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 $subq_joinspec = 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 my $subq = $self->_select_args_to_query (
157 -alias => $attrs->{alias},
158 -rsrc => $inner_from->[0]{-rsrc},
159 $attrs->{alias} => $subq,
163 # Generate the outer from - this is relatively easy (really just replace
164 # the join slot with the subquery), with a major caveat - we can not
165 # join anything that is non-selecting (not part of the prefetch), but at
166 # the same time is a multi-type relationship, as it will explode the result.
168 # There are two possibilities here
169 # - either the join is non-restricting, in which case we simply throw it away
170 # - it is part of the restrictions, in which case we need to collapse the outer
171 # result by tackling yet another group_by to the outside of the query
175 # so first generate the outer_from, up to the substitution point
177 while (my $j = shift @$from) {
178 $j = [ $j ] unless ref $j eq 'ARRAY'; # promote the head-from to an AoH
180 if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap
185 last; # we'll take care of what's left in $from below
188 push @outer_from, $j;
192 # scan the *remaining* from spec against different attributes, and see which joins are needed
194 my $outer_aliastypes =
195 $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs );
197 # see what's left - throw away if not selecting/restricting
198 # also throw in a group_by if restricting to guard against
199 # cross-join explosions
201 my $need_outer_group_by;
202 while (my $j = shift @$from) {
203 my $alias = $j->[0]{-alias};
205 if ($outer_aliastypes->{selecting}{$alias}) {
206 push @outer_from, $j;
208 elsif ($outer_aliastypes->{restricting}{$alias}) {
209 push @outer_from, $j;
210 $need_outer_group_by ||= ! $j->[0]{-is_single};
214 # demote the outer_from head
215 $outer_from[0] = $outer_from[0][0];
217 if ($need_outer_group_by and ! $outer_attrs->{group_by}) {
219 my $unprocessed_order_chunks;
220 ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection (
221 \@outer_from, $outer_select, $outer_attrs->{order_by}
224 $self->throw_exception (
225 'A required group_by clause could not be constructed automatically due to a complex '
226 . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable '
228 ) if $unprocessed_order_chunks;
232 # This is totally horrific - the $where ends up in both the inner and outer query
233 # Unfortunately not much can be done until SQLA2 introspection arrives, and even
234 # then if where conditions apply to the *right* side of the prefetch, you may have
235 # to both filter the inner select (e.g. to apply a limit) and then have to re-filter
236 # the outer select to exclude joins you didin't want in the first place
238 # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
239 return (\@outer_from, $outer_select, $where, $outer_attrs);
243 # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE!
245 # Due to a lack of SQLA2 we fall back to crude scans of all the
246 # select/where/order/group attributes, in order to determine what
247 # aliases are neded to fulfill the query. This information is used
248 # throughout the code to prune unnecessary JOINs from the queries
249 # in an attempt to reduce the execution time.
250 # Although the method is pretty horrific, the worst thing that can
251 # happen is for it to fail due to some scalar SQL, which in turn will
252 # result in a vocal exception.
253 sub _resolve_aliastypes_from_select_args {
254 my ( $self, $from, $select, $where, $attrs ) = @_;
256 $self->throw_exception ('Unable to analyze custom {from}')
257 if ref $from ne 'ARRAY';
259 # what we will return
262 # see what aliases are there to work with
266 $j = $j->[0] if ref $j eq 'ARRAY';
267 my $al = $j->{-alias}
270 $alias_list->{$al} = $j;
271 $aliases_by_type->{multiplying}{$al} = 1
272 if ref($_) eq 'ARRAY' and ! $j->{-is_single}; # not array == {from} head == can't be multiplying
275 # get a column to source/alias map (including unqualified ones)
276 my $colinfo = $self->_resolve_column_info ($from);
278 # set up a botched SQLA
279 my $sql_maker = $self->sql_maker;
281 # these are throw away results, do not pollute the bind stack
282 local $sql_maker->{select_bind};
283 local $sql_maker->{where_bind};
284 local $sql_maker->{group_bind};
285 local $sql_maker->{having_bind};
287 # we can't scan properly without any quoting (\b doesn't cut it
288 # everywhere), so unless there is proper quoting set - use our
289 # own weird impossible character.
290 # Also in the case of no quoting, we need to explicitly disable
291 # name_sep, otherwise sorry nasty legacy syntax like
292 # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
293 local $sql_maker->{quote_char} = $sql_maker->{quote_char};
294 local $sql_maker->{name_sep} = $sql_maker->{name_sep};
296 unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
297 $sql_maker->{quote_char} = ["\x00", "\xFF"];
298 # if we don't unset it we screw up retarded but unfortunately working
299 # 'MAX(foo.bar)' => { '>', 3 }
300 $sql_maker->{name_sep} = '';
303 my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);
305 # generate sql chunks
308 $sql_maker->_recurse_where ($where),
309 $sql_maker->_parse_rs_attrs ({
310 map { $_ => $attrs->{$_} } (qw/group_by having/)
314 $sql_maker->_recurse_fields ($select),
315 ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ),
319 # throw away empty chunks
320 $_ = [ map { $_ || () } @$_ ] for values %$to_scan;
322 # first loop through all fully qualified columns and get the corresponding
323 # alias (should work even if they are in scalarrefs)
324 for my $alias (keys %$alias_list) {
326 $lquote $alias $rquote $sep
331 for my $type (keys %$to_scan) {
332 for my $piece (@{$to_scan->{$type}}) {
333 $aliases_by_type->{$type}{$alias} = 1 if ($piece =~ $al_re);
338 # now loop through unqualified column names, and try to locate them within
340 for my $col (keys %$colinfo) {
341 next if $col =~ / \. /x; # if column is qualified it was caught by the above
343 my $col_re = qr/ $lquote $col $rquote /x;
345 for my $type (keys %$to_scan) {
346 for my $piece (@{$to_scan->{$type}}) {
347 $aliases_by_type->{$type}{$colinfo->{$col}{-source_alias}} = 1 if ($piece =~ $col_re);
352 # Add any non-left joins to the restriction list (such joins are indeed restrictions)
353 for my $j (values %$alias_list) {
354 my $alias = $j->{-alias} or next;
355 $aliases_by_type->{restricting}{$alias} = 1 if (
356 (not $j->{-join_type})
358 ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
362 # mark all restricting/selecting join parents as such
363 # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too )
364 for my $type (qw/restricting selecting/) {
365 for my $alias (keys %{$aliases_by_type->{$type}||{}}) {
366 $aliases_by_type->{$type}{$_} = 1
367 for (map { values %$_ } @{ $alias_list->{$alias}{-join_path} || [] });
371 return $aliases_by_type;
374 # This is the engine behind { distinct => 1 }
375 sub _group_over_selection {
376 my ($self, $from, $select, $order_by) = @_;
378 my $rs_column_list = $self->_resolve_column_info ($from);
380 my (@group_by, %group_index);
382 # the logic is: if it is a { func => val } we assume an aggregate,
383 # otherwise if \'...' or \[...] we assume the user knows what is
384 # going on thus group over it
386 if (! ref($_) or ref ($_) ne 'HASH' ) {
389 if ($rs_column_list->{$_} and $_ !~ /\./ ) {
390 # add a fully qualified version as well
391 $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++;
396 # add any order_by parts that are not already present in the group_by
397 # we need to be careful not to add any named functions/aggregates
398 # i.e. order_by => [ ... { count => 'foo' } ... ]
400 for ($self->_extract_order_criteria($order_by)) {
401 # only consider real columns (for functions the user got to do an explicit group_by)
407 my $colinfo = $rs_column_list->{$chunk} or do {
412 $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./;
413 push @group_by, $chunk unless $group_index{$chunk}++;
417 ? (\@group_by, (@leftovers ? \@leftovers : undef) )
422 sub _resolve_ident_sources {
423 my ($self, $ident) = @_;
425 my $alias2source = {};
428 # the reason this is so contrived is that $ident may be a {from}
429 # structure, specifying multiple tables to join
430 if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
431 # this is compat mode for insert/update/delete which do not deal with aliases
432 $alias2source->{me} = $ident;
435 elsif (ref $ident eq 'ARRAY') {
439 if (ref $_ eq 'HASH') {
441 $rs_alias = $tabinfo->{-alias};
443 if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
447 $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
448 if ($tabinfo->{-rsrc});
452 return ($alias2source, $rs_alias);
455 # Takes $ident, \@column_names
457 # returns { $column_name => \%column_info, ... }
458 # also note: this adds -result_source => $rsrc to the column info
460 # If no columns_names are supplied returns info about *all* columns
462 sub _resolve_column_info {
463 my ($self, $ident, $colnames) = @_;
464 my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident);
466 my (%seen_cols, @auto_colnames);
468 # compile a global list of column names, to be able to properly
469 # disambiguate unqualified column names (if at all possible)
470 for my $alias (keys %$alias2src) {
471 my $rsrc = $alias2src->{$alias};
472 for my $colname ($rsrc->columns) {
473 push @{$seen_cols{$colname}}, $alias;
474 push @auto_colnames, "$alias.$colname" unless $colnames;
480 grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
483 my (%return, $colinfos);
484 foreach my $col (@$colnames) {
485 my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;
487 # if the column was seen exactly once - we know which rsrc it came from
488 $source_alias ||= $seen_cols{$colname}[0]
489 if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);
491 next unless $source_alias;
493 my $rsrc = $alias2src->{$source_alias}
497 %{ ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} },
498 -result_source => $rsrc,
499 -source_alias => $source_alias,
506 # The DBIC relationship chaining implementation is pretty simple - every
507 # new related_relationship is pushed onto the {from} stack, and the {select}
508 # window simply slides further in. This means that when we count somewhere
509 # in the middle, we got to make sure that everything in the join chain is an
510 # actual inner join, otherwise the count will come back with unpredictable
511 # results (a resultset may be generated with _some_ rows regardless of if
512 # the relation which the $rs currently selects has rows or not). E.g.
513 # $artist_rs->cds->count - normally generates:
514 # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
515 # which actually returns the number of artists * (number of cds || 1)
517 # So what we do here is crawl {from}, determine if the current alias is at
518 # the top of the stack, and if not - make sure the chain is inner-joined down
521 sub _inner_join_to_node {
522 my ($self, $from, $alias) = @_;
524 # subqueries and other oddness are naturally not supported
530 ref $from->[0] ne 'HASH'
534 $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do
537 # find the current $alias in the $from structure
540 for my $j (@{$from}[1 .. $#$from]) {
541 if ($j->[0]{-alias} eq $alias) {
542 $switch_branch = $j->[0]{-join_path};
547 # something else went quite wrong
548 return $from unless $switch_branch;
550 # So it looks like we will have to switch some stuff around.
551 # local() is useless here as we will be leaving the scope
552 # anyway, and deep cloning is just too fucking expensive
553 # So replace the first hashref in the node arrayref manually
554 my @new_from = ($from->[0]);
555 my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path
557 for my $j (@{$from}[1 .. $#$from]) {
558 my $jalias = $j->[0]{-alias};
560 if ($sw_idx->{$jalias}) {
561 my %attrs = %{$j->[0]};
562 delete $attrs{-join_type};
576 # Most databases do not allow aliasing of tables in UPDATE/DELETE. Thus
577 # a condition containing 'me' or other table prefixes will not work
578 # at all. What this code tries to do (badly) is introspect the condition
579 # and remove all column qualifiers. If it bails out early (returns undef)
580 # the calling code should try another approach (e.g. a subquery)
582 sub _strip_cond_qualifiers_from_array {
583 my ($self, $where) = @_;
585 for (my $i = 0; $i < @$where; $i++) {
586 my $entry = $where->[$i];
588 my $ref = ref $entry;
589 if ($ref eq 'HASH' or $ref eq 'ARRAY') {
590 $hash = $self->_strip_cond_qualifiers($entry);
593 $entry =~ /([^.]+)$/;
594 $hash->{$1} = $where->[++$i];
601 sub _strip_cond_qualifiers {
602 my ($self, $where) = @_;
606 # No-op. No condition, we're updating/deleting everything
607 return $cond unless $where;
609 if (ref $where eq 'ARRAY') {
610 $cond = $self->_strip_cond_qualifiers_from_array($where);
612 elsif (ref $where eq 'HASH') {
613 if ( (keys %$where) == 1 && ( (keys %{$where})[0] eq '-and' )) {
615 $self->_strip_cond_qualifiers_from_array($where->{-and});
618 foreach my $key (keys %$where) {
619 if ($key eq '-or' && ref $where->{$key} eq 'ARRAY') {
620 $cond->{$key} = $self->_strip_cond_qualifiers($where->{$key});
624 $cond->{$1} = $where->{$key};
636 sub _extract_order_criteria {
637 my ($self, $order_by, $sql_maker) = @_;
640 my ($sql_maker, $order_by) = @_;
642 return scalar $sql_maker->_order_by_chunks ($order_by)
646 for ($sql_maker->_order_by_chunks ($order_by) ) {
647 my $chunk = ref $_ ? $_ : [ $_ ];
648 $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix;
649 push @chunks, $chunk;
656 return $parser->($sql_maker, $order_by);
659 $sql_maker = $self->sql_maker;
660 local $sql_maker->{quote_char};
661 return $parser->($sql_maker, $order_by);