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); |
6298a324 |
20 | use namespace::clean; |
d28bb90d |
21 | |
22 | # |
052e8431 |
23 | # This code will remove non-selecting/non-restricting joins from |
4b1b5ea3 |
24 | # {from} specs, aiding the RDBMS query optimizer |
052e8431 |
25 | # |
26 | sub _prune_unused_joins { |
e1861c2c |
27 | my ($self, $attrs) = @_; |
ea95892e |
28 | |
e1861c2c |
29 | # only standard {from} specs are supported, and we could be disabled in general |
30 | return ($attrs->{from}, {}) unless ( |
31 | ref $attrs->{from} eq 'ARRAY' |
32 | and |
33 | @{$attrs->{from}} > 1 |
34 | and |
35 | ref $attrs->{from}[0] eq 'HASH' |
36 | and |
37 | ref $attrs->{from}[1] eq 'ARRAY' |
38 | and |
39 | $self->_use_join_optimizer |
40 | ); |
052e8431 |
41 | |
eb58c082 |
42 | my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs); |
4b1b5ea3 |
43 | |
eb58c082 |
44 | my $new_aliastypes = { %$orig_aliastypes }; |
45 | |
46 | # we will be recreating this entirely |
47 | my @reclassify = 'joining'; |
97e130fa |
48 | |
4b1b5ea3 |
49 | # a grouped set will not be affected by amount of rows. Thus any |
eb58c082 |
50 | # purely multiplicator classifications can go |
51 | # (will be reintroduced below if needed by something else) |
52 | push @reclassify, qw(multiplying premultiplied) |
437a9cfa |
53 | if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by}; |
4b1b5ea3 |
54 | |
eb58c082 |
55 | # nuke what will be recalculated |
56 | delete @{$new_aliastypes}{@reclassify}; |
57 | |
e1861c2c |
58 | my @newfrom = $attrs->{from}[0]; # FROM head is always present |
052e8431 |
59 | |
eb58c082 |
60 | # recalculate what we need once the multipliers are potentially gone |
61 | # ignore premultiplies, since they do not add any value to anything |
a4812caa |
62 | my %need_joins; |
eb58c082 |
63 | for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) { |
a4812caa |
64 | # add all requested aliases |
65 | $need_joins{$_} = 1 for keys %$_; |
66 | |
67 | # add all their parents (as per joinpath which is an AoH { table => alias }) |
97e130fa |
68 | $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; |
a4812caa |
69 | } |
97e130fa |
70 | |
e1861c2c |
71 | for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) { |
539ffe87 |
72 | push @newfrom, $j if ( |
a6ef93cb |
73 | (! defined $j->[0]{-alias}) # legacy crap |
539ffe87 |
74 | || |
75 | $need_joins{$j->[0]{-alias}} |
76 | ); |
052e8431 |
77 | } |
78 | |
eb58c082 |
79 | # we have a new set of joiners - for everything we nuked pull the classification |
80 | # off the original stack |
81 | for my $ctype (@reclassify) { |
82 | $new_aliastypes->{$ctype} = { map |
83 | { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () } |
84 | keys %{$orig_aliastypes->{$ctype}} |
85 | } |
86 | } |
87 | |
88 | return ( \@newfrom, $new_aliastypes ); |
052e8431 |
89 | } |
90 | |
052e8431 |
91 | # |
d28bb90d |
92 | # This is the code producing joined subqueries like: |
8273e845 |
93 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
d28bb90d |
94 | # |
95 | sub _adjust_select_args_for_complex_prefetch { |
e1861c2c |
96 | my ($self, $attrs) = @_; |
d28bb90d |
97 | |
e1861c2c |
98 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless ( |
99 | ref $attrs->{from} eq 'ARRAY' |
100 | and |
101 | @{$attrs->{from}} > 1 |
102 | and |
103 | ref $attrs->{from}[0] eq 'HASH' |
104 | and |
105 | ref $attrs->{from}[1] eq 'ARRAY' |
106 | ); |
d28bb90d |
107 | |
1e4f9fb3 |
108 | my $root_alias = $attrs->{alias}; |
109 | |
d28bb90d |
110 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
111 | my $outer_attrs = { %$attrs }; |
e1861c2c |
112 | delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)}; |
d28bb90d |
113 | |
6aa93928 |
114 | my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 }; |
115 | delete @{$inner_attrs}{qw(for collapse select as)}; |
d28bb90d |
116 | |
4df1400e |
117 | # there is no point of ordering the insides if there is no limit |
118 | delete $inner_attrs->{order_by} if ( |
119 | delete $inner_attrs->{_order_is_artificial} |
120 | or |
121 | ! $inner_attrs->{rows} |
122 | ); |
946f6260 |
123 | |
d28bb90d |
124 | # generate the inner/outer select lists |
125 | # for inside we consider only stuff *not* brought in by the prefetch |
126 | # on the outside we substitute any function for its alias |
e1861c2c |
127 | $outer_attrs->{select} = [ @{$attrs->{select}} ]; |
36fd7f07 |
128 | |
97e130fa |
129 | my ($root_node, $root_node_offset); |
27e0370d |
130 | |
e1861c2c |
131 | for my $i (0 .. $#{$inner_attrs->{from}}) { |
132 | my $node = $inner_attrs->{from}[$i]; |
27e0370d |
133 | my $h = (ref $node eq 'HASH') ? $node |
134 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
135 | : next |
136 | ; |
137 | |
1e4f9fb3 |
138 | if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) { |
97e130fa |
139 | $root_node = $h; |
140 | $root_node_offset = $i; |
27e0370d |
141 | last; |
142 | } |
143 | } |
144 | |
145 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
146 | unless $root_node; |
27e0370d |
147 | |
148 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
e1861c2c |
149 | my $colinfo = $self->_resolve_column_info($inner_attrs->{from}); |
27e0370d |
150 | my $selected_root_columns; |
151 | |
e1861c2c |
152 | for my $i (0 .. $#{$outer_attrs->{select}}) { |
153 | my $sel = $outer_attrs->{select}->[$i]; |
d28bb90d |
154 | |
1e4f9fb3 |
155 | next if ( |
156 | $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias |
157 | ); |
158 | |
d28bb90d |
159 | if (ref $sel eq 'HASH' ) { |
160 | $sel->{-as} ||= $attrs->{as}[$i]; |
e1861c2c |
161 | $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); |
d28bb90d |
162 | } |
27e0370d |
163 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
164 | $selected_root_columns->{$ci->{-colname}} = 1; |
165 | } |
d28bb90d |
166 | |
e1861c2c |
167 | push @{$inner_attrs->{select}}, $sel; |
bb9bffea |
168 | |
169 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
170 | } |
171 | |
97e130fa |
172 | # We will need to fetch all native columns in the inner subquery, which may |
173 | # be a part of an *outer* join condition, or an order_by (which needs to be |
e1861c2c |
174 | # preserved outside), or wheres. In other words everything but the inner |
175 | # selector |
97e130fa |
176 | # We can not just fetch everything because a potential has_many restricting |
177 | # join collapse *will not work* on heavy data types. |
e1861c2c |
178 | my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({ |
179 | %$inner_attrs, |
180 | select => [], |
181 | }); |
97e130fa |
182 | |
183 | for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { |
184 | my $ci = $colinfo->{$_} or next; |
185 | if ( |
1e4f9fb3 |
186 | $ci->{-source_alias} eq $root_alias |
97e130fa |
187 | and |
188 | ! $selected_root_columns->{$ci->{-colname}}++ |
189 | ) { |
190 | # adding it to both to keep limits not supporting dark selectors happy |
e1861c2c |
191 | push @{$inner_attrs->{select}}, $ci->{-fq_colname}; |
97e130fa |
192 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
193 | } |
194 | } |
195 | |
e1861c2c |
196 | # construct the inner {from} and lock it in a subquery |
48580715 |
197 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
198 | # throw away all non-selecting, non-restricting multijoins |
eb58c082 |
199 | # (since we def. do not care about multiplication of the contents of the subquery) |
6395604e |
200 | my $inner_subq = do { |
ea95892e |
201 | |
eb58c082 |
202 | # must use it here regardless of user requests (vastly gentler on optimizer) |
ea95892e |
203 | local $self->{_use_join_optimizer} = 1; |
204 | |
97e130fa |
205 | # throw away multijoins since we def. do not care about those inside the subquery |
e1861c2c |
206 | ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({ |
437a9cfa |
207 | %$inner_attrs, _force_prune_multiplying_joins => 1 |
208 | }); |
ea95892e |
209 | |
eb58c082 |
210 | # uh-oh a multiplier (which is not us) left in, this is a problem for limits |
211 | # we will need to add a group_by to collapse the resultset for proper counts |
0a3441ee |
212 | if ( |
eb58c082 |
213 | grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} } |
1e4f9fb3 |
214 | and |
560978e2 |
215 | # if there are user-supplied groups - assume user knows wtf they are up to |
216 | ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} ) |
0a3441ee |
217 | ) { |
1e4f9fb3 |
218 | |
eb58c082 |
219 | my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} }; |
1e4f9fb3 |
220 | |
eb58c082 |
221 | # *possibly* supplement the main selection with pks if not already |
222 | # there, as they will have to be a part of the group_by to collapse |
223 | # things properly |
224 | my $inner_select_with_extras; |
225 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
226 | or $self->throw_exception( sprintf |
227 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
228 | $root_node->{-rsrc}->source_name, |
e1861c2c |
229 | ); |
eb58c082 |
230 | for my $col (@pks) { |
231 | push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col |
232 | unless $cur_sel->{$col}++; |
1e4f9fb3 |
233 | } |
eb58c082 |
234 | |
235 | ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({ |
236 | %$inner_attrs, |
237 | $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (), |
238 | _aliastypes => $inner_aliastypes, |
239 | }); |
0a3441ee |
240 | } |
d28bb90d |
241 | |
e1861c2c |
242 | # we already optimized $inner_attrs->{from} above |
97e130fa |
243 | # and already local()ized |
244 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
245 | |
ea95892e |
246 | # generate the subquery |
6395604e |
247 | $self->_select_args_to_query ( |
e1861c2c |
248 | @{$inner_attrs}{qw(from select where)}, |
ea95892e |
249 | $inner_attrs, |
250 | ); |
d28bb90d |
251 | }; |
252 | |
253 | # Generate the outer from - this is relatively easy (really just replace |
254 | # the join slot with the subquery), with a major caveat - we can not |
255 | # join anything that is non-selecting (not part of the prefetch), but at |
256 | # the same time is a multi-type relationship, as it will explode the result. |
257 | # |
258 | # There are two possibilities here |
259 | # - either the join is non-restricting, in which case we simply throw it away |
260 | # - it is part of the restrictions, in which case we need to collapse the outer |
261 | # result by tackling yet another group_by to the outside of the query |
262 | |
27e0370d |
263 | # work on a shallow copy |
e1861c2c |
264 | my @orig_from = @{$attrs->{from}}; |
265 | |
052e8431 |
266 | |
e1861c2c |
267 | $outer_attrs->{from} = \ my @outer_from; |
53c29913 |
268 | |
27e0370d |
269 | # we may not be the head |
97e130fa |
270 | if ($root_node_offset) { |
e1861c2c |
271 | # first generate the outer_from, up to the substitution point |
272 | @outer_from = splice @orig_from, 0, $root_node_offset; |
27e0370d |
273 | |
e1861c2c |
274 | # substitute the subq at the right spot |
27e0370d |
275 | push @outer_from, [ |
276 | { |
1e4f9fb3 |
277 | -alias => $root_alias, |
97e130fa |
278 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
279 | $root_alias => $inner_subq, |
27e0370d |
280 | }, |
e1861c2c |
281 | # preserve attrs from what is now the head of the from after the splice |
282 | @{$orig_from[0]}[1 .. $#{$orig_from[0]}], |
27e0370d |
283 | ]; |
284 | } |
285 | else { |
27e0370d |
286 | @outer_from = { |
1e4f9fb3 |
287 | -alias => $root_alias, |
27e0370d |
288 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
289 | $root_alias => $inner_subq, |
27e0370d |
290 | }; |
d28bb90d |
291 | } |
292 | |
e1861c2c |
293 | shift @orig_from; # what we just replaced above |
97e130fa |
294 | |
ea95892e |
295 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
296 | # in what role |
975b573a |
297 | my $outer_aliastypes = $outer_attrs->{_aliastypes} = |
e1861c2c |
298 | $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from }); |
052e8431 |
299 | |
a4812caa |
300 | # unroll parents |
1e4f9fb3 |
301 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
302 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
303 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
304 | |
d28bb90d |
305 | # see what's left - throw away if not selecting/restricting |
eb58c082 |
306 | my $may_need_outer_group_by; |
e1861c2c |
307 | while (my $j = shift @orig_from) { |
d28bb90d |
308 | my $alias = $j->[0]{-alias}; |
309 | |
a4812caa |
310 | if ( |
311 | $outer_select_chain->{$alias} |
312 | ) { |
313 | push @outer_from, $j |
d28bb90d |
314 | } |
1e4f9fb3 |
315 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
316 | push @outer_from, $j; |
eb58c082 |
317 | $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
318 | } |
319 | } |
320 | |
eb58c082 |
321 | # also throw in a synthetic group_by if a non-selecting multiplier, |
322 | # to guard against cross-join explosions |
323 | # the logic is somewhat fragile, but relies on the idea that if a user supplied |
324 | # a group by on their own - they know what they were doing |
325 | if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
326 | ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({ |
560978e2 |
327 | %$outer_attrs, |
328 | from => \@outer_from, |
560978e2 |
329 | }); |
36fd7f07 |
330 | } |
331 | |
e1861c2c |
332 | # This is totally horrific - the {where} ends up in both the inner and outer query |
d28bb90d |
333 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
334 | # then if where conditions apply to the *right* side of the prefetch, you may have |
335 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
4a0eed52 |
336 | # the outer select to exclude joins you didn't want in the first place |
d28bb90d |
337 | # |
338 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
e1861c2c |
339 | return $outer_attrs; |
d28bb90d |
340 | } |
341 | |
1a736efb |
342 | # |
343 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
344 | # |
ad630f4b |
345 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
346 | # select/where/order/group attributes, in order to determine what |
4a0eed52 |
347 | # aliases are needed to fulfill the query. This information is used |
ad630f4b |
348 | # throughout the code to prune unnecessary JOINs from the queries |
349 | # in an attempt to reduce the execution time. |
350 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
351 | # happen is for it to fail due to some scalar SQL, which in turn will |
352 | # result in a vocal exception. |
539ffe87 |
353 | sub _resolve_aliastypes_from_select_args { |
e1861c2c |
354 | my ( $self, $attrs ) = @_; |
546f1cd9 |
355 | |
ad630f4b |
356 | $self->throw_exception ('Unable to analyze custom {from}') |
e1861c2c |
357 | if ref $attrs->{from} ne 'ARRAY'; |
546f1cd9 |
358 | |
ad630f4b |
359 | # what we will return |
964a3c71 |
360 | my $aliases_by_type; |
546f1cd9 |
361 | |
ad630f4b |
362 | # see what aliases are there to work with |
eb58c082 |
363 | # and record who is a multiplier and who is premultiplied |
ad630f4b |
364 | my $alias_list; |
e1861c2c |
365 | for my $node (@{$attrs->{from}}) { |
366 | |
367 | my $j = $node; |
ad630f4b |
368 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
369 | my $al = $j->{-alias} |
370 | or next; |
371 | |
372 | $alias_list->{$al} = $j; |
eb58c082 |
373 | |
374 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } |
a4812caa |
375 | # not array == {from} head == can't be multiplying |
eb58c082 |
376 | if ref($node) eq 'ARRAY' and ! $j->{-is_single}; |
377 | |
378 | $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] } |
379 | # parts of the path that are not us but are multiplying |
380 | if grep { $aliases_by_type->{multiplying}{$_} } |
381 | grep { $_ ne $al } |
382 | map { values %$_ } |
383 | @{ $j->{-join_path}||[] } |
546f1cd9 |
384 | } |
546f1cd9 |
385 | |
318e3d94 |
386 | # get a column to source/alias map (including unambiguous unqualified ones) |
e1861c2c |
387 | my $colinfo = $self->_resolve_column_info ($attrs->{from}); |
1a736efb |
388 | |
ad630f4b |
389 | # set up a botched SQLA |
390 | my $sql_maker = $self->sql_maker; |
07f31d19 |
391 | |
4c2b30d6 |
392 | # these are throw away results, do not pollute the bind stack |
0542ec57 |
393 | local $sql_maker->{where_bind}; |
394 | local $sql_maker->{group_bind}; |
395 | local $sql_maker->{having_bind}; |
97e130fa |
396 | local $sql_maker->{from_bind}; |
3f5b99fe |
397 | |
398 | # we can't scan properly without any quoting (\b doesn't cut it |
399 | # everywhere), so unless there is proper quoting set - use our |
400 | # own weird impossible character. |
401 | # Also in the case of no quoting, we need to explicitly disable |
402 | # name_sep, otherwise sorry nasty legacy syntax like |
403 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
404 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
405 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
406 | |
407 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
408 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
409 | # if we don't unset it we screw up retarded but unfortunately working |
410 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
411 | $sql_maker->{name_sep} = ''; |
412 | } |
413 | |
414 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
415 | |
1a736efb |
416 | # generate sql chunks |
417 | my $to_scan = { |
418 | restricting => [ |
a9e985b7 |
419 | ($sql_maker->_recurse_where ($attrs->{where}))[0], |
1e4f9fb3 |
420 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
421 | ], |
422 | grouping => [ |
423 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
424 | ], |
97e130fa |
425 | joining => [ |
426 | $sql_maker->_recurse_from ( |
e1861c2c |
427 | ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0], |
428 | @{$attrs->{from}}[1 .. $#{$attrs->{from}}], |
97e130fa |
429 | ), |
430 | ], |
1a736efb |
431 | selecting => [ |
ad1d374e |
432 | map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}}, |
1e4f9fb3 |
433 | ], |
434 | ordering => [ |
435 | map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker), |
1a736efb |
436 | ], |
437 | }; |
07f31d19 |
438 | |
0dadd60d |
439 | # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are |
440 | # bind value specs left in by the sloppy renderer above. It is ok to do this |
441 | # at this point, since we are going to end up rewriting this crap anyway |
442 | for my $v (values %$to_scan) { |
443 | my @nv; |
444 | for (@$v) { |
445 | next if ( |
446 | ! defined $_ |
447 | or |
448 | ( |
449 | ref $_ eq 'ARRAY' |
450 | and |
451 | ( @$_ == 0 or @$_ == 2 ) |
452 | ) |
453 | ); |
454 | |
455 | if (ref $_) { |
456 | require Data::Dumper::Concise; |
457 | $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) ); |
458 | } |
459 | |
460 | push @nv, $_; |
461 | } |
462 | |
463 | $v = \@nv; |
464 | } |
465 | |
466 | # kill all selectors which look like a proper subquery |
467 | # this is a sucky heuristic *BUT* - if we get it wrong the query will simply |
468 | # fail to run, so we are relatively safe |
469 | $to_scan->{selecting} = [ grep { |
470 | $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi |
471 | } @{ $to_scan->{selecting} || [] } ]; |
07f31d19 |
472 | |
318e3d94 |
473 | # first see if we have any exact matches (qualified or unqualified) |
474 | for my $type (keys %$to_scan) { |
475 | for my $piece (@{$to_scan->{$type}}) { |
476 | if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) { |
477 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
478 | $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece; |
479 | } |
480 | } |
481 | } |
482 | |
483 | # now loop through all fully qualified columns and get the corresponding |
1a736efb |
484 | # alias (should work even if they are in scalarrefs) |
ad630f4b |
485 | for my $alias (keys %$alias_list) { |
1a736efb |
486 | my $al_re = qr/ |
97e130fa |
487 | $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
1a736efb |
488 | | |
97e130fa |
489 | \b $alias \. ([^\s\)\($rquote]+)? |
1a736efb |
490 | /x; |
491 | |
1a736efb |
492 | for my $type (keys %$to_scan) { |
493 | for my $piece (@{$to_scan->{$type}}) { |
97e130fa |
494 | if (my @matches = $piece =~ /$al_re/g) { |
495 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
496 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_" |
97e130fa |
497 | for grep { defined $_ } @matches; |
498 | } |
1a736efb |
499 | } |
ad630f4b |
500 | } |
1a736efb |
501 | } |
502 | |
503 | # now loop through unqualified column names, and try to locate them within |
504 | # the chunks |
505 | for my $col (keys %$colinfo) { |
3f5b99fe |
506 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
507 | |
97e130fa |
508 | my $col_re = qr/ $lquote ($col) $rquote /x; |
07f31d19 |
509 | |
1a736efb |
510 | for my $type (keys %$to_scan) { |
511 | for my $piece (@{$to_scan->{$type}}) { |
318e3d94 |
512 | if ( my @matches = $piece =~ /$col_re/g) { |
a4812caa |
513 | my $alias = $colinfo->{$col}{-source_alias}; |
97e130fa |
514 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
515 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
97e130fa |
516 | for grep { defined $_ } @matches; |
a4812caa |
517 | } |
1a736efb |
518 | } |
07f31d19 |
519 | } |
520 | } |
521 | |
522 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
523 | for my $j (values %$alias_list) { |
07f31d19 |
524 | my $alias = $j->{-alias} or next; |
97e130fa |
525 | $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if ( |
07f31d19 |
526 | (not $j->{-join_type}) |
527 | or |
528 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
529 | ); |
530 | } |
531 | |
1e4f9fb3 |
532 | for (keys %$aliases_by_type) { |
533 | delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}}; |
534 | } |
535 | |
964a3c71 |
536 | return $aliases_by_type; |
07f31d19 |
537 | } |
538 | |
eb58c082 |
539 | # This is the engine behind { distinct => 1 } and the general |
540 | # complex prefetch grouper |
0a3441ee |
541 | sub _group_over_selection { |
560978e2 |
542 | my ($self, $attrs) = @_; |
0a3441ee |
543 | |
560978e2 |
544 | my $colinfos = $self->_resolve_column_info ($attrs->{from}); |
0a3441ee |
545 | |
546 | my (@group_by, %group_index); |
547 | |
36fd7f07 |
548 | # the logic is: if it is a { func => val } we assume an aggregate, |
549 | # otherwise if \'...' or \[...] we assume the user knows what is |
550 | # going on thus group over it |
560978e2 |
551 | for (@{$attrs->{select}}) { |
0a3441ee |
552 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
553 | push @group_by, $_; |
554 | $group_index{$_}++; |
560978e2 |
555 | if ($colinfos->{$_} and $_ !~ /\./ ) { |
0a3441ee |
556 | # add a fully qualified version as well |
560978e2 |
557 | $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; |
0a3441ee |
558 | } |
07f31d19 |
559 | } |
560 | } |
ad630f4b |
561 | |
eb58c082 |
562 | my @order_by = $self->_extract_order_criteria($attrs->{order_by}) |
563 | or return (\@group_by, $attrs->{order_by}); |
564 | |
565 | # add any order_by parts that are not already present in the group_by |
566 | # to maintain SQL cross-compatibility and general sanity |
567 | # |
568 | # also in case the original selection is *not* unique, or in case part |
569 | # of the ORDER BY refers to a multiplier - we will need to replace the |
570 | # skipped order_by elements with their MIN/MAX equivalents as to maintain |
571 | # the proper overall order without polluting the group criteria (and |
572 | # possibly changing the outcome entirely) |
573 | |
574 | my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes); |
575 | |
576 | my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by); |
577 | |
578 | for my $o_idx (0 .. $#order_by) { |
579 | |
580 | # if the chunk is already a min/max function - there is nothing left to touch |
581 | next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix; |
582 | |
0a3441ee |
583 | # only consider real columns (for functions the user got to do an explicit group_by) |
eb58c082 |
584 | my $chunk_ci; |
585 | if ( |
586 | @{$order_by[$o_idx]} != 1 |
587 | or |
588 | # only declare an unknown *plain* identifier as "leftover" if we are called with |
589 | # aliastypes to examine. If there are none - we are still in _resolve_attrs, and |
590 | # can just assume the user knows what they want |
591 | ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} ) |
592 | ) { |
593 | push @$leftovers, $order_by[$o_idx][0]; |
14e26c5f |
594 | } |
560978e2 |
595 | |
eb58c082 |
596 | next unless $chunk_ci; |
597 | |
598 | # no duplication of group criteria |
599 | next if $group_index{$chunk_ci->{-fq_colname}}; |
600 | |
601 | $aliastypes ||= ( |
602 | $attrs->{_aliastypes} |
560978e2 |
603 | or |
eb58c082 |
604 | $self->_resolve_aliastypes_from_select_args({ |
605 | from => $attrs->{from}, |
606 | order_by => $attrs->{order_by}, |
607 | }) |
608 | ) if $group_already_unique; |
609 | |
610 | # check that we are not ordering by a multiplier (if a check is requested at all) |
611 | if ( |
612 | $group_already_unique |
613 | and |
614 | ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}} |
615 | and |
616 | ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}} |
560978e2 |
617 | ) { |
eb58c082 |
618 | push @group_by, $chunk_ci->{-fq_colname}; |
619 | $group_index{$chunk_ci->{-fq_colname}}++ |
560978e2 |
620 | } |
eb58c082 |
621 | else { |
622 | # We need to order by external columns without adding them to the group |
623 | # (eiehter a non-unique selection, or a multi-external) |
624 | # |
625 | # This doesn't really make sense in SQL, however from DBICs point |
626 | # of view is rather valid (e.g. order the leftmost objects by whatever |
627 | # criteria and get the offset/rows many). There is a way around |
628 | # this however in SQL - we simply tae the direction of each piece |
629 | # of the external order and convert them to MIN(X) for ASC or MAX(X) |
630 | # for DESC, and group_by the root columns. The end result should be |
631 | # exactly what we expect |
632 | |
633 | # FIXME - this code is a joke, will need to be completely rewritten in |
634 | # the DQ branch. But I need to push a POC here, otherwise the |
635 | # pesky tests won't pass |
636 | # wrap any part of the order_by that "responds" to an ordering alias |
637 | # into a MIN/MAX |
638 | $sql_maker ||= $self->sql_maker; |
639 | $order_chunks ||= [ |
640 | map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}) |
641 | ]; |
0a3441ee |
642 | |
eb58c082 |
643 | my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]); |
644 | |
645 | $new_order_by[$o_idx] = \[ |
646 | sprintf( '%s( %s )%s', |
647 | ($is_desc ? 'MAX' : 'MIN'), |
648 | $chunk, |
649 | ($is_desc ? ' DESC' : ''), |
650 | ), |
651 | @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ] |
652 | ]; |
653 | } |
0a3441ee |
654 | } |
655 | |
eb58c082 |
656 | $self->throw_exception ( sprintf |
9736be65 |
657 | 'Unable to programatically derive a required group_by from the supplied ' |
658 | . 'order_by criteria. To proceed either add an explicit group_by, or ' |
659 | . 'simplify your order_by to only include plain columns ' |
660 | . '(supplied order_by: %s)', |
eb58c082 |
661 | join ', ', map { "'$_'" } @$leftovers, |
662 | ) if $leftovers; |
663 | |
664 | # recreate the untouched order parts |
665 | if (@new_order_by) { |
666 | $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks ); |
667 | } |
668 | |
669 | return ( |
670 | \@group_by, |
671 | (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged |
672 | ); |
07f31d19 |
673 | } |
674 | |
d28bb90d |
675 | sub _resolve_ident_sources { |
676 | my ($self, $ident) = @_; |
677 | |
678 | my $alias2source = {}; |
d28bb90d |
679 | |
680 | # the reason this is so contrived is that $ident may be a {from} |
681 | # structure, specifying multiple tables to join |
6298a324 |
682 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
683 | # this is compat mode for insert/update/delete which do not deal with aliases |
684 | $alias2source->{me} = $ident; |
d28bb90d |
685 | } |
686 | elsif (ref $ident eq 'ARRAY') { |
687 | |
688 | for (@$ident) { |
689 | my $tabinfo; |
690 | if (ref $_ eq 'HASH') { |
691 | $tabinfo = $_; |
d28bb90d |
692 | } |
693 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
694 | $tabinfo = $_->[0]; |
695 | } |
696 | |
4376a157 |
697 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
698 | if ($tabinfo->{-rsrc}); |
d28bb90d |
699 | } |
700 | } |
701 | |
90f10b5a |
702 | return $alias2source; |
d28bb90d |
703 | } |
704 | |
705 | # Takes $ident, \@column_names |
706 | # |
707 | # returns { $column_name => \%column_info, ... } |
708 | # also note: this adds -result_source => $rsrc to the column info |
709 | # |
09e14fdc |
710 | # If no columns_names are supplied returns info about *all* columns |
711 | # for all sources |
d28bb90d |
712 | sub _resolve_column_info { |
713 | my ($self, $ident, $colnames) = @_; |
8d005ad9 |
714 | |
715 | return {} if $colnames and ! @$colnames; |
716 | |
90f10b5a |
717 | my $alias2src = $self->_resolve_ident_sources($ident); |
d28bb90d |
718 | |
52416317 |
719 | my (%seen_cols, @auto_colnames); |
d28bb90d |
720 | |
721 | # compile a global list of column names, to be able to properly |
722 | # disambiguate unqualified column names (if at all possible) |
723 | for my $alias (keys %$alias2src) { |
724 | my $rsrc = $alias2src->{$alias}; |
725 | for my $colname ($rsrc->columns) { |
726 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
727 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
728 | } |
729 | } |
730 | |
09e14fdc |
731 | $colnames ||= [ |
732 | @auto_colnames, |
733 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
734 | ]; |
735 | |
52416317 |
736 | my (%return, $colinfos); |
d28bb90d |
737 | foreach my $col (@$colnames) { |
52416317 |
738 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
739 | |
52416317 |
740 | # if the column was seen exactly once - we know which rsrc it came from |
741 | $source_alias ||= $seen_cols{$colname}[0] |
742 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
743 | |
52416317 |
744 | next unless $source_alias; |
745 | |
746 | my $rsrc = $alias2src->{$source_alias} |
747 | or next; |
748 | |
749 | $return{$col} = { |
6395604e |
750 | %{ |
751 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
752 | || |
753 | $self->throw_exception( |
754 | "No such column '$colname' on source " . $rsrc->source_name |
755 | ); |
756 | }, |
d28bb90d |
757 | -result_source => $rsrc, |
52416317 |
758 | -source_alias => $source_alias, |
81bf295c |
759 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
760 | -colname => $colname, |
d28bb90d |
761 | }; |
81bf295c |
762 | |
763 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
764 | } |
765 | |
766 | return \%return; |
767 | } |
768 | |
289ac713 |
769 | # The DBIC relationship chaining implementation is pretty simple - every |
770 | # new related_relationship is pushed onto the {from} stack, and the {select} |
771 | # window simply slides further in. This means that when we count somewhere |
772 | # in the middle, we got to make sure that everything in the join chain is an |
773 | # actual inner join, otherwise the count will come back with unpredictable |
774 | # results (a resultset may be generated with _some_ rows regardless of if |
775 | # the relation which the $rs currently selects has rows or not). E.g. |
776 | # $artist_rs->cds->count - normally generates: |
777 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
778 | # which actually returns the number of artists * (number of cds || 1) |
779 | # |
780 | # So what we do here is crawl {from}, determine if the current alias is at |
781 | # the top of the stack, and if not - make sure the chain is inner-joined down |
782 | # to the root. |
783 | # |
31a8aaaf |
784 | sub _inner_join_to_node { |
289ac713 |
785 | my ($self, $from, $alias) = @_; |
786 | |
302d35f8 |
787 | my $switch_branch = $self->_find_join_path_to_node($from, $alias); |
289ac713 |
788 | |
302d35f8 |
789 | return $from unless @{$switch_branch||[]}; |
289ac713 |
790 | |
791 | # So it looks like we will have to switch some stuff around. |
792 | # local() is useless here as we will be leaving the scope |
793 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
794 | # So replace the first hashref in the node arrayref manually |
289ac713 |
795 | my @new_from = ($from->[0]); |
faeb2407 |
796 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
797 | |
798 | for my $j (@{$from}[1 .. $#$from]) { |
799 | my $jalias = $j->[0]{-alias}; |
800 | |
801 | if ($sw_idx->{$jalias}) { |
802 | my %attrs = %{$j->[0]}; |
803 | delete $attrs{-join_type}; |
804 | push @new_from, [ |
805 | \%attrs, |
806 | @{$j}[ 1 .. $#$j ], |
807 | ]; |
808 | } |
809 | else { |
810 | push @new_from, $j; |
811 | } |
812 | } |
813 | |
814 | return \@new_from; |
815 | } |
816 | |
302d35f8 |
817 | sub _find_join_path_to_node { |
818 | my ($self, $from, $target_alias) = @_; |
819 | |
820 | # subqueries and other oddness are naturally not supported |
821 | return undef if ( |
822 | ref $from ne 'ARRAY' |
823 | || |
824 | ref $from->[0] ne 'HASH' |
825 | || |
826 | ! defined $from->[0]{-alias} |
827 | ); |
828 | |
829 | # no path - the head is the alias |
830 | return [] if $from->[0]{-alias} eq $target_alias; |
831 | |
832 | for my $i (1 .. $#$from) { |
833 | return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias ); |
834 | } |
835 | |
836 | # something else went quite wrong |
837 | return undef; |
838 | } |
839 | |
bac358c9 |
840 | sub _extract_order_criteria { |
1a736efb |
841 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
842 | |
1a736efb |
843 | my $parser = sub { |
e6977bbb |
844 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
845 | |
1a736efb |
846 | return scalar $sql_maker->_order_by_chunks ($order_by) |
847 | unless wantarray; |
c0748280 |
848 | |
e6977bbb |
849 | my ($lq, $rq, $sep) = map { quotemeta($_) } ( |
850 | ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars), |
851 | $sql_maker->name_sep |
852 | ); |
853 | |
1a736efb |
854 | my @chunks; |
bac358c9 |
855 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
856 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
857 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
858 | |
859 | # order criteria may have come back pre-quoted (literals and whatnot) |
860 | # this is fragile, but the best we can currently do |
861 | $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe |
862 | or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x; |
863 | |
1a736efb |
864 | push @chunks, $chunk; |
bac6c4fb |
865 | } |
1a736efb |
866 | |
867 | return @chunks; |
868 | }; |
869 | |
870 | if ($sql_maker) { |
871 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
872 | } |
873 | else { |
1a736efb |
874 | $sql_maker = $self->sql_maker; |
e6977bbb |
875 | |
876 | # pass these in to deal with literals coming from |
877 | # the user or the deep guts of prefetch |
878 | my $orig_quote_chars = [$sql_maker->_quote_chars]; |
879 | |
1a736efb |
880 | local $sql_maker->{quote_char}; |
e6977bbb |
881 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
882 | } |
bac6c4fb |
883 | } |
884 | |
7cec4356 |
885 | sub _order_by_is_stable { |
5f11e54f |
886 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
887 | |
eb58c082 |
888 | my @cols = ( |
8d005ad9 |
889 | ( map { $_->[0] } $self->_extract_order_criteria($order_by) ), |
8e40a627 |
890 | ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ), |
df4312bc |
891 | ) or return 0; |
eb58c082 |
892 | |
893 | my $colinfo = $self->_resolve_column_info($ident, \@cols); |
894 | |
895 | return keys %$colinfo |
896 | ? $self->_columns_comprise_identifying_set( $colinfo, \@cols ) |
df4312bc |
897 | : 0 |
eb58c082 |
898 | ; |
899 | } |
c0748280 |
900 | |
eb58c082 |
901 | sub _columns_comprise_identifying_set { |
902 | my ($self, $colinfo, $columns) = @_; |
7cec4356 |
903 | |
904 | my $cols_per_src; |
eb58c082 |
905 | $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_ |
906 | for grep { defined $_ } @{$colinfo}{@$columns}; |
7cec4356 |
907 | |
908 | for (values %$cols_per_src) { |
909 | my $src = (values %$_)[0]->{-result_source}; |
910 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
911 | } |
912 | |
df4312bc |
913 | return 0; |
7cec4356 |
914 | } |
915 | |
df4312bc |
916 | # this is almost similar to _order_by_is_stable, except it takes |
0e81e691 |
917 | # a single rsrc, and will succeed only if the first portion of the order |
918 | # by is stable. |
919 | # returns that portion as a colinfo hashref on success |
df4312bc |
920 | sub _extract_colinfo_of_stable_main_source_order_by_portion { |
302d35f8 |
921 | my ($self, $attrs) = @_; |
0e81e691 |
922 | |
302d35f8 |
923 | my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias}); |
924 | |
925 | return unless defined $nodes; |
0e81e691 |
926 | |
927 | my @ord_cols = map |
928 | { $_->[0] } |
302d35f8 |
929 | ( $self->_extract_order_criteria($attrs->{order_by}) ) |
0e81e691 |
930 | ; |
931 | return unless @ord_cols; |
932 | |
302d35f8 |
933 | my $valid_aliases = { map { $_ => 1 } ( |
934 | $attrs->{from}[0]{-alias}, |
935 | map { values %$_ } @$nodes, |
936 | ) }; |
318e3d94 |
937 | |
302d35f8 |
938 | my $colinfos = $self->_resolve_column_info($attrs->{from}); |
939 | |
940 | my ($colinfos_to_return, $seen_main_src_cols); |
941 | |
942 | for my $col (@ord_cols) { |
943 | # if order criteria is unresolvable - there is nothing we can do |
944 | my $colinfo = $colinfos->{$col} or last; |
945 | |
946 | # if we reached the end of the allowed aliases - also nothing we can do |
947 | last unless $valid_aliases->{$colinfo->{-source_alias}}; |
948 | |
949 | $colinfos_to_return->{$col} = $colinfo; |
950 | |
951 | $seen_main_src_cols->{$colinfo->{-colname}} = 1 |
952 | if $colinfo->{-source_alias} eq $attrs->{alias}; |
0e81e691 |
953 | } |
954 | |
302d35f8 |
955 | # FIXME the condition may be singling out things on its own, so we |
956 | # conceivable could come back wi "stable-ordered by nothing" |
957 | # not confient enough in the parser yet, so punt for the time being |
958 | return unless $seen_main_src_cols; |
0e81e691 |
959 | |
302d35f8 |
960 | my $main_src_fixed_cols_from_cond = [ $attrs->{where} |
961 | ? ( |
962 | map |
963 | { |
964 | ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} ) |
965 | ? $colinfos->{$_}{-colname} |
966 | : () |
967 | } |
8e40a627 |
968 | keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) } |
302d35f8 |
969 | ) |
970 | : () |
971 | ]; |
0e81e691 |
972 | |
302d35f8 |
973 | return $attrs->{result_source}->_identifying_column_set([ |
974 | keys %$seen_main_src_cols, |
975 | @$main_src_fixed_cols_from_cond, |
976 | ]) ? $colinfos_to_return : (); |
0e81e691 |
977 | } |
978 | |
8d005ad9 |
979 | # Attempts to flatten a passed in SQLA condition as much as possible towards |
980 | # a plain hashref, *without* altering its semantics. Required by |
981 | # create/populate being able to extract definitive conditions from preexisting |
982 | # resultset {where} stacks |
983 | # |
984 | # FIXME - while relatively robust, this is still imperfect, one of the first |
985 | # things to tackle with DQ |
986 | sub _collapse_cond { |
987 | my ($self, $where, $where_is_anded_array) = @_; |
988 | |
135ac69d |
989 | my $fin; |
990 | |
8d005ad9 |
991 | if (! $where) { |
992 | return; |
993 | } |
994 | elsif ($where_is_anded_array or ref $where eq 'HASH') { |
995 | |
996 | my @pairs; |
997 | |
998 | my @pieces = $where_is_anded_array ? @$where : $where; |
999 | while (@pieces) { |
1000 | my $chunk = shift @pieces; |
1001 | |
1002 | if (ref $chunk eq 'HASH') { |
6565d2c3 |
1003 | push @pairs, map { $_ => $chunk->{$_} } sort keys %$chunk; |
8d005ad9 |
1004 | } |
1005 | elsif (ref $chunk eq 'ARRAY') { |
6565d2c3 |
1006 | push @pairs, -or => $chunk |
8d005ad9 |
1007 | if @$chunk; |
1008 | } |
b34d9331 |
1009 | elsif ( ! length ref $chunk) { |
6565d2c3 |
1010 | push @pairs, $chunk, shift @pieces; |
8d005ad9 |
1011 | } |
1012 | else { |
6565d2c3 |
1013 | push @pairs, '', $chunk; |
8d005ad9 |
1014 | } |
1015 | } |
1016 | |
1017 | return unless @pairs; |
1018 | |
1019 | my @conds = $self->_collapse_cond_unroll_pairs(\@pairs) |
1020 | or return; |
1021 | |
1022 | # Consolidate various @conds back into something more compact |
8d005ad9 |
1023 | for my $c (@conds) { |
1024 | if (ref $c ne 'HASH') { |
1025 | push @{$fin->{-and}}, $c; |
1026 | } |
1027 | else { |
1028 | for my $col (sort keys %$c) { |
8d005ad9 |
1029 | |
135ac69d |
1030 | # consolidate all -and nodes |
1031 | if ($col =~ /^\-and$/i) { |
1032 | push @{$fin->{-and}}, |
1033 | ref $c->{$col} eq 'ARRAY' ? @{$c->{$col}} |
1034 | : ref $c->{$col} eq 'HASH' ? %{$c->{$col}} |
1035 | : { $col => $c->{$col} } |
1036 | ; |
1037 | } |
1038 | elsif ($col =~ /^\-/) { |
1039 | push @{$fin->{-and}}, { $col => $c->{$col} }; |
1040 | } |
1041 | elsif (exists $fin->{$col}) { |
1042 | $fin->{$col} = [ -and => map { |
1043 | (ref $_ eq 'ARRAY' and ($_->[0]||'') =~ /^\-and$/i ) |
1044 | ? @{$_}[1..$#$_] |
1045 | : $_ |
1046 | ; |
1047 | } ($fin->{$col}, $c->{$col}) ]; |
8d005ad9 |
1048 | } |
1049 | else { |
1050 | $fin->{$col} = $c->{$col}; |
1051 | } |
1052 | } |
1053 | } |
1054 | } |
8d005ad9 |
1055 | } |
1056 | elsif (ref $where eq 'ARRAY') { |
22485a7e |
1057 | # we are always at top-level here, it is safe to dump empty *standalone* pieces |
1058 | my $fin_idx; |
8d005ad9 |
1059 | |
22485a7e |
1060 | for (my $i = 0; $i <= $#$where; $i++ ) { |
8d005ad9 |
1061 | |
22485a7e |
1062 | my $logic_mod = lc ( ($where->[$i] =~ /^(\-(?:and|or))$/i)[0] || '' ); |
1063 | |
1064 | if ($logic_mod) { |
1065 | $i++; |
1066 | $self->throw_exception("Unsupported top-level op/arg pair: [ $logic_mod => $where->[$i] ]") |
1067 | unless ref $where->[$i] eq 'HASH' or ref $where->[$i] eq 'ARRAY'; |
1068 | |
1069 | my $sub_elt = $self->_collapse_cond({ $logic_mod => $where->[$i] }) |
1070 | or next; |
1071 | |
135ac69d |
1072 | $fin_idx->{ "SER_" . serialize $sub_elt } = $sub_elt; |
22485a7e |
1073 | } |
1074 | elsif (! length ref $where->[$i] ) { |
135ac69d |
1075 | my $sub_elt = $self->_collapse_cond({ @{$where}[$i, $i+1] }) |
1076 | or next; |
1077 | |
1078 | $fin_idx->{ "COL_$where->[$i]_" . serialize $sub_elt } = $sub_elt; |
22485a7e |
1079 | $i++; |
8d005ad9 |
1080 | } |
1081 | else { |
135ac69d |
1082 | $fin_idx->{ "SER_" . serialize $where->[$i] } = $self->_collapse_cond( $where->[$i] ) || next; |
8d005ad9 |
1083 | } |
1084 | } |
22485a7e |
1085 | |
07add744 |
1086 | if (! $fin_idx) { |
1087 | return; |
1088 | } |
1089 | elsif ( keys %$fin_idx == 1 ) { |
1090 | $fin = (values %$fin_idx)[0]; |
1091 | } |
1092 | else { |
1093 | my @or; |
1094 | |
1095 | # at this point everything is at most one level deep - unroll if needed |
1096 | for (sort keys %$fin_idx) { |
1097 | if ( ref $fin_idx->{$_} eq 'HASH' and keys %{$fin_idx->{$_}} == 1 ) { |
1098 | my ($l, $r) = %{$fin_idx->{$_}}; |
1099 | |
1100 | if ( |
1101 | ref $r eq 'ARRAY' |
1102 | and |
1103 | ( |
1104 | ( @$r == 1 and $l =~ /^\-and$/i ) |
1105 | or |
1106 | $l =~ /^\-or$/i |
1107 | ) |
1108 | ) { |
1109 | push @or, @$r |
1110 | } |
1111 | |
1112 | elsif ( |
1113 | ref $r eq 'HASH' |
1114 | and |
1115 | keys %$r == 1 |
1116 | and |
1117 | $l =~ /^\-(?:and|or)$/i |
1118 | ) { |
1119 | push @or, %$r; |
1120 | } |
1121 | |
1122 | else { |
1123 | push @or, $l, $r; |
1124 | } |
1125 | } |
1126 | else { |
1127 | push @or, $fin_idx->{$_}; |
1128 | } |
1129 | } |
1130 | |
1131 | $fin->{-or} = \@or; |
1132 | } |
8d005ad9 |
1133 | } |
1134 | else { |
1135 | # not a hash not an array |
07add744 |
1136 | $fin = { -and => [ $where ] }; |
135ac69d |
1137 | } |
1138 | |
1139 | # unroll single-element -and's |
1140 | while ( |
1141 | $fin->{-and} |
1142 | and |
1143 | @{$fin->{-and}} < 2 |
1144 | ) { |
1145 | my $and = delete $fin->{-and}; |
1146 | last if @$and == 0; |
1147 | |
1148 | # at this point we have @$and == 1 |
1149 | if ( |
1150 | ref $and->[0] eq 'HASH' |
1151 | and |
1152 | ! grep { exists $fin->{$_} } keys %{$and->[0]} |
1153 | ) { |
1154 | $fin = { |
1155 | %$fin, %{$and->[0]} |
1156 | }; |
1157 | } |
07add744 |
1158 | else { |
1159 | $fin->{-and} = $and; |
1160 | last; |
1161 | } |
135ac69d |
1162 | } |
1163 | |
1164 | # compress same-column conds found in $fin |
1165 | for my $col ( grep { $_ !~ /^\-/ } keys %$fin ) { |
1166 | next unless ref $fin->{$col} eq 'ARRAY' and ($fin->{$col}[0]||'') =~ /^\-and$/i; |
1167 | my $val_bag = { map { |
5379386e |
1168 | (! defined $_ ) ? ( UNDEF => undef ) |
1169 | : ( ! length ref $_ or is_plain_value $_ ) ? ( "VAL_$_" => $_ ) |
135ac69d |
1170 | : ( ( 'SER_' . serialize $_ ) => $_ ) |
1171 | } @{$fin->{$col}}[1 .. $#{$fin->{$col}}] }; |
1172 | |
1173 | if (keys %$val_bag == 1 ) { |
1174 | ($fin->{$col}) = values %$val_bag; |
1175 | } |
1176 | else { |
1177 | $fin->{$col} = [ -and => map { $val_bag->{$_} } sort keys %$val_bag ]; |
1178 | } |
8d005ad9 |
1179 | } |
1180 | |
135ac69d |
1181 | return keys %$fin ? $fin : (); |
8d005ad9 |
1182 | } |
1183 | |
1184 | sub _collapse_cond_unroll_pairs { |
1185 | my ($self, $pairs) = @_; |
1186 | |
1187 | my @conds; |
1188 | |
1189 | while (@$pairs) { |
6565d2c3 |
1190 | my ($lhs, $rhs) = splice @$pairs, 0, 2; |
8d005ad9 |
1191 | |
1192 | if ($lhs eq '') { |
1193 | push @conds, $self->_collapse_cond($rhs); |
1194 | } |
1195 | elsif ( $lhs =~ /^\-and$/i ) { |
1196 | push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY')); |
1197 | } |
1198 | elsif ( $lhs =~ /^\-or$/i ) { |
1199 | push @conds, $self->_collapse_cond( |
1200 | (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs |
1201 | ); |
1202 | } |
1203 | else { |
1204 | if (ref $rhs eq 'HASH' and ! keys %$rhs) { |
1205 | # FIXME - SQLA seems to be doing... nothing...? |
1206 | } |
1d29e7e2 |
1207 | # normalize top level -ident, for saner extract_fixed_condition_columns code |
5f35ba0f |
1208 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) { |
1209 | push @conds, { $lhs => { '=', $rhs } }; |
1210 | } |
1211 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-value} and is_plain_value $rhs->{-value}) { |
1212 | push @conds, { $lhs => $rhs->{-value} }; |
1213 | } |
8d005ad9 |
1214 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) { |
1d29e7e2 |
1215 | if ( length ref $rhs->{'='} and is_literal_value $rhs->{'='} ) { |
5f35ba0f |
1216 | push @conds, { $lhs => $rhs }; |
1217 | } |
1218 | else { |
6565d2c3 |
1219 | for my $p ($self->_collapse_cond_unroll_pairs([ $lhs => $rhs->{'='} ])) { |
5f35ba0f |
1220 | |
1221 | # extra sanity check |
1222 | if (keys %$p > 1) { |
1223 | require Data::Dumper::Concise; |
1224 | local $Data::Dumper::Deepcopy = 1; |
1225 | $self->throw_exception( |
1226 | "Internal error: unexpected collapse unroll:" |
1227 | . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p } |
1228 | ); |
1229 | } |
8d005ad9 |
1230 | |
5f35ba0f |
1231 | my ($l, $r) = %$p; |
8d005ad9 |
1232 | |
1d29e7e2 |
1233 | push @conds, ( |
1234 | ! length ref $r |
1235 | or |
1236 | # the unroller recursion may return a '=' prepended value already |
1237 | ref $r eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='} |
1238 | or |
1239 | is_plain_value($r) |
1240 | ) |
5f35ba0f |
1241 | ? { $l => $r } |
1242 | : { $l => { '=' => $r } } |
1243 | ; |
1244 | } |
8d005ad9 |
1245 | } |
1246 | } |
1247 | elsif (ref $rhs eq 'ARRAY') { |
1248 | # some of these conditionals encounter multi-values - roll them out using |
1249 | # an unshift, which will cause extra looping in the while{} above |
1250 | if (! @$rhs ) { |
1251 | push @conds, { $lhs => [] }; |
1252 | } |
1253 | elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) { |
1254 | $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ") |
1255 | if @$rhs == 1; |
1256 | |
1257 | if( $rhs->[0] =~ /^\-and$/i ) { |
6565d2c3 |
1258 | unshift @$pairs, map { $lhs => $_ } @{$rhs}[1..$#$rhs]; |
8d005ad9 |
1259 | } |
1260 | # if not an AND then it's an OR |
1261 | elsif(@$rhs == 2) { |
6565d2c3 |
1262 | unshift @$pairs, $lhs => $rhs->[1]; |
8d005ad9 |
1263 | } |
1264 | else { |
953d5b7d |
1265 | push @conds, { $lhs => [ @{$rhs}[1..$#$rhs] ] }; |
8d005ad9 |
1266 | } |
1267 | } |
1268 | elsif (@$rhs == 1) { |
6565d2c3 |
1269 | unshift @$pairs, $lhs => $rhs->[0]; |
8d005ad9 |
1270 | } |
1271 | else { |
1272 | push @conds, { $lhs => $rhs }; |
1273 | } |
1274 | } |
c1f3f2e8 |
1275 | # unroll func + { -value => ... } |
1276 | elsif ( |
1277 | ref $rhs eq 'HASH' |
1278 | and |
1279 | ( my ($subop) = keys %$rhs ) == 1 |
1280 | and |
1281 | length ref ((values %$rhs)[0]) |
1282 | and |
1283 | my $vref = is_plain_value( (values %$rhs)[0] ) |
1284 | ) { |
5379386e |
1285 | push @conds, { $lhs => { $subop => $$vref } } |
c1f3f2e8 |
1286 | } |
8d005ad9 |
1287 | else { |
1288 | push @conds, { $lhs => $rhs }; |
1289 | } |
1290 | } |
1291 | } |
1292 | |
1293 | return @conds; |
1294 | } |
1295 | |
8e40a627 |
1296 | # Analyzes a given condition and attempts to extract all columns |
1297 | # with a definitive fixed-condition criteria. Returns a hashref |
1298 | # of k/v pairs suitable to be passed to set_columns(), with a |
1299 | # MAJOR CAVEAT - multi-value (contradictory) equalities are still |
1300 | # represented as a reference to the UNRESOVABLE_CONDITION constant |
1301 | # The reason we do this is that some codepaths only care about the |
1302 | # codition being stable, as opposed to actually making sense |
5f11e54f |
1303 | # |
8e40a627 |
1304 | # The normal mode is used to figure out if a resultset is constrained |
1305 | # to a column which is part of a unique constraint, which in turn |
1306 | # allows us to better predict how ordering will behave etc. |
1307 | # |
1308 | # With the optional "consider_nulls" boolean argument, the function |
1309 | # is instead used to infer inambiguous values from conditions |
1310 | # (e.g. the inheritance of resultset conditions on new_result) |
1311 | # |
5f11e54f |
1312 | sub _extract_fixed_condition_columns { |
8e40a627 |
1313 | my ($self, $where, $consider_nulls) = @_; |
1314 | my $where_hash = $self->_collapse_cond($_[1]); |
1315 | |
1316 | my $res = {}; |
1317 | my ($c, $v); |
1318 | for $c (keys %$where_hash) { |
1319 | my $vals; |
1320 | |
1321 | if (!defined ($v = $where_hash->{$c}) ) { |
b34d9331 |
1322 | $vals->{UNDEF} = $v if $consider_nulls |
8e40a627 |
1323 | } |
1324 | elsif ( |
8e40a627 |
1325 | ref $v eq 'HASH' |
1326 | and |
1327 | keys %$v == 1 |
5f35ba0f |
1328 | ) { |
1329 | if (exists $v->{-value}) { |
1330 | if (defined $v->{-value}) { |
b34d9331 |
1331 | $vals->{"VAL_$v->{-value}"} = $v->{-value} |
5f35ba0f |
1332 | } |
1333 | elsif( $consider_nulls ) { |
b34d9331 |
1334 | $vals->{UNDEF} = $v->{-value}; |
5f35ba0f |
1335 | } |
1336 | } |
8e40a627 |
1337 | # do not need to check for plain values - _collapse_cond did it for us |
1d29e7e2 |
1338 | elsif( |
1339 | length ref $v->{'='} |
1340 | and |
1341 | ( |
1342 | ( ref $v->{'='} eq 'HASH' and keys %{$v->{'='}} == 1 and exists $v->{'='}{-ident} ) |
1343 | or |
1344 | is_literal_value($v->{'='}) |
1345 | ) |
1346 | ) { |
b34d9331 |
1347 | $vals->{ 'SER_' . serialize $v->{'='} } = $v->{'='}; |
5f35ba0f |
1348 | } |
1349 | } |
1350 | elsif ( |
1351 | ! length ref $v |
1352 | or |
1353 | is_plain_value ($v) |
8e40a627 |
1354 | ) { |
b34d9331 |
1355 | $vals->{"VAL_$v"} = $v; |
8e40a627 |
1356 | } |
1357 | elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') { |
1358 | for ( @{$v}[1..$#$v] ) { |
1359 | my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls'); # always fish nulls out on recursion |
1360 | next unless exists $subval->{$c}; # didn't find anything |
b34d9331 |
1361 | $vals->{ |
1362 | ! defined $subval->{$c} ? 'UNDEF' |
1363 | : ( ! length ref $subval->{$c} or is_plain_value $subval->{$c} ) ? "VAL_$subval->{$c}" |
1364 | : ( 'SER_' . serialize $subval->{$c} ) |
1365 | } = $subval->{$c}; |
8d005ad9 |
1366 | } |
5f11e54f |
1367 | } |
8e40a627 |
1368 | |
1369 | if (keys %$vals == 1) { |
1370 | ($res->{$c}) = (values %$vals) |
b34d9331 |
1371 | unless !$consider_nulls and exists $vals->{UNDEF}; |
8e40a627 |
1372 | } |
1373 | elsif (keys %$vals > 1) { |
1374 | $res->{$c} = UNRESOLVABLE_CONDITION; |
1375 | } |
5f11e54f |
1376 | } |
8d005ad9 |
1377 | |
8e40a627 |
1378 | $res; |
c0748280 |
1379 | } |
bac6c4fb |
1380 | |
d28bb90d |
1381 | 1; |