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 | |
75 | $self->throw_exception ('Nothing to prefetch... how did we get here?!') |
a59246c3 |
76 | if not @{$attrs->{_prefetch_selector_range}||[]}; |
d28bb90d |
77 | |
78 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
79 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY'); |
80 | |
d28bb90d |
81 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
82 | my $outer_attrs = { %$attrs }; |
83 | delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/; |
84 | |
186ba34c |
85 | my $inner_attrs = { %$attrs }; |
908aa1bb |
86 | delete $inner_attrs->{$_} for qw/for collapse _prefetch_selector_range select as/; |
d28bb90d |
87 | |
4df1400e |
88 | # there is no point of ordering the insides if there is no limit |
89 | delete $inner_attrs->{order_by} if ( |
90 | delete $inner_attrs->{_order_is_artificial} |
91 | or |
92 | ! $inner_attrs->{rows} |
93 | ); |
946f6260 |
94 | |
d28bb90d |
95 | # generate the inner/outer select lists |
96 | # for inside we consider only stuff *not* brought in by the prefetch |
97 | # on the outside we substitute any function for its alias |
98 | my $outer_select = [ @$select ]; |
97e130fa |
99 | my $inner_select; |
36fd7f07 |
100 | |
97e130fa |
101 | my ($root_node, $root_node_offset); |
27e0370d |
102 | |
103 | for my $i (0 .. $#$from) { |
104 | my $node = $from->[$i]; |
105 | my $h = (ref $node eq 'HASH') ? $node |
106 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
107 | : next |
108 | ; |
109 | |
97e130fa |
110 | if ( ($h->{-alias}||'') eq $attrs->{alias} and $h->{-rsrc} ) { |
111 | $root_node = $h; |
112 | $root_node_offset = $i; |
27e0370d |
113 | last; |
114 | } |
115 | } |
116 | |
117 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
118 | unless $root_node; |
27e0370d |
119 | |
120 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
121 | my $colinfo = $self->_resolve_column_info($from); |
122 | my $selected_root_columns; |
123 | |
36fd7f07 |
124 | my ($p_start, $p_end) = @{$outer_attrs->{_prefetch_selector_range}}; |
125 | for my $i (0 .. $p_start - 1, $p_end + 1 .. $#$outer_select) { |
d28bb90d |
126 | my $sel = $outer_select->[$i]; |
127 | |
128 | if (ref $sel eq 'HASH' ) { |
129 | $sel->{-as} ||= $attrs->{as}[$i]; |
130 | $outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") ); |
131 | } |
27e0370d |
132 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
133 | $selected_root_columns->{$ci->{-colname}} = 1; |
134 | } |
d28bb90d |
135 | |
136 | push @$inner_select, $sel; |
bb9bffea |
137 | |
138 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
139 | } |
140 | |
97e130fa |
141 | # We will need to fetch all native columns in the inner subquery, which may |
142 | # be a part of an *outer* join condition, or an order_by (which needs to be |
143 | # preserved outside) |
144 | # We can not just fetch everything because a potential has_many restricting |
145 | # join collapse *will not work* on heavy data types. |
146 | my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args( |
147 | [grep { ref($_) eq 'ARRAY' or ref($_) eq 'HASH' } @{$from}[$root_node_offset .. $#$from]], |
148 | [], |
149 | $where, |
150 | $inner_attrs |
151 | ); |
152 | |
153 | for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { |
154 | my $ci = $colinfo->{$_} or next; |
155 | if ( |
156 | $ci->{-source_alias} eq $attrs->{alias} |
157 | and |
158 | ! $selected_root_columns->{$ci->{-colname}}++ |
159 | ) { |
160 | # adding it to both to keep limits not supporting dark selectors happy |
161 | push @$inner_select, $ci->{-fq_colname}; |
162 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
163 | } |
164 | } |
165 | |
ea95892e |
166 | # construct the inner $from and lock it in a subquery |
48580715 |
167 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
168 | # throw away all non-selecting, non-restricting multijoins |
169 | # (since we def. do not care about multiplication those inside the subquery) |
6395604e |
170 | my $inner_subq = do { |
ea95892e |
171 | |
172 | # must use it here regardless of user requests |
173 | local $self->{_use_join_optimizer} = 1; |
174 | |
97e130fa |
175 | # throw away multijoins since we def. do not care about those inside the subquery |
176 | my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, $inner_attrs, 'ignore_multiplication'); |
ea95892e |
177 | |
887a0aef |
178 | my $inner_aliastypes = |
179 | $self->_resolve_aliastypes_from_select_args( $inner_from, $inner_select, $where, $inner_attrs ); |
180 | |
a4812caa |
181 | # we need to simulate collapse in the subq if a multiplying join is pulled |
182 | # by being a non-selecting restrictor |
0a3441ee |
183 | if ( |
184 | ! $inner_attrs->{group_by} |
185 | and |
887a0aef |
186 | first { |
187 | $inner_aliastypes->{restricting}{$_} |
188 | and |
189 | ! $inner_aliastypes->{selecting}{$_} |
190 | } ( keys %{$inner_aliastypes->{multiplying}||{}} ) |
0a3441ee |
191 | ) { |
14e26c5f |
192 | my $unprocessed_order_chunks; |
193 | ($inner_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( |
0a3441ee |
194 | $inner_from, $inner_select, $inner_attrs->{order_by} |
195 | ); |
14e26c5f |
196 | |
197 | $self->throw_exception ( |
198 | 'A required group_by clause could not be constructed automatically due to a complex ' |
199 | . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' |
200 | . 'group_by by hand' |
201 | ) if $unprocessed_order_chunks; |
0a3441ee |
202 | } |
d28bb90d |
203 | |
ea95892e |
204 | # we already optimized $inner_from above |
97e130fa |
205 | # and already local()ized |
206 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
207 | |
ea95892e |
208 | # generate the subquery |
6395604e |
209 | $self->_select_args_to_query ( |
ea95892e |
210 | $inner_from, |
211 | $inner_select, |
212 | $where, |
213 | $inner_attrs, |
214 | ); |
d28bb90d |
215 | }; |
216 | |
217 | # Generate the outer from - this is relatively easy (really just replace |
218 | # the join slot with the subquery), with a major caveat - we can not |
219 | # join anything that is non-selecting (not part of the prefetch), but at |
220 | # the same time is a multi-type relationship, as it will explode the result. |
221 | # |
222 | # There are two possibilities here |
223 | # - either the join is non-restricting, in which case we simply throw it away |
224 | # - it is part of the restrictions, in which case we need to collapse the outer |
225 | # result by tackling yet another group_by to the outside of the query |
226 | |
27e0370d |
227 | # work on a shallow copy |
052e8431 |
228 | $from = [ @$from ]; |
052e8431 |
229 | |
d28bb90d |
230 | my @outer_from; |
53c29913 |
231 | |
27e0370d |
232 | # we may not be the head |
97e130fa |
233 | if ($root_node_offset) { |
27e0370d |
234 | # first generate the outer_from, up to the substitution point |
97e130fa |
235 | @outer_from = splice @$from, 0, $root_node_offset; |
27e0370d |
236 | |
237 | push @outer_from, [ |
238 | { |
239 | -alias => $attrs->{alias}, |
97e130fa |
240 | -rsrc => $root_node->{-rsrc}, |
27e0370d |
241 | $attrs->{alias} => $inner_subq, |
242 | }, |
97e130fa |
243 | @{$from->[0]}[1 .. $#{$from->[0]}], |
27e0370d |
244 | ]; |
245 | } |
246 | else { |
27e0370d |
247 | @outer_from = { |
248 | -alias => $attrs->{alias}, |
249 | -rsrc => $root_node->{-rsrc}, |
250 | $attrs->{alias} => $inner_subq, |
251 | }; |
d28bb90d |
252 | } |
253 | |
97e130fa |
254 | shift @$from; # it's replaced in @outer_from already |
255 | |
ea95892e |
256 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
257 | # in what role |
258 | my $outer_aliastypes = |
539ffe87 |
259 | $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs ); |
052e8431 |
260 | |
a4812caa |
261 | # unroll parents |
262 | my ($outer_select_chain, $outer_restrict_chain) = map { +{ |
97e130fa |
263 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} } |
a4812caa |
264 | } } qw/selecting restricting/; |
265 | |
d28bb90d |
266 | # see what's left - throw away if not selecting/restricting |
a4812caa |
267 | # also throw in a group_by if a non-selecting multiplier, |
268 | # to guard against cross-join explosions |
36fd7f07 |
269 | my $need_outer_group_by; |
d28bb90d |
270 | while (my $j = shift @$from) { |
271 | my $alias = $j->[0]{-alias}; |
272 | |
a4812caa |
273 | if ( |
274 | $outer_select_chain->{$alias} |
275 | ) { |
276 | push @outer_from, $j |
d28bb90d |
277 | } |
a4812caa |
278 | elsif ($outer_restrict_chain->{$alias}) { |
d28bb90d |
279 | push @outer_from, $j; |
a4812caa |
280 | $need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
281 | } |
282 | } |
283 | |
36fd7f07 |
284 | if ($need_outer_group_by and ! $outer_attrs->{group_by}) { |
285 | |
286 | my $unprocessed_order_chunks; |
287 | ($outer_attrs->{group_by}, $unprocessed_order_chunks) = $self->_group_over_selection ( |
288 | \@outer_from, $outer_select, $outer_attrs->{order_by} |
289 | ); |
290 | |
291 | $self->throw_exception ( |
292 | 'A required group_by clause could not be constructed automatically due to a complex ' |
293 | . 'order_by criteria. Either order_by columns only (no functions) or construct a suitable ' |
294 | . 'group_by by hand' |
295 | ) if $unprocessed_order_chunks; |
296 | |
297 | } |
298 | |
d28bb90d |
299 | # This is totally horrific - the $where ends up in both the inner and outer query |
300 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
301 | # then if where conditions apply to the *right* side of the prefetch, you may have |
302 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
303 | # the outer select to exclude joins you didin't want in the first place |
304 | # |
305 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
306 | return (\@outer_from, $outer_select, $where, $outer_attrs); |
307 | } |
308 | |
1a736efb |
309 | # |
310 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
311 | # |
ad630f4b |
312 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
313 | # select/where/order/group attributes, in order to determine what |
314 | # aliases are neded to fulfill the query. This information is used |
315 | # throughout the code to prune unnecessary JOINs from the queries |
316 | # in an attempt to reduce the execution time. |
317 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
318 | # happen is for it to fail due to some scalar SQL, which in turn will |
319 | # result in a vocal exception. |
539ffe87 |
320 | sub _resolve_aliastypes_from_select_args { |
052e8431 |
321 | my ( $self, $from, $select, $where, $attrs ) = @_; |
546f1cd9 |
322 | |
ad630f4b |
323 | $self->throw_exception ('Unable to analyze custom {from}') |
324 | if ref $from ne 'ARRAY'; |
546f1cd9 |
325 | |
ad630f4b |
326 | # what we will return |
964a3c71 |
327 | my $aliases_by_type; |
546f1cd9 |
328 | |
ad630f4b |
329 | # see what aliases are there to work with |
330 | my $alias_list; |
539ffe87 |
331 | for (@$from) { |
332 | my $j = $_; |
ad630f4b |
333 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
334 | my $al = $j->{-alias} |
335 | or next; |
336 | |
337 | $alias_list->{$al} = $j; |
97e130fa |
338 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } if ( |
a4812caa |
339 | # not array == {from} head == can't be multiplying |
340 | ( ref($_) eq 'ARRAY' and ! $j->{-is_single} ) |
341 | or |
342 | # a parent of ours is already a multiplier |
343 | ( grep { $aliases_by_type->{multiplying}{$_} } @{ $j->{-join_path}||[] } ) |
344 | ); |
546f1cd9 |
345 | } |
546f1cd9 |
346 | |
1a736efb |
347 | # get a column to source/alias map (including unqualified ones) |
348 | my $colinfo = $self->_resolve_column_info ($from); |
349 | |
ad630f4b |
350 | # set up a botched SQLA |
351 | my $sql_maker = $self->sql_maker; |
07f31d19 |
352 | |
4c2b30d6 |
353 | # these are throw away results, do not pollute the bind stack |
4c2b30d6 |
354 | local $sql_maker->{select_bind}; |
0542ec57 |
355 | local $sql_maker->{where_bind}; |
356 | local $sql_maker->{group_bind}; |
357 | local $sql_maker->{having_bind}; |
97e130fa |
358 | local $sql_maker->{from_bind}; |
3f5b99fe |
359 | |
360 | # we can't scan properly without any quoting (\b doesn't cut it |
361 | # everywhere), so unless there is proper quoting set - use our |
362 | # own weird impossible character. |
363 | # Also in the case of no quoting, we need to explicitly disable |
364 | # name_sep, otherwise sorry nasty legacy syntax like |
365 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
366 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
367 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
368 | |
369 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
370 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
371 | # if we don't unset it we screw up retarded but unfortunately working |
372 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
373 | $sql_maker->{name_sep} = ''; |
374 | } |
375 | |
376 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
377 | |
1a736efb |
378 | # generate sql chunks |
379 | my $to_scan = { |
380 | restricting => [ |
381 | $sql_maker->_recurse_where ($where), |
a7e643b1 |
382 | $sql_maker->_parse_rs_attrs ({ |
1a736efb |
383 | map { $_ => $attrs->{$_} } (qw/group_by having/) |
384 | }), |
385 | ], |
97e130fa |
386 | joining => [ |
387 | $sql_maker->_recurse_from ( |
388 | ref $from->[0] eq 'ARRAY' ? $from->[0][0] : $from->[0], |
389 | @{$from}[1 .. $#$from], |
390 | ), |
391 | ], |
1a736efb |
392 | selecting => [ |
1a736efb |
393 | $sql_maker->_recurse_fields ($select), |
bac358c9 |
394 | ( map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker) ), |
1a736efb |
395 | ], |
396 | }; |
07f31d19 |
397 | |
1a736efb |
398 | # throw away empty chunks |
399 | $_ = [ map { $_ || () } @$_ ] for values %$to_scan; |
07f31d19 |
400 | |
1a736efb |
401 | # first loop through all fully qualified columns and get the corresponding |
402 | # alias (should work even if they are in scalarrefs) |
ad630f4b |
403 | for my $alias (keys %$alias_list) { |
1a736efb |
404 | my $al_re = qr/ |
97e130fa |
405 | $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
1a736efb |
406 | | |
97e130fa |
407 | \b $alias \. ([^\s\)\($rquote]+)? |
1a736efb |
408 | /x; |
409 | |
1a736efb |
410 | for my $type (keys %$to_scan) { |
411 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
412 | if (my @matches = $piece =~ /$al_re/g) { |
413 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
414 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = 1 |
415 | for grep { defined $_ } @matches; |
416 | } |
1a736efb |
417 | } |
ad630f4b |
418 | } |
1a736efb |
419 | } |
420 | |
421 | # now loop through unqualified column names, and try to locate them within |
422 | # the chunks |
423 | for my $col (keys %$colinfo) { |
3f5b99fe |
424 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
425 | |
97e130fa |
426 | my $col_re = qr/ $lquote ($col) $rquote /x; |
07f31d19 |
427 | |
1a736efb |
428 | for my $type (keys %$to_scan) { |
429 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
430 | if (my @matches = $piece =~ /$col_re/g) { |
a4812caa |
431 | my $alias = $colinfo->{$col}{-source_alias}; |
97e130fa |
432 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
433 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = 1 |
434 | for grep { defined $_ } @matches; |
a4812caa |
435 | } |
1a736efb |
436 | } |
07f31d19 |
437 | } |
438 | } |
439 | |
440 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
441 | for my $j (values %$alias_list) { |
07f31d19 |
442 | my $alias = $j->{-alias} or next; |
97e130fa |
443 | $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if ( |
07f31d19 |
444 | (not $j->{-join_type}) |
445 | or |
446 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
447 | ); |
448 | } |
449 | |
964a3c71 |
450 | return $aliases_by_type; |
07f31d19 |
451 | } |
452 | |
bac358c9 |
453 | # This is the engine behind { distinct => 1 } |
0a3441ee |
454 | sub _group_over_selection { |
455 | my ($self, $from, $select, $order_by) = @_; |
456 | |
457 | my $rs_column_list = $self->_resolve_column_info ($from); |
458 | |
459 | my (@group_by, %group_index); |
460 | |
36fd7f07 |
461 | # the logic is: if it is a { func => val } we assume an aggregate, |
462 | # otherwise if \'...' or \[...] we assume the user knows what is |
463 | # going on thus group over it |
0a3441ee |
464 | for (@$select) { |
465 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
466 | push @group_by, $_; |
467 | $group_index{$_}++; |
468 | if ($rs_column_list->{$_} and $_ !~ /\./ ) { |
469 | # add a fully qualified version as well |
470 | $group_index{"$rs_column_list->{$_}{-source_alias}.$_"}++; |
471 | } |
07f31d19 |
472 | } |
473 | } |
ad630f4b |
474 | |
0a3441ee |
475 | # add any order_by parts that are not already present in the group_by |
476 | # we need to be careful not to add any named functions/aggregates |
bac358c9 |
477 | # i.e. order_by => [ ... { count => 'foo' } ... ] |
14e26c5f |
478 | my @leftovers; |
bac358c9 |
479 | for ($self->_extract_order_criteria($order_by)) { |
0a3441ee |
480 | # only consider real columns (for functions the user got to do an explicit group_by) |
14e26c5f |
481 | if (@$_ != 1) { |
482 | push @leftovers, $_; |
483 | next; |
484 | } |
bac358c9 |
485 | my $chunk = $_->[0]; |
14e26c5f |
486 | my $colinfo = $rs_column_list->{$chunk} or do { |
487 | push @leftovers, $_; |
488 | next; |
489 | }; |
0a3441ee |
490 | |
491 | $chunk = "$colinfo->{-source_alias}.$chunk" if $chunk !~ /\./; |
492 | push @group_by, $chunk unless $group_index{$chunk}++; |
493 | } |
494 | |
14e26c5f |
495 | return wantarray |
496 | ? (\@group_by, (@leftovers ? \@leftovers : undef) ) |
497 | : \@group_by |
498 | ; |
07f31d19 |
499 | } |
500 | |
d28bb90d |
501 | sub _resolve_ident_sources { |
502 | my ($self, $ident) = @_; |
503 | |
504 | my $alias2source = {}; |
505 | my $rs_alias; |
506 | |
507 | # the reason this is so contrived is that $ident may be a {from} |
508 | # structure, specifying multiple tables to join |
6298a324 |
509 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
510 | # this is compat mode for insert/update/delete which do not deal with aliases |
511 | $alias2source->{me} = $ident; |
512 | $rs_alias = 'me'; |
513 | } |
514 | elsif (ref $ident eq 'ARRAY') { |
515 | |
516 | for (@$ident) { |
517 | my $tabinfo; |
518 | if (ref $_ eq 'HASH') { |
519 | $tabinfo = $_; |
520 | $rs_alias = $tabinfo->{-alias}; |
521 | } |
522 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
523 | $tabinfo = $_->[0]; |
524 | } |
525 | |
4376a157 |
526 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
527 | if ($tabinfo->{-rsrc}); |
d28bb90d |
528 | } |
529 | } |
530 | |
531 | return ($alias2source, $rs_alias); |
532 | } |
533 | |
534 | # Takes $ident, \@column_names |
535 | # |
536 | # returns { $column_name => \%column_info, ... } |
537 | # also note: this adds -result_source => $rsrc to the column info |
538 | # |
09e14fdc |
539 | # If no columns_names are supplied returns info about *all* columns |
540 | # for all sources |
d28bb90d |
541 | sub _resolve_column_info { |
542 | my ($self, $ident, $colnames) = @_; |
543 | my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); |
544 | |
52416317 |
545 | my (%seen_cols, @auto_colnames); |
d28bb90d |
546 | |
547 | # compile a global list of column names, to be able to properly |
548 | # disambiguate unqualified column names (if at all possible) |
549 | for my $alias (keys %$alias2src) { |
550 | my $rsrc = $alias2src->{$alias}; |
551 | for my $colname ($rsrc->columns) { |
552 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
553 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
554 | } |
555 | } |
556 | |
09e14fdc |
557 | $colnames ||= [ |
558 | @auto_colnames, |
559 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
560 | ]; |
561 | |
52416317 |
562 | my (%return, $colinfos); |
d28bb90d |
563 | foreach my $col (@$colnames) { |
52416317 |
564 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
565 | |
52416317 |
566 | # if the column was seen exactly once - we know which rsrc it came from |
567 | $source_alias ||= $seen_cols{$colname}[0] |
568 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
569 | |
52416317 |
570 | next unless $source_alias; |
571 | |
572 | my $rsrc = $alias2src->{$source_alias} |
573 | or next; |
574 | |
575 | $return{$col} = { |
6395604e |
576 | %{ |
577 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
578 | || |
579 | $self->throw_exception( |
580 | "No such column '$colname' on source " . $rsrc->source_name |
581 | ); |
582 | }, |
d28bb90d |
583 | -result_source => $rsrc, |
52416317 |
584 | -source_alias => $source_alias, |
81bf295c |
585 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
586 | -colname => $colname, |
d28bb90d |
587 | }; |
81bf295c |
588 | |
589 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
590 | } |
591 | |
592 | return \%return; |
593 | } |
594 | |
289ac713 |
595 | # The DBIC relationship chaining implementation is pretty simple - every |
596 | # new related_relationship is pushed onto the {from} stack, and the {select} |
597 | # window simply slides further in. This means that when we count somewhere |
598 | # in the middle, we got to make sure that everything in the join chain is an |
599 | # actual inner join, otherwise the count will come back with unpredictable |
600 | # results (a resultset may be generated with _some_ rows regardless of if |
601 | # the relation which the $rs currently selects has rows or not). E.g. |
602 | # $artist_rs->cds->count - normally generates: |
603 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
604 | # which actually returns the number of artists * (number of cds || 1) |
605 | # |
606 | # So what we do here is crawl {from}, determine if the current alias is at |
607 | # the top of the stack, and if not - make sure the chain is inner-joined down |
608 | # to the root. |
609 | # |
31a8aaaf |
610 | sub _inner_join_to_node { |
289ac713 |
611 | my ($self, $from, $alias) = @_; |
612 | |
613 | # subqueries and other oddness are naturally not supported |
614 | return $from if ( |
615 | ref $from ne 'ARRAY' |
616 | || |
617 | @$from <= 1 |
618 | || |
619 | ref $from->[0] ne 'HASH' |
620 | || |
621 | ! $from->[0]{-alias} |
622 | || |
7eb76996 |
623 | $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do |
289ac713 |
624 | ); |
625 | |
626 | # find the current $alias in the $from structure |
627 | my $switch_branch; |
628 | JOINSCAN: |
629 | for my $j (@{$from}[1 .. $#$from]) { |
630 | if ($j->[0]{-alias} eq $alias) { |
631 | $switch_branch = $j->[0]{-join_path}; |
632 | last JOINSCAN; |
633 | } |
634 | } |
635 | |
7eb76996 |
636 | # something else went quite wrong |
289ac713 |
637 | return $from unless $switch_branch; |
638 | |
639 | # So it looks like we will have to switch some stuff around. |
640 | # local() is useless here as we will be leaving the scope |
641 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
642 | # So replace the first hashref in the node arrayref manually |
289ac713 |
643 | my @new_from = ($from->[0]); |
faeb2407 |
644 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
645 | |
646 | for my $j (@{$from}[1 .. $#$from]) { |
647 | my $jalias = $j->[0]{-alias}; |
648 | |
649 | if ($sw_idx->{$jalias}) { |
650 | my %attrs = %{$j->[0]}; |
651 | delete $attrs{-join_type}; |
652 | push @new_from, [ |
653 | \%attrs, |
654 | @{$j}[ 1 .. $#$j ], |
655 | ]; |
656 | } |
657 | else { |
658 | push @new_from, $j; |
659 | } |
660 | } |
661 | |
662 | return \@new_from; |
663 | } |
664 | |
bac358c9 |
665 | sub _extract_order_criteria { |
1a736efb |
666 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
667 | |
1a736efb |
668 | my $parser = sub { |
669 | my ($sql_maker, $order_by) = @_; |
c0748280 |
670 | |
1a736efb |
671 | return scalar $sql_maker->_order_by_chunks ($order_by) |
672 | unless wantarray; |
c0748280 |
673 | |
1a736efb |
674 | my @chunks; |
bac358c9 |
675 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
676 | my $chunk = ref $_ ? $_ : [ $_ ]; |
677 | $chunk->[0] =~ s/\s+ (?: ASC|DESC ) \s* $//ix; |
1a736efb |
678 | push @chunks, $chunk; |
bac6c4fb |
679 | } |
1a736efb |
680 | |
681 | return @chunks; |
682 | }; |
683 | |
684 | if ($sql_maker) { |
685 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
686 | } |
687 | else { |
1a736efb |
688 | $sql_maker = $self->sql_maker; |
689 | local $sql_maker->{quote_char}; |
690 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
691 | } |
bac6c4fb |
692 | } |
693 | |
7cec4356 |
694 | sub _order_by_is_stable { |
5f11e54f |
695 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
696 | |
5f11e54f |
697 | my $colinfo = $self->_resolve_column_info($ident, [ |
698 | (map { $_->[0] } $self->_extract_order_criteria($order_by)), |
699 | $where ? @{$self->_extract_fixed_condition_columns($where)} :(), |
700 | ]); |
c0748280 |
701 | |
7cec4356 |
702 | return undef unless keys %$colinfo; |
703 | |
704 | my $cols_per_src; |
705 | $cols_per_src->{$_->{-source_alias}}{$_->{-colname}} = $_ for values %$colinfo; |
706 | |
707 | for (values %$cols_per_src) { |
708 | my $src = (values %$_)[0]->{-result_source}; |
709 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
710 | } |
711 | |
7cec4356 |
712 | return undef; |
713 | } |
714 | |
5f11e54f |
715 | # returns an arrayref of column names which *definitely* have som |
716 | # sort of non-nullable equality requested in the given condition |
717 | # specification. This is used to figure out if a resultset is |
718 | # constrained to a column which is part of a unique constraint, |
719 | # which in turn allows us to better predict how ordering will behave |
720 | # etc. |
721 | # |
722 | # this is a rudimentary, incomplete, and error-prone extractor |
723 | # however this is OK - it is conservative, and if we can not find |
724 | # something that is in fact there - the stack will recover gracefully |
725 | # Also - DQ and the mst it rode in on will save us all RSN!!! |
726 | sub _extract_fixed_condition_columns { |
727 | my ($self, $where, $nested) = @_; |
728 | |
729 | return unless ref $where eq 'HASH'; |
730 | |
731 | my @cols; |
732 | for my $lhs (keys %$where) { |
733 | if ($lhs =~ /^\-and$/i) { |
734 | push @cols, ref $where->{$lhs} eq 'ARRAY' |
735 | ? ( map { $self->_extract_fixed_condition_columns($_, 1) } @{$where->{$lhs}} ) |
736 | : $self->_extract_fixed_condition_columns($where->{$lhs}, 1) |
737 | ; |
738 | } |
739 | elsif ($lhs !~ /^\-/) { |
740 | my $val = $where->{$lhs}; |
741 | |
742 | push @cols, $lhs if (defined $val and ( |
743 | ! ref $val |
744 | or |
745 | (ref $val eq 'HASH' and keys %$val == 1 and defined $val->{'='}) |
746 | )); |
747 | } |
748 | } |
749 | return $nested ? @cols : \@cols; |
c0748280 |
750 | } |
bac6c4fb |
751 | |
d28bb90d |
752 | 1; |