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';
19 use DBIx::Class::_Util qw(is_plain_value is_literal_value);
23 # This code will remove non-selecting/non-restricting joins from
24 # {from} specs, aiding the RDBMS query optimizer
26 sub _prune_unused_joins {
27 my ($self, $attrs) = @_;
29 # only standard {from} specs are supported, and we could be disabled in general
30 return ($attrs->{from}, {}) unless (
31 ref $attrs->{from} eq 'ARRAY'
35 ref $attrs->{from}[0] eq 'HASH'
37 ref $attrs->{from}[1] eq 'ARRAY'
39 $self->_use_join_optimizer
42 my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs);
44 my $new_aliastypes = { %$orig_aliastypes };
46 # we will be recreating this entirely
47 my @reclassify = 'joining';
49 # a grouped set will not be affected by amount of rows. Thus any
50 # purely multiplicator classifications can go
51 # (will be reintroduced below if needed by something else)
52 push @reclassify, qw(multiplying premultiplied)
53 if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by};
55 # nuke what will be recalculated
56 delete @{$new_aliastypes}{@reclassify};
58 my @newfrom = $attrs->{from}[0]; # FROM head is always present
60 # recalculate what we need once the multipliers are potentially gone
61 # ignore premultiplies, since they do not add any value to anything
63 for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) {
64 # add all requested aliases
65 $need_joins{$_} = 1 for keys %$_;
67 # add all their parents (as per joinpath which is an AoH { table => alias })
68 $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_;
71 for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) {
72 push @newfrom, $j if (
73 (! defined $j->[0]{-alias}) # legacy crap
75 $need_joins{$j->[0]{-alias}}
79 # we have a new set of joiners - for everything we nuked pull the classification
80 # off the original stack
81 for my $ctype (@reclassify) {
82 $new_aliastypes->{$ctype} = { map
83 { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () }
84 keys %{$orig_aliastypes->{$ctype}}
88 return ( \@newfrom, $new_aliastypes );
92 # This is the code producing joined subqueries like:
93 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
95 sub _adjust_select_args_for_complex_prefetch {
96 my ($self, $attrs) = @_;
98 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless (
99 ref $attrs->{from} eq 'ARRAY'
101 @{$attrs->{from}} > 1
103 ref $attrs->{from}[0] eq 'HASH'
105 ref $attrs->{from}[1] eq 'ARRAY'
108 my $root_alias = $attrs->{alias};
110 # generate inner/outer attribute lists, remove stuff that doesn't apply
111 my $outer_attrs = { %$attrs };
112 delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)};
114 my $inner_attrs = { %$attrs };
115 delete @{$inner_attrs}{qw(for collapse select as _related_results_construction)};
117 # there is no point of ordering the insides if there is no limit
118 delete $inner_attrs->{order_by} if (
119 delete $inner_attrs->{_order_is_artificial}
121 ! $inner_attrs->{rows}
124 # generate the inner/outer select lists
125 # for inside we consider only stuff *not* brought in by the prefetch
126 # on the outside we substitute any function for its alias
127 $outer_attrs->{select} = [ @{$attrs->{select}} ];
129 my ($root_node, $root_node_offset);
131 for my $i (0 .. $#{$inner_attrs->{from}}) {
132 my $node = $inner_attrs->{from}[$i];
133 my $h = (ref $node eq 'HASH') ? $node
134 : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
138 if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) {
140 $root_node_offset = $i;
145 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
148 # use the heavy duty resolver to take care of aliased/nonaliased naming
149 my $colinfo = $self->_resolve_column_info($inner_attrs->{from});
150 my $selected_root_columns;
152 for my $i (0 .. $#{$outer_attrs->{select}}) {
153 my $sel = $outer_attrs->{select}->[$i];
156 $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias
159 if (ref $sel eq 'HASH' ) {
160 $sel->{-as} ||= $attrs->{as}[$i];
161 $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") );
163 elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
164 $selected_root_columns->{$ci->{-colname}} = 1;
167 push @{$inner_attrs->{select}}, $sel;
169 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
172 # We will need to fetch all native columns in the inner subquery, which may
173 # be a part of an *outer* join condition, or an order_by (which needs to be
174 # preserved outside), or wheres. In other words everything but the inner
176 # We can not just fetch everything because a potential has_many restricting
177 # join collapse *will not work* on heavy data types.
178 my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({
183 for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) {
184 my $ci = $colinfo->{$_} or next;
186 $ci->{-source_alias} eq $root_alias
188 ! $selected_root_columns->{$ci->{-colname}}++
190 # adding it to both to keep limits not supporting dark selectors happy
191 push @{$inner_attrs->{select}}, $ci->{-fq_colname};
192 push @{$inner_attrs->{as}}, $ci->{-fq_colname};
196 # construct the inner {from} and lock it in a subquery
197 # we need to prune first, because this will determine if we need a group_by below
198 # throw away all non-selecting, non-restricting multijoins
199 # (since we def. do not care about multiplication of the contents of the subquery)
200 my $inner_subq = do {
202 # must use it here regardless of user requests (vastly gentler on optimizer)
203 local $self->{_use_join_optimizer} = 1;
205 # throw away multijoins since we def. do not care about those inside the subquery
206 ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({
207 %$inner_attrs, _force_prune_multiplying_joins => 1
210 # uh-oh a multiplier (which is not us) left in, this is a problem for limits
211 # we will need to add a group_by to collapse the resultset for proper counts
213 grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} }
215 # if there are user-supplied groups - assume user knows wtf they are up to
216 ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} )
219 my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} };
221 # *possibly* supplement the main selection with pks if not already
222 # there, as they will have to be a part of the group_by to collapse
224 my $inner_select_with_extras;
225 my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns
226 or $self->throw_exception( sprintf
227 'Unable to perform complex limited prefetch off %s without declared primary key',
228 $root_node->{-rsrc}->source_name,
231 push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col
232 unless $cur_sel->{$col}++;
235 ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({
237 $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (),
238 _aliastypes => $inner_aliastypes,
242 # we already optimized $inner_attrs->{from} above
243 # and already local()ized
244 $self->{_use_join_optimizer} = 0;
246 # generate the subquery
247 $self->_select_args_to_query (
248 @{$inner_attrs}{qw(from select where)},
253 # Generate the outer from - this is relatively easy (really just replace
254 # the join slot with the subquery), with a major caveat - we can not
255 # join anything that is non-selecting (not part of the prefetch), but at
256 # the same time is a multi-type relationship, as it will explode the result.
258 # There are two possibilities here
259 # - either the join is non-restricting, in which case we simply throw it away
260 # - it is part of the restrictions, in which case we need to collapse the outer
261 # result by tackling yet another group_by to the outside of the query
263 # work on a shallow copy
264 my @orig_from = @{$attrs->{from}};
267 $outer_attrs->{from} = \ my @outer_from;
269 # we may not be the head
270 if ($root_node_offset) {
271 # first generate the outer_from, up to the substitution point
272 @outer_from = splice @orig_from, 0, $root_node_offset;
274 # substitute the subq at the right spot
277 -alias => $root_alias,
278 -rsrc => $root_node->{-rsrc},
279 $root_alias => $inner_subq,
281 # preserve attrs from what is now the head of the from after the splice
282 @{$orig_from[0]}[1 .. $#{$orig_from[0]}],
287 -alias => $root_alias,
288 -rsrc => $root_node->{-rsrc},
289 $root_alias => $inner_subq,
293 shift @orig_from; # what we just replaced above
295 # scan the *remaining* from spec against different attributes, and see which joins are needed
297 my $outer_aliastypes = $outer_attrs->{_aliastypes} =
298 $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from });
301 my ($outer_select_chain, @outer_nonselecting_chains) = map { +{
302 map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} }
303 } } qw/selecting restricting grouping ordering/;
305 # see what's left - throw away if not selecting/restricting
306 my $may_need_outer_group_by;
307 while (my $j = shift @orig_from) {
308 my $alias = $j->[0]{-alias};
311 $outer_select_chain->{$alias}
315 elsif (first { $_->{$alias} } @outer_nonselecting_chains ) {
316 push @outer_from, $j;
317 $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
321 # also throw in a synthetic group_by if a non-selecting multiplier,
322 # to guard against cross-join explosions
323 # the logic is somewhat fragile, but relies on the idea that if a user supplied
324 # a group by on their own - they know what they were doing
325 if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) {
326 ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({
328 from => \@outer_from,
332 # This is totally horrific - the {where} ends up in both the inner and outer query
333 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
334 # then if where conditions apply to the *right* side of the prefetch, you may have
335 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
336 # the outer select to exclude joins you didn't want in the first place
338 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
343 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
345 # Due to a lack of SQLA2 we fall back to crude scans of all the
346 # select/where/order/group attributes, in order to determine what
347 # aliases are needed to fulfill the query. This information is used
348 # throughout the code to prune unnecessary JOINs from the queries
349 # in an attempt to reduce the execution time.
350 # Although the method is pretty horrific, the worst thing that can
351 # happen is for it to fail due to some scalar SQL, which in turn will
352 # result in a vocal exception.
353 sub _resolve_aliastypes_from_select_args {
354 my ( $self, $attrs ) = @_;
356 $self->throw_exception ('Unable to analyze custom {from}')
357 if ref $attrs->{from} ne 'ARRAY';
359 # what we will return
362 # see what aliases are there to work with
363 # and record who is a multiplier and who is premultiplied
365 for my $node (@{$attrs->{from}}) {
368 $j = $j->[0] if ref $j eq 'ARRAY';
369 my $al = $j->{-alias}
372 $alias_list->{$al} = $j;
374 $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] }
375 # not array == {from} head == can't be multiplying
376 if ref($node) eq 'ARRAY' and ! $j->{-is_single};
378 $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] }
379 # parts of the path that are not us but are multiplying
380 if grep { $aliases_by_type->{multiplying}{$_} }
383 @{ $j->{-join_path}||[] }
386 # get a column to source/alias map (including unambiguous unqualified ones)
387 my $colinfo = $self->_resolve_column_info ($attrs->{from});
389 # set up a botched SQLA
390 my $sql_maker = $self->sql_maker;
392 # these are throw away results, do not pollute the bind stack
393 local $sql_maker->{where_bind};
394 local $sql_maker->{group_bind};
395 local $sql_maker->{having_bind};
396 local $sql_maker->{from_bind};
398 # we can't scan properly without any quoting (\b doesn't cut it
399 # everywhere), so unless there is proper quoting set - use our
400 # own weird impossible character.
401 # Also in the case of no quoting, we need to explicitly disable
402 # name_sep, otherwise sorry nasty legacy syntax like
403 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
404 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
405 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
407 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
408 $sql_maker->{quote_char} = ["\x00", "\xFF"];
409 # if we don't unset it we screw up retarded but unfortunately working
410 # 'MAX(foo.bar)' => { '>', 3 }
411 $sql_maker->{name_sep} = '';
414 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
416 # generate sql chunks
419 $sql_maker->_recurse_where ($attrs->{where}),
420 $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }),
423 $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }),
426 $sql_maker->_recurse_from (
427 ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0],
428 @{$attrs->{from}}[1 .. $#{$attrs->{from}}],
432 map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}},
435 map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker),
439 # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are
440 # bind value specs left in by the sloppy renderer above. It is ok to do this
441 # at this point, since we are going to end up rewriting this crap anyway
442 for my $v (values %$to_scan) {
451 ( @$_ == 0 or @$_ == 2 )
456 require Data::Dumper::Concise;
457 $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) );
466 # kill all selectors which look like a proper subquery
467 # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
468 # fail to run, so we are relatively safe
469 $to_scan->{selecting} = [ grep {
470 $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi
471 } @{ $to_scan->{selecting} || [] } ];
473 # first see if we have any exact matches (qualified or unqualified)
474 for my $type (keys %$to_scan) {
475 for my $piece (@{$to_scan->{$type}}) {
476 if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) {
477 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
478 $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece;
483 # now loop through all fully qualified columns and get the corresponding
484 # alias (should work even if they are in scalarrefs)
485 for my $alias (keys %$alias_list) {
487 $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
489 \b $alias \. ([^\s\)\($rquote]+)?
492 for my $type (keys %$to_scan) {
493 for my $piece (@{$to_scan->{$type}}) {
494 if (my @matches = $piece =~ /$al_re/g) {
495 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
496 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_"
497 for grep { defined $_ } @matches;
503 # now loop through unqualified column names, and try to locate them within
505 for my $col (keys %$colinfo) {
506 next if $col =~ / \. /x; # if column is qualified it was caught by the above
508 my $col_re = qr/ $lquote ($col) $rquote /x;
510 for my $type (keys %$to_scan) {
511 for my $piece (@{$to_scan->{$type}}) {
512 if ( my @matches = $piece =~ /$col_re/g) {
513 my $alias = $colinfo->{$col}{-source_alias};
514 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
515 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
516 for grep { defined $_ } @matches;
522 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
523 for my $j (values %$alias_list) {
524 my $alias = $j->{-alias} or next;
525 $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if (
526 (not $j->{-join_type})
528 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
532 for (keys %$aliases_by_type) {
533 delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}};
536 return $aliases_by_type;
539 # This is the engine behind { distinct => 1 } and the general
540 # complex prefetch grouper
541 sub _group_over_selection {
542 my ($self, $attrs) = @_;
544 my $colinfos = $self->_resolve_column_info ($attrs->{from});
546 my (@group_by, %group_index);
548 # the logic is: if it is a { func => val } we assume an aggregate,
549 # otherwise if \'...' or \[...] we assume the user knows what is
550 # going on thus group over it
551 for (@{$attrs->{select}}) {
552 if (! ref($_) or ref ($_) ne 'HASH' ) {
555 if ($colinfos->{$_} and $_ !~ /\./ ) {
556 # add a fully qualified version as well
557 $group_index{"$colinfos->{$_}{-source_alias}.$_"}++;
562 my @order_by = $self->_extract_order_criteria($attrs->{order_by})
563 or return (\@group_by, $attrs->{order_by});
565 # add any order_by parts that are not already present in the group_by
566 # to maintain SQL cross-compatibility and general sanity
568 # also in case the original selection is *not* unique, or in case part
569 # of the ORDER BY refers to a multiplier - we will need to replace the
570 # skipped order_by elements with their MIN/MAX equivalents as to maintain
571 # the proper overall order without polluting the group criteria (and
572 # possibly changing the outcome entirely)
574 my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes);
576 my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by);
578 for my $o_idx (0 .. $#order_by) {
580 # if the chunk is already a min/max function - there is nothing left to touch
581 next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix;
583 # only consider real columns (for functions the user got to do an explicit group_by)
586 @{$order_by[$o_idx]} != 1
588 # only declare an unknown *plain* identifier as "leftover" if we are called with
589 # aliastypes to examine. If there are none - we are still in _resolve_attrs, and
590 # can just assume the user knows what they want
591 ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} )
593 push @$leftovers, $order_by[$o_idx][0];
596 next unless $chunk_ci;
598 # no duplication of group criteria
599 next if $group_index{$chunk_ci->{-fq_colname}};
602 $attrs->{_aliastypes}
604 $self->_resolve_aliastypes_from_select_args({
605 from => $attrs->{from},
606 order_by => $attrs->{order_by},
608 ) if $group_already_unique;
610 # check that we are not ordering by a multiplier (if a check is requested at all)
612 $group_already_unique
614 ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}}
616 ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}}
618 push @group_by, $chunk_ci->{-fq_colname};
619 $group_index{$chunk_ci->{-fq_colname}}++
622 # We need to order by external columns without adding them to the group
623 # (eiehter a non-unique selection, or a multi-external)
625 # This doesn't really make sense in SQL, however from DBICs point
626 # of view is rather valid (e.g. order the leftmost objects by whatever
627 # criteria and get the offset/rows many). There is a way around
628 # this however in SQL - we simply tae the direction of each piece
629 # of the external order and convert them to MIN(X) for ASC or MAX(X)
630 # for DESC, and group_by the root columns. The end result should be
631 # exactly what we expect
633 # FIXME - this code is a joke, will need to be completely rewritten in
634 # the DQ branch. But I need to push a POC here, otherwise the
635 # pesky tests won't pass
636 # wrap any part of the order_by that "responds" to an ordering alias
638 $sql_maker ||= $self->sql_maker;
640 map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by})
643 my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]);
645 $new_order_by[$o_idx] = \[
646 sprintf( '%s( %s )%s',
647 ($is_desc ? 'MAX' : 'MIN'),
649 ($is_desc ? ' DESC' : ''),
651 @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ]
656 $self->throw_exception ( sprintf
657 'Unable to programatically derive a required group_by from the supplied '
658 . 'order_by criteria. To proceed either add an explicit group_by, or '
659 . 'simplify your order_by to only include plain columns '
660 . '(supplied order_by: %s)',
661 join ', ', map { "'$_'" } @$leftovers,
664 # recreate the untouched order parts
666 $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks );
671 (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged
675 sub _resolve_ident_sources {
676 my ($self, $ident) = @_;
678 my $alias2source = {};
680 # the reason this is so contrived is that $ident may be a {from}
681 # structure, specifying multiple tables to join
682 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
683 # this is compat mode for insert/update/delete which do not deal with aliases
684 $alias2source->{me} = $ident;
686 elsif (ref $ident eq 'ARRAY') {
690 if (ref $_ eq 'HASH') {
693 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
697 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
698 if ($tabinfo->{-rsrc});
702 return $alias2source;
705 # Takes $ident, \@column_names
707 # returns { $column_name => \%column_info, ... }
708 # also note: this adds -result_source => $rsrc to the column info
710 # If no columns_names are supplied returns info about *all* columns
712 sub _resolve_column_info {
713 my ($self, $ident, $colnames) = @_;
715 return {} if $colnames and ! @$colnames;
717 my $alias2src = $self->_resolve_ident_sources($ident);
719 my (%seen_cols, @auto_colnames);
721 # compile a global list of column names, to be able to properly
722 # disambiguate unqualified column names (if at all possible)
723 for my $alias (keys %$alias2src) {
724 my $rsrc = $alias2src->{$alias};
725 for my $colname ($rsrc->columns) {
726 push @{$seen_cols{$colname}}, $alias;
727 push @auto_colnames, "$alias.$colname" unless $colnames;
733 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
736 my (%return, $colinfos);
737 foreach my $col (@$colnames) {
738 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
740 # if the column was seen exactly once - we know which rsrc it came from
741 $source_alias ||= $seen_cols{$colname}[0]
742 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
744 next unless $source_alias;
746 my $rsrc = $alias2src->{$source_alias}
751 ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
753 $self->throw_exception(
754 "No such column '$colname' on source " . $rsrc->source_name
757 -result_source => $rsrc,
758 -source_alias => $source_alias,
759 -fq_colname => $col eq $colname ? "$source_alias.$col" : $col,
760 -colname => $colname,
763 $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname;
769 # The DBIC relationship chaining implementation is pretty simple - every
770 # new related_relationship is pushed onto the {from} stack, and the {select}
771 # window simply slides further in. This means that when we count somewhere
772 # in the middle, we got to make sure that everything in the join chain is an
773 # actual inner join, otherwise the count will come back with unpredictable
774 # results (a resultset may be generated with _some_ rows regardless of if
775 # the relation which the $rs currently selects has rows or not). E.g.
776 # $artist_rs->cds->count - normally generates:
777 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
778 # which actually returns the number of artists * (number of cds || 1)
780 # So what we do here is crawl {from}, determine if the current alias is at
781 # the top of the stack, and if not - make sure the chain is inner-joined down
784 sub _inner_join_to_node {
785 my ($self, $from, $alias) = @_;
787 # subqueries and other oddness are naturally not supported
793 ref $from->[0] ne 'HASH'
797 $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do
800 # find the current $alias in the $from structure
803 for my $j (@{$from}[1 .. $#$from]) {
804 if ($j->[0]{-alias} eq $alias) {
805 $switch_branch = $j->[0]{-join_path};
810 # something else went quite wrong
811 return $from unless $switch_branch;
813 # So it looks like we will have to switch some stuff around.
814 # local() is useless here as we will be leaving the scope
815 # anyway, and deep cloning is just too fucking expensive
816 # So replace the first hashref in the node arrayref manually
817 my @new_from = ($from->[0]);
818 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
820 for my $j (@{$from}[1 .. $#$from]) {
821 my $jalias = $j->[0]{-alias};
823 if ($sw_idx->{$jalias}) {
824 my %attrs = %{$j->[0]};
825 delete $attrs{-join_type};
839 sub _extract_order_criteria {
840 my ($self, $order_by, $sql_maker) = @_;
843 my ($sql_maker, $order_by, $orig_quote_chars) = @_;
845 return scalar $sql_maker->_order_by_chunks ($order_by)
848 my ($lq, $rq, $sep) = map { quotemeta($_) } (
849 ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars),
854 for ($sql_maker->_order_by_chunks ($order_by) ) {
855 my $chunk = ref $_ ? [ @$_ ] : [ $_ ];
856 ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]);
858 # order criteria may have come back pre-quoted (literals and whatnot)
859 # this is fragile, but the best we can currently do
860 $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe
861 or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x;
863 push @chunks, $chunk;
870 return $parser->($sql_maker, $order_by);
873 $sql_maker = $self->sql_maker;
875 # pass these in to deal with literals coming from
876 # the user or the deep guts of prefetch
877 my $orig_quote_chars = [$sql_maker->_quote_chars];
879 local $sql_maker->{quote_char};
880 return $parser->($sql_maker, $order_by, $orig_quote_chars);
884 sub _order_by_is_stable {
885 my ($self, $ident, $order_by, $where) = @_;
888 ( map { $_->[0] } $self->_extract_order_criteria($order_by) ),
889 ( $where ? @{ $self->_extract_fixed_condition_columns($where) || [] } : () ),
892 my $colinfo = $self->_resolve_column_info($ident, \@cols);
894 return keys %$colinfo
895 ? $self->_columns_comprise_identifying_set( $colinfo, \@cols )
900 sub _columns_comprise_identifying_set {
901 my ($self, $colinfo, $columns) = @_;
904 $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_
905 for grep { defined $_ } @{$colinfo}{@$columns};
907 for (values %$cols_per_src) {
908 my $src = (values %$_)[0]->{-result_source};
909 return 1 if $src->_identifying_column_set($_);
915 # this is almost identical to the above, except it accepts only
916 # a single rsrc, and will succeed only if the first portion of the order
918 # returns that portion as a colinfo hashref on success
919 sub _main_source_order_by_portion_is_stable {
920 my ($self, $main_rsrc, $order_by, $where) = @_;
922 die "Huh... I expect a blessed result_source..."
923 if ref($main_rsrc) eq 'ARRAY';
927 ( $self->_extract_order_criteria($order_by) )
929 return unless @ord_cols;
931 my $colinfos = $self->_resolve_column_info($main_rsrc);
933 for (0 .. $#ord_cols) {
935 ! $colinfos->{$ord_cols[$_]}
937 $colinfos->{$ord_cols[$_]}{-result_source} != $main_rsrc
944 # we just truncated it above
945 return unless @ord_cols;
947 my $order_portion_ci = { map {
948 $colinfos->{$_}{-colname} => $colinfos->{$_},
949 $colinfos->{$_}{-fq_colname} => $colinfos->{$_},
952 # since all we check here are the start of the order_by belonging to the
953 # top level $rsrc, a present identifying set will mean that the resultset
954 # is ordered by its leftmost table in a stable manner
956 # RV of _identifying_column_set contains unqualified names only
957 my $unqualified_idset = $main_rsrc->_identifying_column_set({
959 $self->_resolve_column_info(
960 $main_rsrc, $self->_extract_fixed_condition_columns($where)||[]
967 my %unqualified_idcols_from_order = map {
968 $order_portion_ci->{$_} ? ( $_ => $order_portion_ci->{$_} ) : ()
969 } @$unqualified_idset;
971 # extra optimization - cut the order_by at the end of the identifying set
972 # (just in case the user was stupid and overlooked the obvious)
973 for my $i (0 .. $#ord_cols) {
974 my $col = $ord_cols[$i];
975 my $unqualified_colname = $order_portion_ci->{$col}{-colname};
976 $ret_info->{$col} = { %{$order_portion_ci->{$col}}, -idx_in_order_subset => $i };
977 delete $unqualified_idcols_from_order{$ret_info->{$col}{-colname}};
979 # we didn't reach the end of the identifying portion yet
980 return $ret_info unless keys %unqualified_idcols_from_order;
983 die 'How did we get here...';
986 # Attempts to flatten a passed in SQLA condition as much as possible towards
987 # a plain hashref, *without* altering its semantics. Required by
988 # create/populate being able to extract definitive conditions from preexisting
989 # resultset {where} stacks
991 # FIXME - while relatively robust, this is still imperfect, one of the first
992 # things to tackle with DQ
994 my ($self, $where, $where_is_anded_array) = @_;
999 elsif ($where_is_anded_array or ref $where eq 'HASH') {
1003 my @pieces = $where_is_anded_array ? @$where : $where;
1005 my $chunk = shift @pieces;
1007 if (ref $chunk eq 'HASH') {
1008 push @pairs, map { [ $_ => $chunk->{$_} ] } sort keys %$chunk;
1010 elsif (ref $chunk eq 'ARRAY') {
1011 push @pairs, [ -or => $chunk ]
1014 elsif ( ! ref $chunk) {
1015 push @pairs, [ $chunk, shift @pieces ];
1018 push @pairs, [ '', $chunk ];
1022 return unless @pairs;
1024 my @conds = $self->_collapse_cond_unroll_pairs(\@pairs)
1027 # Consolidate various @conds back into something more compact
1030 for my $c (@conds) {
1031 if (ref $c ne 'HASH') {
1032 push @{$fin->{-and}}, $c;
1035 for my $col (sort keys %$c) {
1036 if (exists $fin->{$col}) {
1037 my ($l, $r) = ($fin->{$col}, $c->{$col});
1039 (ref $_ ne 'ARRAY' or !@$_) and $_ = [ -and => $_ ] for ($l, $r);
1041 if (@$l and @$r and $l->[0] eq $r->[0] and $l->[0] eq '-and') {
1042 $fin->{$col} = [ -and => map { @$_[1..$#$_] } ($l, $r) ];
1045 $fin->{$col} = [ -and => $fin->{$col}, $c->{$col} ];
1049 $fin->{$col} = $c->{$col};
1055 if ( ref $fin->{-and} eq 'ARRAY' and @{$fin->{-and}} == 1 ) {
1056 my $piece = (delete $fin->{-and})->[0];
1057 if (ref $piece eq 'ARRAY') {
1058 $fin->{-or} = $fin->{-or} ? [ $piece, $fin->{-or} ] : $piece;
1060 elsif (! exists $fin->{''}) {
1061 $fin->{''} = $piece;
1067 elsif (ref $where eq 'ARRAY') {
1071 (ref $w[0] eq 'ARRAY' and ! @{$w[0]} )
1073 (ref $w[0] eq 'HASH' and ! keys %{$w[0]})
1079 return ( ref $w[0] )
1080 ? $self->_collapse_cond($w[0])
1081 : { $w[0] => undef }
1084 elsif ( ref $w[0] ) {
1088 if ( ( $w[0]||'' ) =~ /^\-and$/i ) {
1089 return (ref $w[1] eq 'HASH' or ref $w[1] eq 'ARRAY')
1090 ? $self->_collapse_cond($w[1], (ref $w[1] eq 'ARRAY') )
1091 : $self->throw_exception("Unsupported top-level op/arg pair: [ $w[0] => $w[1] ]")
1095 return $self->_collapse_cond({ @w });
1100 # not a hash not an array
1101 return { '' => $where };
1104 # catchall, some of the things above fall through
1108 sub _collapse_cond_unroll_pairs {
1109 my ($self, $pairs) = @_;
1114 my ($lhs, $rhs) = @{ shift @$pairs };
1117 push @conds, $self->_collapse_cond($rhs);
1119 elsif ( $lhs =~ /^\-and$/i ) {
1120 push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY'));
1122 elsif ( $lhs =~ /^\-or$/i ) {
1123 push @conds, $self->_collapse_cond(
1124 (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs
1128 if (ref $rhs eq 'HASH' and ! keys %$rhs) {
1129 # FIXME - SQLA seems to be doing... nothing...?
1131 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) {
1132 for my $p ($self->_collapse_cond_unroll_pairs([ [ $lhs => $rhs->{'='} ] ])) {
1134 # extra sanity check
1136 require Data::Dumper::Concise;
1137 local $Data::Dumper::Deepcopy = 1;
1138 $self->throw_exception(
1139 "Internal error: unexpected collapse unroll:"
1140 . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p }
1146 push @conds, ( ! length ref $r or is_plain_value($r) )
1148 : { $l => { '=' => $r } }
1152 elsif (ref $rhs eq 'ARRAY') {
1153 # some of these conditionals encounter multi-values - roll them out using
1154 # an unshift, which will cause extra looping in the while{} above
1156 push @conds, { $lhs => [] };
1158 elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) {
1159 $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ")
1162 if( $rhs->[0] =~ /^\-and$/i ) {
1163 unshift @$pairs, map { [ $lhs => $_ ] } @{$rhs}[1..$#$rhs];
1165 # if not an AND then it's an OR
1167 unshift @$pairs, [ $lhs => $rhs->[1] ];
1170 push @conds, { $lhs => $rhs };
1173 elsif (@$rhs == 1) {
1174 unshift @$pairs, [ $lhs => $rhs->[0] ];
1177 push @conds, { $lhs => $rhs };
1181 push @conds, { $lhs => $rhs };
1190 # returns an arrayref of column names which *definitely* have some
1191 # sort of non-nullable *single* equality requested in the given condition
1192 # specification. This is used to figure out if a resultset is
1193 # constrained to a column which is part of a unique constraint,
1194 # which in turn allows us to better predict how ordering will behave
1197 # this is a rudimentary, incomplete, and error-prone extractor
1198 # however this is OK - it is conservative, and if we can not find
1199 # something that is in fact there - the stack will recover gracefully
1200 sub _extract_fixed_condition_columns {
1202 my $where_hash = $self->_collapse_cond(shift);
1205 for my $c (keys %$where_hash) {
1206 if (defined (my $v = $where_hash->{$c}) ) {
1210 (ref $v eq 'HASH' and keys %$v == 1 and defined $v->{'='} and (
1211 is_literal_value($v->{'='}) or is_plain_value( $v->{'='})
1216 elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') {
1217 $res->{$_} = 1 for map { @{ $self->_extract_fixed_condition_columns({ $c => $_ }) } } @{$v}[1..$#$v];
1222 return [ sort keys %$res ];