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 | ) { |
318e3d94 |
199 | |
1e4f9fb3 |
200 | my $unprocessed_order_chunks; |
201 | ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( |
202 | $inner_from, $inner_select, $inner_attrs->{order_by} |
203 | ); |
204 | |
205 | $self->throw_exception ( |
206 | 'A required group_by clause could not be constructed automatically due to a complex ' |
207 | . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' |
208 | . 'group_by by hand' |
209 | ) if $unprocessed_order_chunks; |
210 | } |
211 | else { |
212 | # We need to order by external columns and group at the same time |
213 | # so we can calculate the proper limit |
214 | # This doesn't really make sense in SQL, however from DBICs point |
215 | # of view is rather valid (order the leftmost objects by whatever |
216 | # criteria and get the offset/rows many). There is a way around |
217 | # this however in SQL - we simply tae the direction of each piece |
218 | # of the foreign order and convert them to MIN(X) for ASC or MAX(X) |
219 | # for DESC, and group_by the root columns. The end result should be |
220 | # exactly what we expect |
221 | |
222 | # FIXME REMOVE LATER - (just a sanity check) |
223 | if (defined ( my $impostor = first |
224 | { $_ ne $root_alias } |
225 | keys %{ $inner_aliastypes->{selecting} } |
226 | ) ) { |
227 | $self->throw_exception(sprintf |
228 | 'Unexpected inner selection during complex prefetch (%s)...', |
229 | join ', ', keys %{ $inner_aliastypes->{joining}{$impostor}{-seen_columns} || {} } |
230 | ); |
231 | } |
232 | |
233 | # supplement the main selection with pks if not already there, |
234 | # as they will have to be a part of the group_by to colapse |
235 | # things properly |
236 | my $cur_sel = { map { $_ => 1 } @$inner_select }; |
318e3d94 |
237 | |
1e4f9fb3 |
238 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
239 | or $self->throw_exception( sprintf |
240 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
241 | $root_node->{-rsrc}->source_name, |
242 | ); |
243 | for my $col (@pks) { |
244 | push @$inner_select, $col |
245 | unless $cur_sel->{$col}++; |
246 | } |
247 | |
248 | # wrap any part of the order_by that "responds" to an ordering alias |
249 | # into a MIN/MAX |
250 | # FIXME - this code is a joke, will need to be completely rewritten in |
251 | # the DQ branch. But I need to push a POC here, otherwise the |
252 | # pesky tests won't pass |
253 | my $sql_maker = $self->sql_maker; |
254 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
255 | my $own_re = qr/ $lquote \Q$root_alias\E $rquote $sep | \b \Q$root_alias\E $sep /x; |
e6977bbb |
256 | my @order_chunks = map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}); |
257 | my @new_order = map { \$_ } @order_chunks; |
258 | my $inner_columns_info = $self->_resolve_column_info($inner_from); |
259 | |
260 | # loop through and replace stuff that is not "ours" with a min/max func |
261 | # everything is a literal at this point, since we are likely properly |
262 | # quoted and stuff |
263 | for my $i (0 .. $#new_order) { |
264 | my $chunk = $order_chunks[$i][0]; |
265 | |
266 | # skip ourselves |
267 | next if $chunk =~ $own_re; |
268 | |
cb3e87f5 |
269 | ($chunk, my $is_desc) = $sql_maker->_split_order_chunk($chunk); |
e6977bbb |
270 | |
271 | # maybe our own unqualified column |
318e3d94 |
272 | my $ord_bit = ( |
273 | $lquote and $sep and $chunk =~ /^ $lquote ([^$sep]+) $rquote $/x |
274 | ) ? $1 : $chunk; |
275 | |
e6977bbb |
276 | next if ( |
277 | $ord_bit |
278 | and |
279 | $inner_columns_info->{$ord_bit} |
280 | and |
281 | $inner_columns_info->{$ord_bit}{-source_alias} eq $root_alias |
282 | ); |
283 | |
284 | $new_order[$i] = \[ |
1e4f9fb3 |
285 | sprintf( |
286 | '%s(%s)%s', |
287 | ($is_desc ? 'MAX' : 'MIN'), |
e6977bbb |
288 | $chunk, |
1e4f9fb3 |
289 | ($is_desc ? ' DESC' : ''), |
290 | ), |
291 | @ {$order_chunks[$i]} [ 1 .. $#{$order_chunks[$i]} ] |
292 | ]; |
293 | } |
294 | |
e6977bbb |
295 | $inner_attrs->{order_by} = \@new_order; |
296 | |
297 | # do not care about leftovers here - it will be all the functions |
298 | # we just created |
1e4f9fb3 |
299 | ($inner_attrs->{group_by}) = $self->_group_over_selection ( |
300 | $inner_from, $inner_select, $inner_attrs->{order_by} |
301 | ); |
302 | } |
0a3441ee |
303 | } |
d28bb90d |
304 | |
ea95892e |
305 | # we already optimized $inner_from above |
97e130fa |
306 | # and already local()ized |
307 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
308 | |
ea95892e |
309 | # generate the subquery |
6395604e |
310 | $self->_select_args_to_query ( |
ea95892e |
311 | $inner_from, |
312 | $inner_select, |
313 | $where, |
314 | $inner_attrs, |
315 | ); |
d28bb90d |
316 | }; |
317 | |
318 | # Generate the outer from - this is relatively easy (really just replace |
319 | # the join slot with the subquery), with a major caveat - we can not |
320 | # join anything that is non-selecting (not part of the prefetch), but at |
321 | # the same time is a multi-type relationship, as it will explode the result. |
322 | # |
323 | # There are two possibilities here |
324 | # - either the join is non-restricting, in which case we simply throw it away |
325 | # - it is part of the restrictions, in which case we need to collapse the outer |
326 | # result by tackling yet another group_by to the outside of the query |
327 | |
27e0370d |
328 | # work on a shallow copy |
052e8431 |
329 | $from = [ @$from ]; |
052e8431 |
330 | |
d28bb90d |
331 | my @outer_from; |
53c29913 |
332 | |
27e0370d |
333 | # we may not be the head |
97e130fa |
334 | if ($root_node_offset) { |
27e0370d |
335 | # first generate the outer_from, up to the substitution point |
97e130fa |
336 | @outer_from = splice @$from, 0, $root_node_offset; |
27e0370d |
337 | |
338 | push @outer_from, [ |
339 | { |
1e4f9fb3 |
340 | -alias => $root_alias, |
97e130fa |
341 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
342 | $root_alias => $inner_subq, |
27e0370d |
343 | }, |
97e130fa |
344 | @{$from->[0]}[1 .. $#{$from->[0]}], |
27e0370d |
345 | ]; |
346 | } |
347 | else { |
27e0370d |
348 | @outer_from = { |
1e4f9fb3 |
349 | -alias => $root_alias, |
27e0370d |
350 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
351 | $root_alias => $inner_subq, |
27e0370d |
352 | }; |
d28bb90d |
353 | } |
354 | |
97e130fa |
355 | shift @$from; # it's replaced in @outer_from already |
356 | |
ea95892e |
357 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
358 | # in what role |
359 | my $outer_aliastypes = |
539ffe87 |
360 | $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs ); |
052e8431 |
361 | |
a4812caa |
362 | # unroll parents |
1e4f9fb3 |
363 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
364 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
365 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
366 | |
d28bb90d |
367 | # see what's left - throw away if not selecting/restricting |
a4812caa |
368 | # also throw in a group_by if a non-selecting multiplier, |
369 | # to guard against cross-join explosions |
36fd7f07 |
370 | my $need_outer_group_by; |
d28bb90d |
371 | while (my $j = shift @$from) { |
372 | my $alias = $j->[0]{-alias}; |
373 | |
a4812caa |
374 | if ( |
375 | $outer_select_chain->{$alias} |
376 | ) { |
377 | push @outer_from, $j |
d28bb90d |
378 | } |
1e4f9fb3 |
379 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
380 | push @outer_from, $j; |
a4812caa |
381 | $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
382 | } |
383 | } |
384 | |
1e4f9fb3 |
385 | if ( $need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
36fd7f07 |
386 | my $unprocessed_order_chunks; |
387 | ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( |
388 | \@outer_from, $outer_select, $outer_attrs->{order_by} |
389 | ); |
390 | |
391 | $self->throw_exception ( |
392 | 'A required group_by clause could not be constructed automatically due to a complex ' |
393 | . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' |
394 | . 'group_by by hand' |
395 | ) if $unprocessed_order_chunks; |
396 | |
397 | } |
398 | |
d28bb90d |
399 | # This is totally horrific - the $where ends up in both the inner and outer query |
400 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
401 | # then if where conditions apply to the *right* side of the prefetch, you may have |
402 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
403 | # the outer select to exclude joins you didin't want in the first place |
404 | # |
405 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
406 | return (\@outer_from, $outer_select, $where, $outer_attrs); |
407 | } |
408 | |
1a736efb |
409 | # |
410 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
411 | # |
ad630f4b |
412 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
413 | # select/where/order/group attributes, in order to determine what |
414 | # aliases are neded to fulfill the query. This information is used |
415 | # throughout the code to prune unnecessary JOINs from the queries |
416 | # in an attempt to reduce the execution time. |
417 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
418 | # happen is for it to fail due to some scalar SQL, which in turn will |
419 | # result in a vocal exception. |
539ffe87 |
420 | sub _resolve_aliastypes_from_select_args { |
052e8431 |
421 | my ( $self, $from, $select, $where, $attrs ) = @_; |
546f1cd9 |
422 | |
ad630f4b |
423 | $self->throw_exception ('Unable to analyze custom {from}') |
424 | if ref $from ne 'ARRAY'; |
546f1cd9 |
425 | |
ad630f4b |
426 | # what we will return |
964a3c71 |
427 | my $aliases_by_type; |
546f1cd9 |
428 | |
ad630f4b |
429 | # see what aliases are there to work with |
430 | my $alias_list; |
539ffe87 |
431 | for (@$from) { |
432 | my $j = $_; |
ad630f4b |
433 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
434 | my $al = $j->{-alias} |
435 | or next; |
436 | |
437 | $alias_list->{$al} = $j; |
97e130fa |
438 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } if ( |
a4812caa |
439 | # not array == {from} head == can't be multiplying |
440 | ( ref($_) eq 'ARRAY' and ! $j->{-is_single} ) |
441 | or |
442 | # a parent of ours is already a multiplier |
443 | ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } ) |
444 | ); |
546f1cd9 |
445 | } |
546f1cd9 |
446 | |
318e3d94 |
447 | # get a column to source/alias map (including unambiguous unqualified ones) |
1a736efb |
448 | my $colinfo = $self->_resolve_column_info ($from); |
449 | |
ad630f4b |
450 | # set up a botched SQLA |
451 | my $sql_maker = $self->sql_maker; |
07f31d19 |
452 | |
4c2b30d6 |
453 | # these are throw away results, do not pollute the bind stack |
4c2b30d6 |
454 | local $sql_maker->{select_bind}; |
0542ec57 |
455 | local $sql_maker->{where_bind}; |
456 | local $sql_maker->{group_bind}; |
457 | local $sql_maker->{having_bind}; |
97e130fa |
458 | local $sql_maker->{from_bind}; |
3f5b99fe |
459 | |
460 | # we can't scan properly without any quoting (\b doesn't cut it |
461 | # everywhere), so unless there is proper quoting set - use our |
462 | # own weird impossible character. |
463 | # Also in the case of no quoting, we need to explicitly disable |
464 | # name_sep, otherwise sorry nasty legacy syntax like |
465 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
466 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
467 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
468 | |
469 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
470 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
471 | # if we don't unset it we screw up retarded but unfortunately working |
472 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
473 | $sql_maker->{name_sep} = ''; |
474 | } |
475 | |
476 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
477 | |
1a736efb |
478 | # generate sql chunks |
479 | my $to_scan = { |
480 | restricting => [ |
481 | $sql_maker->_recurse_where ($where), |
1e4f9fb3 |
482 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
483 | ], |
484 | grouping => [ |
485 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
486 | ], |
97e130fa |
487 | joining => [ |
488 | $sql_maker->_recurse_from ( |
489 | ref $from->[0] eq 'ARRAY' ? $from->[0][0] : $from->[0], |
490 | @{$from}[1 .. $#$from], |
491 | ), |
492 | ], |
1a736efb |
493 | selecting => [ |
1a736efb |
494 | $sql_maker->_recurse_fields ($select), |
1e4f9fb3 |
495 | ], |
496 | ordering => [ |
497 | map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker), |
1a736efb |
498 | ], |
499 | }; |
07f31d19 |
500 | |
1a736efb |
501 | # throw away empty chunks |
502 | $_ = [ map { $_ || () } @$_ ] for values %$to_scan; |
07f31d19 |
503 | |
318e3d94 |
504 | # first see if we have any exact matches (qualified or unqualified) |
505 | for my $type (keys %$to_scan) { |
506 | for my $piece (@{$to_scan->{$type}}) { |
507 | if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) { |
508 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
509 | $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece; |
510 | } |
511 | } |
512 | } |
513 | |
514 | # now loop through all fully qualified columns and get the corresponding |
1a736efb |
515 | # alias (should work even if they are in scalarrefs) |
ad630f4b |
516 | for my $alias (keys %$alias_list) { |
1a736efb |
517 | my $al_re = qr/ |
97e130fa |
518 | $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
1a736efb |
519 | | |
97e130fa |
520 | \b $alias \. ([^\s\)\($rquote]+)? |
1a736efb |
521 | /x; |
522 | |
1a736efb |
523 | for my $type (keys %$to_scan) { |
524 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
525 | if (my @matches = $piece =~ /$al_re/g) { |
526 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
527 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_" |
97e130fa |
528 | for grep { defined $_ } @matches; |
529 | } |
1a736efb |
530 | } |
ad630f4b |
531 | } |
1a736efb |
532 | } |
533 | |
534 | # now loop through unqualified column names, and try to locate them within |
535 | # the chunks |
536 | for my $col (keys %$colinfo) { |
3f5b99fe |
537 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
538 | |
97e130fa |
539 | my $col_re = qr/ $lquote ($col) $rquote /x; |
07f31d19 |
540 | |
1a736efb |
541 | for my $type (keys %$to_scan) { |
542 | for my $piece (@{$to_scan->{$type}}) { |
318e3d94 |
543 | if ( my @matches = $piece =~ /$col_re/g) { |
a4812caa |
544 | my $alias = $colinfo->{$col}{-source_alias}; |
97e130fa |
545 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
546 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
97e130fa |
547 | for grep { defined $_ } @matches; |
a4812caa |
548 | } |
1a736efb |
549 | } |
07f31d19 |
550 | } |
551 | } |
552 | |
553 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
554 | for my $j (values %$alias_list) { |
07f31d19 |
555 | my $alias = $j->{-alias} or next; |
97e130fa |
556 | $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if ( |
07f31d19 |
557 | (not $j->{-join_type}) |
558 | or |
559 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
560 | ); |
561 | } |
562 | |
1e4f9fb3 |
563 | for (keys %$aliases_by_type) { |
564 | delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}}; |
565 | } |
566 | |
964a3c71 |
567 | return $aliases_by_type; |
07f31d19 |
568 | } |
569 | |
bac358c9 |
570 | # This is the engine behind { distinct => 1 } |
0a3441ee |
571 | sub _group_over_selection { |
572 | my ($self, $from, $select, $order_by) = @_; |
573 | |
574 | my $rs_column_list = $self->_resolve_column_info ($from); |
575 | |
576 | my (@group_by, %group_index); |
577 | |
36fd7f07 |
578 | # the logic is: if it is a { func => val } we assume an aggregate, |
579 | # otherwise if \'...' or \[...] we assume the user knows what is |
580 | # going on thus group over it |
0a3441ee |
581 | for (@$select) { |
582 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
583 | push @group_by, $_; |
584 | $group_index{$_}++; |
585 | if ($rs_column_list->{$_} and $_ !~ /\./ ) { |
586 | # add a fully qualified version as well |
587 | $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++; |
588 | } |
07f31d19 |
589 | } |
590 | } |
ad630f4b |
591 | |
0a3441ee |
592 | # add any order_by parts that are not already present in the group_by |
593 | # we need to be careful not to add any named functions/aggregates |
bac358c9 |
594 | # i.e. order_by => [ ... { count => 'foo' } ... ] |
14e26c5f |
595 | my @leftovers; |
bac358c9 |
596 | for ($self->_extract_order_criteria($order_by)) { |
0a3441ee |
597 | # only consider real columns (for functions the user got to do an explicit group_by) |
14e26c5f |
598 | if (@$_ != 1) { |
599 | push @leftovers, $_; |
600 | next; |
601 | } |
bac358c9 |
602 | my $chunk = $_->[0]; |
14e26c5f |
603 | my $colinfo = $rs_column_list->{$chunk} or do { |
604 | push @leftovers, $_; |
605 | next; |
606 | }; |
0a3441ee |
607 | |
608 | $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./; |
609 | push @group_by, $chunk unless $group_index{$chunk}++; |
610 | } |
611 | |
14e26c5f |
612 | return wantarray |
613 | ? (\@group_by, (@leftovers ? \@leftovers : undef) ) |
614 | : \@group_by |
615 | ; |
07f31d19 |
616 | } |
617 | |
d28bb90d |
618 | sub _resolve_ident_sources { |
619 | my ($self, $ident) = @_; |
620 | |
621 | my $alias2source = {}; |
622 | my $rs_alias; |
623 | |
624 | # the reason this is so contrived is that $ident may be a {from} |
625 | # structure, specifying multiple tables to join |
6298a324 |
626 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
627 | # this is compat mode for insert/update/delete which do not deal with aliases |
628 | $alias2source->{me} = $ident; |
629 | $rs_alias = 'me'; |
630 | } |
631 | elsif (ref $ident eq 'ARRAY') { |
632 | |
633 | for (@$ident) { |
634 | my $tabinfo; |
635 | if (ref $_ eq 'HASH') { |
636 | $tabinfo = $_; |
637 | $rs_alias = $tabinfo->{-alias}; |
638 | } |
639 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
640 | $tabinfo = $_->[0]; |
641 | } |
642 | |
4376a157 |
643 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
644 | if ($tabinfo->{-rsrc}); |
d28bb90d |
645 | } |
646 | } |
647 | |
648 | return ($alias2source, $rs_alias); |
649 | } |
650 | |
651 | # Takes $ident, \@column_names |
652 | # |
653 | # returns { $column_name => \%column_info, ... } |
654 | # also note: this adds -result_source => $rsrc to the column info |
655 | # |
09e14fdc |
656 | # If no columns_names are supplied returns info about *all* columns |
657 | # for all sources |
d28bb90d |
658 | sub _resolve_column_info { |
659 | my ($self, $ident, $colnames) = @_; |
660 | my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); |
661 | |
52416317 |
662 | my (%seen_cols, @auto_colnames); |
d28bb90d |
663 | |
664 | # compile a global list of column names, to be able to properly |
665 | # disambiguate unqualified column names (if at all possible) |
666 | for my $alias (keys %$alias2src) { |
667 | my $rsrc = $alias2src->{$alias}; |
668 | for my $colname ($rsrc->columns) { |
669 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
670 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
671 | } |
672 | } |
673 | |
09e14fdc |
674 | $colnames ||= [ |
675 | @auto_colnames, |
676 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
677 | ]; |
678 | |
52416317 |
679 | my (%return, $colinfos); |
d28bb90d |
680 | foreach my $col (@$colnames) { |
52416317 |
681 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
682 | |
52416317 |
683 | # if the column was seen exactly once - we know which rsrc it came from |
684 | $source_alias ||= $seen_cols{$colname}[0] |
685 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
686 | |
52416317 |
687 | next unless $source_alias; |
688 | |
689 | my $rsrc = $alias2src->{$source_alias} |
690 | or next; |
691 | |
692 | $return{$col} = { |
6395604e |
693 | %{ |
694 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
695 | || |
696 | $self->throw_exception( |
697 | "No such column '$colname' on source " . $rsrc->source_name |
698 | ); |
699 | }, |
d28bb90d |
700 | -result_source => $rsrc, |
52416317 |
701 | -source_alias => $source_alias, |
81bf295c |
702 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
703 | -colname => $colname, |
d28bb90d |
704 | }; |
81bf295c |
705 | |
706 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
707 | } |
708 | |
709 | return \%return; |
710 | } |
711 | |
289ac713 |
712 | # The DBIC relationship chaining implementation is pretty simple - every |
713 | # new related_relationship is pushed onto the {from} stack, and the {select} |
714 | # window simply slides further in. This means that when we count somewhere |
715 | # in the middle, we got to make sure that everything in the join chain is an |
716 | # actual inner join, otherwise the count will come back with unpredictable |
717 | # results (a resultset may be generated with _some_ rows regardless of if |
718 | # the relation which the $rs currently selects has rows or not). E.g. |
719 | # $artist_rs->cds->count - normally generates: |
720 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
721 | # which actually returns the number of artists * (number of cds || 1) |
722 | # |
723 | # So what we do here is crawl {from}, determine if the current alias is at |
724 | # the top of the stack, and if not - make sure the chain is inner-joined down |
725 | # to the root. |
726 | # |
31a8aaaf |
727 | sub _inner_join_to_node { |
289ac713 |
728 | my ($self, $from, $alias) = @_; |
729 | |
730 | # subqueries and other oddness are naturally not supported |
731 | return $from if ( |
732 | ref $from ne 'ARRAY' |
733 | || |
734 | @$from <= 1 |
735 | || |
736 | ref $from->[0] ne 'HASH' |
737 | || |
738 | ! $from->[0]{-alias} |
739 | || |
7eb76996 |
740 | $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do |
289ac713 |
741 | ); |
742 | |
743 | # find the current $alias in the $from structure |
744 | my $switch_branch; |
745 | JOINSCAN: |
746 | for my $j (@{$from}[1 .. $#$from]) { |
747 | if ($j->[0]{-alias} eq $alias) { |
748 | $switch_branch = $j->[0]{-join_path}; |
749 | last JOINSCAN; |
750 | } |
751 | } |
752 | |
7eb76996 |
753 | # something else went quite wrong |
289ac713 |
754 | return $from unless $switch_branch; |
755 | |
756 | # So it looks like we will have to switch some stuff around. |
757 | # local() is useless here as we will be leaving the scope |
758 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
759 | # So replace the first hashref in the node arrayref manually |
289ac713 |
760 | my @new_from = ($from->[0]); |
faeb2407 |
761 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
762 | |
763 | for my $j (@{$from}[1 .. $#$from]) { |
764 | my $jalias = $j->[0]{-alias}; |
765 | |
766 | if ($sw_idx->{$jalias}) { |
767 | my %attrs = %{$j->[0]}; |
768 | delete $attrs{-join_type}; |
769 | push @new_from, [ |
770 | \%attrs, |
771 | @{$j}[ 1 .. $#$j ], |
772 | ]; |
773 | } |
774 | else { |
775 | push @new_from, $j; |
776 | } |
777 | } |
778 | |
779 | return \@new_from; |
780 | } |
781 | |
bac358c9 |
782 | sub _extract_order_criteria { |
1a736efb |
783 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
784 | |
1a736efb |
785 | my $parser = sub { |
e6977bbb |
786 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
787 | |
1a736efb |
788 | return scalar $sql_maker->_order_by_chunks ($order_by) |
789 | unless wantarray; |
c0748280 |
790 | |
e6977bbb |
791 | my ($lq, $rq, $sep) = map { quotemeta($_) } ( |
792 | ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars), |
793 | $sql_maker->name_sep |
794 | ); |
795 | |
1a736efb |
796 | my @chunks; |
bac358c9 |
797 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
798 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
799 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
800 | |
801 | # order criteria may have come back pre-quoted (literals and whatnot) |
802 | # this is fragile, but the best we can currently do |
803 | $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe |
804 | or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x; |
805 | |
1a736efb |
806 | push @chunks, $chunk; |
bac6c4fb |
807 | } |
1a736efb |
808 | |
809 | return @chunks; |
810 | }; |
811 | |
812 | if ($sql_maker) { |
813 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
814 | } |
815 | else { |
1a736efb |
816 | $sql_maker = $self->sql_maker; |
e6977bbb |
817 | |
818 | # pass these in to deal with literals coming from |
819 | # the user or the deep guts of prefetch |
820 | my $orig_quote_chars = [$sql_maker->_quote_chars]; |
821 | |
1a736efb |
822 | local $sql_maker->{quote_char}; |
e6977bbb |
823 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
824 | } |
bac6c4fb |
825 | } |
826 | |
7cec4356 |
827 | sub _order_by_is_stable { |
5f11e54f |
828 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
829 | |
5f11e54f |
830 | my $colinfo = $self->_resolve_column_info($ident, [ |
831 | (map { $_->[0] } $self->_extract_order_criteria($order_by)), |
832 | $where ? @{$self->_extract_fixed_condition_columns($where)} :(), |
833 | ]); |
c0748280 |
834 | |
7cec4356 |
835 | return undef unless keys %$colinfo; |
836 | |
837 | my $cols_per_src; |
838 | $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo; |
839 | |
840 | for (values %$cols_per_src) { |
841 | my $src = (values %$_)[0]->{-result_source}; |
842 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
843 | } |
844 | |
7cec4356 |
845 | return undef; |
846 | } |
847 | |
0e81e691 |
848 | # this is almost identical to the above, except it accepts only |
849 | # a single rsrc, and will succeed only if the first portion of the order |
850 | # by is stable. |
851 | # returns that portion as a colinfo hashref on success |
852 | sub _main_source_order_by_portion_is_stable { |
853 | my ($self, $main_rsrc, $order_by, $where) = @_; |
854 | |
855 | die "Huh... I expect a blessed result_source..." |
856 | if ref($main_rsrc) eq 'ARRAY'; |
857 | |
858 | my @ord_cols = map |
859 | { $_->[0] } |
860 | ( $self->_extract_order_criteria($order_by) ) |
861 | ; |
862 | return unless @ord_cols; |
863 | |
318e3d94 |
864 | my $colinfos = $self->_resolve_column_info($main_rsrc); |
865 | |
0e81e691 |
866 | for (0 .. $#ord_cols) { |
867 | if ( |
868 | ! $colinfos->{$ord_cols[$_]} |
869 | or |
870 | $colinfos->{$ord_cols[$_]}{-result_source} != $main_rsrc |
871 | ) { |
872 | $#ord_cols = $_ - 1; |
873 | last; |
874 | } |
875 | } |
876 | |
877 | # we just truncated it above |
878 | return unless @ord_cols; |
879 | |
0e81e691 |
880 | my $order_portion_ci = { map { |
881 | $colinfos->{$_}{-colname} => $colinfos->{$_}, |
882 | $colinfos->{$_}{-fq_colname} => $colinfos->{$_}, |
883 | } @ord_cols }; |
884 | |
318e3d94 |
885 | # since all we check here are the start of the order_by belonging to the |
886 | # top level $rsrc, a present identifying set will mean that the resultset |
887 | # is ordered by its leftmost table in a stable manner |
888 | # |
889 | # RV of _identifying_column_set contains unqualified names only |
890 | my $unqualified_idset = $main_rsrc->_identifying_column_set({ |
891 | ( $where ? %{ |
892 | $self->_resolve_column_info( |
893 | $main_rsrc, $self->_extract_fixed_condition_columns($where) |
894 | ) |
895 | } : () ), |
896 | %$order_portion_ci |
897 | }) or return; |
898 | |
899 | my $ret_info; |
900 | my %unqualified_idcols_from_order = map { |
901 | $order_portion_ci->{$_} ? ( $_ => $order_portion_ci->{$_} ) : () |
902 | } @$unqualified_idset; |
903 | |
904 | # extra optimization - cut the order_by at the end of the identifying set |
905 | # (just in case the user was stupid and overlooked the obvious) |
906 | for my $i (0 .. $#ord_cols) { |
907 | my $col = $ord_cols[$i]; |
908 | my $unqualified_colname = $order_portion_ci->{$col}{-colname}; |
909 | $ret_info->{$col} = { %{$order_portion_ci->{$col}}, -idx_in_order_subset => $i }; |
910 | delete $unqualified_idcols_from_order{$ret_info->{$col}{-colname}}; |
911 | |
912 | # we didn't reach the end of the identifying portion yet |
913 | return $ret_info unless keys %unqualified_idcols_from_order; |
914 | } |
0e81e691 |
915 | |
318e3d94 |
916 | die 'How did we get here...'; |
0e81e691 |
917 | } |
918 | |
5f11e54f |
919 | # returns an arrayref of column names which *definitely* have som |
920 | # sort of non-nullable equality requested in the given condition |
921 | # specification. This is used to figure out if a resultset is |
922 | # constrained to a column which is part of a unique constraint, |
923 | # which in turn allows us to better predict how ordering will behave |
924 | # etc. |
925 | # |
926 | # this is a rudimentary, incomplete, and error-prone extractor |
927 | # however this is OK - it is conservative, and if we can not find |
928 | # something that is in fact there - the stack will recover gracefully |
929 | # Also - DQ and the mst it rode in on will save us all RSN!!! |
930 | sub _extract_fixed_condition_columns { |
931 | my ($self, $where, $nested) = @_; |
932 | |
933 | return unless ref $where eq 'HASH'; |
934 | |
935 | my @cols; |
936 | for my $lhs (keys %$where) { |
937 | if ($lhs =~ /^\-and$/i) { |
938 | push @cols, ref $where->{$lhs} eq 'ARRAY' |
939 | ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} ) |
940 | : $self->_extract_fixed_condition_columns($where->{$lhs}, 1) |
941 | ; |
942 | } |
943 | elsif ($lhs !~ /^\-/) { |
944 | my $val = $where->{$lhs}; |
945 | |
946 | push @cols, $lhs if (defined $val and ( |
947 | ! ref $val |
948 | or |
949 | (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='}) |
950 | )); |
951 | } |
952 | } |
953 | return $nested ? @cols : \@cols; |
c0748280 |
954 | } |
bac6c4fb |
955 | |
d28bb90d |
956 | 1; |