1 package #hide from PAUSE
2 DBIx::Class::Storage::DBIHacks;
5 # This module contains code that should never have seen the light of day,
6 # does not belong in the Storage, or is otherwise unfit for public
7 # display. The arrival of SQLA2 should immediately obsolete 90% of this
13 use base 'DBIx::Class::Storage';
16 use List::Util 'first';
17 use Scalar::Util 'blessed';
18 use DBIx::Class::_Util qw(UNRESOLVABLE_CONDITION serialize);
19 use SQL::Abstract qw(is_plain_value is_literal_value);
20 use DBIx::Class::Carp;
24 # This code will remove non-selecting/non-restricting joins from
25 # {from} specs, aiding the RDBMS query optimizer
27 sub _prune_unused_joins {
28 my ($self, $attrs) = @_;
30 # only standard {from} specs are supported, and we could be disabled in general
31 return ($attrs->{from}, {}) unless (
32 ref $attrs->{from} eq 'ARRAY'
36 ref $attrs->{from}[0] eq 'HASH'
38 ref $attrs->{from}[1] eq 'ARRAY'
40 $self->_use_join_optimizer
43 my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs);
45 my $new_aliastypes = { %$orig_aliastypes };
47 # we will be recreating this entirely
48 my @reclassify = 'joining';
50 # a grouped set will not be affected by amount of rows. Thus any
51 # purely multiplicator classifications can go
52 # (will be reintroduced below if needed by something else)
53 push @reclassify, qw(multiplying premultiplied)
54 if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by};
56 # nuke what will be recalculated
57 delete @{$new_aliastypes}{@reclassify};
59 my @newfrom = $attrs->{from}[0]; # FROM head is always present
61 # recalculate what we need once the multipliers are potentially gone
62 # ignore premultiplies, since they do not add any value to anything
64 for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) {
65 # add all requested aliases
66 $need_joins{$_} = 1 for keys %$_;
68 # add all their parents (as per joinpath which is an AoH { table => alias })
69 $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_;
72 for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) {
73 push @newfrom, $j if (
74 (! defined $j->[0]{-alias}) # legacy crap
76 $need_joins{$j->[0]{-alias}}
80 # we have a new set of joiners - for everything we nuked pull the classification
81 # off the original stack
82 for my $ctype (@reclassify) {
83 $new_aliastypes->{$ctype} = { map
84 { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () }
85 keys %{$orig_aliastypes->{$ctype}}
89 return ( \@newfrom, $new_aliastypes );
93 # This is the code producing joined subqueries like:
94 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
96 sub _adjust_select_args_for_complex_prefetch {
97 my ($self, $attrs) = @_;
99 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless (
100 ref $attrs->{from} eq 'ARRAY'
102 @{$attrs->{from}} > 1
104 ref $attrs->{from}[0] eq 'HASH'
106 ref $attrs->{from}[1] eq 'ARRAY'
109 my $root_alias = $attrs->{alias};
111 # generate inner/outer attribute lists, remove stuff that doesn't apply
112 my $outer_attrs = { %$attrs };
113 delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)};
115 my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 };
116 delete @{$inner_attrs}{qw(for collapse select as)};
118 # there is no point of ordering the insides if there is no limit
119 delete $inner_attrs->{order_by} if (
120 delete $inner_attrs->{_order_is_artificial}
122 ! $inner_attrs->{rows}
125 # generate the inner/outer select lists
126 # for inside we consider only stuff *not* brought in by the prefetch
127 # on the outside we substitute any function for its alias
128 $outer_attrs->{select} = [ @{$attrs->{select}} ];
130 my ($root_node, $root_node_offset);
132 for my $i (0 .. $#{$inner_attrs->{from}}) {
133 my $node = $inner_attrs->{from}[$i];
134 my $h = (ref $node eq 'HASH') ? $node
135 : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
139 if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) {
141 $root_node_offset = $i;
146 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
149 # use the heavy duty resolver to take care of aliased/nonaliased naming
150 my $colinfo = $self->_resolve_column_info($inner_attrs->{from});
151 my $selected_root_columns;
153 for my $i (0 .. $#{$outer_attrs->{select}}) {
154 my $sel = $outer_attrs->{select}->[$i];
157 $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias
160 if (ref $sel eq 'HASH' ) {
161 $sel->{-as} ||= $attrs->{as}[$i];
162 $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") );
164 elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
165 $selected_root_columns->{$ci->{-colname}} = 1;
168 push @{$inner_attrs->{select}}, $sel;
170 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
173 # We will need to fetch all native columns in the inner subquery, which may
174 # be a part of an *outer* join condition, or an order_by (which needs to be
175 # preserved outside), or wheres. In other words everything but the inner
177 # We can not just fetch everything because a potential has_many restricting
178 # join collapse *will not work* on heavy data types.
179 my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({
184 for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) {
185 my $ci = $colinfo->{$_} or next;
187 $ci->{-source_alias} eq $root_alias
189 ! $selected_root_columns->{$ci->{-colname}}++
191 # adding it to both to keep limits not supporting dark selectors happy
192 push @{$inner_attrs->{select}}, $ci->{-fq_colname};
193 push @{$inner_attrs->{as}}, $ci->{-fq_colname};
197 # construct the inner {from} and lock it in a subquery
198 # we need to prune first, because this will determine if we need a group_by below
199 # throw away all non-selecting, non-restricting multijoins
200 # (since we def. do not care about multiplication of the contents of the subquery)
201 my $inner_subq = do {
203 # must use it here regardless of user requests (vastly gentler on optimizer)
204 local $self->{_use_join_optimizer} = 1;
206 # throw away multijoins since we def. do not care about those inside the subquery
207 ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({
208 %$inner_attrs, _force_prune_multiplying_joins => 1
211 # uh-oh a multiplier (which is not us) left in, this is a problem for limits
212 # we will need to add a group_by to collapse the resultset for proper counts
214 grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} }
216 # if there are user-supplied groups - assume user knows wtf they are up to
217 ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} )
220 my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} };
222 # *possibly* supplement the main selection with pks if not already
223 # there, as they will have to be a part of the group_by to collapse
225 my $inner_select_with_extras;
226 my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns
227 or $self->throw_exception( sprintf
228 'Unable to perform complex limited prefetch off %s without declared primary key',
229 $root_node->{-rsrc}->source_name,
232 push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col
233 unless $cur_sel->{$col}++;
236 ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({
238 $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (),
239 _aliastypes => $inner_aliastypes,
243 # we already optimized $inner_attrs->{from} above
244 # and already local()ized
245 $self->{_use_join_optimizer} = 0;
247 # generate the subquery
248 $self->_select_args_to_query (
249 @{$inner_attrs}{qw(from select where)},
254 # Generate the outer from - this is relatively easy (really just replace
255 # the join slot with the subquery), with a major caveat - we can not
256 # join anything that is non-selecting (not part of the prefetch), but at
257 # the same time is a multi-type relationship, as it will explode the result.
259 # There are two possibilities here
260 # - either the join is non-restricting, in which case we simply throw it away
261 # - it is part of the restrictions, in which case we need to collapse the outer
262 # result by tackling yet another group_by to the outside of the query
264 # work on a shallow copy
265 my @orig_from = @{$attrs->{from}};
268 $outer_attrs->{from} = \ my @outer_from;
270 # we may not be the head
271 if ($root_node_offset) {
272 # first generate the outer_from, up to the substitution point
273 @outer_from = splice @orig_from, 0, $root_node_offset;
275 # substitute the subq at the right spot
278 -alias => $root_alias,
279 -rsrc => $root_node->{-rsrc},
280 $root_alias => $inner_subq,
282 # preserve attrs from what is now the head of the from after the splice
283 @{$orig_from[0]}[1 .. $#{$orig_from[0]}],
288 -alias => $root_alias,
289 -rsrc => $root_node->{-rsrc},
290 $root_alias => $inner_subq,
294 shift @orig_from; # what we just replaced above
296 # scan the *remaining* from spec against different attributes, and see which joins are needed
298 my $outer_aliastypes = $outer_attrs->{_aliastypes} =
299 $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from });
302 my ($outer_select_chain, @outer_nonselecting_chains) = map { +{
303 map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} }
304 } } qw/selecting restricting grouping ordering/;
306 # see what's left - throw away if not selecting/restricting
307 my $may_need_outer_group_by;
308 while (my $j = shift @orig_from) {
309 my $alias = $j->[0]{-alias};
312 $outer_select_chain->{$alias}
316 elsif (first { $_->{$alias} } @outer_nonselecting_chains ) {
317 push @outer_from, $j;
318 $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
322 # also throw in a synthetic group_by if a non-selecting multiplier,
323 # to guard against cross-join explosions
324 # the logic is somewhat fragile, but relies on the idea that if a user supplied
325 # a group by on their own - they know what they were doing
326 if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) {
327 ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({
329 from => \@outer_from,
333 # This is totally horrific - the {where} ends up in both the inner and outer query
334 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
335 # then if where conditions apply to the *right* side of the prefetch, you may have
336 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
337 # the outer select to exclude joins you didn't want in the first place
339 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
344 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
346 # Due to a lack of SQLA2 we fall back to crude scans of all the
347 # select/where/order/group attributes, in order to determine what
348 # aliases are needed to fulfill the query. This information is used
349 # throughout the code to prune unnecessary JOINs from the queries
350 # in an attempt to reduce the execution time.
351 # Although the method is pretty horrific, the worst thing that can
352 # happen is for it to fail due to some scalar SQL, which in turn will
353 # result in a vocal exception.
354 sub _resolve_aliastypes_from_select_args {
355 my ( $self, $attrs ) = @_;
357 $self->throw_exception ('Unable to analyze custom {from}')
358 if ref $attrs->{from} ne 'ARRAY';
360 # what we will return
363 # see what aliases are there to work with
364 # and record who is a multiplier and who is premultiplied
366 for my $node (@{$attrs->{from}}) {
369 $j = $j->[0] if ref $j eq 'ARRAY';
370 my $al = $j->{-alias}
373 $alias_list->{$al} = $j;
375 $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] }
376 # not array == {from} head == can't be multiplying
377 if ref($node) eq 'ARRAY' and ! $j->{-is_single};
379 $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] }
380 # parts of the path that are not us but are multiplying
381 if grep { $aliases_by_type->{multiplying}{$_} }
384 @{ $j->{-join_path}||[] }
387 # get a column to source/alias map (including unambiguous unqualified ones)
388 my $colinfo = $self->_resolve_column_info ($attrs->{from});
390 # set up a botched SQLA
391 my $sql_maker = $self->sql_maker;
393 # these are throw away results, do not pollute the bind stack
394 local $sql_maker->{where_bind};
395 local $sql_maker->{group_bind};
396 local $sql_maker->{having_bind};
397 local $sql_maker->{from_bind};
399 # we can't scan properly without any quoting (\b doesn't cut it
400 # everywhere), so unless there is proper quoting set - use our
401 # own weird impossible character.
402 # Also in the case of no quoting, we need to explicitly disable
403 # name_sep, otherwise sorry nasty legacy syntax like
404 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
405 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
406 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
408 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
409 $sql_maker->{quote_char} = ["\x00", "\xFF"];
410 # if we don't unset it we screw up retarded but unfortunately working
411 # 'MAX(foo.bar)' => { '>', 3 }
412 $sql_maker->{name_sep} = '';
415 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
417 # generate sql chunks
420 ($sql_maker->_recurse_where ($attrs->{where}))[0],
421 $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }),
424 $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }),
427 $sql_maker->_recurse_from (
428 ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0],
429 @{$attrs->{from}}[1 .. $#{$attrs->{from}}],
433 # kill all selectors which look like a proper subquery
434 # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
435 # fail to run, so we are relatively safe
437 { $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi }
439 { ($sql_maker->_recurse_fields($_))[0] }
443 map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker),
447 # throw away empty-string chunks, and make sure no binds snuck in
448 # note that we operate over @{$to_scan->{$type}}, hence the
449 # semi-mindbending ... map ... for values ...
453 : (length ref $_) ? (require Data::Dumper::Concise && $self->throw_exception(
454 "Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($_)
458 } @$_ ] ) for values %$to_scan;
460 # throw away empty to-scan's
464 delete $to_scan->{$_}
465 ) for keys %$to_scan;
468 # first see if we have any exact matches (qualified or unqualified)
469 for my $type (keys %$to_scan) {
470 for my $piece (@{$to_scan->{$type}}) {
471 if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) {
472 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
473 $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece;
478 # now loop through all fully qualified columns and get the corresponding
479 # alias (should work even if they are in scalarrefs)
480 for my $alias (keys %$alias_list) {
482 $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
484 \b $alias \. ([^\s\)\($rquote]+)?
487 for my $type (keys %$to_scan) {
488 for my $piece (@{$to_scan->{$type}}) {
489 if (my @matches = $piece =~ /$al_re/g) {
490 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
491 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_"
492 for grep { defined $_ } @matches;
498 # now loop through unqualified column names, and try to locate them within
500 for my $col (keys %$colinfo) {
501 next if $col =~ / \. /x; # if column is qualified it was caught by the above
503 my $col_re = qr/ $lquote ($col) $rquote /x;
505 for my $type (keys %$to_scan) {
506 for my $piece (@{$to_scan->{$type}}) {
507 if ( my @matches = $piece =~ /$col_re/g) {
508 my $alias = $colinfo->{$col}{-source_alias};
509 $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
510 $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
511 for grep { defined $_ } @matches;
517 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
518 for my $j (values %$alias_list) {
519 my $alias = $j->{-alias} or next;
520 $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if (
521 (not $j->{-join_type})
523 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
527 for (keys %$aliases_by_type) {
528 delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}};
531 return $aliases_by_type;
534 # This is the engine behind { distinct => 1 } and the general
535 # complex prefetch grouper
536 sub _group_over_selection {
537 my ($self, $attrs) = @_;
539 my $colinfos = $self->_resolve_column_info ($attrs->{from});
541 my (@group_by, %group_index);
543 # the logic is: if it is a { func => val } we assume an aggregate,
544 # otherwise if \'...' or \[...] we assume the user knows what is
545 # going on thus group over it
546 for (@{$attrs->{select}}) {
547 if (! ref($_) or ref ($_) ne 'HASH' ) {
550 if ($colinfos->{$_} and $_ !~ /\./ ) {
551 # add a fully qualified version as well
552 $group_index{"$colinfos->{$_}{-source_alias}.$_"}++;
557 my @order_by = $self->_extract_order_criteria($attrs->{order_by})
558 or return (\@group_by, $attrs->{order_by});
560 # add any order_by parts that are not already present in the group_by
561 # to maintain SQL cross-compatibility and general sanity
563 # also in case the original selection is *not* unique, or in case part
564 # of the ORDER BY refers to a multiplier - we will need to replace the
565 # skipped order_by elements with their MIN/MAX equivalents as to maintain
566 # the proper overall order without polluting the group criteria (and
567 # possibly changing the outcome entirely)
569 my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes);
571 my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by);
573 for my $o_idx (0 .. $#order_by) {
575 # if the chunk is already a min/max function - there is nothing left to touch
576 next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix;
578 # only consider real columns (for functions the user got to do an explicit group_by)
581 @{$order_by[$o_idx]} != 1
583 # only declare an unknown *plain* identifier as "leftover" if we are called with
584 # aliastypes to examine. If there are none - we are still in _resolve_attrs, and
585 # can just assume the user knows what they want
586 ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} )
588 push @$leftovers, $order_by[$o_idx][0];
591 next unless $chunk_ci;
593 # no duplication of group criteria
594 next if $group_index{$chunk_ci->{-fq_colname}};
597 $attrs->{_aliastypes}
599 $self->_resolve_aliastypes_from_select_args({
600 from => $attrs->{from},
601 order_by => $attrs->{order_by},
603 ) if $group_already_unique;
605 # check that we are not ordering by a multiplier (if a check is requested at all)
607 $group_already_unique
609 ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}}
611 ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}}
613 push @group_by, $chunk_ci->{-fq_colname};
614 $group_index{$chunk_ci->{-fq_colname}}++
617 # We need to order by external columns without adding them to the group
618 # (eiehter a non-unique selection, or a multi-external)
620 # This doesn't really make sense in SQL, however from DBICs point
621 # of view is rather valid (e.g. order the leftmost objects by whatever
622 # criteria and get the offset/rows many). There is a way around
623 # this however in SQL - we simply tae the direction of each piece
624 # of the external order and convert them to MIN(X) for ASC or MAX(X)
625 # for DESC, and group_by the root columns. The end result should be
626 # exactly what we expect
628 # FIXME - this code is a joke, will need to be completely rewritten in
629 # the DQ branch. But I need to push a POC here, otherwise the
630 # pesky tests won't pass
631 # wrap any part of the order_by that "responds" to an ordering alias
633 $sql_maker ||= $self->sql_maker;
635 map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by})
638 my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]);
640 $new_order_by[$o_idx] = \[
641 sprintf( '%s( %s )%s',
642 ($is_desc ? 'MAX' : 'MIN'),
644 ($is_desc ? ' DESC' : ''),
646 @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ]
651 $self->throw_exception ( sprintf
652 'Unable to programatically derive a required group_by from the supplied '
653 . 'order_by criteria. To proceed either add an explicit group_by, or '
654 . 'simplify your order_by to only include plain columns '
655 . '(supplied order_by: %s)',
656 join ', ', map { "'$_'" } @$leftovers,
659 # recreate the untouched order parts
661 $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks );
666 (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged
670 sub _resolve_ident_sources {
671 my ($self, $ident) = @_;
673 my $alias2source = {};
675 # the reason this is so contrived is that $ident may be a {from}
676 # structure, specifying multiple tables to join
677 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
678 # this is compat mode for insert/update/delete which do not deal with aliases
679 $alias2source->{me} = $ident;
681 elsif (ref $ident eq 'ARRAY') {
685 if (ref $_ eq 'HASH') {
688 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
692 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
693 if ($tabinfo->{-rsrc});
697 return $alias2source;
700 # Takes $ident, \@column_names
702 # returns { $column_name => \%column_info, ... }
703 # also note: this adds -result_source => $rsrc to the column info
705 # If no columns_names are supplied returns info about *all* columns
707 sub _resolve_column_info {
708 my ($self, $ident, $colnames) = @_;
710 return {} if $colnames and ! @$colnames;
712 my $alias2src = $self->_resolve_ident_sources($ident);
714 my (%seen_cols, @auto_colnames);
716 # compile a global list of column names, to be able to properly
717 # disambiguate unqualified column names (if at all possible)
718 for my $alias (keys %$alias2src) {
719 my $rsrc = $alias2src->{$alias};
720 for my $colname ($rsrc->columns) {
721 push @{$seen_cols{$colname}}, $alias;
722 push @auto_colnames, "$alias.$colname" unless $colnames;
728 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
731 my (%return, $colinfos);
732 foreach my $col (@$colnames) {
733 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
735 # if the column was seen exactly once - we know which rsrc it came from
736 $source_alias ||= $seen_cols{$colname}[0]
737 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
739 next unless $source_alias;
741 my $rsrc = $alias2src->{$source_alias}
746 ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
748 $self->throw_exception(
749 "No such column '$colname' on source " . $rsrc->source_name
752 -result_source => $rsrc,
753 -source_alias => $source_alias,
754 -fq_colname => $col eq $colname ? "$source_alias.$col" : $col,
755 -colname => $colname,
758 $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname;
764 # The DBIC relationship chaining implementation is pretty simple - every
765 # new related_relationship is pushed onto the {from} stack, and the {select}
766 # window simply slides further in. This means that when we count somewhere
767 # in the middle, we got to make sure that everything in the join chain is an
768 # actual inner join, otherwise the count will come back with unpredictable
769 # results (a resultset may be generated with _some_ rows regardless of if
770 # the relation which the $rs currently selects has rows or not). E.g.
771 # $artist_rs->cds->count - normally generates:
772 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
773 # which actually returns the number of artists * (number of cds || 1)
775 # So what we do here is crawl {from}, determine if the current alias is at
776 # the top of the stack, and if not - make sure the chain is inner-joined down
779 sub _inner_join_to_node {
780 my ($self, $from, $alias) = @_;
782 my $switch_branch = $self->_find_join_path_to_node($from, $alias);
784 return $from unless @{$switch_branch||[]};
786 # So it looks like we will have to switch some stuff around.
787 # local() is useless here as we will be leaving the scope
788 # anyway, and deep cloning is just too fucking expensive
789 # So replace the first hashref in the node arrayref manually
790 my @new_from = ($from->[0]);
791 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
793 for my $j (@{$from}[1 .. $#$from]) {
794 my $jalias = $j->[0]{-alias};
796 if ($sw_idx->{$jalias}) {
797 my %attrs = %{$j->[0]};
798 delete $attrs{-join_type};
812 sub _find_join_path_to_node {
813 my ($self, $from, $target_alias) = @_;
815 # subqueries and other oddness are naturally not supported
819 ref $from->[0] ne 'HASH'
821 ! defined $from->[0]{-alias}
824 # no path - the head is the alias
825 return [] if $from->[0]{-alias} eq $target_alias;
827 for my $i (1 .. $#$from) {
828 return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias );
831 # something else went quite wrong
835 sub _extract_order_criteria {
836 my ($self, $order_by, $sql_maker) = @_;
839 my ($sql_maker, $order_by, $orig_quote_chars) = @_;
841 return scalar $sql_maker->_order_by_chunks ($order_by)
844 my ($lq, $rq, $sep) = map { quotemeta($_) } (
845 ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars),
850 for ($sql_maker->_order_by_chunks ($order_by) ) {
851 my $chunk = ref $_ ? [ @$_ ] : [ $_ ];
852 ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]);
854 # order criteria may have come back pre-quoted (literals and whatnot)
855 # this is fragile, but the best we can currently do
856 $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe
857 or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x;
859 push @chunks, $chunk;
866 return $parser->($sql_maker, $order_by);
869 $sql_maker = $self->sql_maker;
871 # pass these in to deal with literals coming from
872 # the user or the deep guts of prefetch
873 my $orig_quote_chars = [$sql_maker->_quote_chars];
875 local $sql_maker->{quote_char};
876 return $parser->($sql_maker, $order_by, $orig_quote_chars);
880 sub _order_by_is_stable {
881 my ($self, $ident, $order_by, $where) = @_;
884 ( map { $_->[0] } $self->_extract_order_criteria($order_by) ),
885 ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ),
888 my $colinfo = $self->_resolve_column_info($ident, \@cols);
890 return keys %$colinfo
891 ? $self->_columns_comprise_identifying_set( $colinfo, \@cols )
896 sub _columns_comprise_identifying_set {
897 my ($self, $colinfo, $columns) = @_;
900 $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_
901 for grep { defined $_ } @{$colinfo}{@$columns};
903 for (values %$cols_per_src) {
904 my $src = (values %$_)[0]->{-result_source};
905 return 1 if $src->_identifying_column_set($_);
911 # this is almost similar to _order_by_is_stable, except it takes
912 # a single rsrc, and will succeed only if the first portion of the order
914 # returns that portion as a colinfo hashref on success
915 sub _extract_colinfo_of_stable_main_source_order_by_portion {
916 my ($self, $attrs) = @_;
918 my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias});
920 return unless defined $nodes;
924 ( $self->_extract_order_criteria($attrs->{order_by}) )
926 return unless @ord_cols;
928 my $valid_aliases = { map { $_ => 1 } (
929 $attrs->{from}[0]{-alias},
930 map { values %$_ } @$nodes,
933 my $colinfos = $self->_resolve_column_info($attrs->{from});
935 my ($colinfos_to_return, $seen_main_src_cols);
937 for my $col (@ord_cols) {
938 # if order criteria is unresolvable - there is nothing we can do
939 my $colinfo = $colinfos->{$col} or last;
941 # if we reached the end of the allowed aliases - also nothing we can do
942 last unless $valid_aliases->{$colinfo->{-source_alias}};
944 $colinfos_to_return->{$col} = $colinfo;
946 $seen_main_src_cols->{$colinfo->{-colname}} = 1
947 if $colinfo->{-source_alias} eq $attrs->{alias};
950 # FIXME the condition may be singling out things on its own, so we
951 # conceivable could come back wi "stable-ordered by nothing"
952 # not confient enough in the parser yet, so punt for the time being
953 return unless $seen_main_src_cols;
955 my $main_src_fixed_cols_from_cond = [ $attrs->{where}
959 ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} )
960 ? $colinfos->{$_}{-colname}
963 keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) }
968 return $attrs->{result_source}->_identifying_column_set([
969 keys %$seen_main_src_cols,
970 @$main_src_fixed_cols_from_cond,
971 ]) ? $colinfos_to_return : ();
974 # Attempts to flatten a passed in SQLA condition as much as possible towards
975 # a plain hashref, *without* altering its semantics. Required by
976 # create/populate being able to extract definitive conditions from preexisting
977 # resultset {where} stacks
979 # FIXME - while relatively robust, this is still imperfect, one of the first
980 # things to tackle with DQ
982 my ($self, $where, $where_is_anded_array) = @_;
989 elsif ($where_is_anded_array or ref $where eq 'HASH') {
993 my @pieces = $where_is_anded_array ? @$where : $where;
995 my $chunk = shift @pieces;
997 if (ref $chunk eq 'HASH') {
998 for (sort keys %$chunk) {
1000 # Match SQLA 1.79 behavior
1002 is_literal_value($chunk->{$_})
1003 ? carp 'Hash-pairs consisting of an empty string with a literal are deprecated, use -and => [ $literal ] instead'
1004 : $self->throw_exception("Supplying an empty left hand side argument is not supported in hash-pairs")
1008 push @pairs, $_ => $chunk->{$_};
1011 elsif (ref $chunk eq 'ARRAY') {
1012 push @pairs, -or => $chunk
1015 elsif ( ! length ref $chunk) {
1017 # Match SQLA 1.79 behavior
1018 $self->throw_exception("Supplying an empty left hand side argument is not supported in array-pairs")
1019 if $where_is_anded_array and (! defined $chunk or $chunk eq '');
1021 push @pairs, $chunk, shift @pieces;
1024 push @pairs, '', $chunk;
1028 return unless @pairs;
1030 my @conds = $self->_collapse_cond_unroll_pairs(\@pairs)
1033 # Consolidate various @conds back into something more compact
1034 for my $c (@conds) {
1035 if (ref $c ne 'HASH') {
1036 push @{$fin->{-and}}, $c;
1039 for my $col (sort keys %$c) {
1041 # consolidate all -and nodes
1042 if ($col =~ /^\-and$/i) {
1043 push @{$fin->{-and}},
1044 ref $c->{$col} eq 'ARRAY' ? @{$c->{$col}}
1045 : ref $c->{$col} eq 'HASH' ? %{$c->{$col}}
1046 : { $col => $c->{$col} }
1049 elsif ($col =~ /^\-/) {
1050 push @{$fin->{-and}}, { $col => $c->{$col} };
1052 elsif (exists $fin->{$col}) {
1053 $fin->{$col} = [ -and => map {
1054 (ref $_ eq 'ARRAY' and ($_->[0]||'') =~ /^\-and$/i )
1058 } ($fin->{$col}, $c->{$col}) ];
1061 $fin->{$col} = $c->{$col};
1067 elsif (ref $where eq 'ARRAY') {
1068 # we are always at top-level here, it is safe to dump empty *standalone* pieces
1071 for (my $i = 0; $i <= $#$where; $i++ ) {
1073 # Match SQLA 1.79 behavior
1074 $self->throw_exception(
1075 "Supplying an empty left hand side argument is not supported in array-pairs"
1076 ) if (! defined $where->[$i] or ! length $where->[$i]);
1078 my $logic_mod = lc ( ($where->[$i] =~ /^(\-(?:and|or))$/i)[0] || '' );
1082 $self->throw_exception("Unsupported top-level op/arg pair: [ $logic_mod => $where->[$i] ]")
1083 unless ref $where->[$i] eq 'HASH' or ref $where->[$i] eq 'ARRAY';
1085 my $sub_elt = $self->_collapse_cond({ $logic_mod => $where->[$i] })
1088 my @keys = keys %$sub_elt;
1089 if ( @keys == 1 and $keys[0] !~ /^\-/ ) {
1090 $fin_idx->{ "COL_$keys[0]_" . serialize $sub_elt } = $sub_elt;
1093 $fin_idx->{ "SER_" . serialize $sub_elt } = $sub_elt;
1096 elsif (! length ref $where->[$i] ) {
1097 my $sub_elt = $self->_collapse_cond({ @{$where}[$i, $i+1] })
1100 $fin_idx->{ "COL_$where->[$i]_" . serialize $sub_elt } = $sub_elt;
1104 $fin_idx->{ "SER_" . serialize $where->[$i] } = $self->_collapse_cond( $where->[$i] ) || next;
1111 elsif ( keys %$fin_idx == 1 ) {
1112 $fin = (values %$fin_idx)[0];
1117 # at this point everything is at most one level deep - unroll if needed
1118 for (sort keys %$fin_idx) {
1119 if ( ref $fin_idx->{$_} eq 'HASH' and keys %{$fin_idx->{$_}} == 1 ) {
1120 my ($l, $r) = %{$fin_idx->{$_}};
1126 ( @$r == 1 and $l =~ /^\-and$/i )
1139 $l =~ /^\-(?:and|or)$/i
1149 push @or, $fin_idx->{$_};
1157 # not a hash not an array
1158 $fin = { -and => [ $where ] };
1161 # unroll single-element -and's
1167 my $and = delete $fin->{-and};
1170 # at this point we have @$and == 1
1172 ref $and->[0] eq 'HASH'
1174 ! grep { exists $fin->{$_} } keys %{$and->[0]}
1181 $fin->{-and} = $and;
1186 # compress same-column conds found in $fin
1187 for my $col ( grep { $_ !~ /^\-/ } keys %$fin ) {
1188 next unless ref $fin->{$col} eq 'ARRAY' and ($fin->{$col}[0]||'') =~ /^\-and$/i;
1189 my $val_bag = { map {
1190 (! defined $_ ) ? ( UNDEF => undef )
1191 : ( ! length ref $_ or is_plain_value $_ ) ? ( "VAL_$_" => $_ )
1192 : ( ( 'SER_' . serialize $_ ) => $_ )
1193 } @{$fin->{$col}}[1 .. $#{$fin->{$col}}] };
1195 if (keys %$val_bag == 1 ) {
1196 ($fin->{$col}) = values %$val_bag;
1199 $fin->{$col} = [ -and => map { $val_bag->{$_} } sort keys %$val_bag ];
1203 return keys %$fin ? $fin : ();
1206 sub _collapse_cond_unroll_pairs {
1207 my ($self, $pairs) = @_;
1212 my ($lhs, $rhs) = splice @$pairs, 0, 2;
1215 push @conds, $self->_collapse_cond($rhs);
1217 elsif ( $lhs =~ /^\-and$/i ) {
1218 push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY'));
1220 elsif ( $lhs =~ /^\-or$/i ) {
1221 push @conds, $self->_collapse_cond(
1222 (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs
1226 if (ref $rhs eq 'HASH' and ! keys %$rhs) {
1227 # FIXME - SQLA seems to be doing... nothing...?
1229 # normalize top level -ident, for saner extract_fixed_condition_columns code
1230 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) {
1231 push @conds, { $lhs => { '=', $rhs } };
1233 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-value} and is_plain_value $rhs->{-value}) {
1234 push @conds, { $lhs => $rhs->{-value} };
1236 elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) {
1237 if ( length ref $rhs->{'='} and is_literal_value $rhs->{'='} ) {
1238 push @conds, { $lhs => $rhs };
1241 for my $p ($self->_collapse_cond_unroll_pairs([ $lhs => $rhs->{'='} ])) {
1243 # extra sanity check
1245 require Data::Dumper::Concise;
1246 local $Data::Dumper::Deepcopy = 1;
1247 $self->throw_exception(
1248 "Internal error: unexpected collapse unroll:"
1249 . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p }
1258 # the unroller recursion may return a '=' prepended value already
1259 ref $r eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}
1264 : { $l => { '=' => $r } }
1269 elsif (ref $rhs eq 'ARRAY') {
1270 # some of these conditionals encounter multi-values - roll them out using
1271 # an unshift, which will cause extra looping in the while{} above
1273 push @conds, { $lhs => [] };
1275 elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) {
1276 $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ")
1279 if( $rhs->[0] =~ /^\-and$/i ) {
1280 unshift @$pairs, map { $lhs => $_ } @{$rhs}[1..$#$rhs];
1282 # if not an AND then it's an OR
1284 unshift @$pairs, $lhs => $rhs->[1];
1287 push @conds, { $lhs => [ @{$rhs}[1..$#$rhs] ] };
1290 elsif (@$rhs == 1) {
1291 unshift @$pairs, $lhs => $rhs->[0];
1294 push @conds, { $lhs => $rhs };
1297 # unroll func + { -value => ... }
1301 ( my ($subop) = keys %$rhs ) == 1
1303 length ref ((values %$rhs)[0])
1305 my $vref = is_plain_value( (values %$rhs)[0] )
1307 push @conds, { $lhs => { $subop => $$vref } }
1310 push @conds, { $lhs => $rhs };
1318 # Analyzes a given condition and attempts to extract all columns
1319 # with a definitive fixed-condition criteria. Returns a hashref
1320 # of k/v pairs suitable to be passed to set_columns(), with a
1321 # MAJOR CAVEAT - multi-value (contradictory) equalities are still
1322 # represented as a reference to the UNRESOVABLE_CONDITION constant
1323 # The reason we do this is that some codepaths only care about the
1324 # codition being stable, as opposed to actually making sense
1326 # The normal mode is used to figure out if a resultset is constrained
1327 # to a column which is part of a unique constraint, which in turn
1328 # allows us to better predict how ordering will behave etc.
1330 # With the optional "consider_nulls" boolean argument, the function
1331 # is instead used to infer inambiguous values from conditions
1332 # (e.g. the inheritance of resultset conditions on new_result)
1334 sub _extract_fixed_condition_columns {
1335 my ($self, $where, $consider_nulls) = @_;
1336 my $where_hash = $self->_collapse_cond($_[1]);
1340 for $c (keys %$where_hash) {
1343 if (!defined ($v = $where_hash->{$c}) ) {
1344 $vals->{UNDEF} = $v if $consider_nulls
1351 if (exists $v->{-value}) {
1352 if (defined $v->{-value}) {
1353 $vals->{"VAL_$v->{-value}"} = $v->{-value}
1355 elsif( $consider_nulls ) {
1356 $vals->{UNDEF} = $v->{-value};
1359 # do not need to check for plain values - _collapse_cond did it for us
1361 length ref $v->{'='}
1364 ( ref $v->{'='} eq 'HASH' and keys %{$v->{'='}} == 1 and exists $v->{'='}{-ident} )
1366 is_literal_value($v->{'='})
1369 $vals->{ 'SER_' . serialize $v->{'='} } = $v->{'='};
1377 $vals->{"VAL_$v"} = $v;
1379 elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') {
1380 for ( @{$v}[1..$#$v] ) {
1381 my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls'); # always fish nulls out on recursion
1382 next unless exists $subval->{$c}; # didn't find anything
1384 ! defined $subval->{$c} ? 'UNDEF'
1385 : ( ! length ref $subval->{$c} or is_plain_value $subval->{$c} ) ? "VAL_$subval->{$c}"
1386 : ( 'SER_' . serialize $subval->{$c} )
1391 if (keys %$vals == 1) {
1392 ($res->{$c}) = (values %$vals)
1393 unless !$consider_nulls and exists $vals->{UNDEF};
1395 elsif (keys %$vals > 1) {
1396 $res->{$c} = UNRESOLVABLE_CONDITION;