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'; |
b5ce6748 |
18 | use DBIx::Class::_Util 'UNRESOLVABLE_CONDITION'; |
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}; |
cb913aa5 |
405 | local $sql_maker->{escape_char} = $sql_maker->{escape_char}; |
3f5b99fe |
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"]; |
cb913aa5 |
410 | $sql_maker->{escape_char} = "\xFF"; |
e493ecb2 |
411 | # if we don't unset it we screw up retarded but unfortunately working |
412 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
413 | $sql_maker->{name_sep} = ''; |
414 | } |
415 | |
416 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
417 | |
1a736efb |
418 | # generate sql chunks |
419 | my $to_scan = { |
420 | restricting => [ |
e1861c2c |
421 | $sql_maker->_recurse_where ($attrs->{where}), |
1e4f9fb3 |
422 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
423 | ], |
424 | grouping => [ |
425 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
426 | ], |
97e130fa |
427 | joining => [ |
428 | $sql_maker->_recurse_from ( |
e1861c2c |
429 | ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0], |
430 | @{$attrs->{from}}[1 .. $#{$attrs->{from}}], |
97e130fa |
431 | ), |
432 | ], |
1a736efb |
433 | selecting => [ |
ad1d374e |
434 | map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}}, |
1e4f9fb3 |
435 | ], |
436 | ordering => [ |
437 | map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker), |
1a736efb |
438 | ], |
439 | }; |
07f31d19 |
440 | |
0dadd60d |
441 | # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are |
442 | # bind value specs left in by the sloppy renderer above. It is ok to do this |
443 | # at this point, since we are going to end up rewriting this crap anyway |
444 | for my $v (values %$to_scan) { |
445 | my @nv; |
446 | for (@$v) { |
447 | next if ( |
448 | ! defined $_ |
449 | or |
450 | ( |
451 | ref $_ eq 'ARRAY' |
452 | and |
453 | ( @$_ == 0 or @$_ == 2 ) |
454 | ) |
455 | ); |
456 | |
457 | if (ref $_) { |
458 | require Data::Dumper::Concise; |
459 | $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) ); |
460 | } |
461 | |
462 | push @nv, $_; |
463 | } |
464 | |
465 | $v = \@nv; |
466 | } |
467 | |
468 | # kill all selectors which look like a proper subquery |
469 | # this is a sucky heuristic *BUT* - if we get it wrong the query will simply |
470 | # fail to run, so we are relatively safe |
471 | $to_scan->{selecting} = [ grep { |
472 | $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi |
473 | } @{ $to_scan->{selecting} || [] } ]; |
07f31d19 |
474 | |
318e3d94 |
475 | # first see if we have any exact matches (qualified or unqualified) |
476 | for my $type (keys %$to_scan) { |
477 | for my $piece (@{$to_scan->{$type}}) { |
478 | if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) { |
479 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
480 | $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece; |
481 | } |
482 | } |
483 | } |
484 | |
485 | # now loop through all fully qualified columns and get the corresponding |
1a736efb |
486 | # alias (should work even if they are in scalarrefs) |
cb913aa5 |
487 | my $ident_re = $sql_maker->_quoted_ident_re; |
ad630f4b |
488 | for my $alias (keys %$alias_list) { |
1a736efb |
489 | my $al_re = qr/ |
cb913aa5 |
490 | $lquote \Q$alias\E $rquote $sep ($ident_re)? |
1a736efb |
491 | | |
a1228381 |
492 | \b \Q$alias\E \. ([^\s\)\($rquote]+)? |
1a736efb |
493 | /x; |
494 | |
1a736efb |
495 | for my $type (keys %$to_scan) { |
496 | for my $piece (@{$to_scan->{$type}}) { |
cb913aa5 |
497 | if (my @matches = map { $sql_maker->_unquote($_) } $piece =~ /$al_re/g) { |
97e130fa |
498 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
499 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_" |
97e130fa |
500 | for grep { defined $_ } @matches; |
501 | } |
1a736efb |
502 | } |
ad630f4b |
503 | } |
1a736efb |
504 | } |
505 | |
506 | # now loop through unqualified column names, and try to locate them within |
507 | # the chunks |
508 | for my $col (keys %$colinfo) { |
3f5b99fe |
509 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
510 | |
a1228381 |
511 | my $col_re = qr/ $lquote (\Q$col\E) $rquote /x; |
07f31d19 |
512 | |
1a736efb |
513 | for my $type (keys %$to_scan) { |
514 | for my $piece (@{$to_scan->{$type}}) { |
318e3d94 |
515 | if ( my @matches = $piece =~ /$col_re/g) { |
a4812caa |
516 | my $alias = $colinfo->{$col}{-source_alias}; |
97e130fa |
517 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
1e4f9fb3 |
518 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
97e130fa |
519 | for grep { defined $_ } @matches; |
a4812caa |
520 | } |
1a736efb |
521 | } |
07f31d19 |
522 | } |
523 | } |
524 | |
525 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
526 | for my $j (values %$alias_list) { |
07f31d19 |
527 | my $alias = $j->{-alias} or next; |
97e130fa |
528 | $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if ( |
07f31d19 |
529 | (not $j->{-join_type}) |
530 | or |
531 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
532 | ); |
533 | } |
534 | |
1e4f9fb3 |
535 | for (keys %$aliases_by_type) { |
536 | delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}}; |
537 | } |
538 | |
964a3c71 |
539 | return $aliases_by_type; |
07f31d19 |
540 | } |
541 | |
eb58c082 |
542 | # This is the engine behind { distinct => 1 } and the general |
543 | # complex prefetch grouper |
0a3441ee |
544 | sub _group_over_selection { |
560978e2 |
545 | my ($self, $attrs) = @_; |
0a3441ee |
546 | |
560978e2 |
547 | my $colinfos = $self->_resolve_column_info ($attrs->{from}); |
0a3441ee |
548 | |
549 | my (@group_by, %group_index); |
550 | |
36fd7f07 |
551 | # the logic is: if it is a { func => val } we assume an aggregate, |
552 | # otherwise if \'...' or \[...] we assume the user knows what is |
553 | # going on thus group over it |
560978e2 |
554 | for (@{$attrs->{select}}) { |
0a3441ee |
555 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
556 | push @group_by, $_; |
557 | $group_index{$_}++; |
560978e2 |
558 | if ($colinfos->{$_} and $_ !~ /\./ ) { |
0a3441ee |
559 | # add a fully qualified version as well |
560978e2 |
560 | $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; |
0a3441ee |
561 | } |
07f31d19 |
562 | } |
563 | } |
ad630f4b |
564 | |
eb58c082 |
565 | my @order_by = $self->_extract_order_criteria($attrs->{order_by}) |
566 | or return (\@group_by, $attrs->{order_by}); |
567 | |
568 | # add any order_by parts that are not already present in the group_by |
569 | # to maintain SQL cross-compatibility and general sanity |
570 | # |
571 | # also in case the original selection is *not* unique, or in case part |
572 | # of the ORDER BY refers to a multiplier - we will need to replace the |
573 | # skipped order_by elements with their MIN/MAX equivalents as to maintain |
574 | # the proper overall order without polluting the group criteria (and |
575 | # possibly changing the outcome entirely) |
576 | |
577 | my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes); |
578 | |
579 | my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by); |
580 | |
581 | for my $o_idx (0 .. $#order_by) { |
582 | |
583 | # if the chunk is already a min/max function - there is nothing left to touch |
584 | next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix; |
585 | |
0a3441ee |
586 | # only consider real columns (for functions the user got to do an explicit group_by) |
eb58c082 |
587 | my $chunk_ci; |
588 | if ( |
589 | @{$order_by[$o_idx]} != 1 |
590 | or |
591 | # only declare an unknown *plain* identifier as "leftover" if we are called with |
592 | # aliastypes to examine. If there are none - we are still in _resolve_attrs, and |
593 | # can just assume the user knows what they want |
594 | ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} ) |
595 | ) { |
596 | push @$leftovers, $order_by[$o_idx][0]; |
14e26c5f |
597 | } |
560978e2 |
598 | |
eb58c082 |
599 | next unless $chunk_ci; |
600 | |
601 | # no duplication of group criteria |
602 | next if $group_index{$chunk_ci->{-fq_colname}}; |
603 | |
604 | $aliastypes ||= ( |
605 | $attrs->{_aliastypes} |
560978e2 |
606 | or |
eb58c082 |
607 | $self->_resolve_aliastypes_from_select_args({ |
608 | from => $attrs->{from}, |
609 | order_by => $attrs->{order_by}, |
610 | }) |
611 | ) if $group_already_unique; |
612 | |
613 | # check that we are not ordering by a multiplier (if a check is requested at all) |
614 | if ( |
615 | $group_already_unique |
616 | and |
617 | ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}} |
618 | and |
619 | ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}} |
560978e2 |
620 | ) { |
eb58c082 |
621 | push @group_by, $chunk_ci->{-fq_colname}; |
622 | $group_index{$chunk_ci->{-fq_colname}}++ |
560978e2 |
623 | } |
eb58c082 |
624 | else { |
625 | # We need to order by external columns without adding them to the group |
626 | # (eiehter a non-unique selection, or a multi-external) |
627 | # |
628 | # This doesn't really make sense in SQL, however from DBICs point |
629 | # of view is rather valid (e.g. order the leftmost objects by whatever |
630 | # criteria and get the offset/rows many). There is a way around |
631 | # this however in SQL - we simply tae the direction of each piece |
632 | # of the external order and convert them to MIN(X) for ASC or MAX(X) |
633 | # for DESC, and group_by the root columns. The end result should be |
634 | # exactly what we expect |
635 | |
636 | # FIXME - this code is a joke, will need to be completely rewritten in |
637 | # the DQ branch. But I need to push a POC here, otherwise the |
638 | # pesky tests won't pass |
639 | # wrap any part of the order_by that "responds" to an ordering alias |
640 | # into a MIN/MAX |
641 | $sql_maker ||= $self->sql_maker; |
642 | $order_chunks ||= [ |
643 | map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}) |
644 | ]; |
0a3441ee |
645 | |
eb58c082 |
646 | my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]); |
647 | |
648 | $new_order_by[$o_idx] = \[ |
649 | sprintf( '%s( %s )%s', |
650 | ($is_desc ? 'MAX' : 'MIN'), |
651 | $chunk, |
652 | ($is_desc ? ' DESC' : ''), |
653 | ), |
654 | @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ] |
655 | ]; |
656 | } |
0a3441ee |
657 | } |
658 | |
eb58c082 |
659 | $self->throw_exception ( sprintf |
9736be65 |
660 | 'Unable to programatically derive a required group_by from the supplied ' |
661 | . 'order_by criteria. To proceed either add an explicit group_by, or ' |
662 | . 'simplify your order_by to only include plain columns ' |
663 | . '(supplied order_by: %s)', |
eb58c082 |
664 | join ', ', map { "'$_'" } @$leftovers, |
665 | ) if $leftovers; |
666 | |
667 | # recreate the untouched order parts |
668 | if (@new_order_by) { |
669 | $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks ); |
670 | } |
671 | |
672 | return ( |
673 | \@group_by, |
674 | (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged |
675 | ); |
07f31d19 |
676 | } |
677 | |
d28bb90d |
678 | sub _resolve_ident_sources { |
679 | my ($self, $ident) = @_; |
680 | |
681 | my $alias2source = {}; |
d28bb90d |
682 | |
683 | # the reason this is so contrived is that $ident may be a {from} |
684 | # structure, specifying multiple tables to join |
6298a324 |
685 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
686 | # this is compat mode for insert/update/delete which do not deal with aliases |
687 | $alias2source->{me} = $ident; |
d28bb90d |
688 | } |
689 | elsif (ref $ident eq 'ARRAY') { |
690 | |
691 | for (@$ident) { |
692 | my $tabinfo; |
693 | if (ref $_ eq 'HASH') { |
694 | $tabinfo = $_; |
d28bb90d |
695 | } |
696 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
697 | $tabinfo = $_->[0]; |
698 | } |
699 | |
4376a157 |
700 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
701 | if ($tabinfo->{-rsrc}); |
d28bb90d |
702 | } |
703 | } |
704 | |
90f10b5a |
705 | return $alias2source; |
d28bb90d |
706 | } |
707 | |
708 | # Takes $ident, \@column_names |
709 | # |
710 | # returns { $column_name => \%column_info, ... } |
711 | # also note: this adds -result_source => $rsrc to the column info |
712 | # |
09e14fdc |
713 | # If no columns_names are supplied returns info about *all* columns |
714 | # for all sources |
d28bb90d |
715 | sub _resolve_column_info { |
716 | my ($self, $ident, $colnames) = @_; |
8d005ad9 |
717 | |
718 | return {} if $colnames and ! @$colnames; |
719 | |
90f10b5a |
720 | my $alias2src = $self->_resolve_ident_sources($ident); |
d28bb90d |
721 | |
52416317 |
722 | my (%seen_cols, @auto_colnames); |
d28bb90d |
723 | |
724 | # compile a global list of column names, to be able to properly |
725 | # disambiguate unqualified column names (if at all possible) |
726 | for my $alias (keys %$alias2src) { |
727 | my $rsrc = $alias2src->{$alias}; |
728 | for my $colname ($rsrc->columns) { |
729 | push @{$seen_cols{$colname}}, $alias; |
3f5b99fe |
730 | push @auto_colnames, "$alias.$colname" unless $colnames; |
d28bb90d |
731 | } |
732 | } |
733 | |
09e14fdc |
734 | $colnames ||= [ |
735 | @auto_colnames, |
736 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
737 | ]; |
738 | |
52416317 |
739 | my (%return, $colinfos); |
d28bb90d |
740 | foreach my $col (@$colnames) { |
52416317 |
741 | my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x; |
d28bb90d |
742 | |
52416317 |
743 | # if the column was seen exactly once - we know which rsrc it came from |
744 | $source_alias ||= $seen_cols{$colname}[0] |
745 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1); |
d28bb90d |
746 | |
52416317 |
747 | next unless $source_alias; |
748 | |
749 | my $rsrc = $alias2src->{$source_alias} |
750 | or next; |
751 | |
752 | $return{$col} = { |
6395604e |
753 | %{ |
754 | ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname} |
755 | || |
756 | $self->throw_exception( |
757 | "No such column '$colname' on source " . $rsrc->source_name |
758 | ); |
759 | }, |
d28bb90d |
760 | -result_source => $rsrc, |
52416317 |
761 | -source_alias => $source_alias, |
81bf295c |
762 | -fq_colname => $col eq $colname ? "$source_alias.$col" : $col, |
763 | -colname => $colname, |
d28bb90d |
764 | }; |
81bf295c |
765 | |
766 | $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname; |
d28bb90d |
767 | } |
768 | |
769 | return \%return; |
770 | } |
771 | |
289ac713 |
772 | # The DBIC relationship chaining implementation is pretty simple - every |
773 | # new related_relationship is pushed onto the {from} stack, and the {select} |
774 | # window simply slides further in. This means that when we count somewhere |
775 | # in the middle, we got to make sure that everything in the join chain is an |
776 | # actual inner join, otherwise the count will come back with unpredictable |
777 | # results (a resultset may be generated with _some_ rows regardless of if |
778 | # the relation which the $rs currently selects has rows or not). E.g. |
779 | # $artist_rs->cds->count - normally generates: |
780 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
781 | # which actually returns the number of artists * (number of cds || 1) |
782 | # |
783 | # So what we do here is crawl {from}, determine if the current alias is at |
784 | # the top of the stack, and if not - make sure the chain is inner-joined down |
785 | # to the root. |
786 | # |
31a8aaaf |
787 | sub _inner_join_to_node { |
289ac713 |
788 | my ($self, $from, $alias) = @_; |
789 | |
302d35f8 |
790 | my $switch_branch = $self->_find_join_path_to_node($from, $alias); |
289ac713 |
791 | |
302d35f8 |
792 | return $from unless @{$switch_branch||[]}; |
289ac713 |
793 | |
794 | # So it looks like we will have to switch some stuff around. |
795 | # local() is useless here as we will be leaving the scope |
796 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
797 | # So replace the first hashref in the node arrayref manually |
289ac713 |
798 | my @new_from = ($from->[0]); |
faeb2407 |
799 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
800 | |
801 | for my $j (@{$from}[1 .. $#$from]) { |
802 | my $jalias = $j->[0]{-alias}; |
803 | |
804 | if ($sw_idx->{$jalias}) { |
805 | my %attrs = %{$j->[0]}; |
806 | delete $attrs{-join_type}; |
807 | push @new_from, [ |
808 | \%attrs, |
809 | @{$j}[ 1 .. $#$j ], |
810 | ]; |
811 | } |
812 | else { |
813 | push @new_from, $j; |
814 | } |
815 | } |
816 | |
817 | return \@new_from; |
818 | } |
819 | |
302d35f8 |
820 | sub _find_join_path_to_node { |
821 | my ($self, $from, $target_alias) = @_; |
822 | |
823 | # subqueries and other oddness are naturally not supported |
824 | return undef if ( |
825 | ref $from ne 'ARRAY' |
826 | || |
827 | ref $from->[0] ne 'HASH' |
828 | || |
829 | ! defined $from->[0]{-alias} |
830 | ); |
831 | |
832 | # no path - the head is the alias |
833 | return [] if $from->[0]{-alias} eq $target_alias; |
834 | |
835 | for my $i (1 .. $#$from) { |
836 | return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias ); |
837 | } |
838 | |
839 | # something else went quite wrong |
840 | return undef; |
841 | } |
842 | |
bac358c9 |
843 | sub _extract_order_criteria { |
1a736efb |
844 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
845 | |
1a736efb |
846 | my $parser = sub { |
e6977bbb |
847 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
848 | |
1a736efb |
849 | return scalar $sql_maker->_order_by_chunks ($order_by) |
850 | unless wantarray; |
c0748280 |
851 | |
004b0a97 |
852 | my $sep = quotemeta($sql_maker->name_sep); |
853 | |
854 | my @quotes = map { quotemeta($_) } |
855 | $orig_quote_chars |
856 | ? @$orig_quote_chars |
857 | : ($sql_maker->_quote_chars, $sql_maker->_escape_char); |
858 | |
859 | my $quoted_ident_re = $sql_maker->_quoted_ident_re(@quotes); |
e6977bbb |
860 | |
1a736efb |
861 | my @chunks; |
bac358c9 |
862 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
863 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
864 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
865 | |
866 | # order criteria may have come back pre-quoted (literals and whatnot) |
867 | # this is fragile, but the best we can currently do |
004b0a97 |
868 | if (my @quoted = $chunk->[0] =~ /\A ($quoted_ident_re) (?: $sep ($quoted_ident_re) )? \z/x) { |
869 | $chunk->[0] = join('.', map { $sql_maker->_unquote($_, @quotes) } grep { defined } @quoted); |
870 | } |
e6977bbb |
871 | |
1a736efb |
872 | push @chunks, $chunk; |
bac6c4fb |
873 | } |
1a736efb |
874 | |
875 | return @chunks; |
876 | }; |
877 | |
878 | if ($sql_maker) { |
879 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
880 | } |
881 | else { |
1a736efb |
882 | $sql_maker = $self->sql_maker; |
e6977bbb |
883 | |
884 | # pass these in to deal with literals coming from |
885 | # the user or the deep guts of prefetch |
004b0a97 |
886 | my $orig_quote_chars = [$sql_maker->_quote_chars, $sql_maker->_escape_char]; |
e6977bbb |
887 | |
1a736efb |
888 | local $sql_maker->{quote_char}; |
e6977bbb |
889 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
890 | } |
bac6c4fb |
891 | } |
892 | |
7cec4356 |
893 | sub _order_by_is_stable { |
5f11e54f |
894 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
895 | |
eb58c082 |
896 | my @cols = ( |
8d005ad9 |
897 | ( map { $_->[0] } $self->_extract_order_criteria($order_by) ), |
8e40a627 |
898 | ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ), |
df4312bc |
899 | ) or return 0; |
eb58c082 |
900 | |
901 | my $colinfo = $self->_resolve_column_info($ident, \@cols); |
902 | |
903 | return keys %$colinfo |
904 | ? $self->_columns_comprise_identifying_set( $colinfo, \@cols ) |
df4312bc |
905 | : 0 |
eb58c082 |
906 | ; |
907 | } |
c0748280 |
908 | |
eb58c082 |
909 | sub _columns_comprise_identifying_set { |
910 | my ($self, $colinfo, $columns) = @_; |
7cec4356 |
911 | |
912 | my $cols_per_src; |
eb58c082 |
913 | $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_ |
914 | for grep { defined $_ } @{$colinfo}{@$columns}; |
7cec4356 |
915 | |
916 | for (values %$cols_per_src) { |
917 | my $src = (values %$_)[0]->{-result_source}; |
918 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
919 | } |
920 | |
df4312bc |
921 | return 0; |
7cec4356 |
922 | } |
923 | |
df4312bc |
924 | # this is almost similar to _order_by_is_stable, except it takes |
0e81e691 |
925 | # a single rsrc, and will succeed only if the first portion of the order |
926 | # by is stable. |
927 | # returns that portion as a colinfo hashref on success |
df4312bc |
928 | sub _extract_colinfo_of_stable_main_source_order_by_portion { |
302d35f8 |
929 | my ($self, $attrs) = @_; |
0e81e691 |
930 | |
302d35f8 |
931 | my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias}); |
932 | |
933 | return unless defined $nodes; |
0e81e691 |
934 | |
935 | my @ord_cols = map |
936 | { $_->[0] } |
302d35f8 |
937 | ( $self->_extract_order_criteria($attrs->{order_by}) ) |
0e81e691 |
938 | ; |
939 | return unless @ord_cols; |
940 | |
302d35f8 |
941 | my $valid_aliases = { map { $_ => 1 } ( |
942 | $attrs->{from}[0]{-alias}, |
943 | map { values %$_ } @$nodes, |
944 | ) }; |
318e3d94 |
945 | |
302d35f8 |
946 | my $colinfos = $self->_resolve_column_info($attrs->{from}); |
947 | |
948 | my ($colinfos_to_return, $seen_main_src_cols); |
949 | |
950 | for my $col (@ord_cols) { |
951 | # if order criteria is unresolvable - there is nothing we can do |
952 | my $colinfo = $colinfos->{$col} or last; |
953 | |
954 | # if we reached the end of the allowed aliases - also nothing we can do |
955 | last unless $valid_aliases->{$colinfo->{-source_alias}}; |
956 | |
957 | $colinfos_to_return->{$col} = $colinfo; |
958 | |
959 | $seen_main_src_cols->{$colinfo->{-colname}} = 1 |
960 | if $colinfo->{-source_alias} eq $attrs->{alias}; |
0e81e691 |
961 | } |
962 | |
302d35f8 |
963 | # FIXME the condition may be singling out things on its own, so we |
964 | # conceivable could come back wi "stable-ordered by nothing" |
965 | # not confient enough in the parser yet, so punt for the time being |
966 | return unless $seen_main_src_cols; |
0e81e691 |
967 | |
302d35f8 |
968 | my $main_src_fixed_cols_from_cond = [ $attrs->{where} |
969 | ? ( |
970 | map |
971 | { |
972 | ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} ) |
973 | ? $colinfos->{$_}{-colname} |
974 | : () |
975 | } |
8e40a627 |
976 | keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) } |
302d35f8 |
977 | ) |
978 | : () |
979 | ]; |
0e81e691 |
980 | |
302d35f8 |
981 | return $attrs->{result_source}->_identifying_column_set([ |
982 | keys %$seen_main_src_cols, |
983 | @$main_src_fixed_cols_from_cond, |
984 | ]) ? $colinfos_to_return : (); |
0e81e691 |
985 | } |
986 | |
8d005ad9 |
987 | # Attempts to flatten a passed in SQLA condition as much as possible towards |
988 | # a plain hashref, *without* altering its semantics. Required by |
989 | # create/populate being able to extract definitive conditions from preexisting |
990 | # resultset {where} stacks |
991 | # |
992 | # FIXME - while relatively robust, this is still imperfect, one of the first |
993 | # things to tackle with DQ |
994 | sub _collapse_cond { |
995 | my ($self, $where, $where_is_anded_array) = @_; |
996 | |
997 | if (! $where) { |
998 | return; |
999 | } |
1000 | elsif ($where_is_anded_array or ref $where eq 'HASH') { |
1001 | |
1002 | my @pairs; |
1003 | |
1004 | my @pieces = $where_is_anded_array ? @$where : $where; |
1005 | while (@pieces) { |
1006 | my $chunk = shift @pieces; |
1007 | |
1008 | if (ref $chunk eq 'HASH') { |
1009 | push @pairs, map { [ $_ => $chunk->{$_} ] } sort keys %$chunk; |
1010 | } |
1011 | elsif (ref $chunk eq 'ARRAY') { |
1012 | push @pairs, [ -or => $chunk ] |
1013 | if @$chunk; |
1014 | } |
1015 | elsif ( ! ref $chunk) { |
1016 | push @pairs, [ $chunk, shift @pieces ]; |
1017 | } |
1018 | else { |
1019 | push @pairs, [ '', $chunk ]; |
1020 | } |
1021 | } |
1022 | |
1023 | return unless @pairs; |
1024 | |
1025 | my @conds = $self->_collapse_cond_unroll_pairs(\@pairs) |
1026 | or return; |
1027 | |
1028 | # Consolidate various @conds back into something more compact |
1029 | my $fin; |
1030 | |
1031 | for my $c (@conds) { |
1032 | if (ref $c ne 'HASH') { |
1033 | push @{$fin->{-and}}, $c; |
1034 | } |
1035 | else { |
1036 | for my $col (sort keys %$c) { |
1037 | if (exists $fin->{$col}) { |
1038 | my ($l, $r) = ($fin->{$col}, $c->{$col}); |
1039 | |
1040 | (ref $_ ne 'ARRAY' or !@$_) and $_ = [ -and => $_ ] for ($l, $r); |
1041 | |
1042 | if (@$l and @$r and $l->[0] eq $r->[0] and $l->[0] eq '-and') { |
1043 | $fin->{$col} = [ -and => map { @$_[1..$#$_] } ($l, $r) ]; |
1044 | } |
1045 | else { |
1046 | $fin->{$col} = [ -and => $fin->{$col}, $c->{$col} ]; |
1047 | } |
1048 | } |
1049 | else { |
1050 | $fin->{$col} = $c->{$col}; |
1051 | } |
1052 | } |
1053 | } |
1054 | } |
1055 | |
1056 | if ( ref $fin->{-and} eq 'ARRAY' and @{$fin->{-and}} == 1 ) { |
1057 | my $piece = (delete $fin->{-and})->[0]; |
1058 | if (ref $piece eq 'ARRAY') { |
1059 | $fin->{-or} = $fin->{-or} ? [ $piece, $fin->{-or} ] : $piece; |
1060 | } |
1061 | elsif (! exists $fin->{''}) { |
1062 | $fin->{''} = $piece; |
1063 | } |
1064 | } |
1065 | |
1066 | return $fin; |
1067 | } |
1068 | elsif (ref $where eq 'ARRAY') { |
1069 | my @w = @$where; |
1070 | |
1071 | while ( @w and ( |
1072 | (ref $w[0] eq 'ARRAY' and ! @{$w[0]} ) |
1073 | or |
1074 | (ref $w[0] eq 'HASH' and ! keys %{$w[0]}) |
1075 | )) { shift @w }; |
1076 | |
1077 | return unless @w; |
1078 | |
1079 | if ( @w == 1 ) { |
1080 | return ( ref $w[0] ) |
1081 | ? $self->_collapse_cond($w[0]) |
1082 | : { $w[0] => undef } |
1083 | ; |
1084 | } |
8e40a627 |
1085 | elsif ( @w == 2 and ! ref $w[0]) { |
8d005ad9 |
1086 | if ( ( $w[0]||'' ) =~ /^\-and$/i ) { |
1087 | return (ref $w[1] eq 'HASH' or ref $w[1] eq 'ARRAY') |
1088 | ? $self->_collapse_cond($w[1], (ref $w[1] eq 'ARRAY') ) |
1089 | : $self->throw_exception("Unsupported top-level op/arg pair: [ $w[0] => $w[1] ]") |
1090 | ; |
1091 | } |
1092 | else { |
1093 | return $self->_collapse_cond({ @w }); |
1094 | } |
1095 | } |
8e40a627 |
1096 | else { |
1097 | return { -or => \@w }; |
1098 | } |
8d005ad9 |
1099 | } |
1100 | else { |
1101 | # not a hash not an array |
1102 | return { '' => $where }; |
1103 | } |
1104 | |
8e40a627 |
1105 | die 'should not get here'; |
8d005ad9 |
1106 | } |
1107 | |
1108 | sub _collapse_cond_unroll_pairs { |
1109 | my ($self, $pairs) = @_; |
1110 | |
1111 | my @conds; |
1112 | |
1113 | while (@$pairs) { |
1114 | my ($lhs, $rhs) = @{ shift @$pairs }; |
1115 | |
1116 | if ($lhs eq '') { |
1117 | push @conds, $self->_collapse_cond($rhs); |
1118 | } |
1119 | elsif ( $lhs =~ /^\-and$/i ) { |
1120 | push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY')); |
1121 | } |
1122 | elsif ( $lhs =~ /^\-or$/i ) { |
1123 | push @conds, $self->_collapse_cond( |
1124 | (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs |
1125 | ); |
1126 | } |
1127 | else { |
1128 | if (ref $rhs eq 'HASH' and ! keys %$rhs) { |
1129 | # FIXME - SQLA seems to be doing... nothing...? |
1130 | } |
5f35ba0f |
1131 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) { |
1132 | push @conds, { $lhs => { '=', $rhs } }; |
1133 | } |
1134 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-value} and is_plain_value $rhs->{-value}) { |
1135 | push @conds, { $lhs => $rhs->{-value} }; |
1136 | } |
8d005ad9 |
1137 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) { |
5f35ba0f |
1138 | if( is_literal_value $rhs->{'='}) { |
1139 | push @conds, { $lhs => $rhs }; |
1140 | } |
1141 | else { |
1142 | for my $p ($self->_collapse_cond_unroll_pairs([ [ $lhs => $rhs->{'='} ] ])) { |
1143 | |
1144 | # extra sanity check |
1145 | if (keys %$p > 1) { |
1146 | require Data::Dumper::Concise; |
1147 | local $Data::Dumper::Deepcopy = 1; |
1148 | $self->throw_exception( |
1149 | "Internal error: unexpected collapse unroll:" |
1150 | . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p } |
1151 | ); |
1152 | } |
8d005ad9 |
1153 | |
5f35ba0f |
1154 | my ($l, $r) = %$p; |
8d005ad9 |
1155 | |
5f35ba0f |
1156 | push @conds, ( ! length ref $r or is_plain_value($r) ) |
1157 | ? { $l => $r } |
1158 | : { $l => { '=' => $r } } |
1159 | ; |
1160 | } |
8d005ad9 |
1161 | } |
1162 | } |
1163 | elsif (ref $rhs eq 'ARRAY') { |
1164 | # some of these conditionals encounter multi-values - roll them out using |
1165 | # an unshift, which will cause extra looping in the while{} above |
1166 | if (! @$rhs ) { |
1167 | push @conds, { $lhs => [] }; |
1168 | } |
1169 | elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) { |
1170 | $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ") |
1171 | if @$rhs == 1; |
1172 | |
1173 | if( $rhs->[0] =~ /^\-and$/i ) { |
1174 | unshift @$pairs, map { [ $lhs => $_ ] } @{$rhs}[1..$#$rhs]; |
1175 | } |
1176 | # if not an AND then it's an OR |
1177 | elsif(@$rhs == 2) { |
1178 | unshift @$pairs, [ $lhs => $rhs->[1] ]; |
1179 | } |
1180 | else { |
1181 | push @conds, { $lhs => $rhs }; |
1182 | } |
1183 | } |
1184 | elsif (@$rhs == 1) { |
1185 | unshift @$pairs, [ $lhs => $rhs->[0] ]; |
1186 | } |
1187 | else { |
1188 | push @conds, { $lhs => $rhs }; |
1189 | } |
1190 | } |
1191 | else { |
1192 | push @conds, { $lhs => $rhs }; |
1193 | } |
1194 | } |
1195 | } |
1196 | |
1197 | return @conds; |
1198 | } |
1199 | |
8e40a627 |
1200 | # Analyzes a given condition and attempts to extract all columns |
1201 | # with a definitive fixed-condition criteria. Returns a hashref |
1202 | # of k/v pairs suitable to be passed to set_columns(), with a |
1203 | # MAJOR CAVEAT - multi-value (contradictory) equalities are still |
1204 | # represented as a reference to the UNRESOVABLE_CONDITION constant |
1205 | # The reason we do this is that some codepaths only care about the |
1206 | # codition being stable, as opposed to actually making sense |
5f11e54f |
1207 | # |
8e40a627 |
1208 | # The normal mode is used to figure out if a resultset is constrained |
1209 | # to a column which is part of a unique constraint, which in turn |
1210 | # allows us to better predict how ordering will behave etc. |
1211 | # |
1212 | # With the optional "consider_nulls" boolean argument, the function |
1213 | # is instead used to infer inambiguous values from conditions |
1214 | # (e.g. the inheritance of resultset conditions on new_result) |
1215 | # |
1216 | my $undef_marker = \ do{ my $x = 'undef' }; |
5f11e54f |
1217 | sub _extract_fixed_condition_columns { |
8e40a627 |
1218 | my ($self, $where, $consider_nulls) = @_; |
1219 | my $where_hash = $self->_collapse_cond($_[1]); |
1220 | |
1221 | my $res = {}; |
1222 | my ($c, $v); |
1223 | for $c (keys %$where_hash) { |
1224 | my $vals; |
1225 | |
1226 | if (!defined ($v = $where_hash->{$c}) ) { |
1227 | $vals->{$undef_marker} = $v if $consider_nulls |
1228 | } |
1229 | elsif ( |
8e40a627 |
1230 | ref $v eq 'HASH' |
1231 | and |
1232 | keys %$v == 1 |
5f35ba0f |
1233 | ) { |
1234 | if (exists $v->{-value}) { |
1235 | if (defined $v->{-value}) { |
1236 | $vals->{$v->{-value}} = $v->{-value} |
1237 | } |
1238 | elsif( $consider_nulls ) { |
1239 | $vals->{$undef_marker} = $v->{-value}; |
1240 | } |
1241 | } |
8e40a627 |
1242 | # do not need to check for plain values - _collapse_cond did it for us |
5f35ba0f |
1243 | elsif(ref $v->{'='} and is_literal_value($v->{'='}) ) { |
1244 | $vals->{$v->{'='}} = $v->{'='}; |
1245 | } |
1246 | } |
1247 | elsif ( |
1248 | ! length ref $v |
1249 | or |
1250 | is_plain_value ($v) |
8e40a627 |
1251 | ) { |
5f35ba0f |
1252 | $vals->{$v} = $v; |
8e40a627 |
1253 | } |
1254 | elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') { |
1255 | for ( @{$v}[1..$#$v] ) { |
1256 | my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls'); # always fish nulls out on recursion |
1257 | next unless exists $subval->{$c}; # didn't find anything |
1258 | $vals->{defined $subval->{$c} ? $subval->{$c} : $undef_marker} = $subval->{$c}; |
8d005ad9 |
1259 | } |
5f11e54f |
1260 | } |
8e40a627 |
1261 | |
1262 | if (keys %$vals == 1) { |
1263 | ($res->{$c}) = (values %$vals) |
1264 | unless !$consider_nulls and exists $vals->{$undef_marker}; |
1265 | } |
1266 | elsif (keys %$vals > 1) { |
1267 | $res->{$c} = UNRESOLVABLE_CONDITION; |
1268 | } |
5f11e54f |
1269 | } |
8d005ad9 |
1270 | |
8e40a627 |
1271 | $res; |
c0748280 |
1272 | } |
bac6c4fb |
1273 | |
d28bb90d |
1274 | 1; |