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'; |
10cef607 |
19 | use Data::Query::ExprHelpers; |
6298a324 |
20 | use namespace::clean; |
d28bb90d |
21 | |
22 | # |
052e8431 |
23 | # This code will remove non-selecting/non-restricting joins from |
4b1b5ea3 |
24 | # {from} specs, aiding the RDBMS query optimizer |
052e8431 |
25 | # |
26 | sub _prune_unused_joins { |
ea95892e |
27 | my $self = shift; |
437a9cfa |
28 | my ($from, $select, $where, $attrs) = @_; |
052e8431 |
29 | |
ea95892e |
30 | return $from unless $self->_use_join_optimizer; |
31 | |
052e8431 |
32 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY') { |
33 | return $from; # only standard {from} specs are supported |
34 | } |
35 | |
4b1b5ea3 |
36 | my $aliastypes = $self->_resolve_aliastypes_from_select_args(@_); |
37 | |
a6ef93cb |
38 | my $orig_joins = delete $aliastypes->{joining}; |
39 | my $orig_multiplying = $aliastypes->{multiplying}; |
97e130fa |
40 | |
4b1b5ea3 |
41 | # a grouped set will not be affected by amount of rows. Thus any |
42 | # {multiplying} joins can go |
97e130fa |
43 | delete $aliastypes->{multiplying} |
437a9cfa |
44 | if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by}; |
4b1b5ea3 |
45 | |
052e8431 |
46 | my @newfrom = $from->[0]; # FROM head is always present |
47 | |
a4812caa |
48 | my %need_joins; |
97e130fa |
49 | |
a4812caa |
50 | for (values %$aliastypes) { |
51 | # add all requested aliases |
52 | $need_joins{$_} = 1 for keys %$_; |
53 | |
54 | # add all their parents (as per joinpath which is an AoH { table => alias }) |
97e130fa |
55 | $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; |
a4812caa |
56 | } |
97e130fa |
57 | |
052e8431 |
58 | for my $j (@{$from}[1..$#$from]) { |
539ffe87 |
59 | push @newfrom, $j if ( |
a6ef93cb |
60 | (! defined $j->[0]{-alias}) # legacy crap |
539ffe87 |
61 | || |
62 | $need_joins{$j->[0]{-alias}} |
63 | ); |
052e8431 |
64 | } |
65 | |
a6ef93cb |
66 | return ( \@newfrom, { |
67 | multiplying => { map { $need_joins{$_} ? ($_ => $orig_multiplying->{$_}) : () } keys %$orig_multiplying }, |
68 | %$aliastypes, |
69 | joining => { map { $_ => $orig_joins->{$_} } keys %need_joins }, |
70 | } ); |
052e8431 |
71 | } |
72 | |
052e8431 |
73 | # |
d28bb90d |
74 | # This is the code producing joined subqueries like: |
8273e845 |
75 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
d28bb90d |
76 | # |
77 | sub _adjust_select_args_for_complex_prefetch { |
78 | my ($self, $from, $select, $where, $attrs) = @_; |
79 | |
d28bb90d |
80 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
81 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY'); |
82 | |
1e4f9fb3 |
83 | my $root_alias = $attrs->{alias}; |
84 | |
d28bb90d |
85 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
86 | my $outer_attrs = { %$attrs }; |
5b309063 |
87 | delete @{$outer_attrs}{qw(where bind rows offset group_by _grouped_by_distinct having)}; |
d28bb90d |
88 | |
186ba34c |
89 | my $inner_attrs = { %$attrs }; |
5b309063 |
90 | delete @{$inner_attrs}{qw(from for collapse select as _related_results_construction)}; |
d28bb90d |
91 | |
4df1400e |
92 | # there is no point of ordering the insides if there is no limit |
93 | delete $inner_attrs->{order_by} if ( |
94 | delete $inner_attrs->{_order_is_artificial} |
95 | or |
96 | ! $inner_attrs->{rows} |
97 | ); |
946f6260 |
98 | |
d28bb90d |
99 | # generate the inner/outer select lists |
100 | # for inside we consider only stuff *not* brought in by the prefetch |
101 | # on the outside we substitute any function for its alias |
102 | my $outer_select = [ @$select ]; |
97e130fa |
103 | my $inner_select; |
36fd7f07 |
104 | |
97e130fa |
105 | my ($root_node, $root_node_offset); |
27e0370d |
106 | |
107 | for my $i (0 .. $#$from) { |
108 | my $node = $from->[$i]; |
109 | my $h = (ref $node eq 'HASH') ? $node |
110 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
111 | : next |
112 | ; |
113 | |
1e4f9fb3 |
114 | if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) { |
97e130fa |
115 | $root_node = $h; |
116 | $root_node_offset = $i; |
27e0370d |
117 | last; |
118 | } |
119 | } |
120 | |
121 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
122 | unless $root_node; |
27e0370d |
123 | |
124 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
125 | my $colinfo = $self->_resolve_column_info($from); |
126 | my $selected_root_columns; |
127 | |
1e4f9fb3 |
128 | for my $i (0 .. $#$outer_select) { |
d28bb90d |
129 | my $sel = $outer_select->[$i]; |
130 | |
1e4f9fb3 |
131 | next if ( |
132 | $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias |
133 | ); |
134 | |
d28bb90d |
135 | if (ref $sel eq 'HASH' ) { |
136 | $sel->{-as} ||= $attrs->{as}[$i]; |
1e4f9fb3 |
137 | $outer_select->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); |
d28bb90d |
138 | } |
27e0370d |
139 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
140 | $selected_root_columns->{$ci->{-colname}} = 1; |
141 | } |
d28bb90d |
142 | |
143 | push @$inner_select, $sel; |
bb9bffea |
144 | |
145 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
146 | } |
147 | |
97e130fa |
148 | # We will need to fetch all native columns in the inner subquery, which may |
149 | # be a part of an *outer* join condition, or an order_by (which needs to be |
150 | # preserved outside) |
151 | # We can not just fetch everything because a potential has_many restricting |
152 | # join collapse *will not work* on heavy data types. |
153 | my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args( |
560978e2 |
154 | $from, |
97e130fa |
155 | [], |
156 | $where, |
157 | $inner_attrs |
158 | ); |
159 | |
160 | for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { |
161 | my $ci = $colinfo->{$_} or next; |
162 | if ( |
1e4f9fb3 |
163 | $ci->{-source_alias} eq $root_alias |
97e130fa |
164 | and |
165 | ! $selected_root_columns->{$ci->{-colname}}++ |
166 | ) { |
167 | # adding it to both to keep limits not supporting dark selectors happy |
168 | push @$inner_select, $ci->{-fq_colname}; |
169 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
170 | } |
171 | } |
172 | |
ea95892e |
173 | # construct the inner $from and lock it in a subquery |
48580715 |
174 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
175 | # throw away all non-selecting, non-restricting multijoins |
176 | # (since we def. do not care about multiplication those inside the subquery) |
6395604e |
177 | my $inner_subq = do { |
ea95892e |
178 | |
179 | # must use it here regardless of user requests |
180 | local $self->{_use_join_optimizer} = 1; |
181 | |
97e130fa |
182 | # throw away multijoins since we def. do not care about those inside the subquery |
a6ef93cb |
183 | my ($inner_from, $inner_aliastypes) = $self->_prune_unused_joins ($from, $inner_select, $where, { |
437a9cfa |
184 | %$inner_attrs, _force_prune_multiplying_joins => 1 |
185 | }); |
ea95892e |
186 | |
1e4f9fb3 |
187 | # uh-oh a multiplier (which is not us) left in, this is a problem |
0a3441ee |
188 | if ( |
1e4f9fb3 |
189 | $inner_aliastypes->{multiplying} |
190 | and |
560978e2 |
191 | # if there are user-supplied groups - assume user knows wtf they are up to |
192 | ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} ) |
0a3441ee |
193 | and |
1e4f9fb3 |
194 | my @multipliers = grep { $_ ne $root_alias } keys %{$inner_aliastypes->{multiplying}} |
0a3441ee |
195 | ) { |
1e4f9fb3 |
196 | |
197 | # if none of the multipliers came from an order_by (guaranteed to have been combined |
4a0eed52 |
198 | # with a limit) - easy - just slap a group_by to simulate a collapse and be on our way |
1e4f9fb3 |
199 | if ( |
200 | ! $inner_aliastypes->{ordering} |
201 | or |
202 | ! first { $inner_aliastypes->{ordering}{$_} } @multipliers |
203 | ) { |
318e3d94 |
204 | |
1e4f9fb3 |
205 | my $unprocessed_order_chunks; |
560978e2 |
206 | ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ({ |
207 | %$inner_attrs, |
208 | from => $inner_from, |
209 | select => $inner_select, |
210 | }); |
1e4f9fb3 |
211 | |
212 | $self->throw_exception ( |
213 | 'A required group_by clause could not be constructed automatically due to a complex ' |
214 | . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' |
215 | . 'group_by by hand' |
216 | ) if $unprocessed_order_chunks; |
217 | } |
218 | else { |
219 | # We need to order by external columns and group at the same time |
220 | # so we can calculate the proper limit |
221 | # This doesn't really make sense in SQL, however from DBICs point |
222 | # of view is rather valid (order the leftmost objects by whatever |
223 | # criteria and get the offset/rows many). There is a way around |
224 | # this however in SQL - we simply tae the direction of each piece |
225 | # of the foreign order and convert them to MIN(X) for ASC or MAX(X) |
226 | # for DESC, and group_by the root columns. The end result should be |
227 | # exactly what we expect |
228 | |
1e4f9fb3 |
229 | # supplement the main selection with pks if not already there, |
4a0eed52 |
230 | # as they will have to be a part of the group_by to collapse |
1e4f9fb3 |
231 | # things properly |
232 | my $cur_sel = { map { $_ => 1 } @$inner_select }; |
318e3d94 |
233 | |
1e4f9fb3 |
234 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
235 | or $self->throw_exception( sprintf |
236 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
237 | $root_node->{-rsrc}->source_name, |
238 | ); |
239 | for my $col (@pks) { |
240 | push @$inner_select, $col |
241 | unless $cur_sel->{$col}++; |
242 | } |
243 | |
244 | # wrap any part of the order_by that "responds" to an ordering alias |
245 | # into a MIN/MAX |
246 | # FIXME - this code is a joke, will need to be completely rewritten in |
247 | # the DQ branch. But I need to push a POC here, otherwise the |
248 | # pesky tests won't pass |
249 | my $sql_maker = $self->sql_maker; |
250 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
251 | my $own_re = qr/ $lquote \Q$root_alias\E $rquote $sep | \b \Q$root_alias\E $sep /x; |
e6977bbb |
252 | my @order_chunks = map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}); |
253 | my @new_order = map { \$_ } @order_chunks; |
254 | my $inner_columns_info = $self->_resolve_column_info($inner_from); |
255 | |
256 | # loop through and replace stuff that is not "ours" with a min/max func |
257 | # everything is a literal at this point, since we are likely properly |
258 | # quoted and stuff |
259 | for my $i (0 .. $#new_order) { |
260 | my $chunk = $order_chunks[$i][0]; |
261 | |
262 | # skip ourselves |
263 | next if $chunk =~ $own_re; |
264 | |
cb3e87f5 |
265 | ($chunk, my $is_desc) = $sql_maker->_split_order_chunk($chunk); |
e6977bbb |
266 | |
267 | # maybe our own unqualified column |
318e3d94 |
268 | my $ord_bit = ( |
269 | $lquote and $sep and $chunk =~ /^ $lquote ([^$sep]+) $rquote $/x |
270 | ) ? $1 : $chunk; |
271 | |
e6977bbb |
272 | next if ( |
273 | $ord_bit |
274 | and |
275 | $inner_columns_info->{$ord_bit} |
276 | and |
277 | $inner_columns_info->{$ord_bit}{-source_alias} eq $root_alias |
278 | ); |
279 | |
280 | $new_order[$i] = \[ |
1e4f9fb3 |
281 | sprintf( |
282 | '%s(%s)%s', |
283 | ($is_desc ? 'MAX' : 'MIN'), |
e6977bbb |
284 | $chunk, |
1e4f9fb3 |
285 | ($is_desc ? ' DESC' : ''), |
286 | ), |
287 | @ {$order_chunks[$i]} [ 1 .. $#{$order_chunks[$i]} ] |
288 | ]; |
289 | } |
290 | |
e6977bbb |
291 | $inner_attrs->{order_by} = \@new_order; |
292 | |
293 | # do not care about leftovers here - it will be all the functions |
294 | # we just created |
560978e2 |
295 | ($inner_attrs->{group_by}) = $self->_group_over_selection ({ |
296 | %$inner_attrs, |
297 | from => $inner_from, |
298 | select => $inner_select, |
299 | }); |
1e4f9fb3 |
300 | } |
0a3441ee |
301 | } |
d28bb90d |
302 | |
ea95892e |
303 | # we already optimized $inner_from above |
97e130fa |
304 | # and already local()ized |
305 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
306 | |
ea95892e |
307 | # generate the subquery |
6395604e |
308 | $self->_select_args_to_query ( |
ea95892e |
309 | $inner_from, |
310 | $inner_select, |
311 | $where, |
312 | $inner_attrs, |
313 | ); |
d28bb90d |
314 | }; |
315 | |
316 | # Generate the outer from - this is relatively easy (really just replace |
317 | # the join slot with the subquery), with a major caveat - we can not |
318 | # join anything that is non-selecting (not part of the prefetch), but at |
319 | # the same time is a multi-type relationship, as it will explode the result. |
320 | # |
321 | # There are two possibilities here |
322 | # - either the join is non-restricting, in which case we simply throw it away |
323 | # - it is part of the restrictions, in which case we need to collapse the outer |
324 | # result by tackling yet another group_by to the outside of the query |
325 | |
27e0370d |
326 | # work on a shallow copy |
052e8431 |
327 | $from = [ @$from ]; |
052e8431 |
328 | |
d28bb90d |
329 | my @outer_from; |
53c29913 |
330 | |
27e0370d |
331 | # we may not be the head |
97e130fa |
332 | if ($root_node_offset) { |
560978e2 |
333 | # first generate the outer_from, up and including the substitution point |
97e130fa |
334 | @outer_from = splice @$from, 0, $root_node_offset; |
27e0370d |
335 | |
336 | push @outer_from, [ |
337 | { |
1e4f9fb3 |
338 | -alias => $root_alias, |
97e130fa |
339 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
340 | $root_alias => $inner_subq, |
27e0370d |
341 | }, |
97e130fa |
342 | @{$from->[0]}[1 .. $#{$from->[0]}], |
27e0370d |
343 | ]; |
344 | } |
345 | else { |
27e0370d |
346 | @outer_from = { |
1e4f9fb3 |
347 | -alias => $root_alias, |
27e0370d |
348 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
349 | $root_alias => $inner_subq, |
27e0370d |
350 | }; |
d28bb90d |
351 | } |
352 | |
560978e2 |
353 | shift @$from; # what we just replaced above |
97e130fa |
354 | |
ea95892e |
355 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
356 | # in what role |
975b573a |
357 | my $outer_aliastypes = $outer_attrs->{_aliastypes} = |
539ffe87 |
358 | $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs ); |
052e8431 |
359 | |
a4812caa |
360 | # unroll parents |
1e4f9fb3 |
361 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
362 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
363 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
364 | |
d28bb90d |
365 | # see what's left - throw away if not selecting/restricting |
a4812caa |
366 | # also throw in a group_by if a non-selecting multiplier, |
367 | # to guard against cross-join explosions |
36fd7f07 |
368 | my $need_outer_group_by; |
d28bb90d |
369 | while (my $j = shift @$from) { |
370 | my $alias = $j->[0]{-alias}; |
371 | |
a4812caa |
372 | if ( |
373 | $outer_select_chain->{$alias} |
374 | ) { |
375 | push @outer_from, $j |
d28bb90d |
376 | } |
1e4f9fb3 |
377 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
378 | push @outer_from, $j; |
a4812caa |
379 | $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
380 | } |
381 | } |
382 | |
1e4f9fb3 |
383 | if ( $need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
36fd7f07 |
384 | my $unprocessed_order_chunks; |
560978e2 |
385 | ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ({ |
386 | %$outer_attrs, |
387 | from => \@outer_from, |
388 | select => $outer_select, |
389 | }); |
36fd7f07 |
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 |
4a0eed52 |
403 | # the outer select to exclude joins you didn't want in the first place |
d28bb90d |
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 |
4a0eed52 |
414 | # aliases are needed to fulfill the query. This information is used |
ad630f4b |
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 $aliases_by_type; |
07f31d19 |
568 | } |
569 | |
bac358c9 |
570 | # This is the engine behind { distinct => 1 } |
0a3441ee |
571 | sub _group_over_selection { |
560978e2 |
572 | my ($self, $attrs) = @_; |
0a3441ee |
573 | |
560978e2 |
574 | my $colinfos = $self->_resolve_column_info ($attrs->{from}); |
0a3441ee |
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 |
560978e2 |
581 | for (@{$attrs->{select}}) { |
0a3441ee |
582 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
583 | push @group_by, $_; |
584 | $group_index{$_}++; |
560978e2 |
585 | if ($colinfos->{$_} and $_ !~ /\./ ) { |
0a3441ee |
586 | # add a fully qualified version as well |
560978e2 |
587 | $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; |
0a3441ee |
588 | } |
07f31d19 |
589 | } |
590 | } |
ad630f4b |
591 | |
560978e2 |
592 | # add any order_by parts *from the main source* that are not already |
593 | # present in the group_by |
0a3441ee |
594 | # we need to be careful not to add any named functions/aggregates |
bac358c9 |
595 | # i.e. order_by => [ ... { count => 'foo' } ... ] |
14e26c5f |
596 | my @leftovers; |
560978e2 |
597 | for ($self->_extract_order_criteria($attrs->{order_by})) { |
0a3441ee |
598 | # only consider real columns (for functions the user got to do an explicit group_by) |
14e26c5f |
599 | if (@$_ != 1) { |
600 | push @leftovers, $_; |
601 | next; |
602 | } |
bac358c9 |
603 | my $chunk = $_->[0]; |
560978e2 |
604 | |
605 | if ( |
606 | !$colinfos->{$chunk} |
607 | or |
608 | $colinfos->{$chunk}{-source_alias} ne $attrs->{alias} |
609 | ) { |
14e26c5f |
610 | push @leftovers, $_; |
611 | next; |
560978e2 |
612 | } |
0a3441ee |
613 | |
560978e2 |
614 | $chunk = $colinfos->{$chunk}{-fq_colname}; |
0a3441ee |
615 | push @group_by, $chunk unless $group_index{$chunk}++; |
616 | } |
617 | |
14e26c5f |
618 | return wantarray |
619 | ? (\@group_by, (@leftovers ? \@leftovers : undef) ) |
620 | : \@group_by |
621 | ; |
07f31d19 |
622 | } |
623 | |
d28bb90d |
624 | sub _resolve_ident_sources { |
625 | my ($self, $ident) = @_; |
626 | |
627 | my $alias2source = {}; |
d28bb90d |
628 | |
629 | # the reason this is so contrived is that $ident may be a {from} |
630 | # structure, specifying multiple tables to join |
6298a324 |
631 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
632 | # this is compat mode for insert/update/delete which do not deal with aliases |
633 | $alias2source->{me} = $ident; |
d28bb90d |
634 | } |
635 | elsif (ref $ident eq 'ARRAY') { |
636 | |
637 | for (@$ident) { |
638 | my $tabinfo; |
639 | if (ref $_ eq 'HASH') { |
640 | $tabinfo = $_; |
d28bb90d |
641 | } |
642 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
643 | $tabinfo = $_->[0]; |
644 | } |
645 | |
4376a157 |
646 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
647 | if ($tabinfo->{-rsrc}); |
d28bb90d |
648 | } |
649 | } |
650 | |
90f10b5a |
651 | return $alias2source; |
d28bb90d |
652 | } |
653 | |
654 | # Takes $ident, \@column_names |
655 | # |
656 | # returns { $column_name => \%column_info, ... } |
657 | # also note: this adds -result_source => $rsrc to the column info |
658 | # |
09e14fdc |
659 | # If no columns_names are supplied returns info about *all* columns |
660 | # for all sources |
d28bb90d |
661 | sub _resolve_column_info { |
662 | my ($self, $ident, $colnames) = @_; |
90f10b5a |
663 | my $alias2src = $self->_resolve_ident_sources($ident); |
d28bb90d |
664 | |
52416317 |
665 | my (%seen_cols, @auto_colnames); |
d28bb90d |
666 | |
667 | # compile a global list of column names, to be able to properly |
668 | # disambiguate unqualified column names (if at all possible) |
669 | for my $alias (keys %$alias2src) { |
670 | my $rsrc = $alias2src->{$alias}; |
671 | for my $colname ($rsrc->columns) { |
672 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
673 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
674 | } |
675 | } |
676 | |
09e14fdc |
677 | $colnames ||= [ |
678 | @auto_colnames, |
679 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
680 | ]; |
681 | |
52416317 |
682 | my (%return, $colinfos); |
d28bb90d |
683 | foreach my $col (@$colnames) { |
52416317 |
684 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
685 | |
52416317 |
686 | # if the column was seen exactly once - we know which rsrc it came from |
687 | $source_alias ||= $seen_cols{$colname}[0] |
688 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
689 | |
52416317 |
690 | next unless $source_alias; |
691 | |
692 | my $rsrc = $alias2src->{$source_alias} |
693 | or next; |
694 | |
695 | $return{$col} = { |
6395604e |
696 | %{ |
697 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
698 | || |
699 | $self->throw_exception( |
700 | "No such column '$colname' on source " . $rsrc->source_name |
701 | ); |
702 | }, |
d28bb90d |
703 | -result_source => $rsrc, |
52416317 |
704 | -source_alias => $source_alias, |
81bf295c |
705 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
706 | -colname => $colname, |
d28bb90d |
707 | }; |
81bf295c |
708 | |
709 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
710 | } |
711 | |
712 | return \%return; |
713 | } |
714 | |
289ac713 |
715 | # The DBIC relationship chaining implementation is pretty simple - every |
716 | # new related_relationship is pushed onto the {from} stack, and the {select} |
717 | # window simply slides further in. This means that when we count somewhere |
718 | # in the middle, we got to make sure that everything in the join chain is an |
719 | # actual inner join, otherwise the count will come back with unpredictable |
720 | # results (a resultset may be generated with _some_ rows regardless of if |
721 | # the relation which the $rs currently selects has rows or not). E.g. |
722 | # $artist_rs->cds->count - normally generates: |
723 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
724 | # which actually returns the number of artists * (number of cds || 1) |
725 | # |
726 | # So what we do here is crawl {from}, determine if the current alias is at |
727 | # the top of the stack, and if not - make sure the chain is inner-joined down |
728 | # to the root. |
729 | # |
31a8aaaf |
730 | sub _inner_join_to_node { |
289ac713 |
731 | my ($self, $from, $alias) = @_; |
732 | |
733 | # subqueries and other oddness are naturally not supported |
734 | return $from if ( |
735 | ref $from ne 'ARRAY' |
736 | || |
737 | @$from <= 1 |
738 | || |
739 | ref $from->[0] ne 'HASH' |
740 | || |
741 | ! $from->[0]{-alias} |
742 | || |
7eb76996 |
743 | $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do |
289ac713 |
744 | ); |
745 | |
746 | # find the current $alias in the $from structure |
747 | my $switch_branch; |
748 | JOINSCAN: |
749 | for my $j (@{$from}[1 .. $#$from]) { |
750 | if ($j->[0]{-alias} eq $alias) { |
751 | $switch_branch = $j->[0]{-join_path}; |
752 | last JOINSCAN; |
753 | } |
754 | } |
755 | |
7eb76996 |
756 | # something else went quite wrong |
289ac713 |
757 | return $from unless $switch_branch; |
758 | |
759 | # So it looks like we will have to switch some stuff around. |
760 | # local() is useless here as we will be leaving the scope |
761 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
762 | # So replace the first hashref in the node arrayref manually |
289ac713 |
763 | my @new_from = ($from->[0]); |
faeb2407 |
764 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
765 | |
766 | for my $j (@{$from}[1 .. $#$from]) { |
767 | my $jalias = $j->[0]{-alias}; |
768 | |
769 | if ($sw_idx->{$jalias}) { |
770 | my %attrs = %{$j->[0]}; |
771 | delete $attrs{-join_type}; |
772 | push @new_from, [ |
773 | \%attrs, |
774 | @{$j}[ 1 .. $#$j ], |
775 | ]; |
776 | } |
777 | else { |
778 | push @new_from, $j; |
779 | } |
780 | } |
781 | |
782 | return \@new_from; |
783 | } |
784 | |
bac358c9 |
785 | sub _extract_order_criteria { |
1a736efb |
786 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
787 | |
10cef607 |
788 | $sql_maker ||= $self->sql_maker; |
789 | |
790 | my $order_dq = $sql_maker->converter->_order_by_to_dq($order_by); |
791 | |
792 | my @by; |
793 | while (is_Literal($order_dq)) { |
794 | push @by, $order_dq->{by}; |
795 | $order_dq = $order_dq->{from}; |
796 | } |
797 | |
798 | return map { [ $sql_maker->_render_dq($_) ] } @by; |
799 | |
1a736efb |
800 | my $parser = sub { |
e6977bbb |
801 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
802 | |
1a736efb |
803 | return scalar $sql_maker->_order_by_chunks ($order_by) |
804 | unless wantarray; |
c0748280 |
805 | |
e6977bbb |
806 | my ($lq, $rq, $sep) = map { quotemeta($_) } ( |
807 | ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars), |
808 | $sql_maker->name_sep |
809 | ); |
810 | |
1a736efb |
811 | my @chunks; |
bac358c9 |
812 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
813 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
814 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
815 | |
816 | # order criteria may have come back pre-quoted (literals and whatnot) |
817 | # this is fragile, but the best we can currently do |
818 | $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe |
819 | or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x; |
820 | |
1a736efb |
821 | push @chunks, $chunk; |
bac6c4fb |
822 | } |
1a736efb |
823 | |
824 | return @chunks; |
825 | }; |
826 | |
827 | if ($sql_maker) { |
828 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
829 | } |
830 | else { |
1a736efb |
831 | $sql_maker = $self->sql_maker; |
e6977bbb |
832 | |
833 | # pass these in to deal with literals coming from |
834 | # the user or the deep guts of prefetch |
835 | my $orig_quote_chars = [$sql_maker->_quote_chars]; |
836 | |
1a736efb |
837 | local $sql_maker->{quote_char}; |
e6977bbb |
838 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
839 | } |
bac6c4fb |
840 | } |
841 | |
7cec4356 |
842 | sub _order_by_is_stable { |
5f11e54f |
843 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
844 | |
5f11e54f |
845 | my $colinfo = $self->_resolve_column_info($ident, [ |
846 | (map { $_->[0] } $self->_extract_order_criteria($order_by)), |
847 | $where ? @{$self->_extract_fixed_condition_columns($where)} :(), |
848 | ]); |
c0748280 |
849 | |
7cec4356 |
850 | return undef unless keys %$colinfo; |
851 | |
852 | my $cols_per_src; |
853 | $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo; |
854 | |
855 | for (values %$cols_per_src) { |
856 | my $src = (values %$_)[0]->{-result_source}; |
857 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
858 | } |
859 | |
7cec4356 |
860 | return undef; |
861 | } |
862 | |
0e81e691 |
863 | # this is almost identical to the above, except it accepts only |
864 | # a single rsrc, and will succeed only if the first portion of the order |
865 | # by is stable. |
866 | # returns that portion as a colinfo hashref on success |
867 | sub _main_source_order_by_portion_is_stable { |
868 | my ($self, $main_rsrc, $order_by, $where) = @_; |
869 | |
870 | die "Huh... I expect a blessed result_source..." |
871 | if ref($main_rsrc) eq 'ARRAY'; |
872 | |
873 | my @ord_cols = map |
874 | { $_->[0] } |
875 | ( $self->_extract_order_criteria($order_by) ) |
876 | ; |
877 | return unless @ord_cols; |
878 | |
318e3d94 |
879 | my $colinfos = $self->_resolve_column_info($main_rsrc); |
880 | |
0e81e691 |
881 | for (0 .. $#ord_cols) { |
882 | if ( |
883 | ! $colinfos->{$ord_cols[$_]} |
884 | or |
885 | $colinfos->{$ord_cols[$_]}{-result_source} != $main_rsrc |
886 | ) { |
887 | $#ord_cols = $_ - 1; |
888 | last; |
889 | } |
890 | } |
891 | |
892 | # we just truncated it above |
893 | return unless @ord_cols; |
894 | |
0e81e691 |
895 | my $order_portion_ci = { map { |
896 | $colinfos->{$_}{-colname} => $colinfos->{$_}, |
897 | $colinfos->{$_}{-fq_colname} => $colinfos->{$_}, |
898 | } @ord_cols }; |
899 | |
318e3d94 |
900 | # since all we check here are the start of the order_by belonging to the |
901 | # top level $rsrc, a present identifying set will mean that the resultset |
902 | # is ordered by its leftmost table in a stable manner |
903 | # |
904 | # RV of _identifying_column_set contains unqualified names only |
905 | my $unqualified_idset = $main_rsrc->_identifying_column_set({ |
906 | ( $where ? %{ |
907 | $self->_resolve_column_info( |
908 | $main_rsrc, $self->_extract_fixed_condition_columns($where) |
909 | ) |
910 | } : () ), |
911 | %$order_portion_ci |
912 | }) or return; |
913 | |
914 | my $ret_info; |
915 | my %unqualified_idcols_from_order = map { |
916 | $order_portion_ci->{$_} ? ( $_ => $order_portion_ci->{$_} ) : () |
917 | } @$unqualified_idset; |
918 | |
919 | # extra optimization - cut the order_by at the end of the identifying set |
920 | # (just in case the user was stupid and overlooked the obvious) |
921 | for my $i (0 .. $#ord_cols) { |
922 | my $col = $ord_cols[$i]; |
923 | my $unqualified_colname = $order_portion_ci->{$col}{-colname}; |
924 | $ret_info->{$col} = { %{$order_portion_ci->{$col}}, -idx_in_order_subset => $i }; |
925 | delete $unqualified_idcols_from_order{$ret_info->{$col}{-colname}}; |
926 | |
927 | # we didn't reach the end of the identifying portion yet |
928 | return $ret_info unless keys %unqualified_idcols_from_order; |
929 | } |
0e81e691 |
930 | |
318e3d94 |
931 | die 'How did we get here...'; |
0e81e691 |
932 | } |
933 | |
4a0eed52 |
934 | # returns an arrayref of column names which *definitely* have some |
5f11e54f |
935 | # sort of non-nullable equality requested in the given condition |
936 | # specification. This is used to figure out if a resultset is |
937 | # constrained to a column which is part of a unique constraint, |
938 | # which in turn allows us to better predict how ordering will behave |
939 | # etc. |
940 | # |
941 | # this is a rudimentary, incomplete, and error-prone extractor |
942 | # however this is OK - it is conservative, and if we can not find |
943 | # something that is in fact there - the stack will recover gracefully |
944 | # Also - DQ and the mst it rode in on will save us all RSN!!! |
945 | sub _extract_fixed_condition_columns { |
946 | my ($self, $where, $nested) = @_; |
947 | |
948 | return unless ref $where eq 'HASH'; |
949 | |
950 | my @cols; |
951 | for my $lhs (keys %$where) { |
952 | if ($lhs =~ /^\-and$/i) { |
953 | push @cols, ref $where->{$lhs} eq 'ARRAY' |
954 | ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} ) |
955 | : $self->_extract_fixed_condition_columns($where->{$lhs}, 1) |
956 | ; |
957 | } |
958 | elsif ($lhs !~ /^\-/) { |
959 | my $val = $where->{$lhs}; |
960 | |
961 | push @cols, $lhs if (defined $val and ( |
962 | ! ref $val |
963 | or |
964 | (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='}) |
965 | )); |
966 | } |
967 | } |
968 | return $nested ? @cols : \@cols; |
c0748280 |
969 | } |
bac6c4fb |
970 | |
d28bb90d |
971 | 1; |