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 = $from->[0]; # FROM head is always present
42 my %need_joins = (map { %{$_||{}} } (values %$aliastypes) );
43 for my $j (@{$from}[1..$#$from]) {
44 push @newfrom, $j if (
45 (! $j->[0]{-alias}) # legacy crap
47 $need_joins{$j->[0]{-alias}}
55 # This is the code producing joined subqueries like:
56 # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
58 sub _adjust_select_args_for_complex_prefetch {
59 my ($self, $from, $select, $where, $attrs) = @_;
61 $self->throw_exception ('Nothing to prefetch... how did we get here?!')
62 if not @{$attrs->{_prefetch_selector_range}};
64 $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
65 if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY');
68 # generate inner/outer attribute lists, remove stuff that doesn't apply
69 my $outer_attrs = { %$attrs };
70 delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/;
72 my $inner_attrs = { %$attrs };
73 delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range _collapse_order_by select as/;
76 # bring over all non-collapse-induced order_by into the inner query (if any)
77 # the outer one will have to keep them all
78 delete $inner_attrs->{order_by};
79 if (my $ord_cnt = @{$outer_attrs->{order_by}} - @{$outer_attrs->{_collapse_order_by}} ) {
80 $inner_attrs->{order_by} = [
81 @{$outer_attrs->{order_by}}[ 0 .. $ord_cnt - 1]
85 # generate the inner/outer select lists
86 # for inside we consider only stuff *not* brought in by the prefetch
87 # on the outside we substitute any function for its alias
88 my $outer_select = [ @$select ];
89 my $inner_select = [];
91 my ($p_start, $p_end) = @{$outer_attrs->{_prefetch_selector_range}};
92 for my $i (0 .. $p_start - 1, $p_end + 1 .. $#$outer_select) {
93 my $sel = $outer_select->[$i];
95 if (ref $sel eq 'HASH' ) {
96 $sel->{-as} ||= $attrs->{as}[$i];
97 $outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") );
100 push @$inner_select, $sel;
102 push @{$inner_attrs->{as}}, $attrs->{as}[$i];
105 # construct the inner $from and lock it in a subquery
106 # we need to prune first, because this will determine if we need a group_by below
107 # the fake group_by is so that the pruner throws away all non-selecting, non-restricting
108 # multijoins (since we def. do not care about those inside the subquery)
110 my $inner_subq = do {
112 # must use it here regardless of user requests
113 local $self->{_use_join_optimizer} = 1;
115 my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, {
116 group_by => ['dummy'], %$inner_attrs,
119 my $inner_aliastypes =
120 $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs );
122 # if a multi-type non-selecting (only restricting) join was needed in the subquery
123 # add a group_by to simulate the collapse in the subq
125 ! $inner_attrs->{group_by}
128 $inner_aliastypes->{restricting}{$_}
130 ! $inner_aliastypes->{selecting}{$_}
131 } ( keys %{$inner_aliastypes->{multiplying}||{}} )
133 my $unprocessed_order_chunks;
134 ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
135 $inner_from, $inner_select, $inner_attrs->{order_by}
138 $self->throw_exception (
139 'A required group_by clause could not be constructed automatically due to a complex '
140 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
142 ) if $unprocessed_order_chunks;
145 # we already optimized $inner_from above
146 local $self->{_use_join_optimizer} = 0;
148 # generate the subquery
149 $self->_select_args_to_query (
157 # Generate the outer from - this is relatively easy (really just replace
158 # the join slot with the subquery), with a major caveat - we can not
159 # join anything that is non-selecting (not part of the prefetch), but at
160 # the same time is a multi-type relationship, as it will explode the result.
162 # There are two possibilities here
163 # - either the join is non-restricting, in which case we simply throw it away
164 # - it is part of the restrictions, in which case we need to collapse the outer
165 # result by tackling yet another group_by to the outside of the query
169 # so first generate the outer_from, up to the substitution point
171 while (my $j = shift @$from) {
172 $j = [ $j ] unless ref $j eq 'ARRAY'; # promote the head-from to an AoH
174 if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap
178 -alias => $attrs->{alias},
179 -rsrc => $j->[0]{-rsrc},
180 $attrs->{alias} => $inner_subq,
184 last; # we'll take care of what's left in $from below
187 push @outer_from, $j;
191 # scan the *remaining* from spec against different attributes, and see which joins are needed
193 my $outer_aliastypes =
194 $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs );
196 # see what's left - throw away if not selecting/restricting
197 # also throw in a group_by if restricting to guard against
198 # cross-join explosions
200 my $need_outer_group_by;
201 while (my $j = shift @$from) {
202 my $alias = $j->[0]{-alias};
204 if ($outer_aliastypes->{selecting}{$alias}) {
205 push @outer_from, $j;
207 elsif ($outer_aliastypes->{restricting}{$alias}) {
208 push @outer_from, $j;
209 $need_outer_group_by ||= ! $j->[0]{-is_single};
213 # demote the outer_from head
214 $outer_from[0] = $outer_from[0][0];
216 if ($need_outer_group_by and ! $outer_attrs->{group_by}) {
218 my $unprocessed_order_chunks;
219 ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
220 \@outer_from, $outer_select, $outer_attrs->{order_by}
223 $self->throw_exception (
224 'A required group_by clause could not be constructed automatically due to a complex '
225 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
227 ) if $unprocessed_order_chunks;
231 # This is totally horrific - the $where ends up in both the inner and outer query
232 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
233 # then if where conditions apply to the *right* side of the prefetch, you may have
234 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
235 # the outer select to exclude joins you didin't want in the first place
237 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
238 return (\@outer_from, $outer_select, $where, $outer_attrs);
242 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
244 # Due to a lack of SQLA2 we fall back to crude scans of all the
245 # select/where/order/group attributes, in order to determine what
246 # aliases are neded to fulfill the query. This information is used
247 # throughout the code to prune unnecessary JOINs from the queries
248 # in an attempt to reduce the execution time.
249 # Although the method is pretty horrific, the worst thing that can
250 # happen is for it to fail due to some scalar SQL, which in turn will
251 # result in a vocal exception.
252 sub _resolve_aliastypes_from_select_args {
253 my ( $self, $from, $select, $where, $attrs ) = @_;
255 $self->throw_exception ('Unable to analyze custom {from}')
256 if ref $from ne 'ARRAY';
258 # what we will return
261 # see what aliases are there to work with
265 $j = $j->[0] if ref $j eq 'ARRAY';
266 my $al = $j->{-alias}
269 $alias_list->{$al} = $j;
270 $aliases_by_type->{multiplying}{$al} = 1
271 if ref($_) eq 'ARRAY' and ! $j->{-is_single}; # not array == {from} head == can't be multiplying
274 # get a column to source/alias map (including unqualified ones)
275 my $colinfo = $self->_resolve_column_info ($from);
277 # set up a botched SQLA
278 my $sql_maker = $self->sql_maker;
280 # these are throw away results, do not pollute the bind stack
281 local $sql_maker->{select_bind};
282 local $sql_maker->{where_bind};
283 local $sql_maker->{group_bind};
284 local $sql_maker->{having_bind};
286 # we can't scan properly without any quoting (\b doesn't cut it
287 # everywhere), so unless there is proper quoting set - use our
288 # own weird impossible character.
289 # Also in the case of no quoting, we need to explicitly disable
290 # name_sep, otherwise sorry nasty legacy syntax like
291 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
292 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
293 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
295 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
296 $sql_maker->{quote_char} = ["\x00", "\xFF"];
297 # if we don't unset it we screw up retarded but unfortunately working
298 # 'MAX(foo.bar)' => { '>', 3 }
299 $sql_maker->{name_sep} = '';
302 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
304 # generate sql chunks
307 $sql_maker->_recurse_where ($where),
308 $sql_maker->_parse_rs_attrs ({
309 map { $_ => $attrs->{$_} } (qw/group_by having/)
313 $sql_maker->_recurse_fields ($select),
314 ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ),
318 # throw away empty chunks
319 $_ = [ map { $_ || () } @$_ ] for values %$to_scan;
321 # first loop through all fully qualified columns and get the corresponding
322 # alias (should work even if they are in scalarrefs)
323 for my $alias (keys %$alias_list) {
325 $lquote $alias $rquote $sep
330 for my $type (keys %$to_scan) {
331 for my $piece (@{$to_scan->{$type}}) {
332 $aliases_by_type->{$type}{$alias} = 1 if ($piece =~ $al_re);
337 # now loop through unqualified column names, and try to locate them within
339 for my $col (keys %$colinfo) {
340 next if $col =~ / \. /x; # if column is qualified it was caught by the above
342 my $col_re = qr/ $lquote $col $rquote /x;
344 for my $type (keys %$to_scan) {
345 for my $piece (@{$to_scan->{$type}}) {
346 $aliases_by_type->{$type}{$colinfo->{$col}{-source_alias}} = 1 if ($piece =~ $col_re);
351 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
352 for my $j (values %$alias_list) {
353 my $alias = $j->{-alias} or next;
354 $aliases_by_type->{restricting}{$alias} = 1 if (
355 (not $j->{-join_type})
357 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
361 # mark all restricting/selecting join parents as such
362 # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too )
363 for my $type (qw/restricting selecting/) {
364 for my $alias (keys %{$aliases_by_type->{$type}||{}}) {
365 $aliases_by_type->{$type}{$_} = 1
366 for (map { values %$_ } @{ $alias_list->{$alias}{-join_path} || [] });
370 return $aliases_by_type;
373 # This is the engine behind { distinct => 1 }
374 sub _group_over_selection {
375 my ($self, $from, $select, $order_by) = @_;
377 my $rs_column_list = $self->_resolve_column_info ($from);
379 my (@group_by, %group_index);
381 # the logic is: if it is a { func => val } we assume an aggregate,
382 # otherwise if \'...' or \[...] we assume the user knows what is
383 # going on thus group over it
385 if (! ref($_) or ref ($_) ne 'HASH' ) {
388 if ($rs_column_list->{$_} and $_ !~ /\./ ) {
389 # add a fully qualified version as well
390 $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++;
395 # add any order_by parts that are not already present in the group_by
396 # we need to be careful not to add any named functions/aggregates
397 # i.e. order_by => [ ... { count => 'foo' } ... ]
399 for ($self->_extract_order_criteria($order_by)) {
400 # only consider real columns (for functions the user got to do an explicit group_by)
406 my $colinfo = $rs_column_list->{$chunk} or do {
411 $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./;
412 push @group_by, $chunk unless $group_index{$chunk}++;
416 ? (\@group_by, (@leftovers ? \@leftovers : undef) )
421 sub _resolve_ident_sources {
422 my ($self, $ident) = @_;
424 my $alias2source = {};
427 # the reason this is so contrived is that $ident may be a {from}
428 # structure, specifying multiple tables to join
429 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
430 # this is compat mode for insert/update/delete which do not deal with aliases
431 $alias2source->{me} = $ident;
434 elsif (ref $ident eq 'ARRAY') {
438 if (ref $_ eq 'HASH') {
440 $rs_alias = $tabinfo->{-alias};
442 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
446 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
447 if ($tabinfo->{-rsrc});
451 return ($alias2source, $rs_alias);
454 # Takes $ident, \@column_names
456 # returns { $column_name => \%column_info, ... }
457 # also note: this adds -result_source => $rsrc to the column info
459 # If no columns_names are supplied returns info about *all* columns
461 sub _resolve_column_info {
462 my ($self, $ident, $colnames) = @_;
463 my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident);
465 my (%seen_cols, @auto_colnames);
467 # compile a global list of column names, to be able to properly
468 # disambiguate unqualified column names (if at all possible)
469 for my $alias (keys %$alias2src) {
470 my $rsrc = $alias2src->{$alias};
471 for my $colname ($rsrc->columns) {
472 push @{$seen_cols{$colname}}, $alias;
473 push @auto_colnames, "$alias.$colname" unless $colnames;
479 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
482 my (%return, $colinfos);
483 foreach my $col (@$colnames) {
484 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
486 # if the column was seen exactly once - we know which rsrc it came from
487 $source_alias ||= $seen_cols{$colname}[0]
488 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
490 next unless $source_alias;
492 my $rsrc = $alias2src->{$source_alias}
497 ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
499 $self->throw_exception(
500 "No such column '$colname' on source " . $rsrc->source_name
503 -result_source => $rsrc,
504 -source_alias => $source_alias,
511 # The DBIC relationship chaining implementation is pretty simple - every
512 # new related_relationship is pushed onto the {from} stack, and the {select}
513 # window simply slides further in. This means that when we count somewhere
514 # in the middle, we got to make sure that everything in the join chain is an
515 # actual inner join, otherwise the count will come back with unpredictable
516 # results (a resultset may be generated with _some_ rows regardless of if
517 # the relation which the $rs currently selects has rows or not). E.g.
518 # $artist_rs->cds->count - normally generates:
519 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
520 # which actually returns the number of artists * (number of cds || 1)
522 # So what we do here is crawl {from}, determine if the current alias is at
523 # the top of the stack, and if not - make sure the chain is inner-joined down
526 sub _inner_join_to_node {
527 my ($self, $from, $alias) = @_;
529 # subqueries and other oddness are naturally not supported
535 ref $from->[0] ne 'HASH'
539 $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do
542 # find the current $alias in the $from structure
545 for my $j (@{$from}[1 .. $#$from]) {
546 if ($j->[0]{-alias} eq $alias) {
547 $switch_branch = $j->[0]{-join_path};
552 # something else went quite wrong
553 return $from unless $switch_branch;
555 # So it looks like we will have to switch some stuff around.
556 # local() is useless here as we will be leaving the scope
557 # anyway, and deep cloning is just too fucking expensive
558 # So replace the first hashref in the node arrayref manually
559 my @new_from = ($from->[0]);
560 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
562 for my $j (@{$from}[1 .. $#$from]) {
563 my $jalias = $j->[0]{-alias};
565 if ($sw_idx->{$jalias}) {
566 my %attrs = %{$j->[0]};
567 delete $attrs{-join_type};
581 # Most databases do not allow aliasing of tables in UPDATE/DELETE. Thus
582 # a condition containing 'me' or other table prefixes will not work
583 # at all. What this code tries to do (badly) is introspect the condition
584 # and remove all column qualifiers. If it bails out early (returns undef)
585 # the calling code should try another approach (e.g. a subquery)
587 sub _strip_cond_qualifiers_from_array {
588 my ($self, $where) = @_;
590 for (my $i = 0; $i < @$where; $i++) {
591 my $entry = $where->[$i];
593 my $ref = ref $entry;
594 if ($ref eq 'HASH' or $ref eq 'ARRAY') {
595 $hash = $self->_strip_cond_qualifiers($entry);
598 $entry =~ /([^.]+)$/;
599 $hash->{$1} = $where->[++$i];
606 sub _strip_cond_qualifiers {
607 my ($self, $where) = @_;
611 # No-op. No condition, we're updating/deleting everything
612 return $cond unless $where;
614 if (ref $where eq 'ARRAY') {
615 $cond = $self->_strip_cond_qualifiers_from_array($where);
617 elsif (ref $where eq 'HASH') {
618 if ( (keys %$where) == 1 && ( (keys %{$where})[0] eq '-and' )) {
620 $self->_strip_cond_qualifiers_from_array($where->{-and});
623 foreach my $key (keys %$where) {
624 if ($key eq '-or' && ref $where->{$key} eq 'ARRAY') {
625 $cond->{$key} = $self->_strip_cond_qualifiers($where->{$key});
629 $cond->{$1} = $where->{$key};
641 sub _extract_order_criteria {
642 my ($self, $order_by, $sql_maker) = @_;
645 my ($sql_maker, $order_by) = @_;
647 return scalar $sql_maker->_order_by_chunks ($order_by)
651 for ($sql_maker->_order_by_chunks ($order_by) ) {
652 my $chunk = ref $_ ? $_ : [ $_ ];
653 $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix;
654 push @chunks, $chunk;
661 return $parser->($sql_maker, $order_by);
664 $sql_maker = $self->sql_maker;
665 local $sql_maker->{quote_char};
666 return $parser->($sql_maker, $order_by);