1 package #hide from PAUSE
2 DBIx::Class::Storage::DBIHacks;
5 # This module contains code supporting a battery of special cases and tests for
6 # many corner cases pushing the envelope of what DBIC can do. When work on
7 # these utilities began in mid 2009 (51a296b402c) it wasn't immediately obvious
8 # that these pieces, despite their misleading on-first-sight-flakiness, will
9 # become part of the generic query rewriting machinery of DBIC, allowing it to
10 # both generate and process queries representing incredibly complex sets with
11 # reasonable efficiency.
13 # Now (end of 2015), more than 6 years later the routines in this class have
14 # stabilized enough, and are meticulously covered with tests, to a point where
15 # an effort to formalize them into user-facing APIs might be worthwhile.
17 # An implementor working on publicizing and/or replacing the routines with a
18 # more modern SQL generation framework should keep in mind that pretty much all
19 # existing tests are constructed on the basis of real-world code used in
20 # production somewhere.
22 # Please hack on this responsibly ;)
28 use base 'DBIx::Class::Storage';
31 use Scalar::Util 'blessed';
32 use DBIx::Class::_Util qw(
33 dump_value fail_on_internal_call
35 use DBIx::Class::SQLMaker::Util 'extract_equality_conditions';
36 use DBIx::Class::Carp;
40 # This code will remove non-selecting/non-restricting joins from
41 # {from} specs, aiding the RDBMS query optimizer
43 sub _prune_unused_joins {
44 my ($self, $attrs) = @_;
46 # only standard {from} specs are supported, and we could be disabled in general
47 return ($attrs->{from}, {}) unless (
48 ref $attrs->{from} eq 'ARRAY'
52 ref $attrs->{from}[0] eq 'HASH'
54 ref $attrs->{from}[1] eq 'ARRAY'
56 $self->_use_join_optimizer
60 $attrs->{_precalculated_aliastypes}
62 $self->_resolve_aliastypes_from_select_args($attrs)
65 my $new_aliastypes = { %$orig_aliastypes };
67 # we will be recreating this entirely
68 my @reclassify = 'joining';
70 # a grouped set will not be affected by amount of rows. Thus any
71 # purely multiplicator classifications can go
72 # (will be reintroduced below if needed by something else)
73 push @reclassify, qw(multiplying premultiplied)
74 if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by};
76 # nuke what will be recalculated
77 delete @{$new_aliastypes}{@reclassify};
79 my @newfrom = $attrs->{from}[0]; # FROM head is always present
81 # recalculate what we need once the multipliers are potentially gone
82 # ignore premultiplies, since they do not add any value to anything
84 for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) {
85 # add all requested aliases
86 $need_joins{$_} = 1 for keys %$_;
88 # add all their parents (as per joinpath which is an AoH { table => alias })
89 $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_;
92 for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) {
93 push @newfrom, $j if (
94 (! defined $j->[0]{-alias}) # legacy crap
96 $need_joins{$j->[0]{-alias}}
100 # we have a new set of joiners - for everything we nuked pull the classification
101 # off the original stack
102 for my $ctype (@reclassify) {
103 $new_aliastypes->{$ctype} = { map
104 { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () }
105 keys %{$orig_aliastypes->{$ctype}}
109 return ( \@newfrom, $new_aliastypes );
113 # This is the code producing joined subqueries like:
114 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
116 sub _adjust_select_args_for_complex_prefetch {
117 my ($self, $attrs) = @_;
119 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless (
120 ref $attrs->{from} eq 'ARRAY'
122 @{$attrs->{from}} > 1
124 ref $attrs->{from}[0] eq 'HASH'
126 ref $attrs->{from}[1] eq 'ARRAY'
129 my $root_alias = $attrs->{alias};
131 # generate inner/outer attribute lists, remove stuff that doesn't apply
132 my $outer_attrs = { %$attrs };
133 delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)};
135 my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 };
136 delete @{$inner_attrs}{qw(for collapse select as)};
138 # there is no point of ordering the insides if there is no limit
139 delete $inner_attrs->{order_by} if (
140 delete $inner_attrs->{_order_is_artificial}
142 ! $inner_attrs->{rows}
145 # generate the inner/outer select lists
146 # for inside we consider only stuff *not* brought in by the prefetch
147 # on the outside we substitute any function for its alias
148 $outer_attrs->{select} = [ @{$attrs->{select}} ];
150 my ($root_node, $root_node_offset);
152 for my $i (0 .. $#{$inner_attrs->{from}}) {
153 my $node = $inner_attrs->{from}[$i];
154 my $h = (ref $node eq 'HASH') ? $node
155 : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
159 if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) {
161 $root_node_offset = $i;
166 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
169 # use the heavy duty resolver to take care of aliased/nonaliased naming
170 my $colinfo = $self->_resolve_column_info($inner_attrs->{from});
171 my $selected_root_columns;
173 for my $i (0 .. $#{$outer_attrs->{select}}) {
174 my $sel = $outer_attrs->{select}->[$i];
177 $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias
180 if (ref $sel eq 'HASH' ) {
181 $sel->{-as} ||= $attrs->{as}[$i];
182 $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") );
184 elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
185 $selected_root_columns->{$ci->{-colname}} = 1;
188 push @{$inner_attrs->{select}}, $sel;
190 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
193 my $inner_aliastypes = $self->_resolve_aliastypes_from_select_args($inner_attrs);
195 # In the inner subq we will need to fetch *only* native columns which may
196 # be a part of an *outer* join condition, or an order_by (which needs to be
197 # preserved outside), or wheres. In other words everything but the inner
199 # We can not just fetch everything because a potential has_many restricting
200 # join collapse *will not work* on heavy data types.
202 # essentially a map of all non-selecting seen columns
203 # the sort is there for a nicer select list
207 { keys %{$_->{-seen_columns}||{}} }
209 { values %{$inner_aliastypes->{$_}} }
211 { $_ ne 'selecting' }
212 keys %$inner_aliastypes
214 my $ci = $colinfo->{$_} or next;
216 $ci->{-source_alias} eq $root_alias
218 ! $selected_root_columns->{$ci->{-colname}}++
220 # adding it to both to keep limits not supporting dark selectors happy
221 push @{$inner_attrs->{select}}, $ci->{-fq_colname};
222 push @{$inner_attrs->{as}}, $ci->{-fq_colname};
226 # construct the inner {from} and lock it in a subquery
227 # we need to prune first, because this will determine if we need a group_by below
228 # throw away all non-selecting, non-restricting multijoins
229 # (since we def. do not care about multiplication of the contents of the subquery)
230 my $inner_subq = do {
232 # must use it here regardless of user requests (vastly gentler on optimizer)
233 local $self->{_use_join_optimizer} = 1
234 unless $self->{_use_join_optimizer};
236 # throw away multijoins since we def. do not care about those inside the subquery
237 # $inner_aliastypes *will* be redefined at this point
238 ($inner_attrs->{from}, $inner_aliastypes ) = $self->_prune_unused_joins ({
240 _force_prune_multiplying_joins => 1,
241 _precalculated_aliastypes => $inner_aliastypes,
244 # uh-oh a multiplier (which is not us) left in, this is a problem for limits
245 # we will need to add a group_by to collapse the resultset for proper counts
247 grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} }
249 # if there are user-supplied groups - assume user knows wtf they are up to
250 ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} )
253 my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} };
255 # *possibly* supplement the main selection with pks if not already
256 # there, as they will have to be a part of the group_by to collapse
258 my $inner_select_with_extras;
259 my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns
260 or $self->throw_exception( sprintf
261 'Unable to perform complex limited prefetch off %s without declared primary key',
262 $root_node->{-rsrc}->source_name,
265 push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col
266 unless $cur_sel->{$col}++;
269 ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({
271 $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (),
272 _aliastypes => $inner_aliastypes,
276 # we already optimized $inner_attrs->{from} above
277 # and already local()ized
278 $self->{_use_join_optimizer} = 0;
280 # generate the subquery
281 $self->_select_args_to_query (
282 @{$inner_attrs}{qw(from select where)},
287 # Generate the outer from - this is relatively easy (really just replace
288 # the join slot with the subquery), with a major caveat - we can not
289 # join anything that is non-selecting (not part of the prefetch), but at
290 # the same time is a multi-type relationship, as it will explode the result.
292 # There are two possibilities here
293 # - either the join is non-restricting, in which case we simply throw it away
294 # - it is part of the restrictions, in which case we need to collapse the outer
295 # result by tackling yet another group_by to the outside of the query
297 # work on a shallow copy
298 my @orig_from = @{$attrs->{from}};
301 $outer_attrs->{from} = \ my @outer_from;
303 # we may not be the head
304 if ($root_node_offset) {
305 # first generate the outer_from, up to the substitution point
306 @outer_from = splice @orig_from, 0, $root_node_offset;
308 # substitute the subq at the right spot
311 -alias => $root_alias,
312 -rsrc => $root_node->{-rsrc},
313 $root_alias => $inner_subq,
315 # preserve attrs from what is now the head of the from after the splice
316 @{$orig_from[0]}[1 .. $#{$orig_from[0]}],
321 -alias => $root_alias,
322 -rsrc => $root_node->{-rsrc},
323 $root_alias => $inner_subq,
327 shift @orig_from; # what we just replaced above
329 # scan the *remaining* from spec against different attributes, and see which joins are needed
331 my $outer_aliastypes = $outer_attrs->{_aliastypes} =
332 $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from });
335 my ($outer_select_chain, @outer_nonselecting_chains) = map { +{
336 map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} }
337 } } qw/selecting restricting grouping ordering/;
339 # see what's left - throw away if not selecting/restricting
340 my $may_need_outer_group_by;
341 while (my $j = shift @orig_from) {
342 my $alias = $j->[0]{-alias};
345 $outer_select_chain->{$alias}
349 elsif (grep { $_->{$alias} } @outer_nonselecting_chains ) {
350 push @outer_from, $j;
351 $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
355 # also throw in a synthetic group_by if a non-selecting multiplier,
356 # to guard against cross-join explosions
357 # the logic is somewhat fragile, but relies on the idea that if a user supplied
358 # a group by on their own - they know what they were doing
359 if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) {
360 ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({
362 from => \@outer_from,
366 # FIXME: The {where} ends up in both the inner and outer query, i.e. *twice*
368 # This is rather horrific, and while we currently *do* have enough
369 # introspection tooling available to attempt a stab at properly deciding
370 # whether or not to include the where condition on the outside, the
371 # machinery is still too slow to apply it here.
372 # Thus for the time being we do not attempt any sanitation of the where
373 # clause and just pass it through on both sides of the subquery. This *will*
374 # be addressed at a later stage, most likely after folding the SQL generator
375 # into SQLMaker proper
377 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
382 # This is probably the ickiest, yet most relied upon part of the codebase:
383 # this is the place where we take arbitrary SQL input and break it into its
384 # constituent parts, making sure we know which *sources* are used in what
385 # *capacity* ( selecting / restricting / grouping / ordering / joining, etc )
386 # Although the method is pretty horrific, the worst thing that can happen is
387 # for a classification failure, which in turn will result in a vocal exception,
388 # and will lead to a relatively prompt fix.
389 # The code has been slowly improving and is covered with a formiddable battery
390 # of tests, so can be considered "reliably stable" at this point (Oct 2015).
392 # A note to implementors attempting to "replace" this - keep in mind that while
393 # there are multiple optimization avenues, the actual "scan literal elements"
394 # part *MAY NEVER BE REMOVED*, even if it is limited only ot the (future) AST
395 # nodes that are deemed opaque (i.e. contain literal expressions). The use of
396 # blackbox literals is at this point firmly a user-facing API, and is one of
397 # *the* reasons DBIC remains as flexible as it is. In other words, when working
398 # on this keep in mind that the following is widespread and *encouraged* way
399 # of using DBIC in the wild when push comes to shove:
402 # select => \[ $random, @stuff],
403 # from => \[ $random, @stuff ],
404 # where => \[ $random, @stuff ],
405 # group_by => \[ $random, @stuff ],
406 # order_by => \[ $random, @stuff ],
409 # Various incarnations of the above are reflected in many of the tests. If one
410 # gets to fail, you get to fix it. A "this is crazy, nobody does that" is not
411 # acceptable going forward.
413 sub _resolve_aliastypes_from_select_args {
414 my ( $self, $attrs ) = @_;
416 $self->throw_exception ('Unable to analyze custom {from}')
417 if ref $attrs->{from} ne 'ARRAY';
419 # what we will return
422 # see what aliases are there to work with
423 # and record who is a multiplier and who is premultiplied
425 for my $node (@{$attrs->{from}}) {
428 $j = $j->[0] if ref $j eq 'ARRAY';
429 my $al = $j->{-alias}
432 $alias_list->{$al} = $j;
434 $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] }
435 # not array == {from} head == can't be multiplying
436 if ref($node) eq 'ARRAY' and ! $j->{-is_single};
438 $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] }
439 # parts of the path that are not us but are multiplying
440 if grep { $aliases_by_type->{multiplying}{$_} }
443 @{ $j->{-join_path}||[] }
446 # get a column to source/alias map (including unambiguous unqualified ones)
447 my $colinfo = $self->_resolve_column_info ($attrs->{from});
449 # set up a botched SQLA
450 my $sql_maker = $self->sql_maker;
452 # these are throw away results, do not pollute the bind stack
453 local $sql_maker->{where_bind};
454 local $sql_maker->{group_bind};
455 local $sql_maker->{having_bind};
456 local $sql_maker->{from_bind};
458 # we can't scan properly without any quoting (\b doesn't cut it
459 # everywhere), so unless there is proper quoting set - use our
460 # own weird impossible character.
461 # Also in the case of no quoting, we need to explicitly disable
462 # name_sep, otherwise sorry nasty legacy syntax like
463 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
464 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
465 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
467 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
468 $sql_maker->{quote_char} = ["\x00", "\xFF"];
469 # if we don't unset it we screw up retarded but unfortunately working
470 # 'MAX(foo.bar)' => { '>', 3 }
471 $sql_maker->{name_sep} = '';
474 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
476 # generate sql chunks
479 ($sql_maker->_recurse_where ($attrs->{where}))[0],
480 $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }),
483 $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }),
486 $sql_maker->_recurse_from (
487 ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0],
488 @{$attrs->{from}}[1 .. $#{$attrs->{from}}],
492 # kill all selectors which look like a proper subquery
493 # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
494 # fail to run, so we are relatively safe
496 { $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi }
498 { ($sql_maker->_recurse_fields($_))[0] }
503 ( my $sql = (ref $_ ? $_->[0] : $_) ) =~ s/ \s+ (?: ASC | DESC ) \s* \z //xi;
506 $sql_maker->_order_by_chunks( $attrs->{order_by} ),
510 # we will be bulk-scanning anyway - pieces will not matter in that case,
511 # thus join everything up
512 # throw away empty-string chunks, and make sure no binds snuck in
513 # note that we operate over @{$to_scan->{$type}}, hence the
514 # semi-mindbending ... map ... for values ...
515 ( $_ = join ' ', map {
517 ( ! defined $_ ) ? ()
518 : ( length ref $_ ) ? $self->throw_exception(
519 "Unexpected ref in scan-plan: " . dump_value $_
521 : ( $_ =~ /^\s*$/ ) ? ()
524 } @$_ ) for values %$to_scan;
526 # throw away empty to-scan's
528 length $to_scan->{$_}
530 delete $to_scan->{$_}
531 ) for keys %$to_scan;
535 # these will be used for matching in the loop below
536 my $all_aliases = join ' | ', map { quotemeta $_ } keys %$alias_list;
538 $lquote ( $all_aliases ) $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
540 \b ( $all_aliases ) \. ( [^\s\)\($rquote]+ )?
544 my $all_unq_columns = join ' | ',
548 # using a regex here shows up on profiles, boggle
549 { index( $_, '.') < 0 }
552 my $unq_col_re = $all_unq_columns
554 $lquote ( $all_unq_columns ) $rquote
556 (?: \A | \s ) ( $all_unq_columns ) (?: \s | \z )
562 # the actual scan, per type
563 for my $type (keys %$to_scan) {
566 # now loop through all fully qualified columns and get the corresponding
567 # alias (should work even if they are in scalarrefs)
569 # The regex captures in multiples of 4, with one of the two pairs being
570 # undef. There may be a *lot* of matches, hence the convoluted loop
571 my @matches = $to_scan->{$type} =~ /$fq_col_re/g;
573 while( $i < $#matches ) {
578 $aliases_by_type->{$type}{$matches[$i]} ||= { -parents => $alias_list->{$matches[$i]}{-join_path}||[] };
580 $aliases_by_type->{$type}{$matches[$i]}{-seen_columns}{"$matches[$i].$matches[$i+1]"} = "$matches[$i].$matches[$i+1]"
581 if defined $matches[$i+1];
590 # now loop through unqualified column names, and try to locate them within
591 # the chunks, if there are any unqualified columns in the 1st place
592 next unless $unq_col_re;
594 # The regex captures in multiples of 2, one of the two being undef
595 for ( $to_scan->{$type} =~ /$unq_col_re/g ) {
597 my $alias = $colinfo->{$_}{-source_alias} or next;
598 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
599 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
604 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
608 ! $aliases_by_type->{restricting}{ $_->{-alias} }
613 $_->{-join_type} !~ /^left (?: \s+ outer)? $/xi
616 $aliases_by_type->{restricting}{ $_->{-alias} } = { -parents => $_->{-join_path}||[] }
617 ) for values %$alias_list;
622 keys %{$aliases_by_type->{$_}}
624 delete $aliases_by_type->{$_}
625 ) for keys %$aliases_by_type;
631 # This is the engine behind { distinct => 1 } and the general
632 # complex prefetch grouper
633 sub _group_over_selection {
634 my ($self, $attrs) = @_;
636 my $colinfos = $self->_resolve_column_info ($attrs->{from});
638 my (@group_by, %group_index);
640 # the logic is: if it is a { func => val } we assume an aggregate,
641 # otherwise if \'...' or \[...] we assume the user knows what is
642 # going on thus group over it
643 for (@{$attrs->{select}}) {
644 if (! ref($_) or ref ($_) ne 'HASH' ) {
647 if ($colinfos->{$_} and $_ !~ /\./ ) {
648 # add a fully qualified version as well
649 $group_index{"$colinfos->{$_}{-source_alias}.$_"}++;
654 my @order_by = $self->_extract_order_criteria($attrs->{order_by})
655 or return (\@group_by, $attrs->{order_by});
657 # add any order_by parts that are not already present in the group_by
658 # to maintain SQL cross-compatibility and general sanity
660 # also in case the original selection is *not* unique, or in case part
661 # of the ORDER BY refers to a multiplier - we will need to replace the
662 # skipped order_by elements with their MIN/MAX equivalents as to maintain
663 # the proper overall order without polluting the group criteria (and
664 # possibly changing the outcome entirely)
666 my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes);
668 my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by);
670 for my $o_idx (0 .. $#order_by) {
672 # if the chunk is already a min/max function - there is nothing left to touch
673 next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix;
675 # only consider real columns (for functions the user got to do an explicit group_by)
678 @{$order_by[$o_idx]} != 1
680 # only declare an unknown *plain* identifier as "leftover" if we are called with
681 # aliastypes to examine. If there are none - we are still in _resolve_attrs, and
682 # can just assume the user knows what they want
683 ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} )
685 push @$leftovers, $order_by[$o_idx][0];
688 next unless $chunk_ci;
690 # no duplication of group criteria
691 next if $group_index{$chunk_ci->{-fq_colname}};
694 $attrs->{_aliastypes}
696 $self->_resolve_aliastypes_from_select_args({
697 from => $attrs->{from},
698 order_by => $attrs->{order_by},
700 ) if $group_already_unique;
702 # check that we are not ordering by a multiplier (if a check is requested at all)
704 $group_already_unique
706 ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}}
708 ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}}
710 push @group_by, $chunk_ci->{-fq_colname};
711 $group_index{$chunk_ci->{-fq_colname}}++
714 # We need to order by external columns without adding them to the group
715 # (eiehter a non-unique selection, or a multi-external)
717 # This doesn't really make sense in SQL, however from DBICs point
718 # of view is rather valid (e.g. order the leftmost objects by whatever
719 # criteria and get the offset/rows many). There is a way around
720 # this however in SQL - we simply tae the direction of each piece
721 # of the external order and convert them to MIN(X) for ASC or MAX(X)
722 # for DESC, and group_by the root columns. The end result should be
723 # exactly what we expect
726 # both populated on the first loop over $o_idx
727 $sql_maker ||= $self->sql_maker;
729 map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by})
732 my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]);
734 # we reached that far - wrap any part of the order_by that "responded"
735 # to an ordering alias into a MIN/MAX
736 $new_order_by[$o_idx] = \[
737 sprintf( '%s( %s )%s',
738 $self->_minmax_operator_for_datatype($chunk_ci->{data_type}, $is_desc),
740 ($is_desc ? ' DESC' : ''),
742 @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ]
747 $self->throw_exception ( sprintf
748 'Unable to programatically derive a required group_by from the supplied '
749 . 'order_by criteria. To proceed either add an explicit group_by, or '
750 . 'simplify your order_by to only include plain columns '
751 . '(supplied order_by: %s)',
752 join ', ', map { "'$_'" } @$leftovers,
755 # recreate the untouched order parts
757 $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks );
762 (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged
766 sub _minmax_operator_for_datatype {
767 #my ($self, $datatype, $want_max) = @_;
769 $_[2] ? 'MAX' : 'MIN';
772 sub _resolve_ident_sources {
773 my ($self, $ident) = @_;
775 my $alias2source = {};
777 # the reason this is so contrived is that $ident may be a {from}
778 # structure, specifying multiple tables to join
779 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
780 # this is compat mode for insert/update/delete which do not deal with aliases
781 $alias2source->{me} = $ident;
783 elsif (ref $ident eq 'ARRAY') {
787 if (ref $_ eq 'HASH') {
790 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
794 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
795 if ($tabinfo->{-rsrc});
799 return $alias2source;
802 # Takes $ident, \@column_names
804 # returns { $column_name => \%column_info, ... }
805 # also note: this adds -result_source => $rsrc to the column info
807 # If no columns_names are supplied returns info about *all* columns
809 sub _resolve_column_info {
810 my ($self, $ident, $colnames) = @_;
812 return {} if $colnames and ! @$colnames;
814 my $sources = $self->_resolve_ident_sources($ident);
816 $_ = { rsrc => $_, colinfos => $_->columns_info }
817 for values %$sources;
819 my (%seen_cols, @auto_colnames);
821 # compile a global list of column names, to be able to properly
822 # disambiguate unqualified column names (if at all possible)
823 for my $alias (keys %$sources) {
825 ++$seen_cols{$_}{$alias}
829 push @auto_colnames, "$alias.$_"
830 ) for keys %{ $sources->{$alias}{colinfos} };
835 ( grep { keys %{$seen_cols{$_}} == 1 } keys %seen_cols ),
840 my ($colname, $source_alias) = reverse split /\./, $_;
845 # if the column was seen exactly once - we know which rsrc it came from
849 keys %{$seen_cols{$colname}} == 1
851 ( %{$seen_cols{$colname}} )[0]
857 $self->throw_exception(
858 "No such column '$colname' on source " . $sources->{$assumed_alias}{rsrc}->source_name
859 ) unless $seen_cols{$colname}{$assumed_alias};
862 %{ $sources->{$assumed_alias}{colinfos}{$colname} },
863 -result_source => $sources->{$assumed_alias}{rsrc},
864 -source_alias => $assumed_alias,
865 -fq_colname => "$assumed_alias.$colname",
866 -colname => $colname,
869 $return{"$assumed_alias.$colname"} = $return{$_}
870 unless $source_alias;
876 # The DBIC relationship chaining implementation is pretty simple - every
877 # new related_relationship is pushed onto the {from} stack, and the {select}
878 # window simply slides further in. This means that when we count somewhere
879 # in the middle, we got to make sure that everything in the join chain is an
880 # actual inner join, otherwise the count will come back with unpredictable
881 # results (a resultset may be generated with _some_ rows regardless of if
882 # the relation which the $rs currently selects has rows or not). E.g.
883 # $artist_rs->cds->count - normally generates:
884 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
885 # which actually returns the number of artists * (number of cds || 1)
887 # So what we do here is crawl {from}, determine if the current alias is at
888 # the top of the stack, and if not - make sure the chain is inner-joined down
891 sub _inner_join_to_node {
892 my ($self, $from, $alias) = @_;
894 my $switch_branch = $self->_find_join_path_to_node($from, $alias);
896 return $from unless @{$switch_branch||[]};
898 # So it looks like we will have to switch some stuff around.
899 # local() is useless here as we will be leaving the scope
900 # anyway, and deep cloning is just too fucking expensive
901 # So replace the first hashref in the node arrayref manually
902 my @new_from = ($from->[0]);
903 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
905 for my $j (@{$from}[1 .. $#$from]) {
906 my $jalias = $j->[0]{-alias};
908 if ($sw_idx->{$jalias}) {
909 my %attrs = %{$j->[0]};
910 delete $attrs{-join_type};
924 sub _find_join_path_to_node {
925 my ($self, $from, $target_alias) = @_;
927 # subqueries and other oddness are naturally not supported
931 ref $from->[0] ne 'HASH'
933 ! defined $from->[0]{-alias}
936 # no path - the head is the alias
937 return [] if $from->[0]{-alias} eq $target_alias;
939 for my $i (1 .. $#$from) {
940 return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias );
943 # something else went quite wrong
947 sub _extract_order_criteria {
948 my ($self, $order_by, $sql_maker) = @_;
951 my ($sql_maker, $order_by, $orig_quote_chars) = @_;
953 return scalar $sql_maker->_order_by_chunks ($order_by)
956 my ($lq, $rq, $sep) = map { quotemeta($_) } (
957 ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars),
962 for ($sql_maker->_order_by_chunks ($order_by) ) {
963 my $chunk = ref $_ ? [ @$_ ] : [ $_ ];
964 ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]);
966 # order criteria may have come back pre-quoted (literals and whatnot)
967 # this is fragile, but the best we can currently do
968 $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe
969 or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x;
971 push @chunks, $chunk;
978 return $parser->($sql_maker, $order_by);
981 $sql_maker = $self->sql_maker;
983 # pass these in to deal with literals coming from
984 # the user or the deep guts of prefetch
985 my $orig_quote_chars = [$sql_maker->_quote_chars];
987 local $sql_maker->{quote_char};
988 return $parser->($sql_maker, $order_by, $orig_quote_chars);
992 sub _order_by_is_stable {
993 my ($self, $ident, $order_by, $where) = @_;
996 ( map { $_->[0] } $self->_extract_order_criteria($order_by) ),
997 ( $where ? keys %{ extract_equality_conditions( $where ) } : () ),
1000 my $colinfo = $self->_resolve_column_info($ident, \@cols);
1002 return keys %$colinfo
1003 ? $self->_columns_comprise_identifying_set( $colinfo, \@cols )
1008 sub _columns_comprise_identifying_set {
1009 my ($self, $colinfo, $columns) = @_;
1012 $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_
1013 for grep { defined $_ } @{$colinfo}{@$columns};
1015 for (values %$cols_per_src) {
1016 my $src = (values %$_)[0]->{-result_source};
1017 return 1 if $src->_identifying_column_set($_);
1023 # this is almost similar to _order_by_is_stable, except it takes
1024 # a single rsrc, and will succeed only if the first portion of the order
1026 # returns that portion as a colinfo hashref on success
1027 sub _extract_colinfo_of_stable_main_source_order_by_portion {
1028 my ($self, $attrs) = @_;
1030 my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias});
1032 return unless defined $nodes;
1036 ( $self->_extract_order_criteria($attrs->{order_by}) )
1038 return unless @ord_cols;
1040 my $valid_aliases = { map { $_ => 1 } (
1041 $attrs->{from}[0]{-alias},
1042 map { values %$_ } @$nodes,
1045 my $colinfos = $self->_resolve_column_info($attrs->{from});
1047 my ($colinfos_to_return, $seen_main_src_cols);
1049 for my $col (@ord_cols) {
1050 # if order criteria is unresolvable - there is nothing we can do
1051 my $colinfo = $colinfos->{$col} or last;
1053 # if we reached the end of the allowed aliases - also nothing we can do
1054 last unless $valid_aliases->{$colinfo->{-source_alias}};
1056 $colinfos_to_return->{$col} = $colinfo;
1058 $seen_main_src_cols->{$colinfo->{-colname}} = 1
1059 if $colinfo->{-source_alias} eq $attrs->{alias};
1062 # FIXME: the condition may be singling out things on its own, so we
1063 # conceivably could come back with "stable-ordered by nothing"
1064 # not confident enough in the parser yet, so punt for the time being
1065 return unless $seen_main_src_cols;
1067 my $main_src_fixed_cols_from_cond = [ $attrs->{where}
1071 ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} )
1072 ? $colinfos->{$_}{-colname}
1075 keys %{ extract_equality_conditions( $attrs->{where} ) }
1080 return $attrs->{result_source}->_identifying_column_set([
1081 keys %$seen_main_src_cols,
1082 @$main_src_fixed_cols_from_cond,
1083 ]) ? $colinfos_to_return : ();
1086 sub _collapse_cond :DBIC_method_is_indirect_sugar {
1087 DBIx::Class::_ENV_::ASSERT_NO_INTERNAL_INDIRECT_CALLS and fail_on_internal_call;
1088 carp_unique("_collapse_cond() is deprecated, ask on IRC for a better alternative");
1091 DBIx::Class::SQLMaker::Util::normalize_sqla_condition(@_);
1094 sub _extract_fixed_condition_columns :DBIC_method_is_indirect_sugar {
1095 DBIx::Class::_ENV_::ASSERT_NO_INTERNAL_INDIRECT_CALLS and fail_on_internal_call;
1096 carp_unique("_extract_fixed_condition_columns() is deprecated, ask on IRC for a better alternative");
1099 extract_equality_conditions(@_);