1 package #hide from PAUSE
2 DBIx::Class::Storage::DBIHacks;
5 # This module contains code that should never have seen the light of day,
6 # does not belong in the Storage, or is otherwise unfit for public
7 # display. The arrival of SQLA2 should immediately obsolete 90% of this
13 use base 'DBIx::Class::Storage';
16 use List::Util 'first';
17 use Scalar::Util 'blessed';
18 use Sub::Name 'subname';
22 # This code will remove non-selecting/non-restricting joins from
23 # {from} specs, aiding the RDBMS query optimizer
25 sub _prune_unused_joins {
26 my ($self, $attrs) = @_;
28 # only standard {from} specs are supported, and we could be disabled in general
29 return ($attrs->{from}, {}) unless (
30 ref $attrs->{from} eq 'ARRAY'
34 ref $attrs->{from}[0] eq 'HASH'
36 ref $attrs->{from}[1] eq 'ARRAY'
38 $self->_use_join_optimizer
41 my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs);
43 my $new_aliastypes = { %$orig_aliastypes };
45 # we will be recreating this entirely
46 my @reclassify = 'joining';
48 # a grouped set will not be affected by amount of rows. Thus any
49 # purely multiplicator classifications can go
50 # (will be reintroduced below if needed by something else)
51 push @reclassify, qw(multiplying premultiplied)
52 if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by};
54 # nuke what will be recalculated
55 delete @{$new_aliastypes}{@reclassify};
57 my @newfrom = $attrs->{from}[0]; # FROM head is always present
59 # recalculate what we need once the multipliers are potentially gone
60 # ignore premultiplies, since they do not add any value to anything
62 for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) {
63 # add all requested aliases
64 $need_joins{$_} = 1 for keys %$_;
66 # add all their parents (as per joinpath which is an AoH { table => alias })
67 $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_;
70 for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) {
71 push @newfrom, $j if (
72 (! defined $j->[0]{-alias}) # legacy crap
74 $need_joins{$j->[0]{-alias}}
78 # we have a new set of joiners - for everything we nuked pull the classification
79 # off the original stack
80 for my $ctype (@reclassify) {
81 $new_aliastypes->{$ctype} = { map
82 { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () }
83 keys %{$orig_aliastypes->{$ctype}}
87 return ( \@newfrom, $new_aliastypes );
91 # This is the code producing joined subqueries like:
92 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
94 sub _adjust_select_args_for_complex_prefetch {
95 my ($self, $attrs) = @_;
97 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless (
98 ref $attrs->{from} eq 'ARRAY'
100 @{$attrs->{from}} > 1
102 ref $attrs->{from}[0] eq 'HASH'
104 ref $attrs->{from}[1] eq 'ARRAY'
107 my $root_alias = $attrs->{alias};
109 # generate inner/outer attribute lists, remove stuff that doesn't apply
110 my $outer_attrs = { %$attrs };
111 delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)};
113 my $inner_attrs = { %$attrs };
114 delete @{$inner_attrs}{qw(for collapse select as _related_results_construction)};
116 # there is no point of ordering the insides if there is no limit
117 delete $inner_attrs->{order_by} if (
118 delete $inner_attrs->{_order_is_artificial}
120 ! $inner_attrs->{rows}
123 # generate the inner/outer select lists
124 # for inside we consider only stuff *not* brought in by the prefetch
125 # on the outside we substitute any function for its alias
126 $outer_attrs->{select} = [ @{$attrs->{select}} ];
128 my ($root_node, $root_node_offset);
130 for my $i (0 .. $#{$inner_attrs->{from}}) {
131 my $node = $inner_attrs->{from}[$i];
132 my $h = (ref $node eq 'HASH') ? $node
133 : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
137 if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) {
139 $root_node_offset = $i;
144 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
147 # use the heavy duty resolver to take care of aliased/nonaliased naming
148 my $colinfo = $self->_resolve_column_info($inner_attrs->{from});
149 my $selected_root_columns;
151 for my $i (0 .. $#{$outer_attrs->{select}}) {
152 my $sel = $outer_attrs->{select}->[$i];
155 $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias
158 if (ref $sel eq 'HASH' ) {
159 $sel->{-as} ||= $attrs->{as}[$i];
160 $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") );
162 elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
163 $selected_root_columns->{$ci->{-colname}} = 1;
166 push @{$inner_attrs->{select}}, $sel;
168 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
171 # We will need to fetch all native columns in the inner subquery, which may
172 # be a part of an *outer* join condition, or an order_by (which needs to be
173 # preserved outside), or wheres. In other words everything but the inner
175 # We can not just fetch everything because a potential has_many restricting
176 # join collapse *will not work* on heavy data types.
177 my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({
182 for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) {
183 my $ci = $colinfo->{$_} or next;
185 $ci->{-source_alias} eq $root_alias
187 ! $selected_root_columns->{$ci->{-colname}}++
189 # adding it to both to keep limits not supporting dark selectors happy
190 push @{$inner_attrs->{select}}, $ci->{-fq_colname};
191 push @{$inner_attrs->{as}}, $ci->{-fq_colname};
195 # construct the inner {from} and lock it in a subquery
196 # we need to prune first, because this will determine if we need a group_by below
197 # throw away all non-selecting, non-restricting multijoins
198 # (since we def. do not care about multiplication of the contents of the subquery)
199 my $inner_subq = do {
201 # must use it here regardless of user requests (vastly gentler on optimizer)
202 local $self->{_use_join_optimizer} = 1;
204 # throw away multijoins since we def. do not care about those inside the subquery
205 ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({
206 %$inner_attrs, _force_prune_multiplying_joins => 1
209 # uh-oh a multiplier (which is not us) left in, this is a problem for limits
210 # we will need to add a group_by to collapse the resultset for proper counts
212 grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} }
214 # if there are user-supplied groups - assume user knows wtf they are up to
215 ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} )
218 my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} };
220 # *possibly* supplement the main selection with pks if not already
221 # there, as they will have to be a part of the group_by to collapse
223 my $inner_select_with_extras;
224 my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns
225 or $self->throw_exception( sprintf
226 'Unable to perform complex limited prefetch off %s without declared primary key',
227 $root_node->{-rsrc}->source_name,
230 push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col
231 unless $cur_sel->{$col}++;
234 ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({
236 $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (),
237 _aliastypes => $inner_aliastypes,
241 # we already optimized $inner_attrs->{from} above
242 # and already local()ized
243 $self->{_use_join_optimizer} = 0;
245 # generate the subquery
246 $self->_select_args_to_query (
247 @{$inner_attrs}{qw(from select where)},
252 # Generate the outer from - this is relatively easy (really just replace
253 # the join slot with the subquery), with a major caveat - we can not
254 # join anything that is non-selecting (not part of the prefetch), but at
255 # the same time is a multi-type relationship, as it will explode the result.
257 # There are two possibilities here
258 # - either the join is non-restricting, in which case we simply throw it away
259 # - it is part of the restrictions, in which case we need to collapse the outer
260 # result by tackling yet another group_by to the outside of the query
262 # work on a shallow copy
263 my @orig_from = @{$attrs->{from}};
266 $outer_attrs->{from} = \ my @outer_from;
268 # we may not be the head
269 if ($root_node_offset) {
270 # first generate the outer_from, up to the substitution point
271 @outer_from = splice @orig_from, 0, $root_node_offset;
273 # substitute the subq at the right spot
276 -alias => $root_alias,
277 -rsrc => $root_node->{-rsrc},
278 $root_alias => $inner_subq,
280 # preserve attrs from what is now the head of the from after the splice
281 @{$orig_from[0]}[1 .. $#{$orig_from[0]}],
286 -alias => $root_alias,
287 -rsrc => $root_node->{-rsrc},
288 $root_alias => $inner_subq,
292 shift @orig_from; # what we just replaced above
294 # scan the *remaining* from spec against different attributes, and see which joins are needed
296 my $outer_aliastypes = $outer_attrs->{_aliastypes} =
297 $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from });
300 my ($outer_select_chain, @outer_nonselecting_chains) = map { +{
301 map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} }
302 } } qw/selecting restricting grouping ordering/;
304 # see what's left - throw away if not selecting/restricting
305 my $may_need_outer_group_by;
306 while (my $j = shift @orig_from) {
307 my $alias = $j->[0]{-alias};
310 $outer_select_chain->{$alias}
314 elsif (first { $_->{$alias} } @outer_nonselecting_chains ) {
315 push @outer_from, $j;
316 $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
320 # also throw in a synthetic group_by if a non-selecting multiplier,
321 # to guard against cross-join explosions
322 # the logic is somewhat fragile, but relies on the idea that if a user supplied
323 # a group by on their own - they know what they were doing
324 if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) {
325 ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({
327 from => \@outer_from,
331 # This is totally horrific - the {where} ends up in both the inner and outer query
332 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
333 # then if where conditions apply to the *right* side of the prefetch, you may have
334 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
335 # the outer select to exclude joins you didn't want in the first place
337 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
342 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
344 # Due to a lack of SQLA2 we fall back to crude scans of all the
345 # select/where/order/group attributes, in order to determine what
346 # aliases are needed to fulfill the query. This information is used
347 # throughout the code to prune unnecessary JOINs from the queries
348 # in an attempt to reduce the execution time.
349 # Although the method is pretty horrific, the worst thing that can
350 # happen is for it to fail due to some scalar SQL, which in turn will
351 # result in a vocal exception.
352 sub _resolve_aliastypes_from_select_args {
353 my ( $self, $attrs ) = @_;
355 $self->throw_exception ('Unable to analyze custom {from}')
356 if ref $attrs->{from} ne 'ARRAY';
358 # what we will return
361 # see what aliases are there to work with
362 # and record who is a multiplier and who is premultiplied
364 for my $node (@{$attrs->{from}}) {
367 $j = $j->[0] if ref $j eq 'ARRAY';
368 my $al = $j->{-alias}
371 $alias_list->{$al} = $j;
373 $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] }
374 # not array == {from} head == can't be multiplying
375 if ref($node) eq 'ARRAY' and ! $j->{-is_single};
377 $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] }
378 # parts of the path that are not us but are multiplying
379 if grep { $aliases_by_type->{multiplying}{$_} }
382 @{ $j->{-join_path}||[] }
385 # get a column to source/alias map (including unambiguous unqualified ones)
386 my $colinfo = $self->_resolve_column_info ($attrs->{from});
388 # set up a botched SQLA
389 my $sql_maker = $self->sql_maker;
391 # these are throw away results, do not pollute the bind stack
392 local $sql_maker->{where_bind};
393 local $sql_maker->{group_bind};
394 local $sql_maker->{having_bind};
395 local $sql_maker->{from_bind};
397 # we can't scan properly without any quoting (\b doesn't cut it
398 # everywhere), so unless there is proper quoting set - use our
399 # own weird impossible character.
400 # Also in the case of no quoting, we need to explicitly disable
401 # name_sep, otherwise sorry nasty legacy syntax like
402 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
403 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
404 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
406 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
407 $sql_maker->{quote_char} = ["\x00", "\xFF"];
408 # if we don't unset it we screw up retarded but unfortunately working
409 # 'MAX(foo.bar)' => { '>', 3 }
410 $sql_maker->{name_sep} = '';
413 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
415 # generate sql chunks
418 $sql_maker->_recurse_where ($attrs->{where}),
419 $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }),
422 $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }),
425 $sql_maker->_recurse_from (
426 ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0],
427 @{$attrs->{from}}[1 .. $#{$attrs->{from}}],
431 map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}},
434 map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker),
438 # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are
439 # bind value specs left in by the sloppy renderer above. It is ok to do this
440 # at this point, since we are going to end up rewriting this crap anyway
441 for my $v (values %$to_scan) {
450 ( @$_ == 0 or @$_ == 2 )
455 require Data::Dumper::Concise;
456 $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) );
465 # kill all selectors which look like a proper subquery
466 # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
467 # fail to run, so we are relatively safe
468 $to_scan->{selecting} = [ grep {
469 $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi
470 } @{ $to_scan->{selecting} || [] } ];
472 # first see if we have any exact matches (qualified or unqualified)
473 for my $type (keys %$to_scan) {
474 for my $piece (@{$to_scan->{$type}}) {
475 if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) {
476 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
477 $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece;
482 # now loop through all fully qualified columns and get the corresponding
483 # alias (should work even if they are in scalarrefs)
484 for my $alias (keys %$alias_list) {
486 $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
488 \b $alias \. ([^\s\)\($rquote]+)?
491 for my $type (keys %$to_scan) {
492 for my $piece (@{$to_scan->{$type}}) {
493 if (my @matches = $piece =~ /$al_re/g) {
494 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
495 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_"
496 for grep { defined $_ } @matches;
502 # now loop through unqualified column names, and try to locate them within
504 for my $col (keys %$colinfo) {
505 next if $col =~ / \. /x; # if column is qualified it was caught by the above
507 my $col_re = qr/ $lquote ($col) $rquote /x;
509 for my $type (keys %$to_scan) {
510 for my $piece (@{$to_scan->{$type}}) {
511 if ( my @matches = $piece =~ /$col_re/g) {
512 my $alias = $colinfo->{$col}{-source_alias};
513 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
514 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
515 for grep { defined $_ } @matches;
521 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
522 for my $j (values %$alias_list) {
523 my $alias = $j->{-alias} or next;
524 $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if (
525 (not $j->{-join_type})
527 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
531 for (keys %$aliases_by_type) {
532 delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}};
535 return $aliases_by_type;
538 # This is the engine behind { distinct => 1 } and the general
539 # complex prefetch grouper
540 sub _group_over_selection {
541 my ($self, $attrs) = @_;
543 my $colinfos = $self->_resolve_column_info ($attrs->{from});
545 my (@group_by, %group_index);
547 # the logic is: if it is a { func => val } we assume an aggregate,
548 # otherwise if \'...' or \[...] we assume the user knows what is
549 # going on thus group over it
550 for (@{$attrs->{select}}) {
551 if (! ref($_) or ref ($_) ne 'HASH' ) {
554 if ($colinfos->{$_} and $_ !~ /\./ ) {
555 # add a fully qualified version as well
556 $group_index{"$colinfos->{$_}{-source_alias}.$_"}++;
561 my @order_by = $self->_extract_order_criteria($attrs->{order_by})
562 or return (\@group_by, $attrs->{order_by});
564 # add any order_by parts that are not already present in the group_by
565 # to maintain SQL cross-compatibility and general sanity
567 # also in case the original selection is *not* unique, or in case part
568 # of the ORDER BY refers to a multiplier - we will need to replace the
569 # skipped order_by elements with their MIN/MAX equivalents as to maintain
570 # the proper overall order without polluting the group criteria (and
571 # possibly changing the outcome entirely)
573 my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes);
575 my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by);
577 for my $o_idx (0 .. $#order_by) {
579 # if the chunk is already a min/max function - there is nothing left to touch
580 next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix;
582 # only consider real columns (for functions the user got to do an explicit group_by)
585 @{$order_by[$o_idx]} != 1
587 # only declare an unknown *plain* identifier as "leftover" if we are called with
588 # aliastypes to examine. If there are none - we are still in _resolve_attrs, and
589 # can just assume the user knows what they want
590 ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} )
592 push @$leftovers, $order_by[$o_idx][0];
595 next unless $chunk_ci;
597 # no duplication of group criteria
598 next if $group_index{$chunk_ci->{-fq_colname}};
601 $attrs->{_aliastypes}
603 $self->_resolve_aliastypes_from_select_args({
604 from => $attrs->{from},
605 order_by => $attrs->{order_by},
607 ) if $group_already_unique;
609 # check that we are not ordering by a multiplier (if a check is requested at all)
611 $group_already_unique
613 ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}}
615 ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}}
617 push @group_by, $chunk_ci->{-fq_colname};
618 $group_index{$chunk_ci->{-fq_colname}}++
621 # We need to order by external columns without adding them to the group
622 # (eiehter a non-unique selection, or a multi-external)
624 # This doesn't really make sense in SQL, however from DBICs point
625 # of view is rather valid (e.g. order the leftmost objects by whatever
626 # criteria and get the offset/rows many). There is a way around
627 # this however in SQL - we simply tae the direction of each piece
628 # of the external order and convert them to MIN(X) for ASC or MAX(X)
629 # for DESC, and group_by the root columns. The end result should be
630 # exactly what we expect
632 # FIXME - this code is a joke, will need to be completely rewritten in
633 # the DQ branch. But I need to push a POC here, otherwise the
634 # pesky tests won't pass
635 # wrap any part of the order_by that "responds" to an ordering alias
637 $sql_maker ||= $self->sql_maker;
639 map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by})
642 my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]);
644 $new_order_by[$o_idx] = \[
645 sprintf( '%s( %s )%s',
646 ($is_desc ? 'MAX' : 'MIN'),
648 ($is_desc ? ' DESC' : ''),
650 @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ]
655 $self->throw_exception ( sprintf
656 'A required group_by clause could not be constructed automatically due to a complex '
657 . 'order_by criteria (%s). Either order_by columns only (no functions) or construct a suitable '
658 . 'group_by by hand',
659 join ', ', map { "'$_'" } @$leftovers,
662 # recreate the untouched order parts
664 $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks );
669 (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged
673 sub _resolve_ident_sources {
674 my ($self, $ident) = @_;
676 my $alias2source = {};
678 # the reason this is so contrived is that $ident may be a {from}
679 # structure, specifying multiple tables to join
680 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
681 # this is compat mode for insert/update/delete which do not deal with aliases
682 $alias2source->{me} = $ident;
684 elsif (ref $ident eq 'ARRAY') {
688 if (ref $_ eq 'HASH') {
691 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
695 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
696 if ($tabinfo->{-rsrc});
700 return $alias2source;
703 # Takes $ident, \@column_names
705 # returns { $column_name => \%column_info, ... }
706 # also note: this adds -result_source => $rsrc to the column info
708 # If no columns_names are supplied returns info about *all* columns
710 sub _resolve_column_info {
711 my ($self, $ident, $colnames) = @_;
713 return {} if $colnames and ! @$colnames;
715 my $alias2src = $self->_resolve_ident_sources($ident);
717 my (%seen_cols, @auto_colnames);
719 # compile a global list of column names, to be able to properly
720 # disambiguate unqualified column names (if at all possible)
721 for my $alias (keys %$alias2src) {
722 my $rsrc = $alias2src->{$alias};
723 for my $colname ($rsrc->columns) {
724 push @{$seen_cols{$colname}}, $alias;
725 push @auto_colnames, "$alias.$colname" unless $colnames;
731 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
734 my (%return, $colinfos);
735 foreach my $col (@$colnames) {
736 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
738 # if the column was seen exactly once - we know which rsrc it came from
739 $source_alias ||= $seen_cols{$colname}[0]
740 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
742 next unless $source_alias;
744 my $rsrc = $alias2src->{$source_alias}
749 ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
751 $self->throw_exception(
752 "No such column '$colname' on source " . $rsrc->source_name
755 -result_source => $rsrc,
756 -source_alias => $source_alias,
757 -fq_colname => $col eq $colname ? "$source_alias.$col" : $col,
758 -colname => $colname,
761 $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname;
767 # The DBIC relationship chaining implementation is pretty simple - every
768 # new related_relationship is pushed onto the {from} stack, and the {select}
769 # window simply slides further in. This means that when we count somewhere
770 # in the middle, we got to make sure that everything in the join chain is an
771 # actual inner join, otherwise the count will come back with unpredictable
772 # results (a resultset may be generated with _some_ rows regardless of if
773 # the relation which the $rs currently selects has rows or not). E.g.
774 # $artist_rs->cds->count - normally generates:
775 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
776 # which actually returns the number of artists * (number of cds || 1)
778 # So what we do here is crawl {from}, determine if the current alias is at
779 # the top of the stack, and if not - make sure the chain is inner-joined down
782 sub _inner_join_to_node {
783 my ($self, $from, $alias) = @_;
785 # subqueries and other oddness are naturally not supported
791 ref $from->[0] ne 'HASH'
795 $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do
798 # find the current $alias in the $from structure
801 for my $j (@{$from}[1 .. $#$from]) {
802 if ($j->[0]{-alias} eq $alias) {
803 $switch_branch = $j->[0]{-join_path};
808 # something else went quite wrong
809 return $from unless $switch_branch;
811 # So it looks like we will have to switch some stuff around.
812 # local() is useless here as we will be leaving the scope
813 # anyway, and deep cloning is just too fucking expensive
814 # So replace the first hashref in the node arrayref manually
815 my @new_from = ($from->[0]);
816 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
818 for my $j (@{$from}[1 .. $#$from]) {
819 my $jalias = $j->[0]{-alias};
821 if ($sw_idx->{$jalias}) {
822 my %attrs = %{$j->[0]};
823 delete $attrs{-join_type};
837 sub _extract_order_criteria {
838 my ($self, $order_by, $sql_maker) = @_;
841 my ($sql_maker, $order_by, $orig_quote_chars) = @_;
843 return scalar $sql_maker->_order_by_chunks ($order_by)
846 my ($lq, $rq, $sep) = map { quotemeta($_) } (
847 ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars),
852 for ($sql_maker->_order_by_chunks ($order_by) ) {
853 my $chunk = ref $_ ? [ @$_ ] : [ $_ ];
854 ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]);
856 # order criteria may have come back pre-quoted (literals and whatnot)
857 # this is fragile, but the best we can currently do
858 $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe
859 or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x;
861 push @chunks, $chunk;
868 return $parser->($sql_maker, $order_by);
871 $sql_maker = $self->sql_maker;
873 # pass these in to deal with literals coming from
874 # the user or the deep guts of prefetch
875 my $orig_quote_chars = [$sql_maker->_quote_chars];
877 local $sql_maker->{quote_char};
878 return $parser->($sql_maker, $order_by, $orig_quote_chars);
882 sub _order_by_is_stable {
883 my ($self, $ident, $order_by, $where) = @_;
886 ( map { $_->[0] } $self->_extract_order_criteria($order_by) ),
887 ( $where ? @{ $self->_extract_fixed_condition_columns($where) || [] } : () ),
890 my $colinfo = $self->_resolve_column_info($ident, \@cols);
892 return keys %$colinfo
893 ? $self->_columns_comprise_identifying_set( $colinfo, \@cols )
898 sub _columns_comprise_identifying_set {
899 my ($self, $colinfo, $columns) = @_;
902 $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_
903 for grep { defined $_ } @{$colinfo}{@$columns};
905 for (values %$cols_per_src) {
906 my $src = (values %$_)[0]->{-result_source};
907 return 1 if $src->_identifying_column_set($_);
913 # this is almost identical to the above, except it accepts only
914 # a single rsrc, and will succeed only if the first portion of the order
916 # returns that portion as a colinfo hashref on success
917 sub _main_source_order_by_portion_is_stable {
918 my ($self, $main_rsrc, $order_by, $where) = @_;
920 die "Huh... I expect a blessed result_source..."
921 if ref($main_rsrc) eq 'ARRAY';
925 ( $self->_extract_order_criteria($order_by) )
927 return unless @ord_cols;
929 my $colinfos = $self->_resolve_column_info($main_rsrc);
931 for (0 .. $#ord_cols) {
933 ! $colinfos->{$ord_cols[$_]}
935 $colinfos->{$ord_cols[$_]}{-result_source} != $main_rsrc
942 # we just truncated it above
943 return unless @ord_cols;
945 my $order_portion_ci = { map {
946 $colinfos->{$_}{-colname} => $colinfos->{$_},
947 $colinfos->{$_}{-fq_colname} => $colinfos->{$_},
950 # since all we check here are the start of the order_by belonging to the
951 # top level $rsrc, a present identifying set will mean that the resultset
952 # is ordered by its leftmost table in a stable manner
954 # RV of _identifying_column_set contains unqualified names only
955 my $unqualified_idset = $main_rsrc->_identifying_column_set({
957 $self->_resolve_column_info(
958 $main_rsrc, $self->_extract_fixed_condition_columns($where)||[]
965 my %unqualified_idcols_from_order = map {
966 $order_portion_ci->{$_} ? ( $_ => $order_portion_ci->{$_} ) : ()
967 } @$unqualified_idset;
969 # extra optimization - cut the order_by at the end of the identifying set
970 # (just in case the user was stupid and overlooked the obvious)
971 for my $i (0 .. $#ord_cols) {
972 my $col = $ord_cols[$i];
973 my $unqualified_colname = $order_portion_ci->{$col}{-colname};
974 $ret_info->{$col} = { %{$order_portion_ci->{$col}}, -idx_in_order_subset => $i };
975 delete $unqualified_idcols_from_order{$ret_info->{$col}{-colname}};
977 # we didn't reach the end of the identifying portion yet
978 return $ret_info unless keys %unqualified_idcols_from_order;
981 die 'How did we get here...';
984 # Attempts to flatten a passed in SQLA condition as much as possible towards
985 # a plain hashref, *without* altering its semantics. Required by
986 # create/populate being able to extract definitive conditions from preexisting
987 # resultset {where} stacks
989 # FIXME - while relatively robust, this is still imperfect, one of the first
990 # things to tackle with DQ
992 my ($self, $where, $where_is_anded_array) = @_;
997 elsif ($where_is_anded_array or ref $where eq 'HASH') {
1001 my @pieces = $where_is_anded_array ? @$where : $where;
1003 my $chunk = shift @pieces;
1005 if (ref $chunk eq 'HASH') {
1006 push @pairs, map { [ $_ => $chunk->{$_} ] } sort keys %$chunk;
1008 elsif (ref $chunk eq 'ARRAY') {
1009 push @pairs, [ -or => $chunk ]
1012 elsif ( ! ref $chunk) {
1013 push @pairs, [ $chunk, shift @pieces ];
1016 push @pairs, [ '', $chunk ];
1020 return unless @pairs;
1022 my @conds = $self->_collapse_cond_unroll_pairs(\@pairs)
1025 # Consolidate various @conds back into something more compact
1028 for my $c (@conds) {
1029 if (ref $c ne 'HASH') {
1030 push @{$fin->{-and}}, $c;
1033 for my $col (sort keys %$c) {
1034 if (exists $fin->{$col}) {
1035 my ($l, $r) = ($fin->{$col}, $c->{$col});
1037 (ref $_ ne 'ARRAY' or !@$_) and $_ = [ -and => $_ ] for ($l, $r);
1039 if (@$l and @$r and $l->[0] eq $r->[0] and $l->[0] eq '-and') {
1040 $fin->{$col} = [ -and => map { @$_[1..$#$_] } ($l, $r) ];
1043 $fin->{$col} = [ -and => $fin->{$col}, $c->{$col} ];
1047 $fin->{$col} = $c->{$col};
1053 if ( ref $fin->{-and} eq 'ARRAY' and @{$fin->{-and}} == 1 ) {
1054 my $piece = (delete $fin->{-and})->[0];
1055 if (ref $piece eq 'ARRAY') {
1056 $fin->{-or} = $fin->{-or} ? [ $piece, $fin->{-or} ] : $piece;
1058 elsif (! exists $fin->{''}) {
1059 $fin->{''} = $piece;
1065 elsif (ref $where eq 'ARRAY') {
1069 (ref $w[0] eq 'ARRAY' and ! @{$w[0]} )
1071 (ref $w[0] eq 'HASH' and ! keys %{$w[0]})
1077 return ( ref $w[0] )
1078 ? $self->_collapse_cond($w[0])
1079 : { $w[0] => undef }
1082 elsif ( ref $w[0] ) {
1086 if ( ( $w[0]||'' ) =~ /^\-and$/i ) {
1087 return (ref $w[1] eq 'HASH' or ref $w[1] eq 'ARRAY')
1088 ? $self->_collapse_cond($w[1], (ref $w[1] eq 'ARRAY') )
1089 : $self->throw_exception("Unsupported top-level op/arg pair: [ $w[0] => $w[1] ]")
1093 return $self->_collapse_cond({ @w });
1098 # not a hash not an array
1099 return { '' => $where };
1102 # catchall, some of the things above fall through
1106 sub _collapse_cond_unroll_pairs {
1107 my ($self, $pairs) = @_;
1112 my ($lhs, $rhs) = @{ shift @$pairs };
1115 push @conds, $self->_collapse_cond($rhs);
1117 elsif ( $lhs =~ /^\-and$/i ) {
1118 push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY'));
1120 elsif ( $lhs =~ /^\-or$/i ) {
1121 push @conds, $self->_collapse_cond(
1122 (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs
1126 if (ref $rhs eq 'HASH' and ! keys %$rhs) {
1127 # FIXME - SQLA seems to be doing... nothing...?
1129 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) {
1130 for my $p ($self->_collapse_cond_unroll_pairs([ [ $lhs => $rhs->{'='} ] ])) {
1132 # extra sanity check
1134 require Data::Dumper::Concise;
1135 local $Data::Dumper::Deepcopy = 1;
1136 $self->throw_exception(
1137 "Internal error: unexpected collapse unroll:"
1138 . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p }
1144 push @conds, ( ! ref $r or overload::Method($r, '""' ) )
1146 : { $l => { '=' => $r } }
1150 elsif (ref $rhs eq 'ARRAY') {
1151 # some of these conditionals encounter multi-values - roll them out using
1152 # an unshift, which will cause extra looping in the while{} above
1154 push @conds, { $lhs => [] };
1156 elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) {
1157 $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ")
1160 if( $rhs->[0] =~ /^\-and$/i ) {
1161 unshift @$pairs, map { [ $lhs => $_ ] } @{$rhs}[1..$#$rhs];
1163 # if not an AND then it's an OR
1165 unshift @$pairs, [ $lhs => $rhs->[1] ];
1168 push @conds, { $lhs => $rhs };
1171 elsif (@$rhs == 1) {
1172 unshift @$pairs, [ $lhs => $rhs->[0] ];
1175 push @conds, { $lhs => $rhs };
1179 push @conds, { $lhs => $rhs };
1188 # returns an arrayref of column names which *definitely* have some
1189 # sort of non-nullable *single* equality requested in the given condition
1190 # specification. This is used to figure out if a resultset is
1191 # constrained to a column which is part of a unique constraint,
1192 # which in turn allows us to better predict how ordering will behave
1195 # this is a rudimentary, incomplete, and error-prone extractor
1196 # however this is OK - it is conservative, and if we can not find
1197 # something that is in fact there - the stack will recover gracefully
1198 sub _extract_fixed_condition_columns {
1200 my $where_hash = $self->_collapse_cond(shift);
1203 for my $c (keys %$where_hash) {
1204 if (defined (my $v = $where_hash->{$c}) ) {
1208 (ref $v eq 'HASH' and keys %$v == 1 and defined $v->{'='} and (
1211 ref $v->{'='} eq 'SCALAR'
1213 ( ref $v->{'='} eq 'REF' and ref ${$v->{'='}} eq 'ARRAY' )
1215 overload::Method($v->{'='}, '""')
1220 elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') {
1221 $res->{$_} = 1 for map { @{ $self->_extract_fixed_condition_columns({ $c => $_ }) } } @{$v}[1..$#$v];
1226 return [ sort keys %$res ];