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'; |
b34d9331 |
18 | use DBIx::Class::_Util qw(UNRESOLVABLE_CONDITION serialize); |
b5ce6748 |
19 | use SQL::Abstract qw(is_plain_value is_literal_value); |
e466c62b |
20 | use DBIx::Class::Carp; |
6298a324 |
21 | use namespace::clean; |
d28bb90d |
22 | |
23 | # |
052e8431 |
24 | # This code will remove non-selecting/non-restricting joins from |
4b1b5ea3 |
25 | # {from} specs, aiding the RDBMS query optimizer |
052e8431 |
26 | # |
27 | sub _prune_unused_joins { |
e1861c2c |
28 | my ($self, $attrs) = @_; |
ea95892e |
29 | |
e1861c2c |
30 | # only standard {from} specs are supported, and we could be disabled in general |
31 | return ($attrs->{from}, {}) unless ( |
32 | ref $attrs->{from} eq 'ARRAY' |
33 | and |
34 | @{$attrs->{from}} > 1 |
35 | and |
36 | ref $attrs->{from}[0] eq 'HASH' |
37 | and |
38 | ref $attrs->{from}[1] eq 'ARRAY' |
39 | and |
40 | $self->_use_join_optimizer |
41 | ); |
052e8431 |
42 | |
eb58c082 |
43 | my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs); |
4b1b5ea3 |
44 | |
eb58c082 |
45 | my $new_aliastypes = { %$orig_aliastypes }; |
46 | |
47 | # we will be recreating this entirely |
48 | my @reclassify = 'joining'; |
97e130fa |
49 | |
4b1b5ea3 |
50 | # a grouped set will not be affected by amount of rows. Thus any |
eb58c082 |
51 | # purely multiplicator classifications can go |
52 | # (will be reintroduced below if needed by something else) |
53 | push @reclassify, qw(multiplying premultiplied) |
437a9cfa |
54 | if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by}; |
4b1b5ea3 |
55 | |
eb58c082 |
56 | # nuke what will be recalculated |
57 | delete @{$new_aliastypes}{@reclassify}; |
58 | |
e1861c2c |
59 | my @newfrom = $attrs->{from}[0]; # FROM head is always present |
052e8431 |
60 | |
eb58c082 |
61 | # recalculate what we need once the multipliers are potentially gone |
62 | # ignore premultiplies, since they do not add any value to anything |
a4812caa |
63 | my %need_joins; |
eb58c082 |
64 | for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) { |
a4812caa |
65 | # add all requested aliases |
66 | $need_joins{$_} = 1 for keys %$_; |
67 | |
68 | # add all their parents (as per joinpath which is an AoH { table => alias }) |
97e130fa |
69 | $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; |
a4812caa |
70 | } |
97e130fa |
71 | |
e1861c2c |
72 | for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) { |
539ffe87 |
73 | push @newfrom, $j if ( |
a6ef93cb |
74 | (! defined $j->[0]{-alias}) # legacy crap |
539ffe87 |
75 | || |
76 | $need_joins{$j->[0]{-alias}} |
77 | ); |
052e8431 |
78 | } |
79 | |
eb58c082 |
80 | # we have a new set of joiners - for everything we nuked pull the classification |
81 | # off the original stack |
82 | for my $ctype (@reclassify) { |
83 | $new_aliastypes->{$ctype} = { map |
84 | { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () } |
85 | keys %{$orig_aliastypes->{$ctype}} |
86 | } |
87 | } |
88 | |
89 | return ( \@newfrom, $new_aliastypes ); |
052e8431 |
90 | } |
91 | |
052e8431 |
92 | # |
d28bb90d |
93 | # This is the code producing joined subqueries like: |
8273e845 |
94 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
d28bb90d |
95 | # |
96 | sub _adjust_select_args_for_complex_prefetch { |
e1861c2c |
97 | my ($self, $attrs) = @_; |
d28bb90d |
98 | |
e1861c2c |
99 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless ( |
100 | ref $attrs->{from} eq 'ARRAY' |
101 | and |
102 | @{$attrs->{from}} > 1 |
103 | and |
104 | ref $attrs->{from}[0] eq 'HASH' |
105 | and |
106 | ref $attrs->{from}[1] eq 'ARRAY' |
107 | ); |
d28bb90d |
108 | |
1e4f9fb3 |
109 | my $root_alias = $attrs->{alias}; |
110 | |
d28bb90d |
111 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
112 | my $outer_attrs = { %$attrs }; |
e1861c2c |
113 | delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)}; |
d28bb90d |
114 | |
6aa93928 |
115 | my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 }; |
116 | delete @{$inner_attrs}{qw(for collapse select as)}; |
d28bb90d |
117 | |
4df1400e |
118 | # there is no point of ordering the insides if there is no limit |
119 | delete $inner_attrs->{order_by} if ( |
120 | delete $inner_attrs->{_order_is_artificial} |
121 | or |
122 | ! $inner_attrs->{rows} |
123 | ); |
946f6260 |
124 | |
d28bb90d |
125 | # generate the inner/outer select lists |
126 | # for inside we consider only stuff *not* brought in by the prefetch |
127 | # on the outside we substitute any function for its alias |
e1861c2c |
128 | $outer_attrs->{select} = [ @{$attrs->{select}} ]; |
36fd7f07 |
129 | |
97e130fa |
130 | my ($root_node, $root_node_offset); |
27e0370d |
131 | |
e1861c2c |
132 | for my $i (0 .. $#{$inner_attrs->{from}}) { |
133 | my $node = $inner_attrs->{from}[$i]; |
27e0370d |
134 | my $h = (ref $node eq 'HASH') ? $node |
135 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
136 | : next |
137 | ; |
138 | |
1e4f9fb3 |
139 | if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) { |
97e130fa |
140 | $root_node = $h; |
141 | $root_node_offset = $i; |
27e0370d |
142 | last; |
143 | } |
144 | } |
145 | |
146 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
147 | unless $root_node; |
27e0370d |
148 | |
149 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
e1861c2c |
150 | my $colinfo = $self->_resolve_column_info($inner_attrs->{from}); |
27e0370d |
151 | my $selected_root_columns; |
152 | |
e1861c2c |
153 | for my $i (0 .. $#{$outer_attrs->{select}}) { |
154 | my $sel = $outer_attrs->{select}->[$i]; |
d28bb90d |
155 | |
1e4f9fb3 |
156 | next if ( |
157 | $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias |
158 | ); |
159 | |
d28bb90d |
160 | if (ref $sel eq 'HASH' ) { |
161 | $sel->{-as} ||= $attrs->{as}[$i]; |
e1861c2c |
162 | $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); |
d28bb90d |
163 | } |
27e0370d |
164 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
165 | $selected_root_columns->{$ci->{-colname}} = 1; |
166 | } |
d28bb90d |
167 | |
e1861c2c |
168 | push @{$inner_attrs->{select}}, $sel; |
bb9bffea |
169 | |
170 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
171 | } |
172 | |
97e130fa |
173 | # We will need to fetch all native columns in the inner subquery, which may |
174 | # be a part of an *outer* join condition, or an order_by (which needs to be |
e1861c2c |
175 | # preserved outside), or wheres. In other words everything but the inner |
176 | # selector |
97e130fa |
177 | # We can not just fetch everything because a potential has_many restricting |
178 | # join collapse *will not work* on heavy data types. |
e1861c2c |
179 | my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({ |
180 | %$inner_attrs, |
181 | select => [], |
182 | }); |
97e130fa |
183 | |
184 | for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { |
185 | my $ci = $colinfo->{$_} or next; |
186 | if ( |
1e4f9fb3 |
187 | $ci->{-source_alias} eq $root_alias |
97e130fa |
188 | and |
189 | ! $selected_root_columns->{$ci->{-colname}}++ |
190 | ) { |
191 | # adding it to both to keep limits not supporting dark selectors happy |
e1861c2c |
192 | push @{$inner_attrs->{select}}, $ci->{-fq_colname}; |
97e130fa |
193 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
194 | } |
195 | } |
196 | |
e1861c2c |
197 | # construct the inner {from} and lock it in a subquery |
48580715 |
198 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
199 | # throw away all non-selecting, non-restricting multijoins |
eb58c082 |
200 | # (since we def. do not care about multiplication of the contents of the subquery) |
6395604e |
201 | my $inner_subq = do { |
ea95892e |
202 | |
eb58c082 |
203 | # must use it here regardless of user requests (vastly gentler on optimizer) |
ea95892e |
204 | local $self->{_use_join_optimizer} = 1; |
205 | |
97e130fa |
206 | # throw away multijoins since we def. do not care about those inside the subquery |
e1861c2c |
207 | ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({ |
437a9cfa |
208 | %$inner_attrs, _force_prune_multiplying_joins => 1 |
209 | }); |
ea95892e |
210 | |
eb58c082 |
211 | # uh-oh a multiplier (which is not us) left in, this is a problem for limits |
212 | # we will need to add a group_by to collapse the resultset for proper counts |
0a3441ee |
213 | if ( |
eb58c082 |
214 | grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} } |
1e4f9fb3 |
215 | and |
560978e2 |
216 | # if there are user-supplied groups - assume user knows wtf they are up to |
217 | ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} ) |
0a3441ee |
218 | ) { |
1e4f9fb3 |
219 | |
eb58c082 |
220 | my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} }; |
1e4f9fb3 |
221 | |
eb58c082 |
222 | # *possibly* supplement the main selection with pks if not already |
223 | # there, as they will have to be a part of the group_by to collapse |
224 | # things properly |
225 | my $inner_select_with_extras; |
226 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
227 | or $self->throw_exception( sprintf |
228 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
229 | $root_node->{-rsrc}->source_name, |
e1861c2c |
230 | ); |
eb58c082 |
231 | for my $col (@pks) { |
232 | push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col |
233 | unless $cur_sel->{$col}++; |
1e4f9fb3 |
234 | } |
eb58c082 |
235 | |
236 | ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({ |
237 | %$inner_attrs, |
238 | $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (), |
239 | _aliastypes => $inner_aliastypes, |
240 | }); |
0a3441ee |
241 | } |
d28bb90d |
242 | |
e1861c2c |
243 | # we already optimized $inner_attrs->{from} above |
97e130fa |
244 | # and already local()ized |
245 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
246 | |
ea95892e |
247 | # generate the subquery |
6395604e |
248 | $self->_select_args_to_query ( |
e1861c2c |
249 | @{$inner_attrs}{qw(from select where)}, |
ea95892e |
250 | $inner_attrs, |
251 | ); |
d28bb90d |
252 | }; |
253 | |
254 | # Generate the outer from - this is relatively easy (really just replace |
255 | # the join slot with the subquery), with a major caveat - we can not |
256 | # join anything that is non-selecting (not part of the prefetch), but at |
257 | # the same time is a multi-type relationship, as it will explode the result. |
258 | # |
259 | # There are two possibilities here |
260 | # - either the join is non-restricting, in which case we simply throw it away |
261 | # - it is part of the restrictions, in which case we need to collapse the outer |
262 | # result by tackling yet another group_by to the outside of the query |
263 | |
27e0370d |
264 | # work on a shallow copy |
e1861c2c |
265 | my @orig_from = @{$attrs->{from}}; |
266 | |
052e8431 |
267 | |
e1861c2c |
268 | $outer_attrs->{from} = \ my @outer_from; |
53c29913 |
269 | |
27e0370d |
270 | # we may not be the head |
97e130fa |
271 | if ($root_node_offset) { |
e1861c2c |
272 | # first generate the outer_from, up to the substitution point |
273 | @outer_from = splice @orig_from, 0, $root_node_offset; |
27e0370d |
274 | |
e1861c2c |
275 | # substitute the subq at the right spot |
27e0370d |
276 | push @outer_from, [ |
277 | { |
1e4f9fb3 |
278 | -alias => $root_alias, |
97e130fa |
279 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
280 | $root_alias => $inner_subq, |
27e0370d |
281 | }, |
e1861c2c |
282 | # preserve attrs from what is now the head of the from after the splice |
283 | @{$orig_from[0]}[1 .. $#{$orig_from[0]}], |
27e0370d |
284 | ]; |
285 | } |
286 | else { |
27e0370d |
287 | @outer_from = { |
1e4f9fb3 |
288 | -alias => $root_alias, |
27e0370d |
289 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
290 | $root_alias => $inner_subq, |
27e0370d |
291 | }; |
d28bb90d |
292 | } |
293 | |
e1861c2c |
294 | shift @orig_from; # what we just replaced above |
97e130fa |
295 | |
ea95892e |
296 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
297 | # in what role |
975b573a |
298 | my $outer_aliastypes = $outer_attrs->{_aliastypes} = |
e1861c2c |
299 | $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from }); |
052e8431 |
300 | |
a4812caa |
301 | # unroll parents |
1e4f9fb3 |
302 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
303 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
304 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
305 | |
d28bb90d |
306 | # see what's left - throw away if not selecting/restricting |
eb58c082 |
307 | my $may_need_outer_group_by; |
e1861c2c |
308 | while (my $j = shift @orig_from) { |
d28bb90d |
309 | my $alias = $j->[0]{-alias}; |
310 | |
a4812caa |
311 | if ( |
312 | $outer_select_chain->{$alias} |
313 | ) { |
314 | push @outer_from, $j |
d28bb90d |
315 | } |
1e4f9fb3 |
316 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
317 | push @outer_from, $j; |
eb58c082 |
318 | $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
319 | } |
320 | } |
321 | |
eb58c082 |
322 | # also throw in a synthetic group_by if a non-selecting multiplier, |
323 | # to guard against cross-join explosions |
324 | # the logic is somewhat fragile, but relies on the idea that if a user supplied |
325 | # a group by on their own - they know what they were doing |
326 | if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
327 | ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({ |
560978e2 |
328 | %$outer_attrs, |
329 | from => \@outer_from, |
560978e2 |
330 | }); |
36fd7f07 |
331 | } |
332 | |
e1861c2c |
333 | # This is totally horrific - the {where} ends up in both the inner and outer query |
d28bb90d |
334 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
335 | # then if where conditions apply to the *right* side of the prefetch, you may have |
336 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
4a0eed52 |
337 | # the outer select to exclude joins you didn't want in the first place |
d28bb90d |
338 | # |
339 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
e1861c2c |
340 | return $outer_attrs; |
d28bb90d |
341 | } |
342 | |
1a736efb |
343 | # |
344 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
345 | # |
ad630f4b |
346 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
347 | # select/where/order/group attributes, in order to determine what |
4a0eed52 |
348 | # aliases are needed to fulfill the query. This information is used |
ad630f4b |
349 | # throughout the code to prune unnecessary JOINs from the queries |
350 | # in an attempt to reduce the execution time. |
351 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
352 | # happen is for it to fail due to some scalar SQL, which in turn will |
353 | # result in a vocal exception. |
539ffe87 |
354 | sub _resolve_aliastypes_from_select_args { |
e1861c2c |
355 | my ( $self, $attrs ) = @_; |
546f1cd9 |
356 | |
ad630f4b |
357 | $self->throw_exception ('Unable to analyze custom {from}') |
e1861c2c |
358 | if ref $attrs->{from} ne 'ARRAY'; |
546f1cd9 |
359 | |
ad630f4b |
360 | # what we will return |
964a3c71 |
361 | my $aliases_by_type; |
546f1cd9 |
362 | |
ad630f4b |
363 | # see what aliases are there to work with |
eb58c082 |
364 | # and record who is a multiplier and who is premultiplied |
ad630f4b |
365 | my $alias_list; |
e1861c2c |
366 | for my $node (@{$attrs->{from}}) { |
367 | |
368 | my $j = $node; |
ad630f4b |
369 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
370 | my $al = $j->{-alias} |
371 | or next; |
372 | |
373 | $alias_list->{$al} = $j; |
eb58c082 |
374 | |
375 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } |
a4812caa |
376 | # not array == {from} head == can't be multiplying |
eb58c082 |
377 | if ref($node) eq 'ARRAY' and ! $j->{-is_single}; |
378 | |
379 | $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] } |
380 | # parts of the path that are not us but are multiplying |
381 | if grep { $aliases_by_type->{multiplying}{$_} } |
382 | grep { $_ ne $al } |
383 | map { values %$_ } |
384 | @{ $j->{-join_path}||[] } |
546f1cd9 |
385 | } |
546f1cd9 |
386 | |
318e3d94 |
387 | # get a column to source/alias map (including unambiguous unqualified ones) |
e1861c2c |
388 | my $colinfo = $self->_resolve_column_info ($attrs->{from}); |
1a736efb |
389 | |
ad630f4b |
390 | # set up a botched SQLA |
391 | my $sql_maker = $self->sql_maker; |
07f31d19 |
392 | |
4c2b30d6 |
393 | # these are throw away results, do not pollute the bind stack |
0542ec57 |
394 | local $sql_maker->{where_bind}; |
395 | local $sql_maker->{group_bind}; |
396 | local $sql_maker->{having_bind}; |
97e130fa |
397 | local $sql_maker->{from_bind}; |
3f5b99fe |
398 | |
399 | # we can't scan properly without any quoting (\b doesn't cut it |
400 | # everywhere), so unless there is proper quoting set - use our |
401 | # own weird impossible character. |
402 | # Also in the case of no quoting, we need to explicitly disable |
403 | # name_sep, otherwise sorry nasty legacy syntax like |
404 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
405 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
406 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
407 | |
408 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
409 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
410 | # if we don't unset it we screw up retarded but unfortunately working |
411 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
412 | $sql_maker->{name_sep} = ''; |
413 | } |
414 | |
415 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
416 | |
1a736efb |
417 | # generate sql chunks |
418 | my $to_scan = { |
419 | restricting => [ |
a9e985b7 |
420 | ($sql_maker->_recurse_where ($attrs->{where}))[0], |
1e4f9fb3 |
421 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
422 | ], |
423 | grouping => [ |
424 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
425 | ], |
97e130fa |
426 | joining => [ |
427 | $sql_maker->_recurse_from ( |
e1861c2c |
428 | ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0], |
429 | @{$attrs->{from}}[1 .. $#{$attrs->{from}}], |
97e130fa |
430 | ), |
431 | ], |
1a736efb |
432 | selecting => [ |
ad1d374e |
433 | map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}}, |
1e4f9fb3 |
434 | ], |
435 | ordering => [ |
436 | map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker), |
1a736efb |
437 | ], |
438 | }; |
07f31d19 |
439 | |
0dadd60d |
440 | # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are |
441 | # bind value specs left in by the sloppy renderer above. It is ok to do this |
442 | # at this point, since we are going to end up rewriting this crap anyway |
443 | for my $v (values %$to_scan) { |
444 | my @nv; |
445 | for (@$v) { |
446 | next if ( |
447 | ! defined $_ |
448 | or |
449 | ( |
450 | ref $_ eq 'ARRAY' |
451 | and |
452 | ( @$_ == 0 or @$_ == 2 ) |
453 | ) |
454 | ); |
455 | |
456 | if (ref $_) { |
457 | require Data::Dumper::Concise; |
458 | $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) ); |
459 | } |
460 | |
461 | push @nv, $_; |
462 | } |
463 | |
464 | $v = \@nv; |
465 | } |
466 | |
467 | # kill all selectors which look like a proper subquery |
468 | # this is a sucky heuristic *BUT* - if we get it wrong the query will simply |
469 | # fail to run, so we are relatively safe |
470 | $to_scan->{selecting} = [ grep { |
471 | $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi |
472 | } @{ $to_scan->{selecting} || [] } ]; |
07f31d19 |
473 | |
318e3d94 |
474 | # first see if we have any exact matches (qualified or unqualified) |
475 | for my $type (keys %$to_scan) { |
476 | for my $piece (@{$to_scan->{$type}}) { |
477 | if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) { |
478 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
479 | $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece; |
480 | } |
481 | } |
482 | } |
483 | |
484 | # now loop through all fully qualified columns and get the corresponding |
1a736efb |
485 | # alias (should work even if they are in scalarrefs) |
ad630f4b |
486 | for my $alias (keys %$alias_list) { |
1a736efb |
487 | my $al_re = qr/ |
97e130fa |
488 | $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
1a736efb |
489 | | |
97e130fa |
490 | \b $alias \. ([^\s\)\($rquote]+)? |
1a736efb |
491 | /x; |
492 | |
1a736efb |
493 | for my $type (keys %$to_scan) { |
494 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
495 | if (my @matches = $piece =~ /$al_re/g) { |
496 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
497 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_" |
97e130fa |
498 | for grep { defined $_ } @matches; |
499 | } |
1a736efb |
500 | } |
ad630f4b |
501 | } |
1a736efb |
502 | } |
503 | |
504 | # now loop through unqualified column names, and try to locate them within |
505 | # the chunks |
506 | for my $col (keys %$colinfo) { |
3f5b99fe |
507 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
508 | |
97e130fa |
509 | my $col_re = qr/ $lquote ($col) $rquote /x; |
07f31d19 |
510 | |
1a736efb |
511 | for my $type (keys %$to_scan) { |
512 | for my $piece (@{$to_scan->{$type}}) { |
318e3d94 |
513 | if ( my @matches = $piece =~ /$col_re/g) { |
a4812caa |
514 | my $alias = $colinfo->{$col}{-source_alias}; |
97e130fa |
515 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
516 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
97e130fa |
517 | for grep { defined $_ } @matches; |
a4812caa |
518 | } |
1a736efb |
519 | } |
07f31d19 |
520 | } |
521 | } |
522 | |
523 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
524 | for my $j (values %$alias_list) { |
07f31d19 |
525 | my $alias = $j->{-alias} or next; |
97e130fa |
526 | $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if ( |
07f31d19 |
527 | (not $j->{-join_type}) |
528 | or |
529 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
530 | ); |
531 | } |
532 | |
1e4f9fb3 |
533 | for (keys %$aliases_by_type) { |
534 | delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}}; |
535 | } |
536 | |
964a3c71 |
537 | return $aliases_by_type; |
07f31d19 |
538 | } |
539 | |
eb58c082 |
540 | # This is the engine behind { distinct => 1 } and the general |
541 | # complex prefetch grouper |
0a3441ee |
542 | sub _group_over_selection { |
560978e2 |
543 | my ($self, $attrs) = @_; |
0a3441ee |
544 | |
560978e2 |
545 | my $colinfos = $self->_resolve_column_info ($attrs->{from}); |
0a3441ee |
546 | |
547 | my (@group_by, %group_index); |
548 | |
36fd7f07 |
549 | # the logic is: if it is a { func => val } we assume an aggregate, |
550 | # otherwise if \'...' or \[...] we assume the user knows what is |
551 | # going on thus group over it |
560978e2 |
552 | for (@{$attrs->{select}}) { |
0a3441ee |
553 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
554 | push @group_by, $_; |
555 | $group_index{$_}++; |
560978e2 |
556 | if ($colinfos->{$_} and $_ !~ /\./ ) { |
0a3441ee |
557 | # add a fully qualified version as well |
560978e2 |
558 | $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; |
0a3441ee |
559 | } |
07f31d19 |
560 | } |
561 | } |
ad630f4b |
562 | |
eb58c082 |
563 | my @order_by = $self->_extract_order_criteria($attrs->{order_by}) |
564 | or return (\@group_by, $attrs->{order_by}); |
565 | |
566 | # add any order_by parts that are not already present in the group_by |
567 | # to maintain SQL cross-compatibility and general sanity |
568 | # |
569 | # also in case the original selection is *not* unique, or in case part |
570 | # of the ORDER BY refers to a multiplier - we will need to replace the |
571 | # skipped order_by elements with their MIN/MAX equivalents as to maintain |
572 | # the proper overall order without polluting the group criteria (and |
573 | # possibly changing the outcome entirely) |
574 | |
575 | my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes); |
576 | |
577 | my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by); |
578 | |
579 | for my $o_idx (0 .. $#order_by) { |
580 | |
581 | # if the chunk is already a min/max function - there is nothing left to touch |
582 | next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix; |
583 | |
0a3441ee |
584 | # only consider real columns (for functions the user got to do an explicit group_by) |
eb58c082 |
585 | my $chunk_ci; |
586 | if ( |
587 | @{$order_by[$o_idx]} != 1 |
588 | or |
589 | # only declare an unknown *plain* identifier as "leftover" if we are called with |
590 | # aliastypes to examine. If there are none - we are still in _resolve_attrs, and |
591 | # can just assume the user knows what they want |
592 | ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} ) |
593 | ) { |
594 | push @$leftovers, $order_by[$o_idx][0]; |
14e26c5f |
595 | } |
560978e2 |
596 | |
eb58c082 |
597 | next unless $chunk_ci; |
598 | |
599 | # no duplication of group criteria |
600 | next if $group_index{$chunk_ci->{-fq_colname}}; |
601 | |
602 | $aliastypes ||= ( |
603 | $attrs->{_aliastypes} |
560978e2 |
604 | or |
eb58c082 |
605 | $self->_resolve_aliastypes_from_select_args({ |
606 | from => $attrs->{from}, |
607 | order_by => $attrs->{order_by}, |
608 | }) |
609 | ) if $group_already_unique; |
610 | |
611 | # check that we are not ordering by a multiplier (if a check is requested at all) |
612 | if ( |
613 | $group_already_unique |
614 | and |
615 | ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}} |
616 | and |
617 | ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}} |
560978e2 |
618 | ) { |
eb58c082 |
619 | push @group_by, $chunk_ci->{-fq_colname}; |
620 | $group_index{$chunk_ci->{-fq_colname}}++ |
560978e2 |
621 | } |
eb58c082 |
622 | else { |
623 | # We need to order by external columns without adding them to the group |
624 | # (eiehter a non-unique selection, or a multi-external) |
625 | # |
626 | # This doesn't really make sense in SQL, however from DBICs point |
627 | # of view is rather valid (e.g. order the leftmost objects by whatever |
628 | # criteria and get the offset/rows many). There is a way around |
629 | # this however in SQL - we simply tae the direction of each piece |
630 | # of the external order and convert them to MIN(X) for ASC or MAX(X) |
631 | # for DESC, and group_by the root columns. The end result should be |
632 | # exactly what we expect |
633 | |
634 | # FIXME - this code is a joke, will need to be completely rewritten in |
635 | # the DQ branch. But I need to push a POC here, otherwise the |
636 | # pesky tests won't pass |
637 | # wrap any part of the order_by that "responds" to an ordering alias |
638 | # into a MIN/MAX |
639 | $sql_maker ||= $self->sql_maker; |
640 | $order_chunks ||= [ |
641 | map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}) |
642 | ]; |
0a3441ee |
643 | |
eb58c082 |
644 | my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]); |
645 | |
646 | $new_order_by[$o_idx] = \[ |
647 | sprintf( '%s( %s )%s', |
648 | ($is_desc ? 'MAX' : 'MIN'), |
649 | $chunk, |
650 | ($is_desc ? ' DESC' : ''), |
651 | ), |
652 | @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ] |
653 | ]; |
654 | } |
0a3441ee |
655 | } |
656 | |
eb58c082 |
657 | $self->throw_exception ( sprintf |
9736be65 |
658 | 'Unable to programatically derive a required group_by from the supplied ' |
659 | . 'order_by criteria. To proceed either add an explicit group_by, or ' |
660 | . 'simplify your order_by to only include plain columns ' |
661 | . '(supplied order_by: %s)', |
eb58c082 |
662 | join ', ', map { "'$_'" } @$leftovers, |
663 | ) if $leftovers; |
664 | |
665 | # recreate the untouched order parts |
666 | if (@new_order_by) { |
667 | $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks ); |
668 | } |
669 | |
670 | return ( |
671 | \@group_by, |
672 | (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged |
673 | ); |
07f31d19 |
674 | } |
675 | |
d28bb90d |
676 | sub _resolve_ident_sources { |
677 | my ($self, $ident) = @_; |
678 | |
679 | my $alias2source = {}; |
d28bb90d |
680 | |
681 | # the reason this is so contrived is that $ident may be a {from} |
682 | # structure, specifying multiple tables to join |
6298a324 |
683 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
684 | # this is compat mode for insert/update/delete which do not deal with aliases |
685 | $alias2source->{me} = $ident; |
d28bb90d |
686 | } |
687 | elsif (ref $ident eq 'ARRAY') { |
688 | |
689 | for (@$ident) { |
690 | my $tabinfo; |
691 | if (ref $_ eq 'HASH') { |
692 | $tabinfo = $_; |
d28bb90d |
693 | } |
694 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
695 | $tabinfo = $_->[0]; |
696 | } |
697 | |
4376a157 |
698 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
699 | if ($tabinfo->{-rsrc}); |
d28bb90d |
700 | } |
701 | } |
702 | |
90f10b5a |
703 | return $alias2source; |
d28bb90d |
704 | } |
705 | |
706 | # Takes $ident, \@column_names |
707 | # |
708 | # returns { $column_name => \%column_info, ... } |
709 | # also note: this adds -result_source => $rsrc to the column info |
710 | # |
09e14fdc |
711 | # If no columns_names are supplied returns info about *all* columns |
712 | # for all sources |
d28bb90d |
713 | sub _resolve_column_info { |
714 | my ($self, $ident, $colnames) = @_; |
8d005ad9 |
715 | |
716 | return {} if $colnames and ! @$colnames; |
717 | |
90f10b5a |
718 | my $alias2src = $self->_resolve_ident_sources($ident); |
d28bb90d |
719 | |
52416317 |
720 | my (%seen_cols, @auto_colnames); |
d28bb90d |
721 | |
722 | # compile a global list of column names, to be able to properly |
723 | # disambiguate unqualified column names (if at all possible) |
724 | for my $alias (keys %$alias2src) { |
725 | my $rsrc = $alias2src->{$alias}; |
726 | for my $colname ($rsrc->columns) { |
727 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
728 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
729 | } |
730 | } |
731 | |
09e14fdc |
732 | $colnames ||= [ |
733 | @auto_colnames, |
734 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
735 | ]; |
736 | |
52416317 |
737 | my (%return, $colinfos); |
d28bb90d |
738 | foreach my $col (@$colnames) { |
52416317 |
739 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
740 | |
52416317 |
741 | # if the column was seen exactly once - we know which rsrc it came from |
742 | $source_alias ||= $seen_cols{$colname}[0] |
743 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
744 | |
52416317 |
745 | next unless $source_alias; |
746 | |
747 | my $rsrc = $alias2src->{$source_alias} |
748 | or next; |
749 | |
750 | $return{$col} = { |
6395604e |
751 | %{ |
752 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
753 | || |
754 | $self->throw_exception( |
755 | "No such column '$colname' on source " . $rsrc->source_name |
756 | ); |
757 | }, |
d28bb90d |
758 | -result_source => $rsrc, |
52416317 |
759 | -source_alias => $source_alias, |
81bf295c |
760 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
761 | -colname => $colname, |
d28bb90d |
762 | }; |
81bf295c |
763 | |
764 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
765 | } |
766 | |
767 | return \%return; |
768 | } |
769 | |
289ac713 |
770 | # The DBIC relationship chaining implementation is pretty simple - every |
771 | # new related_relationship is pushed onto the {from} stack, and the {select} |
772 | # window simply slides further in. This means that when we count somewhere |
773 | # in the middle, we got to make sure that everything in the join chain is an |
774 | # actual inner join, otherwise the count will come back with unpredictable |
775 | # results (a resultset may be generated with _some_ rows regardless of if |
776 | # the relation which the $rs currently selects has rows or not). E.g. |
777 | # $artist_rs->cds->count - normally generates: |
778 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
779 | # which actually returns the number of artists * (number of cds || 1) |
780 | # |
781 | # So what we do here is crawl {from}, determine if the current alias is at |
782 | # the top of the stack, and if not - make sure the chain is inner-joined down |
783 | # to the root. |
784 | # |
31a8aaaf |
785 | sub _inner_join_to_node { |
289ac713 |
786 | my ($self, $from, $alias) = @_; |
787 | |
302d35f8 |
788 | my $switch_branch = $self->_find_join_path_to_node($from, $alias); |
289ac713 |
789 | |
302d35f8 |
790 | return $from unless @{$switch_branch||[]}; |
289ac713 |
791 | |
792 | # So it looks like we will have to switch some stuff around. |
793 | # local() is useless here as we will be leaving the scope |
794 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
795 | # So replace the first hashref in the node arrayref manually |
289ac713 |
796 | my @new_from = ($from->[0]); |
faeb2407 |
797 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
798 | |
799 | for my $j (@{$from}[1 .. $#$from]) { |
800 | my $jalias = $j->[0]{-alias}; |
801 | |
802 | if ($sw_idx->{$jalias}) { |
803 | my %attrs = %{$j->[0]}; |
804 | delete $attrs{-join_type}; |
805 | push @new_from, [ |
806 | \%attrs, |
807 | @{$j}[ 1 .. $#$j ], |
808 | ]; |
809 | } |
810 | else { |
811 | push @new_from, $j; |
812 | } |
813 | } |
814 | |
815 | return \@new_from; |
816 | } |
817 | |
302d35f8 |
818 | sub _find_join_path_to_node { |
819 | my ($self, $from, $target_alias) = @_; |
820 | |
821 | # subqueries and other oddness are naturally not supported |
822 | return undef if ( |
823 | ref $from ne 'ARRAY' |
824 | || |
825 | ref $from->[0] ne 'HASH' |
826 | || |
827 | ! defined $from->[0]{-alias} |
828 | ); |
829 | |
830 | # no path - the head is the alias |
831 | return [] if $from->[0]{-alias} eq $target_alias; |
832 | |
833 | for my $i (1 .. $#$from) { |
834 | return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias ); |
835 | } |
836 | |
837 | # something else went quite wrong |
838 | return undef; |
839 | } |
840 | |
bac358c9 |
841 | sub _extract_order_criteria { |
1a736efb |
842 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
843 | |
1a736efb |
844 | my $parser = sub { |
e6977bbb |
845 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
846 | |
1a736efb |
847 | return scalar $sql_maker->_order_by_chunks ($order_by) |
848 | unless wantarray; |
c0748280 |
849 | |
e6977bbb |
850 | my ($lq, $rq, $sep) = map { quotemeta($_) } ( |
851 | ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars), |
852 | $sql_maker->name_sep |
853 | ); |
854 | |
1a736efb |
855 | my @chunks; |
bac358c9 |
856 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
857 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
858 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
859 | |
860 | # order criteria may have come back pre-quoted (literals and whatnot) |
861 | # this is fragile, but the best we can currently do |
862 | $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe |
863 | or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x; |
864 | |
1a736efb |
865 | push @chunks, $chunk; |
bac6c4fb |
866 | } |
1a736efb |
867 | |
868 | return @chunks; |
869 | }; |
870 | |
871 | if ($sql_maker) { |
872 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
873 | } |
874 | else { |
1a736efb |
875 | $sql_maker = $self->sql_maker; |
e6977bbb |
876 | |
877 | # pass these in to deal with literals coming from |
878 | # the user or the deep guts of prefetch |
879 | my $orig_quote_chars = [$sql_maker->_quote_chars]; |
880 | |
1a736efb |
881 | local $sql_maker->{quote_char}; |
e6977bbb |
882 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
883 | } |
bac6c4fb |
884 | } |
885 | |
7cec4356 |
886 | sub _order_by_is_stable { |
5f11e54f |
887 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
888 | |
eb58c082 |
889 | my @cols = ( |
8d005ad9 |
890 | ( map { $_->[0] } $self->_extract_order_criteria($order_by) ), |
8e40a627 |
891 | ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ), |
df4312bc |
892 | ) or return 0; |
eb58c082 |
893 | |
894 | my $colinfo = $self->_resolve_column_info($ident, \@cols); |
895 | |
896 | return keys %$colinfo |
897 | ? $self->_columns_comprise_identifying_set( $colinfo, \@cols ) |
df4312bc |
898 | : 0 |
eb58c082 |
899 | ; |
900 | } |
c0748280 |
901 | |
eb58c082 |
902 | sub _columns_comprise_identifying_set { |
903 | my ($self, $colinfo, $columns) = @_; |
7cec4356 |
904 | |
905 | my $cols_per_src; |
eb58c082 |
906 | $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_ |
907 | for grep { defined $_ } @{$colinfo}{@$columns}; |
7cec4356 |
908 | |
909 | for (values %$cols_per_src) { |
910 | my $src = (values %$_)[0]->{-result_source}; |
911 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
912 | } |
913 | |
df4312bc |
914 | return 0; |
7cec4356 |
915 | } |
916 | |
df4312bc |
917 | # this is almost similar to _order_by_is_stable, except it takes |
0e81e691 |
918 | # a single rsrc, and will succeed only if the first portion of the order |
919 | # by is stable. |
920 | # returns that portion as a colinfo hashref on success |
df4312bc |
921 | sub _extract_colinfo_of_stable_main_source_order_by_portion { |
302d35f8 |
922 | my ($self, $attrs) = @_; |
0e81e691 |
923 | |
302d35f8 |
924 | my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias}); |
925 | |
926 | return unless defined $nodes; |
0e81e691 |
927 | |
928 | my @ord_cols = map |
929 | { $_->[0] } |
302d35f8 |
930 | ( $self->_extract_order_criteria($attrs->{order_by}) ) |
0e81e691 |
931 | ; |
932 | return unless @ord_cols; |
933 | |
302d35f8 |
934 | my $valid_aliases = { map { $_ => 1 } ( |
935 | $attrs->{from}[0]{-alias}, |
936 | map { values %$_ } @$nodes, |
937 | ) }; |
318e3d94 |
938 | |
302d35f8 |
939 | my $colinfos = $self->_resolve_column_info($attrs->{from}); |
940 | |
941 | my ($colinfos_to_return, $seen_main_src_cols); |
942 | |
943 | for my $col (@ord_cols) { |
944 | # if order criteria is unresolvable - there is nothing we can do |
945 | my $colinfo = $colinfos->{$col} or last; |
946 | |
947 | # if we reached the end of the allowed aliases - also nothing we can do |
948 | last unless $valid_aliases->{$colinfo->{-source_alias}}; |
949 | |
950 | $colinfos_to_return->{$col} = $colinfo; |
951 | |
952 | $seen_main_src_cols->{$colinfo->{-colname}} = 1 |
953 | if $colinfo->{-source_alias} eq $attrs->{alias}; |
0e81e691 |
954 | } |
955 | |
302d35f8 |
956 | # FIXME the condition may be singling out things on its own, so we |
957 | # conceivable could come back wi "stable-ordered by nothing" |
958 | # not confient enough in the parser yet, so punt for the time being |
959 | return unless $seen_main_src_cols; |
0e81e691 |
960 | |
302d35f8 |
961 | my $main_src_fixed_cols_from_cond = [ $attrs->{where} |
962 | ? ( |
963 | map |
964 | { |
965 | ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} ) |
966 | ? $colinfos->{$_}{-colname} |
967 | : () |
968 | } |
8e40a627 |
969 | keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) } |
302d35f8 |
970 | ) |
971 | : () |
972 | ]; |
0e81e691 |
973 | |
302d35f8 |
974 | return $attrs->{result_source}->_identifying_column_set([ |
975 | keys %$seen_main_src_cols, |
976 | @$main_src_fixed_cols_from_cond, |
977 | ]) ? $colinfos_to_return : (); |
0e81e691 |
978 | } |
979 | |
8d005ad9 |
980 | # Attempts to flatten a passed in SQLA condition as much as possible towards |
981 | # a plain hashref, *without* altering its semantics. Required by |
982 | # create/populate being able to extract definitive conditions from preexisting |
983 | # resultset {where} stacks |
984 | # |
985 | # FIXME - while relatively robust, this is still imperfect, one of the first |
986 | # things to tackle with DQ |
987 | sub _collapse_cond { |
988 | my ($self, $where, $where_is_anded_array) = @_; |
989 | |
135ac69d |
990 | my $fin; |
991 | |
8d005ad9 |
992 | if (! $where) { |
993 | return; |
994 | } |
995 | elsif ($where_is_anded_array or ref $where eq 'HASH') { |
996 | |
997 | my @pairs; |
998 | |
999 | my @pieces = $where_is_anded_array ? @$where : $where; |
1000 | while (@pieces) { |
1001 | my $chunk = shift @pieces; |
1002 | |
1003 | if (ref $chunk eq 'HASH') { |
e466c62b |
1004 | for (sort keys %$chunk) { |
1005 | |
1006 | # Match SQLA 1.79 behavior |
1007 | if ($_ eq '') { |
1008 | is_literal_value($chunk->{$_}) |
1009 | ? carp 'Hash-pairs consisting of an empty string with a literal are deprecated, use -and => [ $literal ] instead' |
1010 | : $self->throw_exception("Supplying an empty left hand side argument is not supported in hash-pairs") |
1011 | ; |
1012 | } |
1013 | |
1014 | push @pairs, $_ => $chunk->{$_}; |
1015 | } |
8d005ad9 |
1016 | } |
1017 | elsif (ref $chunk eq 'ARRAY') { |
6565d2c3 |
1018 | push @pairs, -or => $chunk |
8d005ad9 |
1019 | if @$chunk; |
1020 | } |
b34d9331 |
1021 | elsif ( ! length ref $chunk) { |
e466c62b |
1022 | |
1023 | # Match SQLA 1.79 behavior |
1024 | $self->throw_exception("Supplying an empty left hand side argument is not supported in array-pairs") |
1025 | if $where_is_anded_array and (! defined $chunk or $chunk eq ''); |
1026 | |
6565d2c3 |
1027 | push @pairs, $chunk, shift @pieces; |
8d005ad9 |
1028 | } |
1029 | else { |
6565d2c3 |
1030 | push @pairs, '', $chunk; |
8d005ad9 |
1031 | } |
1032 | } |
1033 | |
1034 | return unless @pairs; |
1035 | |
1036 | my @conds = $self->_collapse_cond_unroll_pairs(\@pairs) |
1037 | or return; |
1038 | |
1039 | # Consolidate various @conds back into something more compact |
8d005ad9 |
1040 | for my $c (@conds) { |
1041 | if (ref $c ne 'HASH') { |
1042 | push @{$fin->{-and}}, $c; |
1043 | } |
1044 | else { |
1045 | for my $col (sort keys %$c) { |
8d005ad9 |
1046 | |
135ac69d |
1047 | # consolidate all -and nodes |
1048 | if ($col =~ /^\-and$/i) { |
1049 | push @{$fin->{-and}}, |
1050 | ref $c->{$col} eq 'ARRAY' ? @{$c->{$col}} |
1051 | : ref $c->{$col} eq 'HASH' ? %{$c->{$col}} |
1052 | : { $col => $c->{$col} } |
1053 | ; |
1054 | } |
1055 | elsif ($col =~ /^\-/) { |
1056 | push @{$fin->{-and}}, { $col => $c->{$col} }; |
1057 | } |
1058 | elsif (exists $fin->{$col}) { |
1059 | $fin->{$col} = [ -and => map { |
1060 | (ref $_ eq 'ARRAY' and ($_->[0]||'') =~ /^\-and$/i ) |
1061 | ? @{$_}[1..$#$_] |
1062 | : $_ |
1063 | ; |
1064 | } ($fin->{$col}, $c->{$col}) ]; |
8d005ad9 |
1065 | } |
1066 | else { |
1067 | $fin->{$col} = $c->{$col}; |
1068 | } |
1069 | } |
1070 | } |
1071 | } |
8d005ad9 |
1072 | } |
1073 | elsif (ref $where eq 'ARRAY') { |
22485a7e |
1074 | # we are always at top-level here, it is safe to dump empty *standalone* pieces |
1075 | my $fin_idx; |
8d005ad9 |
1076 | |
22485a7e |
1077 | for (my $i = 0; $i <= $#$where; $i++ ) { |
8d005ad9 |
1078 | |
e466c62b |
1079 | # Match SQLA 1.79 behavior |
1080 | $self->throw_exception( |
1081 | "Supplying an empty left hand side argument is not supported in array-pairs" |
1082 | ) if (! defined $where->[$i] or ! length $where->[$i]); |
1083 | |
22485a7e |
1084 | my $logic_mod = lc ( ($where->[$i] =~ /^(\-(?:and|or))$/i)[0] || '' ); |
1085 | |
1086 | if ($logic_mod) { |
1087 | $i++; |
1088 | $self->throw_exception("Unsupported top-level op/arg pair: [ $logic_mod => $where->[$i] ]") |
1089 | unless ref $where->[$i] eq 'HASH' or ref $where->[$i] eq 'ARRAY'; |
1090 | |
1091 | my $sub_elt = $self->_collapse_cond({ $logic_mod => $where->[$i] }) |
1092 | or next; |
1093 | |
e466c62b |
1094 | my @keys = keys %$sub_elt; |
1095 | if ( @keys == 1 and $keys[0] !~ /^\-/ ) { |
1096 | $fin_idx->{ "COL_$keys[0]_" . serialize $sub_elt } = $sub_elt; |
1097 | } |
1098 | else { |
1099 | $fin_idx->{ "SER_" . serialize $sub_elt } = $sub_elt; |
1100 | } |
22485a7e |
1101 | } |
1102 | elsif (! length ref $where->[$i] ) { |
135ac69d |
1103 | my $sub_elt = $self->_collapse_cond({ @{$where}[$i, $i+1] }) |
1104 | or next; |
1105 | |
1106 | $fin_idx->{ "COL_$where->[$i]_" . serialize $sub_elt } = $sub_elt; |
22485a7e |
1107 | $i++; |
8d005ad9 |
1108 | } |
1109 | else { |
135ac69d |
1110 | $fin_idx->{ "SER_" . serialize $where->[$i] } = $self->_collapse_cond( $where->[$i] ) || next; |
8d005ad9 |
1111 | } |
1112 | } |
22485a7e |
1113 | |
07add744 |
1114 | if (! $fin_idx) { |
1115 | return; |
1116 | } |
1117 | elsif ( keys %$fin_idx == 1 ) { |
1118 | $fin = (values %$fin_idx)[0]; |
1119 | } |
1120 | else { |
1121 | my @or; |
1122 | |
1123 | # at this point everything is at most one level deep - unroll if needed |
1124 | for (sort keys %$fin_idx) { |
1125 | if ( ref $fin_idx->{$_} eq 'HASH' and keys %{$fin_idx->{$_}} == 1 ) { |
1126 | my ($l, $r) = %{$fin_idx->{$_}}; |
1127 | |
1128 | if ( |
1129 | ref $r eq 'ARRAY' |
1130 | and |
1131 | ( |
1132 | ( @$r == 1 and $l =~ /^\-and$/i ) |
1133 | or |
1134 | $l =~ /^\-or$/i |
1135 | ) |
1136 | ) { |
1137 | push @or, @$r |
1138 | } |
1139 | |
1140 | elsif ( |
1141 | ref $r eq 'HASH' |
1142 | and |
1143 | keys %$r == 1 |
1144 | and |
1145 | $l =~ /^\-(?:and|or)$/i |
1146 | ) { |
1147 | push @or, %$r; |
1148 | } |
1149 | |
1150 | else { |
1151 | push @or, $l, $r; |
1152 | } |
1153 | } |
1154 | else { |
1155 | push @or, $fin_idx->{$_}; |
1156 | } |
1157 | } |
1158 | |
1159 | $fin->{-or} = \@or; |
1160 | } |
8d005ad9 |
1161 | } |
1162 | else { |
1163 | # not a hash not an array |
07add744 |
1164 | $fin = { -and => [ $where ] }; |
135ac69d |
1165 | } |
1166 | |
1167 | # unroll single-element -and's |
1168 | while ( |
1169 | $fin->{-and} |
1170 | and |
1171 | @{$fin->{-and}} < 2 |
1172 | ) { |
1173 | my $and = delete $fin->{-and}; |
1174 | last if @$and == 0; |
1175 | |
1176 | # at this point we have @$and == 1 |
1177 | if ( |
1178 | ref $and->[0] eq 'HASH' |
1179 | and |
1180 | ! grep { exists $fin->{$_} } keys %{$and->[0]} |
1181 | ) { |
1182 | $fin = { |
1183 | %$fin, %{$and->[0]} |
1184 | }; |
1185 | } |
07add744 |
1186 | else { |
1187 | $fin->{-and} = $and; |
1188 | last; |
1189 | } |
135ac69d |
1190 | } |
1191 | |
1192 | # compress same-column conds found in $fin |
1193 | for my $col ( grep { $_ !~ /^\-/ } keys %$fin ) { |
1194 | next unless ref $fin->{$col} eq 'ARRAY' and ($fin->{$col}[0]||'') =~ /^\-and$/i; |
1195 | my $val_bag = { map { |
5379386e |
1196 | (! defined $_ ) ? ( UNDEF => undef ) |
1197 | : ( ! length ref $_ or is_plain_value $_ ) ? ( "VAL_$_" => $_ ) |
135ac69d |
1198 | : ( ( 'SER_' . serialize $_ ) => $_ ) |
1199 | } @{$fin->{$col}}[1 .. $#{$fin->{$col}}] }; |
1200 | |
1201 | if (keys %$val_bag == 1 ) { |
1202 | ($fin->{$col}) = values %$val_bag; |
1203 | } |
1204 | else { |
1205 | $fin->{$col} = [ -and => map { $val_bag->{$_} } sort keys %$val_bag ]; |
1206 | } |
8d005ad9 |
1207 | } |
1208 | |
135ac69d |
1209 | return keys %$fin ? $fin : (); |
8d005ad9 |
1210 | } |
1211 | |
1212 | sub _collapse_cond_unroll_pairs { |
1213 | my ($self, $pairs) = @_; |
1214 | |
1215 | my @conds; |
1216 | |
1217 | while (@$pairs) { |
6565d2c3 |
1218 | my ($lhs, $rhs) = splice @$pairs, 0, 2; |
8d005ad9 |
1219 | |
1220 | if ($lhs eq '') { |
1221 | push @conds, $self->_collapse_cond($rhs); |
1222 | } |
1223 | elsif ( $lhs =~ /^\-and$/i ) { |
1224 | push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY')); |
1225 | } |
1226 | elsif ( $lhs =~ /^\-or$/i ) { |
1227 | push @conds, $self->_collapse_cond( |
1228 | (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs |
1229 | ); |
1230 | } |
1231 | else { |
1232 | if (ref $rhs eq 'HASH' and ! keys %$rhs) { |
1233 | # FIXME - SQLA seems to be doing... nothing...? |
1234 | } |
f6fff270 |
1235 | # normalize top level -ident, for saner extract_fixed_condition_columns code |
5f35ba0f |
1236 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) { |
1237 | push @conds, { $lhs => { '=', $rhs } }; |
1238 | } |
1239 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-value} and is_plain_value $rhs->{-value}) { |
1240 | push @conds, { $lhs => $rhs->{-value} }; |
1241 | } |
8d005ad9 |
1242 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) { |
f6fff270 |
1243 | if ( length ref $rhs->{'='} and is_literal_value $rhs->{'='} ) { |
5f35ba0f |
1244 | push @conds, { $lhs => $rhs }; |
1245 | } |
1246 | else { |
6565d2c3 |
1247 | for my $p ($self->_collapse_cond_unroll_pairs([ $lhs => $rhs->{'='} ])) { |
5f35ba0f |
1248 | |
1249 | # extra sanity check |
1250 | if (keys %$p > 1) { |
1251 | require Data::Dumper::Concise; |
1252 | local $Data::Dumper::Deepcopy = 1; |
1253 | $self->throw_exception( |
1254 | "Internal error: unexpected collapse unroll:" |
1255 | . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p } |
1256 | ); |
1257 | } |
8d005ad9 |
1258 | |
5f35ba0f |
1259 | my ($l, $r) = %$p; |
8d005ad9 |
1260 | |
f6fff270 |
1261 | push @conds, ( |
1262 | ! length ref $r |
1263 | or |
1264 | # the unroller recursion may return a '=' prepended value already |
1265 | ref $r eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='} |
1266 | or |
1267 | is_plain_value($r) |
1268 | ) |
5f35ba0f |
1269 | ? { $l => $r } |
1270 | : { $l => { '=' => $r } } |
1271 | ; |
1272 | } |
8d005ad9 |
1273 | } |
1274 | } |
1275 | elsif (ref $rhs eq 'ARRAY') { |
1276 | # some of these conditionals encounter multi-values - roll them out using |
1277 | # an unshift, which will cause extra looping in the while{} above |
1278 | if (! @$rhs ) { |
1279 | push @conds, { $lhs => [] }; |
1280 | } |
1281 | elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) { |
1282 | $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ") |
1283 | if @$rhs == 1; |
1284 | |
1285 | if( $rhs->[0] =~ /^\-and$/i ) { |
6565d2c3 |
1286 | unshift @$pairs, map { $lhs => $_ } @{$rhs}[1..$#$rhs]; |
8d005ad9 |
1287 | } |
1288 | # if not an AND then it's an OR |
1289 | elsif(@$rhs == 2) { |
6565d2c3 |
1290 | unshift @$pairs, $lhs => $rhs->[1]; |
8d005ad9 |
1291 | } |
1292 | else { |
953d5b7d |
1293 | push @conds, { $lhs => [ @{$rhs}[1..$#$rhs] ] }; |
8d005ad9 |
1294 | } |
1295 | } |
1296 | elsif (@$rhs == 1) { |
6565d2c3 |
1297 | unshift @$pairs, $lhs => $rhs->[0]; |
8d005ad9 |
1298 | } |
1299 | else { |
1300 | push @conds, { $lhs => $rhs }; |
1301 | } |
1302 | } |
c1f3f2e8 |
1303 | # unroll func + { -value => ... } |
1304 | elsif ( |
1305 | ref $rhs eq 'HASH' |
1306 | and |
1307 | ( my ($subop) = keys %$rhs ) == 1 |
1308 | and |
1309 | length ref ((values %$rhs)[0]) |
1310 | and |
1311 | my $vref = is_plain_value( (values %$rhs)[0] ) |
1312 | ) { |
5379386e |
1313 | push @conds, { $lhs => { $subop => $$vref } } |
c1f3f2e8 |
1314 | } |
8d005ad9 |
1315 | else { |
1316 | push @conds, { $lhs => $rhs }; |
1317 | } |
1318 | } |
1319 | } |
1320 | |
1321 | return @conds; |
1322 | } |
1323 | |
8e40a627 |
1324 | # Analyzes a given condition and attempts to extract all columns |
1325 | # with a definitive fixed-condition criteria. Returns a hashref |
1326 | # of k/v pairs suitable to be passed to set_columns(), with a |
1327 | # MAJOR CAVEAT - multi-value (contradictory) equalities are still |
1328 | # represented as a reference to the UNRESOVABLE_CONDITION constant |
1329 | # The reason we do this is that some codepaths only care about the |
1330 | # codition being stable, as opposed to actually making sense |
5f11e54f |
1331 | # |
8e40a627 |
1332 | # The normal mode is used to figure out if a resultset is constrained |
1333 | # to a column which is part of a unique constraint, which in turn |
1334 | # allows us to better predict how ordering will behave etc. |
1335 | # |
1336 | # With the optional "consider_nulls" boolean argument, the function |
1337 | # is instead used to infer inambiguous values from conditions |
1338 | # (e.g. the inheritance of resultset conditions on new_result) |
1339 | # |
5f11e54f |
1340 | sub _extract_fixed_condition_columns { |
8e40a627 |
1341 | my ($self, $where, $consider_nulls) = @_; |
1342 | my $where_hash = $self->_collapse_cond($_[1]); |
1343 | |
1344 | my $res = {}; |
1345 | my ($c, $v); |
1346 | for $c (keys %$where_hash) { |
1347 | my $vals; |
1348 | |
1349 | if (!defined ($v = $where_hash->{$c}) ) { |
b34d9331 |
1350 | $vals->{UNDEF} = $v if $consider_nulls |
8e40a627 |
1351 | } |
1352 | elsif ( |
8e40a627 |
1353 | ref $v eq 'HASH' |
1354 | and |
1355 | keys %$v == 1 |
5f35ba0f |
1356 | ) { |
1357 | if (exists $v->{-value}) { |
1358 | if (defined $v->{-value}) { |
b34d9331 |
1359 | $vals->{"VAL_$v->{-value}"} = $v->{-value} |
5f35ba0f |
1360 | } |
1361 | elsif( $consider_nulls ) { |
b34d9331 |
1362 | $vals->{UNDEF} = $v->{-value}; |
5f35ba0f |
1363 | } |
1364 | } |
8e40a627 |
1365 | # do not need to check for plain values - _collapse_cond did it for us |
f6fff270 |
1366 | elsif( |
1367 | length ref $v->{'='} |
1368 | and |
1369 | ( |
1370 | ( ref $v->{'='} eq 'HASH' and keys %{$v->{'='}} == 1 and exists $v->{'='}{-ident} ) |
1371 | or |
1372 | is_literal_value($v->{'='}) |
1373 | ) |
1374 | ) { |
b34d9331 |
1375 | $vals->{ 'SER_' . serialize $v->{'='} } = $v->{'='}; |
5f35ba0f |
1376 | } |
1377 | } |
1378 | elsif ( |
1379 | ! length ref $v |
1380 | or |
1381 | is_plain_value ($v) |
8e40a627 |
1382 | ) { |
b34d9331 |
1383 | $vals->{"VAL_$v"} = $v; |
8e40a627 |
1384 | } |
1385 | elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') { |
1386 | for ( @{$v}[1..$#$v] ) { |
1387 | my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls'); # always fish nulls out on recursion |
1388 | next unless exists $subval->{$c}; # didn't find anything |
b34d9331 |
1389 | $vals->{ |
1390 | ! defined $subval->{$c} ? 'UNDEF' |
1391 | : ( ! length ref $subval->{$c} or is_plain_value $subval->{$c} ) ? "VAL_$subval->{$c}" |
1392 | : ( 'SER_' . serialize $subval->{$c} ) |
1393 | } = $subval->{$c}; |
8d005ad9 |
1394 | } |
5f11e54f |
1395 | } |
8e40a627 |
1396 | |
1397 | if (keys %$vals == 1) { |
1398 | ($res->{$c}) = (values %$vals) |
b34d9331 |
1399 | unless !$consider_nulls and exists $vals->{UNDEF}; |
8e40a627 |
1400 | } |
1401 | elsif (keys %$vals > 1) { |
1402 | $res->{$c} = UNRESOLVABLE_CONDITION; |
1403 | } |
5f11e54f |
1404 | } |
8d005ad9 |
1405 | |
8e40a627 |
1406 | $res; |
c0748280 |
1407 | } |
bac6c4fb |
1408 | |
d28bb90d |
1409 | 1; |