columns (for functions the user got to do an explicit group_by)
635 @{$order_by[$o_idx]} != 1
637 # only declare an unknown *plain* identifier as "leftover" if we are called with
638 # aliastypes to examine. If there are none - we are still in _resolve_attrs, and
639 # can just assume the user knows what they want
640 ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} )
642 push @$leftovers, $order_by[$o_idx][0];
645 next unless $chunk_ci;
647 # no duplication of group criteria
648 next if $group_index{$chunk_ci->{-fq_colname}};
651 $attrs->{_aliastypes}
653 $self->_resolve_aliastypes_from_select_args({
654 from => $attrs->{from},
655 order_by => $attrs->{order_by},
657 ) if $group_already_unique;
659 # check that we are not ordering by a multiplier (if a check is requested at all)
661 $group_already_unique
663 ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}}
665 ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}}
667 push @group_by, $chunk_ci->{-fq_colname};
668 $group_index{$chunk_ci->{-fq_colname}}++
671 # We need to order by external columns without adding them to the group
672 # (eiehter a non-unique selection, or a multi-external)
674 # This doesn't really make sense in SQL, however from DBICs point
675 # of view is rather valid (e.g. order the leftmost objects by whatever
676 # criteria and get the offset/rows many). There is a way around
677 # this however in SQL - we simply tae the direction of each piece
678 # of the external order and convert them to MIN(X) for ASC or MAX(X)
679 # for DESC, and group_by the root columns. The end result should be
680 # exactly what we expect
682 # FIXME - this code is a joke, will need to be completely rewritten in
683 # the DQ branch. But I need to push a POC here, otherwise the
684 # pesky tests won't pass
685 # wrap any part of the order_by that "responds" to an ordering alias
687 $sql_maker ||= $self->sql_maker;
689 map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by})
692 my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]);
694 $new_order_by[$o_idx] = \[
695 sprintf( '%s( %s )%s',
696 ($is_desc ? 'MAX' : 'MIN'),
698 ($is_desc ? ' DESC' : ''),
700 @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ]
705 $self->throw_exception ( sprintf
706 'Unable to programatically derive a required group_by from the supplied '
707 . 'order_by criteria. To proceed either add an explicit group_by, or '
708 . 'simplify your order_by to only include plain columns '
709 . '(supplied order_by: %s)',
710 join ', ', map { "'$_'" } @$leftovers,
713 # recreate the untouched order parts
715 $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks );
720 (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged
724 sub _resolve_ident_sources {
725 my ($self, $ident) = @_;
727 my $alias2source = {};
729 # the reason this is so contrived is that $ident may be a {from}
730 # structure, specifying multiple tables to join
731 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
732 # this is compat mode for insert/update/delete which do not deal with aliases
733 $alias2source->{me} = $ident;
735 elsif (ref $ident eq 'ARRAY') {
739 if (ref $_ eq 'HASH') {
742 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
746 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
747 if ($tabinfo->{-rsrc});
751 return $alias2source;
754 # Takes $ident, \@column_names
756 # returns { $column_name => \%column_info, ... }
757 # also note: this adds -result_source => $rsrc to the column info
759 # If no columns_names are supplied returns info about *all* columns
761 sub _resolve_column_info {
762 my ($self, $ident, $colnames) = @_;
764 return {} if $colnames and ! @$colnames;
766 my $sources = $self->_resolve_ident_sources($ident);
768 $_ = { rsrc => $_, colinfos => $_->columns_info }
769 for values %$sources;
771 my (%seen_cols, @auto_colnames);
773 # compile a global list of column names, to be able to properly
774 # disambiguate unqualified column names (if at all possible)
775 for my $alias (keys %$sources) {
777 ++$seen_cols{$_}{$alias}
781 push @auto_colnames, "$alias.$_"
782 ) for keys %{ $sources->{$alias}{colinfos} };
787 ( grep { keys %{$seen_cols{$_}} == 1 } keys %seen_cols ),
792 my ($colname, $source_alias) = reverse split /\./, $_;
797 # if the column was seen exactly once - we know which rsrc it came from
801 keys %{$seen_cols{$colname}} == 1
803 ( %{$seen_cols{$colname}} )[0]
809 $self->throw_exception(
810 "No such column '$colname' on source " . $sources->{$assumed_alias}{rsrc}->source_name
811 ) unless $seen_cols{$colname}{$assumed_alias};
814 %{ $sources->{$assumed_alias}{colinfos}{$colname} },
815 -result_source => $sources->{$assumed_alias}{rsrc},
816 -source_alias => $assumed_alias,
817 -fq_colname => "$assumed_alias.$colname",
818 -colname => $colname,
821 $return{"$assumed_alias.$colname"} = $return{$_}
822 unless $source_alias;
828 # The DBIC relationship chaining implementation is pretty simple - every
829 # new related_relationship is pushed onto the {from} stack, and the {select}
830 # window simply slides further in. This means that when we count somewhere
831 # in the middle, we got to make sure that everything in the join chain is an
832 # actual inner join, otherwise the count will come back with unpredictable
833 # results (a resultset may be generated with _some_ rows regardless of if
834 # the relation which the $rs currently selects has rows or not). E.g.
835 # $artist_rs->cds->count - normally generates:
836 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
837 # which actually returns the number of artists * (number of cds || 1)
839 # So what we do here is crawl {from}, determine if the current alias is at
840 # the top of the stack, and if not - make sure the chain is inner-joined down
843 sub _inner_join_to_node {
844 my ($self, $from, $alias) = @_;
846 my $switch_branch = $self->_find_join_path_to_node($from, $alias);
848 return $from unless @{$switch_branch||[]};
850 # So it looks like we will have to switch some stuff around.
851 # local() is useless here as we will be leaving the scope
852 # anyway, and deep cloning is just too fucking expensive
853 # So replace the first hashref in the node arrayref manually
854 my @new_from = ($from->[0]);
855 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
857 for my $j (@{$from}[1 .. $#$from]) {
858 my $jalias = $j->[0]{-alias};
860 if ($sw_idx->{$jalias}) {
861 my %attrs = %{$j->[0]};
862 delete $attrs{-join_type};
876 sub _find_join_path_to_node {
877 my ($self, $from, $target_alias) = @_;
879 # subqueries and other oddness are naturally not supported
883 ref $from->[0] ne 'HASH'
885 ! defined $from->[0]{-alias}
888 # no path - the head is the alias
889 return [] if $from->[0]{-alias} eq $target_alias;
891 for my $i (1 .. $#$from) {
892 return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias );
895 # something else went quite wrong
899 sub _extract_order_criteria {
900 my ($self, $order_by, $sql_maker) = @_;
903 my ($sql_maker, $order_by, $orig_quote_chars) = @_;
905 return scalar $sql_maker->_order_by_chunks ($order_by)
908 my ($lq, $rq, $sep) = map { quotemeta($_) } (
909 ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars),
914 for ($sql_maker->_order_by_chunks ($order_by) ) {
915 my $chunk = ref $_ ? [ @$_ ] : [ $_ ];
916 ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]);
918 # order criteria may have come back pre-quoted (literals and whatnot)
919 # this is fragile, but the best we can currently do
920 $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe
921 or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x;
923 push @chunks, $chunk;
930 return $parser->($sql_maker, $order_by);
933 $sql_maker = $self->sql_maker;
935 # pass these in to deal with literals coming from
936 # the user or the deep guts of prefetch
937 my $orig_quote_chars = [$sql_maker->_quote_chars];
939 local $sql_maker->{quote_char};
940 return $parser->($sql_maker, $order_by, $orig_quote_chars);
944 sub _order_by_is_stable {
945 my ($self, $ident, $order_by, $where) = @_;
948 ( map { $_->[0] } $self->_extract_order_criteria($order_by) ),
949 ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ),
952 my $colinfo = $self->_resolve_column_info($ident, \@cols);
954 return keys %$colinfo
955 ? $self->_columns_comprise_identifying_set( $colinfo, \@cols )
960 sub _columns_comprise_identifying_set {
961 my ($self, $colinfo, $columns) = @_;
964 $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_
965 for grep { defined $_ } @{$colinfo}{@$columns};
967 for (values %$cols_per_src) {
968 my $src = (values %$_)[0]->{-result_source};
969 return 1 if $src->_identifying_column_set($_);
975 # this is almost similar to _order_by_is_stable, except it takes
976 # a single rsrc, and will succeed only if the first portion of the order
978 # returns that portion as a colinfo hashref on success
979 sub _extract_colinfo_of_stable_main_source_order_by_portion {
980 my ($self, $attrs) = @_;
982 my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias});
984 return unless defined $nodes;
988 ( $self->_extract_order_criteria($attrs->{order_by}) )
990 return unless @ord_cols;
992 my $valid_aliases = { map { $_ => 1 } (
993 $attrs->{from}[0]{-alias},
994 map { values %$_ } @$nodes,
997 my $colinfos = $self->_resolve_column_info($attrs->{from});
999 my ($colinfos_to_return, $seen_main_src_cols);
1001 for my $col (@ord_cols) {
1002 # if order criteria is unresolvable - there is nothing we can do
1003 my $colinfo = $colinfos->{$col} or last;
1005 # if we reached the end of the allowed aliases - also nothing we can do
1006 last unless $valid_aliases->{$colinfo->{-source_alias}};
1008 $colinfos_to_return->{$col} = $colinfo;
1010 $seen_main_src_cols->{$colinfo->{-colname}} = 1
1011 if $colinfo->{-source_alias} eq $attrs->{alias};
1014 # FIXME the condition may be singling out things on its own, so we
1015 # conceivable could come back wi "stable-ordered by nothing"
1016 # not confient enough in the parser yet, so punt for the time being
1017 return unless $seen_main_src_cols;
1019 my $main_src_fixed_cols_from_cond = [ $attrs->{where}
1023 ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} )
1024 ? $colinfos->{$_}{-colname}
1027 keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) }
1032 return $attrs->{result_source}->_identifying_column_set([
1033 keys %$seen_main_src_cols,
1034 @$main_src_fixed_cols_from_cond,
1035 ]) ? $colinfos_to_return : ();
1038 # Attempts to flatten a passed in SQLA condition as much as possible towards
1039 # a plain hashref, *without* altering its semantics. Required by
1040 # create/populate being able to extract definitive conditions from preexisting
1041 # resultset {where} stacks
1043 # FIXME - while relatively robust, this is still imperfect, one of the first
1044 # things to tackle with DQ
1045 sub _collapse_cond {
1046 my ($self, $where, $where_is_anded_array) = @_;
1053 elsif ($where_is_anded_array or ref $where eq 'HASH') {
1057 my @pieces = $where_is_anded_array ? @$where : $where;
1059 my $chunk = shift @pieces;
1061 if (ref $chunk eq 'HASH') {
1062 for (sort keys %$chunk) {
1064 # Match SQLA 1.79 behavior
1066 is_literal_value($chunk->{$_})
1067 ? carp 'Hash-pairs consisting of an empty string with a literal are deprecated, use -and => [ $literal ] instead'
1068 : $self->throw_exception("Supplying an empty left hand side argument is not supported in hash-pairs")
1072 push @pairs, $_ => $chunk->{$_};
1075 elsif (ref $chunk eq 'ARRAY') {
1076 push @pairs, -or => $chunk
1079 elsif ( ! length ref $chunk) {
1081 # Match SQLA 1.79 behavior
1082 $self->throw_exception("Supplying an empty left hand side argument is not supported in array-pairs")
1083 if $where_is_anded_array and (! defined $chunk or $chunk eq '');
1085 push @pairs, $chunk, shift @pieces;
1088 push @pairs, '', $chunk;
1092 return unless @pairs;
1094 my @conds = $self->_collapse_cond_unroll_pairs(\@pairs)
1097 # Consolidate various @conds back into something more compact
1098 for my $c (@conds) {
1099 if (ref $c ne 'HASH') {
1100 push @{$fin->{-and}}, $c;
1103 for my $col (sort keys %$c) {
1105 # consolidate all -and nodes
1106 if ($col =~ /^\-and$/i) {
1107 push @{$fin->{-and}},
1108 ref $c->{$col} eq 'ARRAY' ? @{$c->{$col}}
1109 : ref $c->{$col} eq 'HASH' ? %{$c->{$col}}
1110 : { $col => $c->{$col} }
1113 elsif ($col =~ /^\-/) {
1114 push @{$fin->{-and}}, { $col => $c->{$col} };
1116 elsif (exists $fin->{$col}) {
1117 $fin->{$col} = [ -and => map {
1118 (ref $_ eq 'ARRAY' and ($_->[0]||'') =~ /^\-and$/i )
1122 } ($fin->{$col}, $c->{$col}) ];
1125 $fin->{$col} = $c->{$col};
1131 elsif (ref $where eq 'ARRAY') {
1132 # we are always at top-level here, it is safe to dump empty *standalone* pieces
1135 for (my $i = 0; $i <= $#$where; $i++ ) {
1137 # Match SQLA 1.79 behavior
1138 $self->throw_exception(
1139 "Supplying an empty left hand side argument is not supported in array-pairs"
1140 ) if (! defined $where->[$i] or ! length $where->[$i]);
1142 my $logic_mod = lc ( ($where->[$i] =~ /^(\-(?:and|or))$/i)[0] || '' );
1146 $self->throw_exception("Unsupported top-level op/arg pair: [ $logic_mod => $where->[$i] ]")
1147 unless ref $where->[$i] eq 'HASH' or ref $where->[$i] eq 'ARRAY';
1149 my $sub_elt = $self->_collapse_cond({ $logic_mod => $where->[$i] })
1152 my @keys = keys %$sub_elt;
1153 if ( @keys == 1 and $keys[0] !~ /^\-/ ) {
1154 $fin_idx->{ "COL_$keys[0]_" . serialize $sub_elt } = $sub_elt;
1157 $fin_idx->{ "SER_" . serialize $sub_elt } = $sub_elt;
1160 elsif (! length ref $where->[$i] ) {
1161 my $sub_elt = $self->_collapse_cond({ @{$where}[$i, $i+1] })
1164 $fin_idx->{ "COL_$where->[$i]_" . serialize $sub_elt } = $sub_elt;
1168 $fin_idx->{ "SER_" . serialize $where->[$i] } = $self->_collapse_cond( $where->[$i] ) || next;
1175 elsif ( keys %$fin_idx == 1 ) {
1176 $fin = (values %$fin_idx)[0];
1181 # at this point everything is at most one level deep - unroll if needed
1182 for (sort keys %$fin_idx) {
1183 if ( ref $fin_idx->{$_} eq 'HASH' and keys %{$fin_idx->{$_}} == 1 ) {
1184 my ($l, $r) = %{$fin_idx->{$_}};
1190 ( @$r == 1 and $l =~ /^\-and$/i )
1203 $l =~ /^\-(?:and|or)$/i
1213 push @or, $fin_idx->{$_};
1221 # not a hash not an array
1222 $fin = { -and => [ $where ] };
1225 # unroll single-element -and's
1231 my $and = delete $fin->{-and};
1234 # at this point we have @$and == 1
1236 ref $and->[0] eq 'HASH'
1238 ! grep { exists $fin->{$_} } keys %{$and->[0]}
1245 $fin->{-and} = $and;
1250 # compress same-column conds found in $fin
1251 for my $col ( grep { $_ !~ /^\-/ } keys %$fin ) {
1252 next unless ref $fin->{$col} eq 'ARRAY' and ($fin->{$col}[0]||'') =~ /^\-and$/i;
1253 my $val_bag = { map {
1254 (! defined $_ ) ? ( UNDEF => undef )
1255 : ( ! length ref $_ or is_plain_value $_ ) ? ( "VAL_$_" => $_ )
1256 : ( ( 'SER_' . serialize $_ ) => $_ )
1257 } @{$fin->{$col}}[1 .. $#{$fin->{$col}}] };
1259 if (keys %$val_bag == 1 ) {
1260 ($fin->{$col}) = values %$val_bag;
1263 $fin->{$col} = [ -and => map { $val_bag->{$_} } sort keys %$val_bag ];
1267 return keys %$fin ? $fin : ();
1270 sub _collapse_cond_unroll_pairs {
1271 my ($self, $pairs) = @_;
1276 my ($lhs, $rhs) = splice @$pairs, 0, 2;
1279 push @conds, $self->_collapse_cond($rhs);
1281 elsif ( $lhs =~ /^\-and$/i ) {
1282 push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY'));
1284 elsif ( $lhs =~ /^\-or$/i ) {
1285 push @conds, $self->_collapse_cond(
1286 (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs
1290 if (ref $rhs eq 'HASH' and ! keys %$rhs) {
1291 # FIXME - SQLA seems to be doing... nothing...?
1293 # normalize top level -ident, for saner extract_fixed_condition_columns code
1294 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) {
1295 push @conds, { $lhs => { '=', $rhs } };
1297 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-value} and is_plain_value $rhs->{-value}) {
1298 push @conds, { $lhs => $rhs->{-value} };
1300 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) {
1301 if ( length ref $rhs->{'='} and is_literal_value $rhs->{'='} ) {
1302 push @conds, { $lhs => $rhs };
1305 for my $p ($self->_collapse_cond_unroll_pairs([ $lhs => $rhs->{'='} ])) {
1307 # extra sanity check
1309 require Data::Dumper::Concise;
1310 local $Data::Dumper::Deepcopy = 1;
1311 $self->throw_exception(
1312 "Internal error: unexpected collapse unroll:"
1313 . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p }
1322 # the unroller recursion may return a '=' prepended value already
1323 ref $r eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}
1328 : { $l => { '=' => $r } }
1333 elsif (ref $rhs eq 'ARRAY') {
1334 # some of these conditionals encounter multi-values - roll them out using
1335 # an unshift, which will cause extra looping in the while{} above
1337 push @conds, { $lhs => [] };
1339 elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) {
1340 $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ")
1343 if( $rhs->[0] =~ /^\-and$/i ) {
1344 unshift @$pairs, map { $lhs => $_ } @{$rhs}[1..$#$rhs];
1346 # if not an AND then it's an OR
1348 unshift @$pairs, $lhs => $rhs->[1];
1351 push @conds, { $lhs => [ @{$rhs}[1..$#$rhs] ] };
1354 elsif (@$rhs == 1) {
1355 unshift @$pairs, $lhs => $rhs->[0];
1358 push @conds, { $lhs => $rhs };
1361 # unroll func + { -value => ... }
1365 ( my ($subop) = keys %$rhs ) == 1
1367 length ref ((values %$rhs)[0])
1369 my $vref = is_plain_value( (values %$rhs)[0] )
1371 push @conds, { $lhs => { $subop => $$vref } }
1374 push @conds, { $lhs => $rhs };
1382 # Analyzes a given condition and attempts to extract all columns
1383 # with a definitive fixed-condition criteria. Returns a hashref
1384 # of k/v pairs suitable to be passed to set_columns(), with a
1385 # MAJOR CAVEAT - multi-value (contradictory) equalities are still
1386 # represented as a reference to the UNRESOVABLE_CONDITION constant
1387 # The reason we do this is that some codepaths only care about the
1388 # codition being stable, as opposed to actually making sense
1390 # The normal mode is used to figure out if a resultset is constrained
1391 # to a column which is part of a unique constraint, which in turn
1392 # allows us to better predict how ordering will behave etc.
1394 # With the optional "consider_nulls" boolean argument, the function
1395 # is instead used to infer inambiguous values from conditions
1396 # (e.g. the inheritance of resultset conditions on new_result)
1398 sub _extract_fixed_condition_columns {
1399 my ($self, $where, $consider_nulls) = @_;
1400 my $where_hash = $self->_collapse_cond($_[1]);
1404 for $c (keys %$where_hash) {
1407 if (!defined ($v = $where_hash->{$c}) ) {
1408 $vals->{UNDEF} = $v if $consider_nulls
1415 if (exists $v->{-value}) {
1416 if (defined $v->{-value}) {
1417 $vals->{"VAL_$v->{-value}"} = $v->{-value}
1419 elsif( $consider_nulls ) {
1420 $vals->{UNDEF} = $v->{-value};
1423 # do not need to check for plain values - _collapse_cond did it for us
1425 length ref $v->{'='}
1428 ( ref $v->{'='} eq 'HASH' and keys %{$v->{'='}} == 1 and exists $v->{'='}{-ident} )
1430 is_literal_value($v->{'='})
1433 $vals->{ 'SER_' . serialize $v->{'='} } = $v->{'='};
1441 $vals->{"VAL_$v"} = $v;
1443 elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') {
1444 for ( @{$v}[1..$#$v] ) {
1445 my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls'); # always fish nulls out on recursion
1446 next unless exists $subval->{$c}; # didn't find anything
1448 ! defined $subval->{$c} ? 'UNDEF'
1449 : ( ! length ref $subval->{$c} or is_plain_value $subval->{$c} ) ? "VAL_$subval->{$c}"
1450 : ( 'SER_' . serialize $subval->{$c} )
1455 if (keys %$vals == 1) {
1456 ($res->{$c}) = (values %$vals)
1457 unless !$consider_nulls and exists $vals->{UNDEF};
1459 elsif (keys %$vals > 1) {
1460 $res->{$c} = UNRESOLVABLE_CONDITION;