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 oboslete 90% of this
13 use base 'DBIx::Class::Storage';
16 use List::Util 'first';
17 use Scalar::Util 'blessed';
21 # This code will remove non-selecting/non-restricting joins from
22 # {from} specs, aiding the RDBMS query optimizer
24 sub _prune_unused_joins {
26 my ($from, $select, $where, $attrs) = @_;
28 return $from unless $self->_use_join_optimizer;
30 if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY') {
31 return $from; # only standard {from} specs are supported
34 my $aliastypes = $self->_resolve_aliastypes_from_select_args(@_);
36 # a grouped set will not be affected by amount of rows. Thus any
37 # {multiplying} joins can go
38 delete $aliastypes->{multiplying} if $attrs->{group_by};
40 my @newfrom = $from->[0]; # FROM head is always present
43 for (values %$aliastypes) {
44 # add all requested aliases
45 $need_joins{$_} = 1 for keys %$_;
47 # add all their parents (as per joinpath which is an AoH { table => alias })
48 $need_joins{$_} = 1 for map { values %$_ } map { @$_ } values %$_;
50 for my $j (@{$from}[1..$#$from]) {
51 push @newfrom, $j if (
52 (! $j->[0]{-alias}) # legacy crap
54 $need_joins{$j->[0]{-alias}}
62 # This is the code producing joined subqueries like:
63 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
65 sub _adjust_select_args_for_complex_prefetch {
66 my ($self, $from, $select, $where, $attrs) = @_;
68 $self->throw_exception ('Nothing to prefetch... how did we get here?!')
69 if not @{$attrs->{_prefetch_selector_range}};
71 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
72 if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY');
75 # generate inner/outer attribute lists, remove stuff that doesn't apply
76 my $outer_attrs = { %$attrs };
77 delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/;
79 my $inner_attrs = { %$attrs };
80 delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range _collapse_order_by select as/;
83 # bring over all non-collapse-induced order_by into the inner query (if any)
84 # the outer one will have to keep them all
85 delete $inner_attrs->{order_by};
86 if (my $ord_cnt = @{$outer_attrs->{order_by}} - @{$outer_attrs->{_collapse_order_by}} ) {
87 $inner_attrs->{order_by} = [
88 @{$outer_attrs->{order_by}}[ 0 .. $ord_cnt - 1]
92 # generate the inner/outer select lists
93 # for inside we consider only stuff *not* brought in by the prefetch
94 # on the outside we substitute any function for its alias
95 my $outer_select = [ @$select ];
96 my $inner_select = [];
98 my ($p_start, $p_end) = @{$outer_attrs->{_prefetch_selector_range}};
99 for my $i (0 .. $p_start - 1, $p_end + 1 .. $#$outer_select) {
100 my $sel = $outer_select->[$i];
102 if (ref $sel eq 'HASH' ) {
103 $sel->{-as} ||= $attrs->{as}[$i];
104 $outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") );
107 push @$inner_select, $sel;
109 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
112 # construct the inner $from and lock it in a subquery
113 # we need to prune first, because this will determine if we need a group_by below
114 # the fake group_by is so that the pruner throws away all non-selecting, non-restricting
115 # multijoins (since we def. do not care about those inside the subquery)
117 my $inner_subq = do {
119 # must use it here regardless of user requests
120 local $self->{_use_join_optimizer} = 1;
122 my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, {
123 group_by => ['dummy'], %$inner_attrs,
126 my $inner_aliastypes =
127 $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs );
129 # we need to simulate collapse in the subq if a multiplying join is pulled
130 # by being a non-selecting restrictor
132 ! $inner_attrs->{group_by}
135 $inner_aliastypes->{restricting}{$_}
137 ! $inner_aliastypes->{selecting}{$_}
138 } ( keys %{$inner_aliastypes->{multiplying}||{}} )
140 my $unprocessed_order_chunks;
141 ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
142 $inner_from, $inner_select, $inner_attrs->{order_by}
145 $self->throw_exception (
146 'A required group_by clause could not be constructed automatically due to a complex '
147 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
149 ) if $unprocessed_order_chunks;
152 # we already optimized $inner_from above
153 local $self->{_use_join_optimizer} = 0;
155 # generate the subquery
156 $self->_select_args_to_query (
164 # Generate the outer from - this is relatively easy (really just replace
165 # the join slot with the subquery), with a major caveat - we can not
166 # join anything that is non-selecting (not part of the prefetch), but at
167 # the same time is a multi-type relationship, as it will explode the result.
169 # There are two possibilities here
170 # - either the join is non-restricting, in which case we simply throw it away
171 # - it is part of the restrictions, in which case we need to collapse the outer
172 # result by tackling yet another group_by to the outside of the query
176 # so first generate the outer_from, up to the substitution point
178 while (my $j = shift @$from) {
179 $j = [ $j ] unless ref $j eq 'ARRAY'; # promote the head-from to an AoH
181 if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap
185 -alias => $attrs->{alias},
186 -rsrc => $j->[0]{-rsrc},
187 $attrs->{alias} => $inner_subq,
191 last; # we'll take care of what's left in $from below
194 push @outer_from, $j;
198 # scan the *remaining* from spec against different attributes, and see which joins are needed
200 my $outer_aliastypes =
201 $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs );
204 my ($outer_select_chain, $outer_restrict_chain) = map { +{
205 map { $_ => 1 } map { values %$_} map { @$_ } values %{ $outer_aliastypes->{$_} || {} }
206 } } qw/selecting restricting/;
208 # see what's left - throw away if not selecting/restricting
209 # also throw in a group_by if a non-selecting multiplier,
210 # to guard against cross-join explosions
211 my $need_outer_group_by;
212 while (my $j = shift @$from) {
213 my $alias = $j->[0]{-alias};
216 $outer_select_chain->{$alias}
220 elsif ($outer_restrict_chain->{$alias}) {
221 push @outer_from, $j;
222 $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
226 # demote the outer_from head
227 $outer_from[0] = $outer_from[0][0];
229 if ($need_outer_group_by and ! $outer_attrs->{group_by}) {
231 my $unprocessed_order_chunks;
232 ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
233 \@outer_from, $outer_select, $outer_attrs->{order_by}
236 $self->throw_exception (
237 'A required group_by clause could not be constructed automatically due to a complex '
238 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
240 ) if $unprocessed_order_chunks;
244 # This is totally horrific - the $where ends up in both the inner and outer query
245 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
246 # then if where conditions apply to the *right* side of the prefetch, you may have
247 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
248 # the outer select to exclude joins you didin't want in the first place
250 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
251 return (\@outer_from, $outer_select, $where, $outer_attrs);
255 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
257 # Due to a lack of SQLA2 we fall back to crude scans of all the
258 # select/where/order/group attributes, in order to determine what
259 # aliases are neded to fulfill the query. This information is used
260 # throughout the code to prune unnecessary JOINs from the queries
261 # in an attempt to reduce the execution time.
262 # Although the method is pretty horrific, the worst thing that can
263 # happen is for it to fail due to some scalar SQL, which in turn will
264 # result in a vocal exception.
265 sub _resolve_aliastypes_from_select_args {
266 my ( $self, $from, $select, $where, $attrs ) = @_;
268 $self->throw_exception ('Unable to analyze custom {from}')
269 if ref $from ne 'ARRAY';
271 # what we will return
274 # see what aliases are there to work with
278 $j = $j->[0] if ref $j eq 'ARRAY';
279 my $al = $j->{-alias}
282 $alias_list->{$al} = $j;
283 $aliases_by_type->{multiplying}{$al} ||= $j->{-join_path}||[] if (
284 # not array == {from} head == can't be multiplying
285 ( ref($_) eq 'ARRAY' and ! $j->{-is_single} )
287 # a parent of ours is already a multiplier
288 ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } )
292 # get a column to source/alias map (including unqualified ones)
293 my $colinfo = $self->_resolve_column_info ($from);
295 # set up a botched SQLA
296 my $sql_maker = $self->sql_maker;
298 # these are throw away results, do not pollute the bind stack
299 local $sql_maker->{select_bind};
300 local $sql_maker->{where_bind};
301 local $sql_maker->{group_bind};
302 local $sql_maker->{having_bind};
304 # we can't scan properly without any quoting (\b doesn't cut it
305 # everywhere), so unless there is proper quoting set - use our
306 # own weird impossible character.
307 # Also in the case of no quoting, we need to explicitly disable
308 # name_sep, otherwise sorry nasty legacy syntax like
309 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
310 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
311 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
313 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
314 $sql_maker->{quote_char} = ["\x00", "\xFF"];
315 # if we don't unset it we screw up retarded but unfortunately working
316 # 'MAX(foo.bar)' => { '>', 3 }
317 $sql_maker->{name_sep} = '';
320 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
322 # generate sql chunks
325 $sql_maker->_recurse_where ($where),
326 $sql_maker->_parse_rs_attrs ({
327 map { $_ => $attrs->{$_} } (qw/group_by having/)
331 $sql_maker->_recurse_fields ($select),
332 ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ),
336 # throw away empty chunks
337 $_ = [ map { $_ || () } @$_ ] for values %$to_scan;
339 # first loop through all fully qualified columns and get the corresponding
340 # alias (should work even if they are in scalarrefs)
341 for my $alias (keys %$alias_list) {
343 $lquote $alias $rquote $sep
348 for my $type (keys %$to_scan) {
349 for my $piece (@{$to_scan->{$type}}) {
350 $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[]
351 if ($piece =~ $al_re);
356 # now loop through unqualified column names, and try to locate them within
358 for my $col (keys %$colinfo) {
359 next if $col =~ / \. /x; # if column is qualified it was caught by the above
361 my $col_re = qr/ $lquote $col $rquote /x;
363 for my $type (keys %$to_scan) {
364 for my $piece (@{$to_scan->{$type}}) {
365 if ($piece =~ $col_re) {
366 my $alias = $colinfo->{$col}{-source_alias};
367 $aliases_by_type->{$type}{$alias} ||= $alias_list->{$alias}{-join_path}||[];
373 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
374 for my $j (values %$alias_list) {
375 my $alias = $j->{-alias} or next;
376 $aliases_by_type->{restricting}{$alias} ||= $j->{-join_path}||[] if (
377 (not $j->{-join_type})
379 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
383 return $aliases_by_type;
386 # This is the engine behind { distinct => 1 }
387 sub _group_over_selection {
388 my ($self, $from, $select, $order_by) = @_;
390 my $rs_column_list = $self->_resolve_column_info ($from);
392 my (@group_by, %group_index);
394 # the logic is: if it is a { func => val } we assume an aggregate,
395 # otherwise if \'...' or \[...] we assume the user knows what is
396 # going on thus group over it
398 if (! ref($_) or ref ($_) ne 'HASH' ) {
401 if ($rs_column_list->{$_} and $_ !~ /\./ ) {
402 # add a fully qualified version as well
403 $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++;
408 # add any order_by parts that are not already present in the group_by
409 # we need to be careful not to add any named functions/aggregates
410 # i.e. order_by => [ ... { count => 'foo' } ... ]
412 for ($self->_extract_order_criteria($order_by)) {
413 # only consider real columns (for functions the user got to do an explicit group_by)
419 my $colinfo = $rs_column_list->{$chunk} or do {
424 $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./;
425 push @group_by, $chunk unless $group_index{$chunk}++;
429 ? (\@group_by, (@leftovers ? \@leftovers : undef) )
434 sub _resolve_ident_sources {
435 my ($self, $ident) = @_;
437 my $alias2source = {};
440 # the reason this is so contrived is that $ident may be a {from}
441 # structure, specifying multiple tables to join
442 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
443 # this is compat mode for insert/update/delete which do not deal with aliases
444 $alias2source->{me} = $ident;
447 elsif (ref $ident eq 'ARRAY') {
451 if (ref $_ eq 'HASH') {
453 $rs_alias = $tabinfo->{-alias};
455 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
459 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
460 if ($tabinfo->{-rsrc});
464 return ($alias2source, $rs_alias);
467 # Takes $ident, \@column_names
469 # returns { $column_name => \%column_info, ... }
470 # also note: this adds -result_source => $rsrc to the column info
472 # If no columns_names are supplied returns info about *all* columns
474 sub _resolve_column_info {
475 my ($self, $ident, $colnames) = @_;
476 my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident);
478 my (%seen_cols, @auto_colnames);
480 # compile a global list of column names, to be able to properly
481 # disambiguate unqualified column names (if at all possible)
482 for my $alias (keys %$alias2src) {
483 my $rsrc = $alias2src->{$alias};
484 for my $colname ($rsrc->columns) {
485 push @{$seen_cols{$colname}}, $alias;
486 push @auto_colnames, "$alias.$colname" unless $colnames;
492 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
495 my (%return, $colinfos);
496 foreach my $col (@$colnames) {
497 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
499 # if the column was seen exactly once - we know which rsrc it came from
500 $source_alias ||= $seen_cols{$colname}[0]
501 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
503 next unless $source_alias;
505 my $rsrc = $alias2src->{$source_alias}
510 ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
512 $self->throw_exception(
513 "No such column '$colname' on source " . $rsrc->source_name
516 -result_source => $rsrc,
517 -source_alias => $source_alias,
524 # The DBIC relationship chaining implementation is pretty simple - every
525 # new related_relationship is pushed onto the {from} stack, and the {select}
526 # window simply slides further in. This means that when we count somewhere
527 # in the middle, we got to make sure that everything in the join chain is an
528 # actual inner join, otherwise the count will come back with unpredictable
529 # results (a resultset may be generated with _some_ rows regardless of if
530 # the relation which the $rs currently selects has rows or not). E.g.
531 # $artist_rs->cds->count - normally generates:
532 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
533 # which actually returns the number of artists * (number of cds || 1)
535 # So what we do here is crawl {from}, determine if the current alias is at
536 # the top of the stack, and if not - make sure the chain is inner-joined down
539 sub _inner_join_to_node {
540 my ($self, $from, $alias) = @_;
542 # subqueries and other oddness are naturally not supported
548 ref $from->[0] ne 'HASH'
552 $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do
555 # find the current $alias in the $from structure
558 for my $j (@{$from}[1 .. $#$from]) {
559 if ($j->[0]{-alias} eq $alias) {
560 $switch_branch = $j->[0]{-join_path};
565 # something else went quite wrong
566 return $from unless $switch_branch;
568 # So it looks like we will have to switch some stuff around.
569 # local() is useless here as we will be leaving the scope
570 # anyway, and deep cloning is just too fucking expensive
571 # So replace the first hashref in the node arrayref manually
572 my @new_from = ($from->[0]);
573 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
575 for my $j (@{$from}[1 .. $#$from]) {
576 my $jalias = $j->[0]{-alias};
578 if ($sw_idx->{$jalias}) {
579 my %attrs = %{$j->[0]};
580 delete $attrs{-join_type};
594 sub _extract_order_criteria {
595 my ($self, $order_by, $sql_maker) = @_;
598 my ($sql_maker, $order_by) = @_;
600 return scalar $sql_maker->_order_by_chunks ($order_by)
604 for ($sql_maker->_order_by_chunks ($order_by) ) {
605 my $chunk = ref $_ ? $_ : [ $_ ];
606 $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix;
607 push @chunks, $chunk;
614 return $parser->($sql_maker, $order_by);
617 $sql_maker = $self->sql_maker;
618 local $sql_maker->{quote_char};
619 return $parser->($sql_maker, $order_by);