Commit | Line | Data |
c443438f |
1 | package #hide from PAUSE |
2 | DBIx::Class::Storage::DBIHacks; |
d28bb90d |
3 | |
4 | # |
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 |
6a6394f1 |
7 | # display. The arrival of SQLA2 should immediately obsolete 90% of this |
d28bb90d |
8 | # |
9 | |
10 | use strict; |
11 | use warnings; |
12 | |
13 | use base 'DBIx::Class::Storage'; |
14 | use mro 'c3'; |
15 | |
6298a324 |
16 | use List::Util 'first'; |
17 | use Scalar::Util 'blessed'; |
ea5c7509 |
18 | use Sub::Name 'subname'; |
6298a324 |
19 | use namespace::clean; |
d28bb90d |
20 | |
21 | # |
052e8431 |
22 | # This code will remove non-selecting/non-restricting joins from |
4b1b5ea3 |
23 | # {from} specs, aiding the RDBMS query optimizer |
052e8431 |
24 | # |
25 | sub _prune_unused_joins { |
ea95892e |
26 | my $self = shift; |
97e130fa |
27 | my ($from, $select, $where, $attrs, $ignore_multiplication) = @_; |
052e8431 |
28 | |
ea95892e |
29 | return $from unless $self->_use_join_optimizer; |
30 | |
052e8431 |
31 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY') { |
32 | return $from; # only standard {from} specs are supported |
33 | } |
34 | |
4b1b5ea3 |
35 | my $aliastypes = $self->_resolve_aliastypes_from_select_args(@_); |
36 | |
97e130fa |
37 | # don't care |
38 | delete $aliastypes->{joining}; |
39 | |
4b1b5ea3 |
40 | # a grouped set will not be affected by amount of rows. Thus any |
41 | # {multiplying} joins can go |
97e130fa |
42 | delete $aliastypes->{multiplying} |
43 | if $ignore_multiplication or $attrs->{group_by}; |
4b1b5ea3 |
44 | |
052e8431 |
45 | my @newfrom = $from->[0]; # FROM head is always present |
46 | |
a4812caa |
47 | my %need_joins; |
97e130fa |
48 | |
a4812caa |
49 | for (values %$aliastypes) { |
50 | # add all requested aliases |
51 | $need_joins{$_} = 1 for keys %$_; |
52 | |
53 | # add all their parents (as per joinpath which is an AoH { table => alias }) |
97e130fa |
54 | $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; |
a4812caa |
55 | } |
97e130fa |
56 | |
052e8431 |
57 | for my $j (@{$from}[1..$#$from]) { |
539ffe87 |
58 | push @newfrom, $j if ( |
4b1b5ea3 |
59 | (! $j->[0]{-alias}) # legacy crap |
539ffe87 |
60 | || |
61 | $need_joins{$j->[0]{-alias}} |
62 | ); |
052e8431 |
63 | } |
64 | |
65 | return \@newfrom; |
66 | } |
67 | |
052e8431 |
68 | # |
d28bb90d |
69 | # This is the code producing joined subqueries like: |
8273e845 |
70 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
d28bb90d |
71 | # |
72 | sub _adjust_select_args_for_complex_prefetch { |
73 | my ($self, $from, $select, $where, $attrs) = @_; |
74 | |
d28bb90d |
75 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
76 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY'); |
77 | |
1e4f9fb3 |
78 | my $root_alias = $attrs->{alias}; |
79 | |
d28bb90d |
80 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
81 | my $outer_attrs = { %$attrs }; |
1e4f9fb3 |
82 | delete $outer_attrs->{$_} for qw/where bind rows offset group_by _grouped_by_distinct having/; |
d28bb90d |
83 | |
186ba34c |
84 | my $inner_attrs = { %$attrs }; |
1e4f9fb3 |
85 | delete $inner_attrs->{$_} for qw/from for collapse select as _related_results_construction/; |
d28bb90d |
86 | |
4df1400e |
87 | # there is no point of ordering the insides if there is no limit |
88 | delete $inner_attrs->{order_by} if ( |
89 | delete $inner_attrs->{_order_is_artificial} |
90 | or |
91 | ! $inner_attrs->{rows} |
92 | ); |
946f6260 |
93 | |
d28bb90d |
94 | # generate the inner/outer select lists |
95 | # for inside we consider only stuff *not* brought in by the prefetch |
96 | # on the outside we substitute any function for its alias |
97 | my $outer_select = [ @$select ]; |
97e130fa |
98 | my $inner_select; |
36fd7f07 |
99 | |
97e130fa |
100 | my ($root_node, $root_node_offset); |
27e0370d |
101 | |
102 | for my $i (0 .. $#$from) { |
103 | my $node = $from->[$i]; |
104 | my $h = (ref $node eq 'HASH') ? $node |
105 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
106 | : next |
107 | ; |
108 | |
1e4f9fb3 |
109 | if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) { |
97e130fa |
110 | $root_node = $h; |
111 | $root_node_offset = $i; |
27e0370d |
112 | last; |
113 | } |
114 | } |
115 | |
116 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
117 | unless $root_node; |
27e0370d |
118 | |
119 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
120 | my $colinfo = $self->_resolve_column_info($from); |
121 | my $selected_root_columns; |
122 | |
1e4f9fb3 |
123 | for my $i (0 .. $#$outer_select) { |
d28bb90d |
124 | my $sel = $outer_select->[$i]; |
125 | |
1e4f9fb3 |
126 | next if ( |
127 | $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias |
128 | ); |
129 | |
d28bb90d |
130 | if (ref $sel eq 'HASH' ) { |
131 | $sel->{-as} ||= $attrs->{as}[$i]; |
1e4f9fb3 |
132 | $outer_select->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); |
d28bb90d |
133 | } |
27e0370d |
134 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
135 | $selected_root_columns->{$ci->{-colname}} = 1; |
136 | } |
d28bb90d |
137 | |
138 | push @$inner_select, $sel; |
bb9bffea |
139 | |
140 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
141 | } |
142 | |
97e130fa |
143 | # We will need to fetch all native columns in the inner subquery, which may |
144 | # be a part of an *outer* join condition, or an order_by (which needs to be |
145 | # preserved outside) |
146 | # We can not just fetch everything because a potential has_many restricting |
147 | # join collapse *will not work* on heavy data types. |
148 | my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args( |
149 | [grep { ref($_) eq 'ARRAY' or ref($_) eq 'HASH' } @{$from}[$root_node_offset .. $#$from]], |
150 | [], |
151 | $where, |
152 | $inner_attrs |
153 | ); |
154 | |
155 | for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { |
156 | my $ci = $colinfo->{$_} or next; |
157 | if ( |
1e4f9fb3 |
158 | $ci->{-source_alias} eq $root_alias |
97e130fa |
159 | and |
160 | ! $selected_root_columns->{$ci->{-colname}}++ |
161 | ) { |
162 | # adding it to both to keep limits not supporting dark selectors happy |
163 | push @$inner_select, $ci->{-fq_colname}; |
164 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
165 | } |
166 | } |
167 | |
ea95892e |
168 | # construct the inner $from and lock it in a subquery |
48580715 |
169 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
170 | # throw away all non-selecting, non-restricting multijoins |
171 | # (since we def. do not care about multiplication those inside the subquery) |
6395604e |
172 | my $inner_subq = do { |
ea95892e |
173 | |
174 | # must use it here regardless of user requests |
175 | local $self->{_use_join_optimizer} = 1; |
176 | |
97e130fa |
177 | # throw away multijoins since we def. do not care about those inside the subquery |
178 | my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, $inner_attrs, 'ignore_multiplication'); |
ea95892e |
179 | |
887a0aef |
180 | my $inner_aliastypes = |
181 | $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs ); |
182 | |
1e4f9fb3 |
183 | # uh-oh a multiplier (which is not us) left in, this is a problem |
0a3441ee |
184 | if ( |
1e4f9fb3 |
185 | $inner_aliastypes->{multiplying} |
186 | and |
187 | !$inner_aliastypes->{grouping} # if there are groups - assume user knows wtf they are up to |
0a3441ee |
188 | and |
1e4f9fb3 |
189 | my @multipliers = grep { $_ ne $root_alias } keys %{$inner_aliastypes->{multiplying}} |
0a3441ee |
190 | ) { |
1e4f9fb3 |
191 | |
192 | # if none of the multipliers came from an order_by (guaranteed to have been combined |
193 | # with a limit) - easy - just slap a group_by to simulate a collape and be on our way |
194 | if ( |
195 | ! $inner_aliastypes->{ordering} |
196 | or |
197 | ! first { $inner_aliastypes->{ordering}{$_} } @multipliers |
198 | ) { |
199 | my $unprocessed_order_chunks; |
200 | ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( |
201 | $inner_from, $inner_select, $inner_attrs->{order_by} |
202 | ); |
203 | |
204 | $self->throw_exception ( |
205 | 'A required group_by clause could not be constructed automatically due to a complex ' |
206 | . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' |
207 | . 'group_by by hand' |
208 | ) if $unprocessed_order_chunks; |
209 | } |
210 | else { |
211 | # We need to order by external columns and group at the same time |
212 | # so we can calculate the proper limit |
213 | # This doesn't really make sense in SQL, however from DBICs point |
214 | # of view is rather valid (order the leftmost objects by whatever |
215 | # criteria and get the offset/rows many). There is a way around |
216 | # this however in SQL - we simply tae the direction of each piece |
217 | # of the foreign order and convert them to MIN(X) for ASC or MAX(X) |
218 | # for DESC, and group_by the root columns. The end result should be |
219 | # exactly what we expect |
220 | |
221 | # FIXME REMOVE LATER - (just a sanity check) |
222 | if (defined ( my $impostor = first |
223 | { $_ ne $root_alias } |
224 | keys %{ $inner_aliastypes->{selecting} } |
225 | ) ) { |
226 | $self->throw_exception(sprintf |
227 | 'Unexpected inner selection during complex prefetch (%s)...', |
228 | join ', ', keys %{ $inner_aliastypes->{joining}{$impostor}{-seen_columns} || {} } |
229 | ); |
230 | } |
231 | |
232 | # supplement the main selection with pks if not already there, |
233 | # as they will have to be a part of the group_by to colapse |
234 | # things properly |
235 | my $cur_sel = { map { $_ => 1 } @$inner_select }; |
236 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
237 | or $self->throw_exception( sprintf |
238 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
239 | $root_node->{-rsrc}->source_name, |
240 | ); |
241 | for my $col (@pks) { |
242 | push @$inner_select, $col |
243 | unless $cur_sel->{$col}++; |
244 | } |
245 | |
246 | # wrap any part of the order_by that "responds" to an ordering alias |
247 | # into a MIN/MAX |
248 | # FIXME - this code is a joke, will need to be completely rewritten in |
249 | # the DQ branch. But I need to push a POC here, otherwise the |
250 | # pesky tests won't pass |
251 | my $sql_maker = $self->sql_maker; |
252 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
253 | my $own_re = qr/ $lquote \Q$root_alias\E $rquote $sep | \b \Q$root_alias\E $sep /x; |
254 | my @order = @{$attrs->{order_by}}; |
255 | my @order_chunks = map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks (\@order); |
256 | $self->throw_exception ('Order By parsing failed...') if @order != @order_chunks; |
257 | for my $i (0 .. $#order) { |
258 | # skip ourselves, and anything that looks like a literal |
259 | next if $order_chunks[$i][0] =~ $own_re; |
260 | next if (ref $order[$i] and ref $order[$i] ne 'HASH'); |
261 | |
262 | my $is_desc = $order_chunks[$i][0] =~ s/\sDESC$//i; |
263 | $order_chunks[$i][0] =~ s/\sASC$//i; |
264 | |
265 | $order[$i] = \[ |
266 | sprintf( |
267 | '%s(%s)%s', |
268 | ($is_desc ? 'MAX' : 'MIN'), |
269 | $order_chunks[$i][0], |
270 | ($is_desc ? ' DESC' : ''), |
271 | ), |
272 | @ {$order_chunks[$i]} [ 1 .. $#{$order_chunks[$i]} ] |
273 | ]; |
274 | } |
275 | |
276 | $inner_attrs->{order_by} = \@order; |
277 | ($inner_attrs->{group_by}) = $self->_group_over_selection ( |
278 | $inner_from, $inner_select, $inner_attrs->{order_by} |
279 | ); |
280 | } |
0a3441ee |
281 | } |
d28bb90d |
282 | |
ea95892e |
283 | # we already optimized $inner_from above |
97e130fa |
284 | # and already local()ized |
285 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
286 | |
ea95892e |
287 | # generate the subquery |
6395604e |
288 | $self->_select_args_to_query ( |
ea95892e |
289 | $inner_from, |
290 | $inner_select, |
291 | $where, |
292 | $inner_attrs, |
293 | ); |
d28bb90d |
294 | }; |
295 | |
296 | # Generate the outer from - this is relatively easy (really just replace |
297 | # the join slot with the subquery), with a major caveat - we can not |
298 | # join anything that is non-selecting (not part of the prefetch), but at |
299 | # the same time is a multi-type relationship, as it will explode the result. |
300 | # |
301 | # There are two possibilities here |
302 | # - either the join is non-restricting, in which case we simply throw it away |
303 | # - it is part of the restrictions, in which case we need to collapse the outer |
304 | # result by tackling yet another group_by to the outside of the query |
305 | |
27e0370d |
306 | # work on a shallow copy |
052e8431 |
307 | $from = [ @$from ]; |
052e8431 |
308 | |
d28bb90d |
309 | my @outer_from; |
53c29913 |
310 | |
27e0370d |
311 | # we may not be the head |
97e130fa |
312 | if ($root_node_offset) { |
27e0370d |
313 | # first generate the outer_from, up to the substitution point |
97e130fa |
314 | @outer_from = splice @$from, 0, $root_node_offset; |
27e0370d |
315 | |
316 | push @outer_from, [ |
317 | { |
1e4f9fb3 |
318 | -alias => $root_alias, |
97e130fa |
319 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
320 | $root_alias => $inner_subq, |
27e0370d |
321 | }, |
97e130fa |
322 | @{$from->[0]}[1 .. $#{$from->[0]}], |
27e0370d |
323 | ]; |
324 | } |
325 | else { |
27e0370d |
326 | @outer_from = { |
1e4f9fb3 |
327 | -alias => $root_alias, |
27e0370d |
328 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
329 | $root_alias => $inner_subq, |
27e0370d |
330 | }; |
d28bb90d |
331 | } |
332 | |
97e130fa |
333 | shift @$from; # it's replaced in @outer_from already |
334 | |
ea95892e |
335 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
336 | # in what role |
337 | my $outer_aliastypes = |
539ffe87 |
338 | $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs ); |
052e8431 |
339 | |
a4812caa |
340 | # unroll parents |
1e4f9fb3 |
341 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
342 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
343 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
344 | |
d28bb90d |
345 | # see what's left - throw away if not selecting/restricting |
a4812caa |
346 | # also throw in a group_by if a non-selecting multiplier, |
347 | # to guard against cross-join explosions |
36fd7f07 |
348 | my $need_outer_group_by; |
d28bb90d |
349 | while (my $j = shift @$from) { |
350 | my $alias = $j->[0]{-alias}; |
351 | |
a4812caa |
352 | if ( |
353 | $outer_select_chain->{$alias} |
354 | ) { |
355 | push @outer_from, $j |
d28bb90d |
356 | } |
1e4f9fb3 |
357 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
358 | push @outer_from, $j; |
a4812caa |
359 | $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
360 | } |
361 | } |
362 | |
1e4f9fb3 |
363 | if ( $need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
36fd7f07 |
364 | |
365 | my $unprocessed_order_chunks; |
366 | ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( |
367 | \@outer_from, $outer_select, $outer_attrs->{order_by} |
368 | ); |
369 | |
370 | $self->throw_exception ( |
371 | 'A required group_by clause could not be constructed automatically due to a complex ' |
372 | . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' |
373 | . 'group_by by hand' |
374 | ) if $unprocessed_order_chunks; |
375 | |
376 | } |
377 | |
d28bb90d |
378 | # This is totally horrific - the $where ends up in both the inner and outer query |
379 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
380 | # then if where conditions apply to the *right* side of the prefetch, you may have |
381 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
382 | # the outer select to exclude joins you didin't want in the first place |
383 | # |
384 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
385 | return (\@outer_from, $outer_select, $where, $outer_attrs); |
386 | } |
387 | |
1a736efb |
388 | # |
389 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
390 | # |
ad630f4b |
391 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
392 | # select/where/order/group attributes, in order to determine what |
393 | # aliases are neded to fulfill the query. This information is used |
394 | # throughout the code to prune unnecessary JOINs from the queries |
395 | # in an attempt to reduce the execution time. |
396 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
397 | # happen is for it to fail due to some scalar SQL, which in turn will |
398 | # result in a vocal exception. |
539ffe87 |
399 | sub _resolve_aliastypes_from_select_args { |
052e8431 |
400 | my ( $self, $from, $select, $where, $attrs ) = @_; |
546f1cd9 |
401 | |
ad630f4b |
402 | $self->throw_exception ('Unable to analyze custom {from}') |
403 | if ref $from ne 'ARRAY'; |
546f1cd9 |
404 | |
ad630f4b |
405 | # what we will return |
964a3c71 |
406 | my $aliases_by_type; |
546f1cd9 |
407 | |
ad630f4b |
408 | # see what aliases are there to work with |
409 | my $alias_list; |
539ffe87 |
410 | for (@$from) { |
411 | my $j = $_; |
ad630f4b |
412 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
413 | my $al = $j->{-alias} |
414 | or next; |
415 | |
416 | $alias_list->{$al} = $j; |
97e130fa |
417 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } if ( |
a4812caa |
418 | # not array == {from} head == can't be multiplying |
419 | ( ref($_) eq 'ARRAY' and ! $j->{-is_single} ) |
420 | or |
421 | # a parent of ours is already a multiplier |
422 | ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } ) |
423 | ); |
546f1cd9 |
424 | } |
546f1cd9 |
425 | |
1a736efb |
426 | # get a column to source/alias map (including unqualified ones) |
427 | my $colinfo = $self->_resolve_column_info ($from); |
428 | |
ad630f4b |
429 | # set up a botched SQLA |
430 | my $sql_maker = $self->sql_maker; |
07f31d19 |
431 | |
4c2b30d6 |
432 | # these are throw away results, do not pollute the bind stack |
4c2b30d6 |
433 | local $sql_maker->{select_bind}; |
0542ec57 |
434 | local $sql_maker->{where_bind}; |
435 | local $sql_maker->{group_bind}; |
436 | local $sql_maker->{having_bind}; |
97e130fa |
437 | local $sql_maker->{from_bind}; |
3f5b99fe |
438 | |
439 | # we can't scan properly without any quoting (\b doesn't cut it |
440 | # everywhere), so unless there is proper quoting set - use our |
441 | # own weird impossible character. |
442 | # Also in the case of no quoting, we need to explicitly disable |
443 | # name_sep, otherwise sorry nasty legacy syntax like |
444 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
445 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
446 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
447 | |
448 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
449 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
450 | # if we don't unset it we screw up retarded but unfortunately working |
451 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
452 | $sql_maker->{name_sep} = ''; |
453 | } |
454 | |
455 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
456 | |
1a736efb |
457 | # generate sql chunks |
458 | my $to_scan = { |
459 | restricting => [ |
460 | $sql_maker->_recurse_where ($where), |
1e4f9fb3 |
461 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
462 | ], |
463 | grouping => [ |
464 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
465 | ], |
97e130fa |
466 | joining => [ |
467 | $sql_maker->_recurse_from ( |
468 | ref $from->[0] eq 'ARRAY' ? $from->[0][0] : $from->[0], |
469 | @{$from}[1 .. $#$from], |
470 | ), |
471 | ], |
1a736efb |
472 | selecting => [ |
1a736efb |
473 | $sql_maker->_recurse_fields ($select), |
1e4f9fb3 |
474 | ], |
475 | ordering => [ |
476 | map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker), |
1a736efb |
477 | ], |
478 | }; |
07f31d19 |
479 | |
1a736efb |
480 | # throw away empty chunks |
481 | $_ = [ map { $_ || () } @$_ ] for values %$to_scan; |
07f31d19 |
482 | |
1a736efb |
483 | # first loop through all fully qualified columns and get the corresponding |
484 | # alias (should work even if they are in scalarrefs) |
ad630f4b |
485 | for my $alias (keys %$alias_list) { |
1a736efb |
486 | my $al_re = qr/ |
97e130fa |
487 | $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
1a736efb |
488 | | |
97e130fa |
489 | \b $alias \. ([^\s\)\($rquote]+)? |
1a736efb |
490 | /x; |
491 | |
1a736efb |
492 | for my $type (keys %$to_scan) { |
493 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
494 | if (my @matches = $piece =~ /$al_re/g) { |
495 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
496 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_" |
97e130fa |
497 | for grep { defined $_ } @matches; |
498 | } |
1a736efb |
499 | } |
ad630f4b |
500 | } |
1a736efb |
501 | } |
502 | |
503 | # now loop through unqualified column names, and try to locate them within |
504 | # the chunks |
505 | for my $col (keys %$colinfo) { |
3f5b99fe |
506 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
507 | |
97e130fa |
508 | my $col_re = qr/ $lquote ($col) $rquote /x; |
07f31d19 |
509 | |
1a736efb |
510 | for my $type (keys %$to_scan) { |
511 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
512 | if (my @matches = $piece =~ /$col_re/g) { |
a4812caa |
513 | my $alias = $colinfo->{$col}{-source_alias}; |
97e130fa |
514 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
515 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
97e130fa |
516 | for grep { defined $_ } @matches; |
a4812caa |
517 | } |
1a736efb |
518 | } |
07f31d19 |
519 | } |
520 | } |
521 | |
522 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
523 | for my $j (values %$alias_list) { |
07f31d19 |
524 | my $alias = $j->{-alias} or next; |
97e130fa |
525 | $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if ( |
07f31d19 |
526 | (not $j->{-join_type}) |
527 | or |
528 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
529 | ); |
530 | } |
531 | |
1e4f9fb3 |
532 | for (keys %$aliases_by_type) { |
533 | delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}}; |
534 | } |
535 | |
964a3c71 |
536 | return $aliases_by_type; |
07f31d19 |
537 | } |
538 | |
bac358c9 |
539 | # This is the engine behind { distinct => 1 } |
0a3441ee |
540 | sub _group_over_selection { |
541 | my ($self, $from, $select, $order_by) = @_; |
542 | |
543 | my $rs_column_list = $self->_resolve_column_info ($from); |
544 | |
545 | my (@group_by, %group_index); |
546 | |
36fd7f07 |
547 | # the logic is: if it is a { func => val } we assume an aggregate, |
548 | # otherwise if \'...' or \[...] we assume the user knows what is |
549 | # going on thus group over it |
0a3441ee |
550 | for (@$select) { |
551 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
552 | push @group_by, $_; |
553 | $group_index{$_}++; |
554 | if ($rs_column_list->{$_} and $_ !~ /\./ ) { |
555 | # add a fully qualified version as well |
556 | $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++; |
557 | } |
07f31d19 |
558 | } |
559 | } |
ad630f4b |
560 | |
0a3441ee |
561 | # add any order_by parts that are not already present in the group_by |
562 | # we need to be careful not to add any named functions/aggregates |
bac358c9 |
563 | # i.e. order_by => [ ... { count => 'foo' } ... ] |
14e26c5f |
564 | my @leftovers; |
bac358c9 |
565 | for ($self->_extract_order_criteria($order_by)) { |
0a3441ee |
566 | # only consider real columns (for functions the user got to do an explicit group_by) |
14e26c5f |
567 | if (@$_ != 1) { |
568 | push @leftovers, $_; |
569 | next; |
570 | } |
bac358c9 |
571 | my $chunk = $_->[0]; |
14e26c5f |
572 | my $colinfo = $rs_column_list->{$chunk} or do { |
573 | push @leftovers, $_; |
574 | next; |
575 | }; |
0a3441ee |
576 | |
577 | $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./; |
578 | push @group_by, $chunk unless $group_index{$chunk}++; |
579 | } |
580 | |
14e26c5f |
581 | return wantarray |
582 | ? (\@group_by, (@leftovers ? \@leftovers : undef) ) |
583 | : \@group_by |
584 | ; |
07f31d19 |
585 | } |
586 | |
d28bb90d |
587 | sub _resolve_ident_sources { |
588 | my ($self, $ident) = @_; |
589 | |
590 | my $alias2source = {}; |
591 | my $rs_alias; |
592 | |
593 | # the reason this is so contrived is that $ident may be a {from} |
594 | # structure, specifying multiple tables to join |
6298a324 |
595 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
596 | # this is compat mode for insert/update/delete which do not deal with aliases |
597 | $alias2source->{me} = $ident; |
598 | $rs_alias = 'me'; |
599 | } |
600 | elsif (ref $ident eq 'ARRAY') { |
601 | |
602 | for (@$ident) { |
603 | my $tabinfo; |
604 | if (ref $_ eq 'HASH') { |
605 | $tabinfo = $_; |
606 | $rs_alias = $tabinfo->{-alias}; |
607 | } |
608 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
609 | $tabinfo = $_->[0]; |
610 | } |
611 | |
4376a157 |
612 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
613 | if ($tabinfo->{-rsrc}); |
d28bb90d |
614 | } |
615 | } |
616 | |
617 | return ($alias2source, $rs_alias); |
618 | } |
619 | |
620 | # Takes $ident, \@column_names |
621 | # |
622 | # returns { $column_name => \%column_info, ... } |
623 | # also note: this adds -result_source => $rsrc to the column info |
624 | # |
09e14fdc |
625 | # If no columns_names are supplied returns info about *all* columns |
626 | # for all sources |
d28bb90d |
627 | sub _resolve_column_info { |
628 | my ($self, $ident, $colnames) = @_; |
629 | my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); |
630 | |
52416317 |
631 | my (%seen_cols, @auto_colnames); |
d28bb90d |
632 | |
633 | # compile a global list of column names, to be able to properly |
634 | # disambiguate unqualified column names (if at all possible) |
635 | for my $alias (keys %$alias2src) { |
636 | my $rsrc = $alias2src->{$alias}; |
637 | for my $colname ($rsrc->columns) { |
638 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
639 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
640 | } |
641 | } |
642 | |
09e14fdc |
643 | $colnames ||= [ |
644 | @auto_colnames, |
645 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
646 | ]; |
647 | |
52416317 |
648 | my (%return, $colinfos); |
d28bb90d |
649 | foreach my $col (@$colnames) { |
52416317 |
650 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
651 | |
52416317 |
652 | # if the column was seen exactly once - we know which rsrc it came from |
653 | $source_alias ||= $seen_cols{$colname}[0] |
654 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
655 | |
52416317 |
656 | next unless $source_alias; |
657 | |
658 | my $rsrc = $alias2src->{$source_alias} |
659 | or next; |
660 | |
661 | $return{$col} = { |
6395604e |
662 | %{ |
663 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
664 | || |
665 | $self->throw_exception( |
666 | "No such column '$colname' on source " . $rsrc->source_name |
667 | ); |
668 | }, |
d28bb90d |
669 | -result_source => $rsrc, |
52416317 |
670 | -source_alias => $source_alias, |
81bf295c |
671 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
672 | -colname => $colname, |
d28bb90d |
673 | }; |
81bf295c |
674 | |
675 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
676 | } |
677 | |
678 | return \%return; |
679 | } |
680 | |
289ac713 |
681 | # The DBIC relationship chaining implementation is pretty simple - every |
682 | # new related_relationship is pushed onto the {from} stack, and the {select} |
683 | # window simply slides further in. This means that when we count somewhere |
684 | # in the middle, we got to make sure that everything in the join chain is an |
685 | # actual inner join, otherwise the count will come back with unpredictable |
686 | # results (a resultset may be generated with _some_ rows regardless of if |
687 | # the relation which the $rs currently selects has rows or not). E.g. |
688 | # $artist_rs->cds->count - normally generates: |
689 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
690 | # which actually returns the number of artists * (number of cds || 1) |
691 | # |
692 | # So what we do here is crawl {from}, determine if the current alias is at |
693 | # the top of the stack, and if not - make sure the chain is inner-joined down |
694 | # to the root. |
695 | # |
31a8aaaf |
696 | sub _inner_join_to_node { |
289ac713 |
697 | my ($self, $from, $alias) = @_; |
698 | |
699 | # subqueries and other oddness are naturally not supported |
700 | return $from if ( |
701 | ref $from ne 'ARRAY' |
702 | || |
703 | @$from <= 1 |
704 | || |
705 | ref $from->[0] ne 'HASH' |
706 | || |
707 | ! $from->[0]{-alias} |
708 | || |
7eb76996 |
709 | $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do |
289ac713 |
710 | ); |
711 | |
712 | # find the current $alias in the $from structure |
713 | my $switch_branch; |
714 | JOINSCAN: |
715 | for my $j (@{$from}[1 .. $#$from]) { |
716 | if ($j->[0]{-alias} eq $alias) { |
717 | $switch_branch = $j->[0]{-join_path}; |
718 | last JOINSCAN; |
719 | } |
720 | } |
721 | |
7eb76996 |
722 | # something else went quite wrong |
289ac713 |
723 | return $from unless $switch_branch; |
724 | |
725 | # So it looks like we will have to switch some stuff around. |
726 | # local() is useless here as we will be leaving the scope |
727 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
728 | # So replace the first hashref in the node arrayref manually |
289ac713 |
729 | my @new_from = ($from->[0]); |
faeb2407 |
730 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
731 | |
732 | for my $j (@{$from}[1 .. $#$from]) { |
733 | my $jalias = $j->[0]{-alias}; |
734 | |
735 | if ($sw_idx->{$jalias}) { |
736 | my %attrs = %{$j->[0]}; |
737 | delete $attrs{-join_type}; |
738 | push @new_from, [ |
739 | \%attrs, |
740 | @{$j}[ 1 .. $#$j ], |
741 | ]; |
742 | } |
743 | else { |
744 | push @new_from, $j; |
745 | } |
746 | } |
747 | |
748 | return \@new_from; |
749 | } |
750 | |
bac358c9 |
751 | sub _extract_order_criteria { |
1a736efb |
752 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
753 | |
1a736efb |
754 | my $parser = sub { |
755 | my ($sql_maker, $order_by) = @_; |
c0748280 |
756 | |
1a736efb |
757 | return scalar $sql_maker->_order_by_chunks ($order_by) |
758 | unless wantarray; |
c0748280 |
759 | |
1a736efb |
760 | my @chunks; |
bac358c9 |
761 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
762 | my $chunk = ref $_ ? $_ : [ $_ ]; |
763 | $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix; |
1a736efb |
764 | push @chunks, $chunk; |
bac6c4fb |
765 | } |
1a736efb |
766 | |
767 | return @chunks; |
768 | }; |
769 | |
770 | if ($sql_maker) { |
771 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
772 | } |
773 | else { |
1a736efb |
774 | $sql_maker = $self->sql_maker; |
775 | local $sql_maker->{quote_char}; |
776 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
777 | } |
bac6c4fb |
778 | } |
779 | |
7cec4356 |
780 | sub _order_by_is_stable { |
5f11e54f |
781 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
782 | |
5f11e54f |
783 | my $colinfo = $self->_resolve_column_info($ident, [ |
784 | (map { $_->[0] } $self->_extract_order_criteria($order_by)), |
785 | $where ? @{$self->_extract_fixed_condition_columns($where)} :(), |
786 | ]); |
c0748280 |
787 | |
7cec4356 |
788 | return undef unless keys %$colinfo; |
789 | |
790 | my $cols_per_src; |
791 | $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo; |
792 | |
793 | for (values %$cols_per_src) { |
794 | my $src = (values %$_)[0]->{-result_source}; |
795 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
796 | } |
797 | |
7cec4356 |
798 | return undef; |
799 | } |
800 | |
5f11e54f |
801 | # returns an arrayref of column names which *definitely* have som |
802 | # sort of non-nullable equality requested in the given condition |
803 | # specification. This is used to figure out if a resultset is |
804 | # constrained to a column which is part of a unique constraint, |
805 | # which in turn allows us to better predict how ordering will behave |
806 | # etc. |
807 | # |
808 | # this is a rudimentary, incomplete, and error-prone extractor |
809 | # however this is OK - it is conservative, and if we can not find |
810 | # something that is in fact there - the stack will recover gracefully |
811 | # Also - DQ and the mst it rode in on will save us all RSN!!! |
812 | sub _extract_fixed_condition_columns { |
813 | my ($self, $where, $nested) = @_; |
814 | |
815 | return unless ref $where eq 'HASH'; |
816 | |
817 | my @cols; |
818 | for my $lhs (keys %$where) { |
819 | if ($lhs =~ /^\-and$/i) { |
820 | push @cols, ref $where->{$lhs} eq 'ARRAY' |
821 | ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} ) |
822 | : $self->_extract_fixed_condition_columns($where->{$lhs}, 1) |
823 | ; |
824 | } |
825 | elsif ($lhs !~ /^\-/) { |
826 | my $val = $where->{$lhs}; |
827 | |
828 | push @cols, $lhs if (defined $val and ( |
829 | ! ref $val |
830 | or |
831 | (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='}) |
832 | )); |
833 | } |
834 | } |
835 | return $nested ? @cols : \@cols; |
c0748280 |
836 | } |
bac6c4fb |
837 | |
d28bb90d |
838 | 1; |