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 { |
e1861c2c |
26 | my ($self, $attrs) = @_; |
ea95892e |
27 | |
e1861c2c |
28 | # only standard {from} specs are supported, and we could be disabled in general |
29 | return ($attrs->{from}, {}) unless ( |
30 | ref $attrs->{from} eq 'ARRAY' |
31 | and |
32 | @{$attrs->{from}} > 1 |
33 | and |
34 | ref $attrs->{from}[0] eq 'HASH' |
35 | and |
36 | ref $attrs->{from}[1] eq 'ARRAY' |
37 | and |
38 | $self->_use_join_optimizer |
39 | ); |
052e8431 |
40 | |
eb58c082 |
41 | my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs); |
4b1b5ea3 |
42 | |
eb58c082 |
43 | my $new_aliastypes = { %$orig_aliastypes }; |
44 | |
45 | # we will be recreating this entirely |
46 | my @reclassify = 'joining'; |
97e130fa |
47 | |
4b1b5ea3 |
48 | # a grouped set will not be affected by amount of rows. Thus any |
eb58c082 |
49 | # purely multiplicator classifications can go |
50 | # (will be reintroduced below if needed by something else) |
51 | push @reclassify, qw(multiplying premultiplied) |
437a9cfa |
52 | if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by}; |
4b1b5ea3 |
53 | |
eb58c082 |
54 | # nuke what will be recalculated |
55 | delete @{$new_aliastypes}{@reclassify}; |
56 | |
e1861c2c |
57 | my @newfrom = $attrs->{from}[0]; # FROM head is always present |
052e8431 |
58 | |
eb58c082 |
59 | # recalculate what we need once the multipliers are potentially gone |
60 | # ignore premultiplies, since they do not add any value to anything |
a4812caa |
61 | my %need_joins; |
eb58c082 |
62 | for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) { |
a4812caa |
63 | # add all requested aliases |
64 | $need_joins{$_} = 1 for keys %$_; |
65 | |
66 | # add all their parents (as per joinpath which is an AoH { table => alias }) |
97e130fa |
67 | $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; |
a4812caa |
68 | } |
97e130fa |
69 | |
e1861c2c |
70 | for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) { |
539ffe87 |
71 | push @newfrom, $j if ( |
a6ef93cb |
72 | (! defined $j->[0]{-alias}) # legacy crap |
539ffe87 |
73 | || |
74 | $need_joins{$j->[0]{-alias}} |
75 | ); |
052e8431 |
76 | } |
77 | |
eb58c082 |
78 | # we have a new set of joiners - for everything we nuked pull the classification |
79 | # off the original stack |
80 | for my $ctype (@reclassify) { |
81 | $new_aliastypes->{$ctype} = { map |
82 | { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () } |
83 | keys %{$orig_aliastypes->{$ctype}} |
84 | } |
85 | } |
86 | |
87 | return ( \@newfrom, $new_aliastypes ); |
052e8431 |
88 | } |
89 | |
052e8431 |
90 | # |
d28bb90d |
91 | # This is the code producing joined subqueries like: |
8273e845 |
92 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
d28bb90d |
93 | # |
94 | sub _adjust_select_args_for_complex_prefetch { |
e1861c2c |
95 | my ($self, $attrs) = @_; |
d28bb90d |
96 | |
e1861c2c |
97 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless ( |
98 | ref $attrs->{from} eq 'ARRAY' |
99 | and |
100 | @{$attrs->{from}} > 1 |
101 | and |
102 | ref $attrs->{from}[0] eq 'HASH' |
103 | and |
104 | ref $attrs->{from}[1] eq 'ARRAY' |
105 | ); |
d28bb90d |
106 | |
1e4f9fb3 |
107 | my $root_alias = $attrs->{alias}; |
108 | |
d28bb90d |
109 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
110 | my $outer_attrs = { %$attrs }; |
e1861c2c |
111 | delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)}; |
d28bb90d |
112 | |
186ba34c |
113 | my $inner_attrs = { %$attrs }; |
e1861c2c |
114 | delete @{$inner_attrs}{qw(for collapse select as _related_results_construction)}; |
d28bb90d |
115 | |
4df1400e |
116 | # there is no point of ordering the insides if there is no limit |
117 | delete $inner_attrs->{order_by} if ( |
118 | delete $inner_attrs->{_order_is_artificial} |
119 | or |
120 | ! $inner_attrs->{rows} |
121 | ); |
946f6260 |
122 | |
d28bb90d |
123 | # generate the inner/outer select lists |
124 | # for inside we consider only stuff *not* brought in by the prefetch |
125 | # on the outside we substitute any function for its alias |
e1861c2c |
126 | $outer_attrs->{select} = [ @{$attrs->{select}} ]; |
36fd7f07 |
127 | |
97e130fa |
128 | my ($root_node, $root_node_offset); |
27e0370d |
129 | |
e1861c2c |
130 | for my $i (0 .. $#{$inner_attrs->{from}}) { |
131 | my $node = $inner_attrs->{from}[$i]; |
27e0370d |
132 | my $h = (ref $node eq 'HASH') ? $node |
133 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
134 | : next |
135 | ; |
136 | |
1e4f9fb3 |
137 | if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) { |
97e130fa |
138 | $root_node = $h; |
139 | $root_node_offset = $i; |
27e0370d |
140 | last; |
141 | } |
142 | } |
143 | |
144 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
145 | unless $root_node; |
27e0370d |
146 | |
147 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
e1861c2c |
148 | my $colinfo = $self->_resolve_column_info($inner_attrs->{from}); |
27e0370d |
149 | my $selected_root_columns; |
150 | |
e1861c2c |
151 | for my $i (0 .. $#{$outer_attrs->{select}}) { |
152 | my $sel = $outer_attrs->{select}->[$i]; |
d28bb90d |
153 | |
1e4f9fb3 |
154 | next if ( |
155 | $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias |
156 | ); |
157 | |
d28bb90d |
158 | if (ref $sel eq 'HASH' ) { |
159 | $sel->{-as} ||= $attrs->{as}[$i]; |
e1861c2c |
160 | $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); |
d28bb90d |
161 | } |
27e0370d |
162 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
163 | $selected_root_columns->{$ci->{-colname}} = 1; |
164 | } |
d28bb90d |
165 | |
e1861c2c |
166 | push @{$inner_attrs->{select}}, $sel; |
bb9bffea |
167 | |
168 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
169 | } |
170 | |
97e130fa |
171 | # We will need to fetch all native columns in the inner subquery, which may |
172 | # be a part of an *outer* join condition, or an order_by (which needs to be |
e1861c2c |
173 | # preserved outside), or wheres. In other words everything but the inner |
174 | # selector |
97e130fa |
175 | # We can not just fetch everything because a potential has_many restricting |
176 | # join collapse *will not work* on heavy data types. |
e1861c2c |
177 | my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({ |
178 | %$inner_attrs, |
179 | select => [], |
180 | }); |
97e130fa |
181 | |
182 | for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { |
183 | my $ci = $colinfo->{$_} or next; |
184 | if ( |
1e4f9fb3 |
185 | $ci->{-source_alias} eq $root_alias |
97e130fa |
186 | and |
187 | ! $selected_root_columns->{$ci->{-colname}}++ |
188 | ) { |
189 | # adding it to both to keep limits not supporting dark selectors happy |
e1861c2c |
190 | push @{$inner_attrs->{select}}, $ci->{-fq_colname}; |
97e130fa |
191 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
192 | } |
193 | } |
194 | |
e1861c2c |
195 | # construct the inner {from} and lock it in a subquery |
48580715 |
196 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
197 | # throw away all non-selecting, non-restricting multijoins |
eb58c082 |
198 | # (since we def. do not care about multiplication of the contents of the subquery) |
6395604e |
199 | my $inner_subq = do { |
ea95892e |
200 | |
eb58c082 |
201 | # must use it here regardless of user requests (vastly gentler on optimizer) |
ea95892e |
202 | local $self->{_use_join_optimizer} = 1; |
203 | |
97e130fa |
204 | # throw away multijoins since we def. do not care about those inside the subquery |
e1861c2c |
205 | ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({ |
437a9cfa |
206 | %$inner_attrs, _force_prune_multiplying_joins => 1 |
207 | }); |
ea95892e |
208 | |
eb58c082 |
209 | # uh-oh a multiplier (which is not us) left in, this is a problem for limits |
210 | # we will need to add a group_by to collapse the resultset for proper counts |
0a3441ee |
211 | if ( |
eb58c082 |
212 | grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} } |
1e4f9fb3 |
213 | and |
560978e2 |
214 | # if there are user-supplied groups - assume user knows wtf they are up to |
215 | ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} ) |
0a3441ee |
216 | ) { |
1e4f9fb3 |
217 | |
eb58c082 |
218 | my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} }; |
1e4f9fb3 |
219 | |
eb58c082 |
220 | # *possibly* supplement the main selection with pks if not already |
221 | # there, as they will have to be a part of the group_by to collapse |
222 | # things properly |
223 | my $inner_select_with_extras; |
224 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
225 | or $self->throw_exception( sprintf |
226 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
227 | $root_node->{-rsrc}->source_name, |
e1861c2c |
228 | ); |
eb58c082 |
229 | for my $col (@pks) { |
230 | push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col |
231 | unless $cur_sel->{$col}++; |
1e4f9fb3 |
232 | } |
eb58c082 |
233 | |
234 | ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({ |
235 | %$inner_attrs, |
236 | $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (), |
237 | _aliastypes => $inner_aliastypes, |
238 | }); |
0a3441ee |
239 | } |
d28bb90d |
240 | |
e1861c2c |
241 | # we already optimized $inner_attrs->{from} above |
97e130fa |
242 | # and already local()ized |
243 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
244 | |
ea95892e |
245 | # generate the subquery |
6395604e |
246 | $self->_select_args_to_query ( |
e1861c2c |
247 | @{$inner_attrs}{qw(from select where)}, |
ea95892e |
248 | $inner_attrs, |
249 | ); |
d28bb90d |
250 | }; |
251 | |
252 | # Generate the outer from - this is relatively easy (really just replace |
253 | # the join slot with the subquery), with a major caveat - we can not |
254 | # join anything that is non-selecting (not part of the prefetch), but at |
255 | # the same time is a multi-type relationship, as it will explode the result. |
256 | # |
257 | # There are two possibilities here |
258 | # - either the join is non-restricting, in which case we simply throw it away |
259 | # - it is part of the restrictions, in which case we need to collapse the outer |
260 | # result by tackling yet another group_by to the outside of the query |
261 | |
27e0370d |
262 | # work on a shallow copy |
e1861c2c |
263 | my @orig_from = @{$attrs->{from}}; |
264 | |
052e8431 |
265 | |
e1861c2c |
266 | $outer_attrs->{from} = \ my @outer_from; |
53c29913 |
267 | |
27e0370d |
268 | # we may not be the head |
97e130fa |
269 | if ($root_node_offset) { |
e1861c2c |
270 | # first generate the outer_from, up to the substitution point |
271 | @outer_from = splice @orig_from, 0, $root_node_offset; |
27e0370d |
272 | |
e1861c2c |
273 | # substitute the subq at the right spot |
27e0370d |
274 | push @outer_from, [ |
275 | { |
1e4f9fb3 |
276 | -alias => $root_alias, |
97e130fa |
277 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
278 | $root_alias => $inner_subq, |
27e0370d |
279 | }, |
e1861c2c |
280 | # preserve attrs from what is now the head of the from after the splice |
281 | @{$orig_from[0]}[1 .. $#{$orig_from[0]}], |
27e0370d |
282 | ]; |
283 | } |
284 | else { |
27e0370d |
285 | @outer_from = { |
1e4f9fb3 |
286 | -alias => $root_alias, |
27e0370d |
287 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
288 | $root_alias => $inner_subq, |
27e0370d |
289 | }; |
d28bb90d |
290 | } |
291 | |
e1861c2c |
292 | shift @orig_from; # what we just replaced above |
97e130fa |
293 | |
ea95892e |
294 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
295 | # in what role |
975b573a |
296 | my $outer_aliastypes = $outer_attrs->{_aliastypes} = |
e1861c2c |
297 | $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from }); |
052e8431 |
298 | |
a4812caa |
299 | # unroll parents |
1e4f9fb3 |
300 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
301 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
302 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
303 | |
d28bb90d |
304 | # see what's left - throw away if not selecting/restricting |
eb58c082 |
305 | my $may_need_outer_group_by; |
e1861c2c |
306 | while (my $j = shift @orig_from) { |
d28bb90d |
307 | my $alias = $j->[0]{-alias}; |
308 | |
a4812caa |
309 | if ( |
310 | $outer_select_chain->{$alias} |
311 | ) { |
312 | push @outer_from, $j |
d28bb90d |
313 | } |
1e4f9fb3 |
314 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
315 | push @outer_from, $j; |
eb58c082 |
316 | $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
317 | } |
318 | } |
319 | |
eb58c082 |
320 | # also throw in a synthetic group_by if a non-selecting multiplier, |
321 | # to guard against cross-join explosions |
322 | # the logic is somewhat fragile, but relies on the idea that if a user supplied |
323 | # a group by on their own - they know what they were doing |
324 | if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
325 | ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({ |
560978e2 |
326 | %$outer_attrs, |
327 | from => \@outer_from, |
560978e2 |
328 | }); |
36fd7f07 |
329 | } |
330 | |
e1861c2c |
331 | # This is totally horrific - the {where} ends up in both the inner and outer query |
d28bb90d |
332 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
333 | # then if where conditions apply to the *right* side of the prefetch, you may have |
334 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
4a0eed52 |
335 | # the outer select to exclude joins you didn't want in the first place |
d28bb90d |
336 | # |
337 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
e1861c2c |
338 | return $outer_attrs; |
d28bb90d |
339 | } |
340 | |
1a736efb |
341 | # |
342 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
343 | # |
ad630f4b |
344 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
345 | # select/where/order/group attributes, in order to determine what |
4a0eed52 |
346 | # aliases are needed to fulfill the query. This information is used |
ad630f4b |
347 | # throughout the code to prune unnecessary JOINs from the queries |
348 | # in an attempt to reduce the execution time. |
349 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
350 | # happen is for it to fail due to some scalar SQL, which in turn will |
351 | # result in a vocal exception. |
539ffe87 |
352 | sub _resolve_aliastypes_from_select_args { |
e1861c2c |
353 | my ( $self, $attrs ) = @_; |
546f1cd9 |
354 | |
ad630f4b |
355 | $self->throw_exception ('Unable to analyze custom {from}') |
e1861c2c |
356 | if ref $attrs->{from} ne 'ARRAY'; |
546f1cd9 |
357 | |
ad630f4b |
358 | # what we will return |
964a3c71 |
359 | my $aliases_by_type; |
546f1cd9 |
360 | |
ad630f4b |
361 | # see what aliases are there to work with |
eb58c082 |
362 | # and record who is a multiplier and who is premultiplied |
ad630f4b |
363 | my $alias_list; |
e1861c2c |
364 | for my $node (@{$attrs->{from}}) { |
365 | |
366 | my $j = $node; |
ad630f4b |
367 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
368 | my $al = $j->{-alias} |
369 | or next; |
370 | |
371 | $alias_list->{$al} = $j; |
eb58c082 |
372 | |
373 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } |
a4812caa |
374 | # not array == {from} head == can't be multiplying |
eb58c082 |
375 | if ref($node) eq 'ARRAY' and ! $j->{-is_single}; |
376 | |
377 | $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] } |
378 | # parts of the path that are not us but are multiplying |
379 | if grep { $aliases_by_type->{multiplying}{$_} } |
380 | grep { $_ ne $al } |
381 | map { values %$_ } |
382 | @{ $j->{-join_path}||[] } |
546f1cd9 |
383 | } |
546f1cd9 |
384 | |
318e3d94 |
385 | # get a column to source/alias map (including unambiguous unqualified ones) |
e1861c2c |
386 | my $colinfo = $self->_resolve_column_info ($attrs->{from}); |
1a736efb |
387 | |
ad630f4b |
388 | # set up a botched SQLA |
389 | my $sql_maker = $self->sql_maker; |
07f31d19 |
390 | |
4c2b30d6 |
391 | # these are throw away results, do not pollute the bind stack |
0542ec57 |
392 | local $sql_maker->{where_bind}; |
393 | local $sql_maker->{group_bind}; |
394 | local $sql_maker->{having_bind}; |
97e130fa |
395 | local $sql_maker->{from_bind}; |
3f5b99fe |
396 | |
397 | # we can't scan properly without any quoting (\b doesn't cut it |
398 | # everywhere), so unless there is proper quoting set - use our |
399 | # own weird impossible character. |
400 | # Also in the case of no quoting, we need to explicitly disable |
401 | # name_sep, otherwise sorry nasty legacy syntax like |
402 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
403 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
404 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
405 | |
406 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
407 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
408 | # if we don't unset it we screw up retarded but unfortunately working |
409 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
410 | $sql_maker->{name_sep} = ''; |
411 | } |
412 | |
413 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
414 | |
1a736efb |
415 | # generate sql chunks |
416 | my $to_scan = { |
417 | restricting => [ |
e1861c2c |
418 | $sql_maker->_recurse_where ($attrs->{where}), |
1e4f9fb3 |
419 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
420 | ], |
421 | grouping => [ |
422 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
423 | ], |
97e130fa |
424 | joining => [ |
425 | $sql_maker->_recurse_from ( |
e1861c2c |
426 | ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0], |
427 | @{$attrs->{from}}[1 .. $#{$attrs->{from}}], |
97e130fa |
428 | ), |
429 | ], |
1a736efb |
430 | selecting => [ |
ad1d374e |
431 | map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}}, |
1e4f9fb3 |
432 | ], |
433 | ordering => [ |
434 | map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker), |
1a736efb |
435 | ], |
436 | }; |
07f31d19 |
437 | |
0dadd60d |
438 | # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are |
439 | # bind value specs left in by the sloppy renderer above. It is ok to do this |
440 | # at this point, since we are going to end up rewriting this crap anyway |
441 | for my $v (values %$to_scan) { |
442 | my @nv; |
443 | for (@$v) { |
444 | next if ( |
445 | ! defined $_ |
446 | or |
447 | ( |
448 | ref $_ eq 'ARRAY' |
449 | and |
450 | ( @$_ == 0 or @$_ == 2 ) |
451 | ) |
452 | ); |
453 | |
454 | if (ref $_) { |
455 | require Data::Dumper::Concise; |
456 | $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) ); |
457 | } |
458 | |
459 | push @nv, $_; |
460 | } |
461 | |
462 | $v = \@nv; |
463 | } |
464 | |
465 | # kill all selectors which look like a proper subquery |
466 | # this is a sucky heuristic *BUT* - if we get it wrong the query will simply |
467 | # fail to run, so we are relatively safe |
468 | $to_scan->{selecting} = [ grep { |
469 | $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi |
470 | } @{ $to_scan->{selecting} || [] } ]; |
07f31d19 |
471 | |
318e3d94 |
472 | # first see if we have any exact matches (qualified or unqualified) |
473 | for my $type (keys %$to_scan) { |
474 | for my $piece (@{$to_scan->{$type}}) { |
475 | if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) { |
476 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
477 | $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece; |
478 | } |
479 | } |
480 | } |
481 | |
482 | # now loop through all fully qualified columns and get the corresponding |
1a736efb |
483 | # alias (should work even if they are in scalarrefs) |
ad630f4b |
484 | for my $alias (keys %$alias_list) { |
1a736efb |
485 | my $al_re = qr/ |
97e130fa |
486 | $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
1a736efb |
487 | | |
97e130fa |
488 | \b $alias \. ([^\s\)\($rquote]+)? |
1a736efb |
489 | /x; |
490 | |
1a736efb |
491 | for my $type (keys %$to_scan) { |
492 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
493 | if (my @matches = $piece =~ /$al_re/g) { |
494 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
495 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_" |
97e130fa |
496 | for grep { defined $_ } @matches; |
497 | } |
1a736efb |
498 | } |
ad630f4b |
499 | } |
1a736efb |
500 | } |
501 | |
502 | # now loop through unqualified column names, and try to locate them within |
503 | # the chunks |
504 | for my $col (keys %$colinfo) { |
3f5b99fe |
505 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
506 | |
97e130fa |
507 | my $col_re = qr/ $lquote ($col) $rquote /x; |
07f31d19 |
508 | |
1a736efb |
509 | for my $type (keys %$to_scan) { |
510 | for my $piece (@{$to_scan->{$type}}) { |
318e3d94 |
511 | if ( my @matches = $piece =~ /$col_re/g) { |
a4812caa |
512 | my $alias = $colinfo->{$col}{-source_alias}; |
97e130fa |
513 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
514 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
97e130fa |
515 | for grep { defined $_ } @matches; |
a4812caa |
516 | } |
1a736efb |
517 | } |
07f31d19 |
518 | } |
519 | } |
520 | |
521 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
522 | for my $j (values %$alias_list) { |
07f31d19 |
523 | my $alias = $j->{-alias} or next; |
97e130fa |
524 | $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if ( |
07f31d19 |
525 | (not $j->{-join_type}) |
526 | or |
527 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
528 | ); |
529 | } |
530 | |
1e4f9fb3 |
531 | for (keys %$aliases_by_type) { |
532 | delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}}; |
533 | } |
534 | |
964a3c71 |
535 | return $aliases_by_type; |
07f31d19 |
536 | } |
537 | |
eb58c082 |
538 | # This is the engine behind { distinct => 1 } and the general |
539 | # complex prefetch grouper |
0a3441ee |
540 | sub _group_over_selection { |
560978e2 |
541 | my ($self, $attrs) = @_; |
0a3441ee |
542 | |
560978e2 |
543 | my $colinfos = $self->_resolve_column_info ($attrs->{from}); |
0a3441ee |
544 | |
545 | my (@group_by, %group_index); |
546 | |
36fd7f07 |
547 | # the logic is: if it is a { func => val } we assume an aggregate, |
548 | # otherwise if \'...' or \[...] we assume the user knows what is |
549 | # going on thus group over it |
560978e2 |
550 | for (@{$attrs->{select}}) { |
0a3441ee |
551 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
552 | push @group_by, $_; |
553 | $group_index{$_}++; |
560978e2 |
554 | if ($colinfos->{$_} and $_ !~ /\./ ) { |
0a3441ee |
555 | # add a fully qualified version as well |
560978e2 |
556 | $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; |
0a3441ee |
557 | } |
07f31d19 |
558 | } |
559 | } |
ad630f4b |
560 | |
eb58c082 |
561 | my @order_by = $self->_extract_order_criteria($attrs->{order_by}) |
562 | or return (\@group_by, $attrs->{order_by}); |
563 | |
564 | # add any order_by parts that are not already present in the group_by |
565 | # to maintain SQL cross-compatibility and general sanity |
566 | # |
567 | # also in case the original selection is *not* unique, or in case part |
568 | # of the ORDER BY refers to a multiplier - we will need to replace the |
569 | # skipped order_by elements with their MIN/MAX equivalents as to maintain |
570 | # the proper overall order without polluting the group criteria (and |
571 | # possibly changing the outcome entirely) |
572 | |
573 | my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes); |
574 | |
575 | my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by); |
576 | |
577 | for my $o_idx (0 .. $#order_by) { |
578 | |
579 | # if the chunk is already a min/max function - there is nothing left to touch |
580 | next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix; |
581 | |
0a3441ee |
582 | # only consider real columns (for functions the user got to do an explicit group_by) |
eb58c082 |
583 | my $chunk_ci; |
584 | if ( |
585 | @{$order_by[$o_idx]} != 1 |
586 | or |
587 | # only declare an unknown *plain* identifier as "leftover" if we are called with |
588 | # aliastypes to examine. If there are none - we are still in _resolve_attrs, and |
589 | # can just assume the user knows what they want |
590 | ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} ) |
591 | ) { |
592 | push @$leftovers, $order_by[$o_idx][0]; |
14e26c5f |
593 | } |
560978e2 |
594 | |
eb58c082 |
595 | next unless $chunk_ci; |
596 | |
597 | # no duplication of group criteria |
598 | next if $group_index{$chunk_ci->{-fq_colname}}; |
599 | |
600 | $aliastypes ||= ( |
601 | $attrs->{_aliastypes} |
560978e2 |
602 | or |
eb58c082 |
603 | $self->_resolve_aliastypes_from_select_args({ |
604 | from => $attrs->{from}, |
605 | order_by => $attrs->{order_by}, |
606 | }) |
607 | ) if $group_already_unique; |
608 | |
609 | # check that we are not ordering by a multiplier (if a check is requested at all) |
610 | if ( |
611 | $group_already_unique |
612 | and |
613 | ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}} |
614 | and |
615 | ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}} |
560978e2 |
616 | ) { |
eb58c082 |
617 | push @group_by, $chunk_ci->{-fq_colname}; |
618 | $group_index{$chunk_ci->{-fq_colname}}++ |
560978e2 |
619 | } |
eb58c082 |
620 | else { |
621 | # We need to order by external columns without adding them to the group |
622 | # (eiehter a non-unique selection, or a multi-external) |
623 | # |
624 | # This doesn't really make sense in SQL, however from DBICs point |
625 | # of view is rather valid (e.g. order the leftmost objects by whatever |
626 | # criteria and get the offset/rows many). There is a way around |
627 | # this however in SQL - we simply tae the direction of each piece |
628 | # of the external order and convert them to MIN(X) for ASC or MAX(X) |
629 | # for DESC, and group_by the root columns. The end result should be |
630 | # exactly what we expect |
631 | |
632 | # FIXME - this code is a joke, will need to be completely rewritten in |
633 | # the DQ branch. But I need to push a POC here, otherwise the |
634 | # pesky tests won't pass |
635 | # wrap any part of the order_by that "responds" to an ordering alias |
636 | # into a MIN/MAX |
637 | $sql_maker ||= $self->sql_maker; |
638 | $order_chunks ||= [ |
639 | map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}) |
640 | ]; |
0a3441ee |
641 | |
eb58c082 |
642 | my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]); |
643 | |
644 | $new_order_by[$o_idx] = \[ |
645 | sprintf( '%s( %s )%s', |
646 | ($is_desc ? 'MAX' : 'MIN'), |
647 | $chunk, |
648 | ($is_desc ? ' DESC' : ''), |
649 | ), |
650 | @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ] |
651 | ]; |
652 | } |
0a3441ee |
653 | } |
654 | |
eb58c082 |
655 | $self->throw_exception ( sprintf |
9736be65 |
656 | 'Unable to programatically derive a required group_by from the supplied ' |
657 | . 'order_by criteria. To proceed either add an explicit group_by, or ' |
658 | . 'simplify your order_by to only include plain columns ' |
659 | . '(supplied order_by: %s)', |
eb58c082 |
660 | join ', ', map { "'$_'" } @$leftovers, |
661 | ) if $leftovers; |
662 | |
663 | # recreate the untouched order parts |
664 | if (@new_order_by) { |
665 | $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks ); |
666 | } |
667 | |
668 | return ( |
669 | \@group_by, |
670 | (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged |
671 | ); |
07f31d19 |
672 | } |
673 | |
d28bb90d |
674 | sub _resolve_ident_sources { |
675 | my ($self, $ident) = @_; |
676 | |
677 | my $alias2source = {}; |
d28bb90d |
678 | |
679 | # the reason this is so contrived is that $ident may be a {from} |
680 | # structure, specifying multiple tables to join |
6298a324 |
681 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
682 | # this is compat mode for insert/update/delete which do not deal with aliases |
683 | $alias2source->{me} = $ident; |
d28bb90d |
684 | } |
685 | elsif (ref $ident eq 'ARRAY') { |
686 | |
687 | for (@$ident) { |
688 | my $tabinfo; |
689 | if (ref $_ eq 'HASH') { |
690 | $tabinfo = $_; |
d28bb90d |
691 | } |
692 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
693 | $tabinfo = $_->[0]; |
694 | } |
695 | |
4376a157 |
696 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
697 | if ($tabinfo->{-rsrc}); |
d28bb90d |
698 | } |
699 | } |
700 | |
90f10b5a |
701 | return $alias2source; |
d28bb90d |
702 | } |
703 | |
704 | # Takes $ident, \@column_names |
705 | # |
706 | # returns { $column_name => \%column_info, ... } |
707 | # also note: this adds -result_source => $rsrc to the column info |
708 | # |
09e14fdc |
709 | # If no columns_names are supplied returns info about *all* columns |
710 | # for all sources |
d28bb90d |
711 | sub _resolve_column_info { |
712 | my ($self, $ident, $colnames) = @_; |
8d005ad9 |
713 | |
714 | return {} if $colnames and ! @$colnames; |
715 | |
90f10b5a |
716 | my $alias2src = $self->_resolve_ident_sources($ident); |
d28bb90d |
717 | |
52416317 |
718 | my (%seen_cols, @auto_colnames); |
d28bb90d |
719 | |
720 | # compile a global list of column names, to be able to properly |
721 | # disambiguate unqualified column names (if at all possible) |
722 | for my $alias (keys %$alias2src) { |
723 | my $rsrc = $alias2src->{$alias}; |
724 | for my $colname ($rsrc->columns) { |
725 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
726 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
727 | } |
728 | } |
729 | |
09e14fdc |
730 | $colnames ||= [ |
731 | @auto_colnames, |
732 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
733 | ]; |
734 | |
52416317 |
735 | my (%return, $colinfos); |
d28bb90d |
736 | foreach my $col (@$colnames) { |
52416317 |
737 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
738 | |
52416317 |
739 | # if the column was seen exactly once - we know which rsrc it came from |
740 | $source_alias ||= $seen_cols{$colname}[0] |
741 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
742 | |
52416317 |
743 | next unless $source_alias; |
744 | |
745 | my $rsrc = $alias2src->{$source_alias} |
746 | or next; |
747 | |
748 | $return{$col} = { |
6395604e |
749 | %{ |
750 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
751 | || |
752 | $self->throw_exception( |
753 | "No such column '$colname' on source " . $rsrc->source_name |
754 | ); |
755 | }, |
d28bb90d |
756 | -result_source => $rsrc, |
52416317 |
757 | -source_alias => $source_alias, |
81bf295c |
758 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
759 | -colname => $colname, |
d28bb90d |
760 | }; |
81bf295c |
761 | |
762 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
763 | } |
764 | |
765 | return \%return; |
766 | } |
767 | |
289ac713 |
768 | # The DBIC relationship chaining implementation is pretty simple - every |
769 | # new related_relationship is pushed onto the {from} stack, and the {select} |
770 | # window simply slides further in. This means that when we count somewhere |
771 | # in the middle, we got to make sure that everything in the join chain is an |
772 | # actual inner join, otherwise the count will come back with unpredictable |
773 | # results (a resultset may be generated with _some_ rows regardless of if |
774 | # the relation which the $rs currently selects has rows or not). E.g. |
775 | # $artist_rs->cds->count - normally generates: |
776 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
777 | # which actually returns the number of artists * (number of cds || 1) |
778 | # |
779 | # So what we do here is crawl {from}, determine if the current alias is at |
780 | # the top of the stack, and if not - make sure the chain is inner-joined down |
781 | # to the root. |
782 | # |
31a8aaaf |
783 | sub _inner_join_to_node { |
289ac713 |
784 | my ($self, $from, $alias) = @_; |
785 | |
786 | # subqueries and other oddness are naturally not supported |
787 | return $from if ( |
788 | ref $from ne 'ARRAY' |
789 | || |
790 | @$from <= 1 |
791 | || |
792 | ref $from->[0] ne 'HASH' |
793 | || |
794 | ! $from->[0]{-alias} |
795 | || |
7eb76996 |
796 | $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do |
289ac713 |
797 | ); |
798 | |
799 | # find the current $alias in the $from structure |
800 | my $switch_branch; |
801 | JOINSCAN: |
802 | for my $j (@{$from}[1 .. $#$from]) { |
803 | if ($j->[0]{-alias} eq $alias) { |
804 | $switch_branch = $j->[0]{-join_path}; |
805 | last JOINSCAN; |
806 | } |
807 | } |
808 | |
7eb76996 |
809 | # something else went quite wrong |
289ac713 |
810 | return $from unless $switch_branch; |
811 | |
812 | # So it looks like we will have to switch some stuff around. |
813 | # local() is useless here as we will be leaving the scope |
814 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
815 | # So replace the first hashref in the node arrayref manually |
289ac713 |
816 | my @new_from = ($from->[0]); |
faeb2407 |
817 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
818 | |
819 | for my $j (@{$from}[1 .. $#$from]) { |
820 | my $jalias = $j->[0]{-alias}; |
821 | |
822 | if ($sw_idx->{$jalias}) { |
823 | my %attrs = %{$j->[0]}; |
824 | delete $attrs{-join_type}; |
825 | push @new_from, [ |
826 | \%attrs, |
827 | @{$j}[ 1 .. $#$j ], |
828 | ]; |
829 | } |
830 | else { |
831 | push @new_from, $j; |
832 | } |
833 | } |
834 | |
835 | return \@new_from; |
836 | } |
837 | |
bac358c9 |
838 | sub _extract_order_criteria { |
1a736efb |
839 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
840 | |
1a736efb |
841 | my $parser = sub { |
e6977bbb |
842 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
843 | |
1a736efb |
844 | return scalar $sql_maker->_order_by_chunks ($order_by) |
845 | unless wantarray; |
c0748280 |
846 | |
e6977bbb |
847 | my ($lq, $rq, $sep) = map { quotemeta($_) } ( |
848 | ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars), |
849 | $sql_maker->name_sep |
850 | ); |
851 | |
1a736efb |
852 | my @chunks; |
bac358c9 |
853 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
854 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
855 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
856 | |
857 | # order criteria may have come back pre-quoted (literals and whatnot) |
858 | # this is fragile, but the best we can currently do |
859 | $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe |
860 | or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x; |
861 | |
1a736efb |
862 | push @chunks, $chunk; |
bac6c4fb |
863 | } |
1a736efb |
864 | |
865 | return @chunks; |
866 | }; |
867 | |
868 | if ($sql_maker) { |
869 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
870 | } |
871 | else { |
1a736efb |
872 | $sql_maker = $self->sql_maker; |
e6977bbb |
873 | |
874 | # pass these in to deal with literals coming from |
875 | # the user or the deep guts of prefetch |
876 | my $orig_quote_chars = [$sql_maker->_quote_chars]; |
877 | |
1a736efb |
878 | local $sql_maker->{quote_char}; |
e6977bbb |
879 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
880 | } |
bac6c4fb |
881 | } |
882 | |
7cec4356 |
883 | sub _order_by_is_stable { |
5f11e54f |
884 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
885 | |
eb58c082 |
886 | my @cols = ( |
8d005ad9 |
887 | ( map { $_->[0] } $self->_extract_order_criteria($order_by) ), |
888 | ( $where ? @{ $self->_extract_fixed_condition_columns($where) || [] } : () ), |
eb58c082 |
889 | ) or return undef; |
890 | |
891 | my $colinfo = $self->_resolve_column_info($ident, \@cols); |
892 | |
893 | return keys %$colinfo |
894 | ? $self->_columns_comprise_identifying_set( $colinfo, \@cols ) |
895 | : undef |
896 | ; |
897 | } |
c0748280 |
898 | |
eb58c082 |
899 | sub _columns_comprise_identifying_set { |
900 | my ($self, $colinfo, $columns) = @_; |
7cec4356 |
901 | |
902 | my $cols_per_src; |
eb58c082 |
903 | $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_ |
904 | for grep { defined $_ } @{$colinfo}{@$columns}; |
7cec4356 |
905 | |
906 | for (values %$cols_per_src) { |
907 | my $src = (values %$_)[0]->{-result_source}; |
908 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
909 | } |
910 | |
7cec4356 |
911 | return undef; |
912 | } |
913 | |
0e81e691 |
914 | # this is almost identical to the above, except it accepts only |
915 | # a single rsrc, and will succeed only if the first portion of the order |
916 | # by is stable. |
917 | # returns that portion as a colinfo hashref on success |
918 | sub _main_source_order_by_portion_is_stable { |
919 | my ($self, $main_rsrc, $order_by, $where) = @_; |
920 | |
921 | die "Huh... I expect a blessed result_source..." |
922 | if ref($main_rsrc) eq 'ARRAY'; |
923 | |
924 | my @ord_cols = map |
925 | { $_->[0] } |
926 | ( $self->_extract_order_criteria($order_by) ) |
927 | ; |
928 | return unless @ord_cols; |
929 | |
318e3d94 |
930 | my $colinfos = $self->_resolve_column_info($main_rsrc); |
931 | |
0e81e691 |
932 | for (0 .. $#ord_cols) { |
933 | if ( |
934 | ! $colinfos->{$ord_cols[$_]} |
935 | or |
936 | $colinfos->{$ord_cols[$_]}{-result_source} != $main_rsrc |
937 | ) { |
938 | $#ord_cols = $_ - 1; |
939 | last; |
940 | } |
941 | } |
942 | |
943 | # we just truncated it above |
944 | return unless @ord_cols; |
945 | |
0e81e691 |
946 | my $order_portion_ci = { map { |
947 | $colinfos->{$_}{-colname} => $colinfos->{$_}, |
948 | $colinfos->{$_}{-fq_colname} => $colinfos->{$_}, |
949 | } @ord_cols }; |
950 | |
318e3d94 |
951 | # since all we check here are the start of the order_by belonging to the |
952 | # top level $rsrc, a present identifying set will mean that the resultset |
953 | # is ordered by its leftmost table in a stable manner |
954 | # |
955 | # RV of _identifying_column_set contains unqualified names only |
956 | my $unqualified_idset = $main_rsrc->_identifying_column_set({ |
957 | ( $where ? %{ |
958 | $self->_resolve_column_info( |
8d005ad9 |
959 | $main_rsrc, $self->_extract_fixed_condition_columns($where)||[] |
318e3d94 |
960 | ) |
961 | } : () ), |
962 | %$order_portion_ci |
963 | }) or return; |
964 | |
965 | my $ret_info; |
966 | my %unqualified_idcols_from_order = map { |
967 | $order_portion_ci->{$_} ? ( $_ => $order_portion_ci->{$_} ) : () |
968 | } @$unqualified_idset; |
969 | |
970 | # extra optimization - cut the order_by at the end of the identifying set |
971 | # (just in case the user was stupid and overlooked the obvious) |
972 | for my $i (0 .. $#ord_cols) { |
973 | my $col = $ord_cols[$i]; |
974 | my $unqualified_colname = $order_portion_ci->{$col}{-colname}; |
975 | $ret_info->{$col} = { %{$order_portion_ci->{$col}}, -idx_in_order_subset => $i }; |
976 | delete $unqualified_idcols_from_order{$ret_info->{$col}{-colname}}; |
977 | |
978 | # we didn't reach the end of the identifying portion yet |
979 | return $ret_info unless keys %unqualified_idcols_from_order; |
980 | } |
0e81e691 |
981 | |
318e3d94 |
982 | die 'How did we get here...'; |
0e81e691 |
983 | } |
984 | |
8d005ad9 |
985 | # Attempts to flatten a passed in SQLA condition as much as possible towards |
986 | # a plain hashref, *without* altering its semantics. Required by |
987 | # create/populate being able to extract definitive conditions from preexisting |
988 | # resultset {where} stacks |
989 | # |
990 | # FIXME - while relatively robust, this is still imperfect, one of the first |
991 | # things to tackle with DQ |
992 | sub _collapse_cond { |
993 | my ($self, $where, $where_is_anded_array) = @_; |
994 | |
995 | if (! $where) { |
996 | return; |
997 | } |
998 | elsif ($where_is_anded_array or ref $where eq 'HASH') { |
999 | |
1000 | my @pairs; |
1001 | |
1002 | my @pieces = $where_is_anded_array ? @$where : $where; |
1003 | while (@pieces) { |
1004 | my $chunk = shift @pieces; |
1005 | |
1006 | if (ref $chunk eq 'HASH') { |
1007 | push @pairs, map { [ $_ => $chunk->{$_} ] } sort keys %$chunk; |
1008 | } |
1009 | elsif (ref $chunk eq 'ARRAY') { |
1010 | push @pairs, [ -or => $chunk ] |
1011 | if @$chunk; |
1012 | } |
1013 | elsif ( ! ref $chunk) { |
1014 | push @pairs, [ $chunk, shift @pieces ]; |
1015 | } |
1016 | else { |
1017 | push @pairs, [ '', $chunk ]; |
1018 | } |
1019 | } |
1020 | |
1021 | return unless @pairs; |
1022 | |
1023 | my @conds = $self->_collapse_cond_unroll_pairs(\@pairs) |
1024 | or return; |
1025 | |
1026 | # Consolidate various @conds back into something more compact |
1027 | my $fin; |
1028 | |
1029 | for my $c (@conds) { |
1030 | if (ref $c ne 'HASH') { |
1031 | push @{$fin->{-and}}, $c; |
1032 | } |
1033 | else { |
1034 | for my $col (sort keys %$c) { |
1035 | if (exists $fin->{$col}) { |
1036 | my ($l, $r) = ($fin->{$col}, $c->{$col}); |
1037 | |
1038 | (ref $_ ne 'ARRAY' or !@$_) and $_ = [ -and => $_ ] for ($l, $r); |
1039 | |
1040 | if (@$l and @$r and $l->[0] eq $r->[0] and $l->[0] eq '-and') { |
1041 | $fin->{$col} = [ -and => map { @$_[1..$#$_] } ($l, $r) ]; |
1042 | } |
1043 | else { |
1044 | $fin->{$col} = [ -and => $fin->{$col}, $c->{$col} ]; |
1045 | } |
1046 | } |
1047 | else { |
1048 | $fin->{$col} = $c->{$col}; |
1049 | } |
1050 | } |
1051 | } |
1052 | } |
1053 | |
1054 | if ( ref $fin->{-and} eq 'ARRAY' and @{$fin->{-and}} == 1 ) { |
1055 | my $piece = (delete $fin->{-and})->[0]; |
1056 | if (ref $piece eq 'ARRAY') { |
1057 | $fin->{-or} = $fin->{-or} ? [ $piece, $fin->{-or} ] : $piece; |
1058 | } |
1059 | elsif (! exists $fin->{''}) { |
1060 | $fin->{''} = $piece; |
1061 | } |
1062 | } |
1063 | |
1064 | return $fin; |
1065 | } |
1066 | elsif (ref $where eq 'ARRAY') { |
1067 | my @w = @$where; |
1068 | |
1069 | while ( @w and ( |
1070 | (ref $w[0] eq 'ARRAY' and ! @{$w[0]} ) |
1071 | or |
1072 | (ref $w[0] eq 'HASH' and ! keys %{$w[0]}) |
1073 | )) { shift @w }; |
1074 | |
1075 | return unless @w; |
1076 | |
1077 | if ( @w == 1 ) { |
1078 | return ( ref $w[0] ) |
1079 | ? $self->_collapse_cond($w[0]) |
1080 | : { $w[0] => undef } |
1081 | ; |
1082 | } |
1083 | elsif ( ref $w[0] ) { |
1084 | return \@w; |
1085 | } |
1086 | elsif ( @w == 2 ) { |
1087 | if ( ( $w[0]||'' ) =~ /^\-and$/i ) { |
1088 | return (ref $w[1] eq 'HASH' or ref $w[1] eq 'ARRAY') |
1089 | ? $self->_collapse_cond($w[1], (ref $w[1] eq 'ARRAY') ) |
1090 | : $self->throw_exception("Unsupported top-level op/arg pair: [ $w[0] => $w[1] ]") |
1091 | ; |
1092 | } |
1093 | else { |
1094 | return $self->_collapse_cond({ @w }); |
1095 | } |
1096 | } |
1097 | } |
1098 | else { |
1099 | # not a hash not an array |
1100 | return { '' => $where }; |
1101 | } |
1102 | |
1103 | # catchall, some of the things above fall through |
1104 | return $where; |
1105 | } |
1106 | |
1107 | sub _collapse_cond_unroll_pairs { |
1108 | my ($self, $pairs) = @_; |
1109 | |
1110 | my @conds; |
1111 | |
1112 | while (@$pairs) { |
1113 | my ($lhs, $rhs) = @{ shift @$pairs }; |
1114 | |
1115 | if ($lhs eq '') { |
1116 | push @conds, $self->_collapse_cond($rhs); |
1117 | } |
1118 | elsif ( $lhs =~ /^\-and$/i ) { |
1119 | push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY')); |
1120 | } |
1121 | elsif ( $lhs =~ /^\-or$/i ) { |
1122 | push @conds, $self->_collapse_cond( |
1123 | (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs |
1124 | ); |
1125 | } |
1126 | else { |
1127 | if (ref $rhs eq 'HASH' and ! keys %$rhs) { |
1128 | # FIXME - SQLA seems to be doing... nothing...? |
1129 | } |
1130 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) { |
1131 | for my $p ($self->_collapse_cond_unroll_pairs([ [ $lhs => $rhs->{'='} ] ])) { |
1132 | |
1133 | # extra sanity check |
1134 | if (keys %$p > 1) { |
1135 | require Data::Dumper::Concise; |
1136 | local $Data::Dumper::Deepcopy = 1; |
1137 | $self->throw_exception( |
1138 | "Internal error: unexpected collapse unroll:" |
1139 | . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p } |
1140 | ); |
1141 | } |
1142 | |
1143 | my ($l, $r) = %$p; |
1144 | |
1145 | push @conds, ( ! ref $r or overload::Method($r, '""' ) ) |
1146 | ? { $l => $r } |
1147 | : { $l => { '=' => $r } } |
1148 | ; |
1149 | } |
1150 | } |
1151 | elsif (ref $rhs eq 'ARRAY') { |
1152 | # some of these conditionals encounter multi-values - roll them out using |
1153 | # an unshift, which will cause extra looping in the while{} above |
1154 | if (! @$rhs ) { |
1155 | push @conds, { $lhs => [] }; |
1156 | } |
1157 | elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) { |
1158 | $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ") |
1159 | if @$rhs == 1; |
1160 | |
1161 | if( $rhs->[0] =~ /^\-and$/i ) { |
1162 | unshift @$pairs, map { [ $lhs => $_ ] } @{$rhs}[1..$#$rhs]; |
1163 | } |
1164 | # if not an AND then it's an OR |
1165 | elsif(@$rhs == 2) { |
1166 | unshift @$pairs, [ $lhs => $rhs->[1] ]; |
1167 | } |
1168 | else { |
1169 | push @conds, { $lhs => $rhs }; |
1170 | } |
1171 | } |
1172 | elsif (@$rhs == 1) { |
1173 | unshift @$pairs, [ $lhs => $rhs->[0] ]; |
1174 | } |
1175 | else { |
1176 | push @conds, { $lhs => $rhs }; |
1177 | } |
1178 | } |
1179 | else { |
1180 | push @conds, { $lhs => $rhs }; |
1181 | } |
1182 | } |
1183 | } |
1184 | |
1185 | return @conds; |
1186 | } |
1187 | |
1188 | |
4a0eed52 |
1189 | # returns an arrayref of column names which *definitely* have some |
8d005ad9 |
1190 | # sort of non-nullable *single* equality requested in the given condition |
5f11e54f |
1191 | # specification. This is used to figure out if a resultset is |
1192 | # constrained to a column which is part of a unique constraint, |
1193 | # which in turn allows us to better predict how ordering will behave |
1194 | # etc. |
1195 | # |
1196 | # this is a rudimentary, incomplete, and error-prone extractor |
1197 | # however this is OK - it is conservative, and if we can not find |
1198 | # something that is in fact there - the stack will recover gracefully |
5f11e54f |
1199 | sub _extract_fixed_condition_columns { |
8d005ad9 |
1200 | my $self = shift; |
1201 | my $where_hash = $self->_collapse_cond(shift); |
1202 | |
1203 | my $res; |
1204 | for my $c (keys %$where_hash) { |
1205 | if (defined (my $v = $where_hash->{$c}) ) { |
1206 | if ( |
1207 | ! ref $v |
5f11e54f |
1208 | or |
8d005ad9 |
1209 | (ref $v eq 'HASH' and keys %$v == 1 and defined $v->{'='} and ( |
1210 | ! ref $v->{'='} |
1211 | or |
1212 | ref $v->{'='} eq 'SCALAR' |
1213 | or |
1214 | ( ref $v->{'='} eq 'REF' and ref ${$v->{'='}} eq 'ARRAY' ) |
1215 | or |
1216 | overload::Method($v->{'='}, '""') |
1217 | )) |
1218 | ) { |
1219 | $res->{$c} = 1; |
1220 | } |
1221 | elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') { |
1222 | $res->{$_} = 1 for map { @{ $self->_extract_fixed_condition_columns({ $c => $_ }) } } @{$v}[1..$#$v]; |
1223 | } |
5f11e54f |
1224 | } |
1225 | } |
8d005ad9 |
1226 | |
1227 | return [ sort keys %$res ]; |
c0748280 |
1228 | } |
bac6c4fb |
1229 | |
d28bb90d |
1230 | 1; |