Commit | Line | Data |
c443438f |
1 | package #hide from PAUSE |
2 | DBIx::Class::Storage::DBIHacks; |
d28bb90d |
3 | |
4 | # |
5 | # This module contains code that should never have seen the light of day, |
6 | # does not belong in the Storage, or is otherwise unfit for public |
6a6394f1 |
7 | # display. The arrival of SQLA2 should immediately obsolete 90% of this |
d28bb90d |
8 | # |
9 | |
10 | use strict; |
11 | use warnings; |
12 | |
13 | use base 'DBIx::Class::Storage'; |
14 | use mro 'c3'; |
15 | |
6298a324 |
16 | use List::Util 'first'; |
17 | use Scalar::Util 'blessed'; |
b34d9331 |
18 | use DBIx::Class::_Util qw(UNRESOLVABLE_CONDITION serialize); |
b5ce6748 |
19 | use SQL::Abstract qw(is_plain_value is_literal_value); |
e466c62b |
20 | use DBIx::Class::Carp; |
6298a324 |
21 | use namespace::clean; |
d28bb90d |
22 | |
23 | # |
052e8431 |
24 | # This code will remove non-selecting/non-restricting joins from |
4b1b5ea3 |
25 | # {from} specs, aiding the RDBMS query optimizer |
052e8431 |
26 | # |
27 | sub _prune_unused_joins { |
e1861c2c |
28 | my ($self, $attrs) = @_; |
ea95892e |
29 | |
e1861c2c |
30 | # only standard {from} specs are supported, and we could be disabled in general |
31 | return ($attrs->{from}, {}) unless ( |
32 | ref $attrs->{from} eq 'ARRAY' |
33 | and |
34 | @{$attrs->{from}} > 1 |
35 | and |
36 | ref $attrs->{from}[0] eq 'HASH' |
37 | and |
38 | ref $attrs->{from}[1] eq 'ARRAY' |
39 | and |
40 | $self->_use_join_optimizer |
41 | ); |
052e8431 |
42 | |
eb58c082 |
43 | my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs); |
4b1b5ea3 |
44 | |
eb58c082 |
45 | my $new_aliastypes = { %$orig_aliastypes }; |
46 | |
47 | # we will be recreating this entirely |
48 | my @reclassify = 'joining'; |
97e130fa |
49 | |
4b1b5ea3 |
50 | # a grouped set will not be affected by amount of rows. Thus any |
eb58c082 |
51 | # purely multiplicator classifications can go |
52 | # (will be reintroduced below if needed by something else) |
53 | push @reclassify, qw(multiplying premultiplied) |
437a9cfa |
54 | if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by}; |
4b1b5ea3 |
55 | |
eb58c082 |
56 | # nuke what will be recalculated |
57 | delete @{$new_aliastypes}{@reclassify}; |
58 | |
e1861c2c |
59 | my @newfrom = $attrs->{from}[0]; # FROM head is always present |
052e8431 |
60 | |
eb58c082 |
61 | # recalculate what we need once the multipliers are potentially gone |
62 | # ignore premultiplies, since they do not add any value to anything |
a4812caa |
63 | my %need_joins; |
eb58c082 |
64 | for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) { |
a4812caa |
65 | # add all requested aliases |
66 | $need_joins{$_} = 1 for keys %$_; |
67 | |
68 | # add all their parents (as per joinpath which is an AoH { table => alias }) |
97e130fa |
69 | $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; |
a4812caa |
70 | } |
97e130fa |
71 | |
e1861c2c |
72 | for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) { |
539ffe87 |
73 | push @newfrom, $j if ( |
a6ef93cb |
74 | (! defined $j->[0]{-alias}) # legacy crap |
539ffe87 |
75 | || |
76 | $need_joins{$j->[0]{-alias}} |
77 | ); |
052e8431 |
78 | } |
79 | |
eb58c082 |
80 | # we have a new set of joiners - for everything we nuked pull the classification |
81 | # off the original stack |
82 | for my $ctype (@reclassify) { |
83 | $new_aliastypes->{$ctype} = { map |
84 | { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () } |
85 | keys %{$orig_aliastypes->{$ctype}} |
86 | } |
87 | } |
88 | |
89 | return ( \@newfrom, $new_aliastypes ); |
052e8431 |
90 | } |
91 | |
052e8431 |
92 | # |
d28bb90d |
93 | # This is the code producing joined subqueries like: |
8273e845 |
94 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
d28bb90d |
95 | # |
96 | sub _adjust_select_args_for_complex_prefetch { |
e1861c2c |
97 | my ($self, $attrs) = @_; |
d28bb90d |
98 | |
e1861c2c |
99 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless ( |
100 | ref $attrs->{from} eq 'ARRAY' |
101 | and |
102 | @{$attrs->{from}} > 1 |
103 | and |
104 | ref $attrs->{from}[0] eq 'HASH' |
105 | and |
106 | ref $attrs->{from}[1] eq 'ARRAY' |
107 | ); |
d28bb90d |
108 | |
1e4f9fb3 |
109 | my $root_alias = $attrs->{alias}; |
110 | |
d28bb90d |
111 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
112 | my $outer_attrs = { %$attrs }; |
e1861c2c |
113 | delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)}; |
d28bb90d |
114 | |
6aa93928 |
115 | my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 }; |
116 | delete @{$inner_attrs}{qw(for collapse select as)}; |
d28bb90d |
117 | |
4df1400e |
118 | # there is no point of ordering the insides if there is no limit |
119 | delete $inner_attrs->{order_by} if ( |
120 | delete $inner_attrs->{_order_is_artificial} |
121 | or |
122 | ! $inner_attrs->{rows} |
123 | ); |
946f6260 |
124 | |
d28bb90d |
125 | # generate the inner/outer select lists |
126 | # for inside we consider only stuff *not* brought in by the prefetch |
127 | # on the outside we substitute any function for its alias |
e1861c2c |
128 | $outer_attrs->{select} = [ @{$attrs->{select}} ]; |
36fd7f07 |
129 | |
97e130fa |
130 | my ($root_node, $root_node_offset); |
27e0370d |
131 | |
e1861c2c |
132 | for my $i (0 .. $#{$inner_attrs->{from}}) { |
133 | my $node = $inner_attrs->{from}[$i]; |
27e0370d |
134 | my $h = (ref $node eq 'HASH') ? $node |
135 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
136 | : next |
137 | ; |
138 | |
1e4f9fb3 |
139 | if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) { |
97e130fa |
140 | $root_node = $h; |
141 | $root_node_offset = $i; |
27e0370d |
142 | last; |
143 | } |
144 | } |
145 | |
146 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
147 | unless $root_node; |
27e0370d |
148 | |
149 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
e1861c2c |
150 | my $colinfo = $self->_resolve_column_info($inner_attrs->{from}); |
27e0370d |
151 | my $selected_root_columns; |
152 | |
e1861c2c |
153 | for my $i (0 .. $#{$outer_attrs->{select}}) { |
154 | my $sel = $outer_attrs->{select}->[$i]; |
d28bb90d |
155 | |
1e4f9fb3 |
156 | next if ( |
157 | $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias |
158 | ); |
159 | |
d28bb90d |
160 | if (ref $sel eq 'HASH' ) { |
161 | $sel->{-as} ||= $attrs->{as}[$i]; |
e1861c2c |
162 | $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); |
d28bb90d |
163 | } |
27e0370d |
164 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
165 | $selected_root_columns->{$ci->{-colname}} = 1; |
166 | } |
d28bb90d |
167 | |
e1861c2c |
168 | push @{$inner_attrs->{select}}, $sel; |
bb9bffea |
169 | |
170 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
171 | } |
172 | |
97e130fa |
173 | # We will need to fetch all native columns in the inner subquery, which may |
174 | # be a part of an *outer* join condition, or an order_by (which needs to be |
e1861c2c |
175 | # preserved outside), or wheres. In other words everything but the inner |
176 | # selector |
97e130fa |
177 | # We can not just fetch everything because a potential has_many restricting |
178 | # join collapse *will not work* on heavy data types. |
e1861c2c |
179 | my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({ |
180 | %$inner_attrs, |
181 | select => [], |
182 | }); |
97e130fa |
183 | |
184 | for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) { |
185 | my $ci = $colinfo->{$_} or next; |
186 | if ( |
1e4f9fb3 |
187 | $ci->{-source_alias} eq $root_alias |
97e130fa |
188 | and |
189 | ! $selected_root_columns->{$ci->{-colname}}++ |
190 | ) { |
191 | # adding it to both to keep limits not supporting dark selectors happy |
e1861c2c |
192 | push @{$inner_attrs->{select}}, $ci->{-fq_colname}; |
97e130fa |
193 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
194 | } |
195 | } |
196 | |
e1861c2c |
197 | # construct the inner {from} and lock it in a subquery |
48580715 |
198 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
199 | # throw away all non-selecting, non-restricting multijoins |
eb58c082 |
200 | # (since we def. do not care about multiplication of the contents of the subquery) |
6395604e |
201 | my $inner_subq = do { |
ea95892e |
202 | |
eb58c082 |
203 | # must use it here regardless of user requests (vastly gentler on optimizer) |
ea95892e |
204 | local $self->{_use_join_optimizer} = 1; |
205 | |
97e130fa |
206 | # throw away multijoins since we def. do not care about those inside the subquery |
e1861c2c |
207 | ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({ |
437a9cfa |
208 | %$inner_attrs, _force_prune_multiplying_joins => 1 |
209 | }); |
ea95892e |
210 | |
eb58c082 |
211 | # uh-oh a multiplier (which is not us) left in, this is a problem for limits |
212 | # we will need to add a group_by to collapse the resultset for proper counts |
0a3441ee |
213 | if ( |
eb58c082 |
214 | grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} } |
1e4f9fb3 |
215 | and |
560978e2 |
216 | # if there are user-supplied groups - assume user knows wtf they are up to |
217 | ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} ) |
0a3441ee |
218 | ) { |
1e4f9fb3 |
219 | |
eb58c082 |
220 | my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} }; |
1e4f9fb3 |
221 | |
eb58c082 |
222 | # *possibly* supplement the main selection with pks if not already |
223 | # there, as they will have to be a part of the group_by to collapse |
224 | # things properly |
225 | my $inner_select_with_extras; |
226 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
227 | or $self->throw_exception( sprintf |
228 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
229 | $root_node->{-rsrc}->source_name, |
e1861c2c |
230 | ); |
eb58c082 |
231 | for my $col (@pks) { |
232 | push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col |
233 | unless $cur_sel->{$col}++; |
1e4f9fb3 |
234 | } |
eb58c082 |
235 | |
236 | ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({ |
237 | %$inner_attrs, |
238 | $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (), |
239 | _aliastypes => $inner_aliastypes, |
240 | }); |
0a3441ee |
241 | } |
d28bb90d |
242 | |
e1861c2c |
243 | # we already optimized $inner_attrs->{from} above |
97e130fa |
244 | # and already local()ized |
245 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
246 | |
ea95892e |
247 | # generate the subquery |
6395604e |
248 | $self->_select_args_to_query ( |
e1861c2c |
249 | @{$inner_attrs}{qw(from select where)}, |
ea95892e |
250 | $inner_attrs, |
251 | ); |
d28bb90d |
252 | }; |
253 | |
254 | # Generate the outer from - this is relatively easy (really just replace |
255 | # the join slot with the subquery), with a major caveat - we can not |
256 | # join anything that is non-selecting (not part of the prefetch), but at |
257 | # the same time is a multi-type relationship, as it will explode the result. |
258 | # |
259 | # There are two possibilities here |
260 | # - either the join is non-restricting, in which case we simply throw it away |
261 | # - it is part of the restrictions, in which case we need to collapse the outer |
262 | # result by tackling yet another group_by to the outside of the query |
263 | |
27e0370d |
264 | # work on a shallow copy |
e1861c2c |
265 | my @orig_from = @{$attrs->{from}}; |
266 | |
052e8431 |
267 | |
e1861c2c |
268 | $outer_attrs->{from} = \ my @outer_from; |
53c29913 |
269 | |
27e0370d |
270 | # we may not be the head |
97e130fa |
271 | if ($root_node_offset) { |
e1861c2c |
272 | # first generate the outer_from, up to the substitution point |
273 | @outer_from = splice @orig_from, 0, $root_node_offset; |
27e0370d |
274 | |
e1861c2c |
275 | # substitute the subq at the right spot |
27e0370d |
276 | push @outer_from, [ |
277 | { |
1e4f9fb3 |
278 | -alias => $root_alias, |
97e130fa |
279 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
280 | $root_alias => $inner_subq, |
27e0370d |
281 | }, |
e1861c2c |
282 | # preserve attrs from what is now the head of the from after the splice |
283 | @{$orig_from[0]}[1 .. $#{$orig_from[0]}], |
27e0370d |
284 | ]; |
285 | } |
286 | else { |
27e0370d |
287 | @outer_from = { |
1e4f9fb3 |
288 | -alias => $root_alias, |
27e0370d |
289 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
290 | $root_alias => $inner_subq, |
27e0370d |
291 | }; |
d28bb90d |
292 | } |
293 | |
e1861c2c |
294 | shift @orig_from; # what we just replaced above |
97e130fa |
295 | |
ea95892e |
296 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
297 | # in what role |
975b573a |
298 | my $outer_aliastypes = $outer_attrs->{_aliastypes} = |
e1861c2c |
299 | $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from }); |
052e8431 |
300 | |
a4812caa |
301 | # unroll parents |
1e4f9fb3 |
302 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
303 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
304 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
305 | |
d28bb90d |
306 | # see what's left - throw away if not selecting/restricting |
eb58c082 |
307 | my $may_need_outer_group_by; |
e1861c2c |
308 | while (my $j = shift @orig_from) { |
d28bb90d |
309 | my $alias = $j->[0]{-alias}; |
310 | |
a4812caa |
311 | if ( |
312 | $outer_select_chain->{$alias} |
313 | ) { |
314 | push @outer_from, $j |
d28bb90d |
315 | } |
1e4f9fb3 |
316 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
317 | push @outer_from, $j; |
eb58c082 |
318 | $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
319 | } |
320 | } |
321 | |
eb58c082 |
322 | # also throw in a synthetic group_by if a non-selecting multiplier, |
323 | # to guard against cross-join explosions |
324 | # the logic is somewhat fragile, but relies on the idea that if a user supplied |
325 | # a group by on their own - they know what they were doing |
326 | if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
327 | ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({ |
560978e2 |
328 | %$outer_attrs, |
329 | from => \@outer_from, |
560978e2 |
330 | }); |
36fd7f07 |
331 | } |
332 | |
e1861c2c |
333 | # This is totally horrific - the {where} ends up in both the inner and outer query |
d28bb90d |
334 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
335 | # then if where conditions apply to the *right* side of the prefetch, you may have |
336 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
4a0eed52 |
337 | # the outer select to exclude joins you didn't want in the first place |
d28bb90d |
338 | # |
339 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
e1861c2c |
340 | return $outer_attrs; |
d28bb90d |
341 | } |
342 | |
1a736efb |
343 | # |
344 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
345 | # |
ad630f4b |
346 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
347 | # select/where/order/group attributes, in order to determine what |
4a0eed52 |
348 | # aliases are needed to fulfill the query. This information is used |
ad630f4b |
349 | # throughout the code to prune unnecessary JOINs from the queries |
350 | # in an attempt to reduce the execution time. |
351 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
352 | # happen is for it to fail due to some scalar SQL, which in turn will |
353 | # result in a vocal exception. |
539ffe87 |
354 | sub _resolve_aliastypes_from_select_args { |
e1861c2c |
355 | my ( $self, $attrs ) = @_; |
546f1cd9 |
356 | |
ad630f4b |
357 | $self->throw_exception ('Unable to analyze custom {from}') |
e1861c2c |
358 | if ref $attrs->{from} ne 'ARRAY'; |
546f1cd9 |
359 | |
ad630f4b |
360 | # what we will return |
964a3c71 |
361 | my $aliases_by_type; |
546f1cd9 |
362 | |
ad630f4b |
363 | # see what aliases are there to work with |
eb58c082 |
364 | # and record who is a multiplier and who is premultiplied |
ad630f4b |
365 | my $alias_list; |
e1861c2c |
366 | for my $node (@{$attrs->{from}}) { |
367 | |
368 | my $j = $node; |
ad630f4b |
369 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
370 | my $al = $j->{-alias} |
371 | or next; |
372 | |
373 | $alias_list->{$al} = $j; |
eb58c082 |
374 | |
375 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } |
a4812caa |
376 | # not array == {from} head == can't be multiplying |
eb58c082 |
377 | if ref($node) eq 'ARRAY' and ! $j->{-is_single}; |
378 | |
379 | $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] } |
380 | # parts of the path that are not us but are multiplying |
381 | if grep { $aliases_by_type->{multiplying}{$_} } |
382 | grep { $_ ne $al } |
383 | map { values %$_ } |
384 | @{ $j->{-join_path}||[] } |
546f1cd9 |
385 | } |
546f1cd9 |
386 | |
318e3d94 |
387 | # get a column to source/alias map (including unambiguous unqualified ones) |
e1861c2c |
388 | my $colinfo = $self->_resolve_column_info ($attrs->{from}); |
1a736efb |
389 | |
ad630f4b |
390 | # set up a botched SQLA |
391 | my $sql_maker = $self->sql_maker; |
07f31d19 |
392 | |
4c2b30d6 |
393 | # these are throw away results, do not pollute the bind stack |
0542ec57 |
394 | local $sql_maker->{where_bind}; |
395 | local $sql_maker->{group_bind}; |
396 | local $sql_maker->{having_bind}; |
97e130fa |
397 | local $sql_maker->{from_bind}; |
3f5b99fe |
398 | |
399 | # we can't scan properly without any quoting (\b doesn't cut it |
400 | # everywhere), so unless there is proper quoting set - use our |
401 | # own weird impossible character. |
402 | # Also in the case of no quoting, we need to explicitly disable |
403 | # name_sep, otherwise sorry nasty legacy syntax like |
404 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
405 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
406 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
407 | |
408 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
409 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
410 | # if we don't unset it we screw up retarded but unfortunately working |
411 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
412 | $sql_maker->{name_sep} = ''; |
413 | } |
414 | |
415 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
416 | |
1a736efb |
417 | # generate sql chunks |
418 | my $to_scan = { |
419 | restricting => [ |
a9e985b7 |
420 | ($sql_maker->_recurse_where ($attrs->{where}))[0], |
1e4f9fb3 |
421 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
422 | ], |
423 | grouping => [ |
424 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
425 | ], |
97e130fa |
426 | joining => [ |
427 | $sql_maker->_recurse_from ( |
e1861c2c |
428 | ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0], |
429 | @{$attrs->{from}}[1 .. $#{$attrs->{from}}], |
97e130fa |
430 | ), |
431 | ], |
1a736efb |
432 | selecting => [ |
fdd47fe8 |
433 | # kill all selectors which look like a proper subquery |
434 | # this is a sucky heuristic *BUT* - if we get it wrong the query will simply |
435 | # fail to run, so we are relatively safe |
436 | grep |
437 | { $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi } |
438 | map |
439 | { ($sql_maker->_recurse_fields($_))[0] } |
440 | @{$attrs->{select}} |
1e4f9fb3 |
441 | ], |
66bbb12c |
442 | ordering => [ map |
443 | { |
444 | ( my $sql = (ref $_ ? $_->[0] : $_) ) =~ s/ \s+ (?: ASC | DESC ) \s* \z //xi; |
445 | $sql; |
446 | } |
447 | $sql_maker->_order_by_chunks( $attrs->{order_by} ), |
1a736efb |
448 | ], |
449 | }; |
07f31d19 |
450 | |
fdd47fe8 |
451 | # throw away empty-string chunks, and make sure no binds snuck in |
452 | # note that we operate over @{$to_scan->{$type}}, hence the |
453 | # semi-mindbending ... map ... for values ... |
454 | ( $_ = [ map { |
0dadd60d |
455 | |
fdd47fe8 |
456 | (not $_) ? () |
457 | : (length ref $_) ? (require Data::Dumper::Concise && $self->throw_exception( |
458 | "Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($_) |
459 | )) |
460 | : $_ |
0dadd60d |
461 | |
fdd47fe8 |
462 | } @$_ ] ) for values %$to_scan; |
463 | |
464 | # throw away empty to-scan's |
465 | ( |
466 | @{$to_scan->{$_}} |
467 | or |
468 | delete $to_scan->{$_} |
469 | ) for keys %$to_scan; |
0dadd60d |
470 | |
07f31d19 |
471 | |
19955cdf |
472 | # the actual scan, per type |
318e3d94 |
473 | for my $type (keys %$to_scan) { |
19955cdf |
474 | |
475 | # first see if we have any exact matches (qualified or unqualified) |
318e3d94 |
476 | for my $piece (@{$to_scan->{$type}}) { |
477 | if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) { |
478 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
479 | $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece; |
480 | } |
481 | } |
318e3d94 |
482 | |
753fd57a |
483 | # we will be bulk-scanning anyway - pieces will not matter in that case |
484 | # (unlike in the direct-equivalence above) |
485 | my $scan_string = join ' ', @{$to_scan->{$type}}; |
486 | |
19955cdf |
487 | # now loop through all fully qualified columns and get the corresponding |
488 | # alias (should work even if they are in scalarrefs) |
489 | for my $alias (keys %$alias_list) { |
490 | my $al_re = qr/ |
491 | $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
492 | | |
493 | \b $alias \. ([^\s\)\($rquote]+)? |
494 | /x; |
1a736efb |
495 | |
753fd57a |
496 | if (my @matches = $scan_string =~ /$al_re/g) { |
497 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
498 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_" |
499 | for grep { defined $_ } @matches; |
1a736efb |
500 | } |
ad630f4b |
501 | } |
1a736efb |
502 | |
19955cdf |
503 | # now loop through unqualified column names, and try to locate them within |
504 | # the chunks |
505 | for my $col (keys %$colinfo) { |
506 | next if $col =~ / \. /x; # if column is qualified it was caught by the above |
1a736efb |
507 | |
19955cdf |
508 | my $col_re = qr/ $lquote ($col) $rquote /x; |
07f31d19 |
509 | |
753fd57a |
510 | if ( my @matches = $scan_string =~ /$col_re/g) { |
511 | my $alias = $colinfo->{$col}{-source_alias}; |
512 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
513 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
514 | for grep { defined $_ } @matches; |
1a736efb |
515 | } |
07f31d19 |
516 | } |
517 | } |
518 | |
519 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
19955cdf |
520 | ( |
521 | $_->{-alias} |
522 | and |
523 | ! $aliases_by_type->{restricting}{ $_->{-alias} } |
524 | and |
525 | ( |
526 | not $_->{-join_type} |
07f31d19 |
527 | or |
19955cdf |
528 | $_->{-join_type} !~ /^left (?: \s+ outer)? $/xi |
529 | ) |
530 | and |
531 | $aliases_by_type->{restricting}{ $_->{-alias} } = { -parents => $_->{-join_path}||[] } |
532 | ) for values %$alias_list; |
07f31d19 |
533 | |
19955cdf |
534 | # final cleanup |
535 | ( |
536 | keys %{$aliases_by_type->{$_}} |
537 | or |
538 | delete $aliases_by_type->{$_} |
539 | ) for keys %$aliases_by_type; |
1e4f9fb3 |
540 | |
19955cdf |
541 | $aliases_by_type; |
07f31d19 |
542 | } |
543 | |
eb58c082 |
544 | # This is the engine behind { distinct => 1 } and the general |
545 | # complex prefetch grouper |
0a3441ee |
546 | sub _group_over_selection { |
560978e2 |
547 | my ($self, $attrs) = @_; |
0a3441ee |
548 | |
560978e2 |
549 | my $colinfos = $self->_resolve_column_info ($attrs->{from}); |
0a3441ee |
550 | |
551 | my (@group_by, %group_index); |
552 | |
36fd7f07 |
553 | # the logic is: if it is a { func => val } we assume an aggregate, |
554 | # otherwise if \'...' or \[...] we assume the user knows what is |
555 | # going on thus group over it |
560978e2 |
556 | for (@{$attrs->{select}}) { |
0a3441ee |
557 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
558 | push @group_by, $_; |
559 | $group_index{$_}++; |
560978e2 |
560 | if ($colinfos->{$_} and $_ !~ /\./ ) { |
0a3441ee |
561 | # add a fully qualified version as well |
560978e2 |
562 | $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; |
0a3441ee |
563 | } |
07f31d19 |
564 | } |
565 | } |
ad630f4b |
566 | |
eb58c082 |
567 | my @order_by = $self->_extract_order_criteria($attrs->{order_by}) |
568 | or return (\@group_by, $attrs->{order_by}); |
569 | |
570 | # add any order_by parts that are not already present in the group_by |
571 | # to maintain SQL cross-compatibility and general sanity |
572 | # |
573 | # also in case the original selection is *not* unique, or in case part |
574 | # of the ORDER BY refers to a multiplier - we will need to replace the |
575 | # skipped order_by elements with their MIN/MAX equivalents as to maintain |
576 | # the proper overall order without polluting the group criteria (and |
577 | # possibly changing the outcome entirely) |
578 | |
579 | my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes); |
580 | |
581 | my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by); |
582 | |
583 | for my $o_idx (0 .. $#order_by) { |
584 | |
585 | # if the chunk is already a min/max function - there is nothing left to touch |
586 | next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix; |
587 | |
0a3441ee |
588 | # only consider real columns (for functions the user got to do an explicit group_by) |
eb58c082 |
589 | my $chunk_ci; |
590 | if ( |
591 | @{$order_by[$o_idx]} != 1 |
592 | or |
593 | # only declare an unknown *plain* identifier as "leftover" if we are called with |
594 | # aliastypes to examine. If there are none - we are still in _resolve_attrs, and |
595 | # can just assume the user knows what they want |
596 | ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} ) |
597 | ) { |
598 | push @$leftovers, $order_by[$o_idx][0]; |
14e26c5f |
599 | } |
560978e2 |
600 | |
eb58c082 |
601 | next unless $chunk_ci; |
602 | |
603 | # no duplication of group criteria |
604 | next if $group_index{$chunk_ci->{-fq_colname}}; |
605 | |
606 | $aliastypes ||= ( |
607 | $attrs->{_aliastypes} |
560978e2 |
608 | or |
eb58c082 |
609 | $self->_resolve_aliastypes_from_select_args({ |
610 | from => $attrs->{from}, |
611 | order_by => $attrs->{order_by}, |
612 | }) |
613 | ) if $group_already_unique; |
614 | |
615 | # check that we are not ordering by a multiplier (if a check is requested at all) |
616 | if ( |
617 | $group_already_unique |
618 | and |
619 | ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}} |
620 | and |
621 | ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}} |
560978e2 |
622 | ) { |
eb58c082 |
623 | push @group_by, $chunk_ci->{-fq_colname}; |
624 | $group_index{$chunk_ci->{-fq_colname}}++ |
560978e2 |
625 | } |
eb58c082 |
626 | else { |
627 | # We need to order by external columns without adding them to the group |
628 | # (eiehter a non-unique selection, or a multi-external) |
629 | # |
630 | # This doesn't really make sense in SQL, however from DBICs point |
631 | # of view is rather valid (e.g. order the leftmost objects by whatever |
632 | # criteria and get the offset/rows many). There is a way around |
633 | # this however in SQL - we simply tae the direction of each piece |
634 | # of the external order and convert them to MIN(X) for ASC or MAX(X) |
635 | # for DESC, and group_by the root columns. The end result should be |
636 | # exactly what we expect |
637 | |
638 | # FIXME - this code is a joke, will need to be completely rewritten in |
639 | # the DQ branch. But I need to push a POC here, otherwise the |
640 | # pesky tests won't pass |
641 | # wrap any part of the order_by that "responds" to an ordering alias |
642 | # into a MIN/MAX |
643 | $sql_maker ||= $self->sql_maker; |
644 | $order_chunks ||= [ |
645 | map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}) |
646 | ]; |
0a3441ee |
647 | |
eb58c082 |
648 | my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]); |
649 | |
650 | $new_order_by[$o_idx] = \[ |
651 | sprintf( '%s( %s )%s', |
652 | ($is_desc ? 'MAX' : 'MIN'), |
653 | $chunk, |
654 | ($is_desc ? ' DESC' : ''), |
655 | ), |
656 | @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ] |
657 | ]; |
658 | } |
0a3441ee |
659 | } |
660 | |
eb58c082 |
661 | $self->throw_exception ( sprintf |
9736be65 |
662 | 'Unable to programatically derive a required group_by from the supplied ' |
663 | . 'order_by criteria. To proceed either add an explicit group_by, or ' |
664 | . 'simplify your order_by to only include plain columns ' |
665 | . '(supplied order_by: %s)', |
eb58c082 |
666 | join ', ', map { "'$_'" } @$leftovers, |
667 | ) if $leftovers; |
668 | |
669 | # recreate the untouched order parts |
670 | if (@new_order_by) { |
671 | $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks ); |
672 | } |
673 | |
674 | return ( |
675 | \@group_by, |
676 | (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged |
677 | ); |
07f31d19 |
678 | } |
679 | |
d28bb90d |
680 | sub _resolve_ident_sources { |
681 | my ($self, $ident) = @_; |
682 | |
683 | my $alias2source = {}; |
d28bb90d |
684 | |
685 | # the reason this is so contrived is that $ident may be a {from} |
686 | # structure, specifying multiple tables to join |
6298a324 |
687 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
688 | # this is compat mode for insert/update/delete which do not deal with aliases |
689 | $alias2source->{me} = $ident; |
d28bb90d |
690 | } |
691 | elsif (ref $ident eq 'ARRAY') { |
692 | |
693 | for (@$ident) { |
694 | my $tabinfo; |
695 | if (ref $_ eq 'HASH') { |
696 | $tabinfo = $_; |
d28bb90d |
697 | } |
698 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
699 | $tabinfo = $_->[0]; |
700 | } |
701 | |
4376a157 |
702 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
703 | if ($tabinfo->{-rsrc}); |
d28bb90d |
704 | } |
705 | } |
706 | |
90f10b5a |
707 | return $alias2source; |
d28bb90d |
708 | } |
709 | |
710 | # Takes $ident, \@column_names |
711 | # |
712 | # returns { $column_name => \%column_info, ... } |
713 | # also note: this adds -result_source => $rsrc to the column info |
714 | # |
09e14fdc |
715 | # If no columns_names are supplied returns info about *all* columns |
716 | # for all sources |
d28bb90d |
717 | sub _resolve_column_info { |
718 | my ($self, $ident, $colnames) = @_; |
8d005ad9 |
719 | |
720 | return {} if $colnames and ! @$colnames; |
721 | |
229401a0 |
722 | my $sources = $self->_resolve_ident_sources($ident); |
723 | |
724 | $_ = { rsrc => $_, colinfos => $_->columns_info } |
725 | for values %$sources; |
d28bb90d |
726 | |
52416317 |
727 | my (%seen_cols, @auto_colnames); |
d28bb90d |
728 | |
729 | # compile a global list of column names, to be able to properly |
730 | # disambiguate unqualified column names (if at all possible) |
229401a0 |
731 | for my $alias (keys %$sources) { |
732 | ( |
733 | ++$seen_cols{$_}{$alias} |
734 | and |
735 | ! $colnames |
736 | and |
737 | push @auto_colnames, "$alias.$_" |
738 | ) for keys %{ $sources->{$alias}{colinfos} }; |
d28bb90d |
739 | } |
740 | |
09e14fdc |
741 | $colnames ||= [ |
742 | @auto_colnames, |
229401a0 |
743 | ( grep { keys %{$seen_cols{$_}} == 1 } keys %seen_cols ), |
09e14fdc |
744 | ]; |
745 | |
229401a0 |
746 | my %return; |
747 | for (@$colnames) { |
748 | my ($colname, $source_alias) = reverse split /\./, $_; |
d28bb90d |
749 | |
229401a0 |
750 | my $assumed_alias = |
751 | $source_alias |
752 | || |
753 | # if the column was seen exactly once - we know which rsrc it came from |
754 | ( |
755 | $seen_cols{$colname} |
756 | and |
757 | keys %{$seen_cols{$colname}} == 1 |
758 | and |
759 | ( %{$seen_cols{$colname}} )[0] |
760 | ) |
761 | || |
762 | next |
763 | ; |
52416317 |
764 | |
229401a0 |
765 | $self->throw_exception( |
766 | "No such column '$colname' on source " . $sources->{$assumed_alias}{rsrc}->source_name |
767 | ) unless $seen_cols{$colname}{$assumed_alias}; |
52416317 |
768 | |
229401a0 |
769 | $return{$_} = { |
770 | %{ $sources->{$assumed_alias}{colinfos}{$colname} }, |
771 | -result_source => $sources->{$assumed_alias}{rsrc}, |
772 | -source_alias => $assumed_alias, |
773 | -fq_colname => "$assumed_alias.$colname", |
81bf295c |
774 | -colname => $colname, |
d28bb90d |
775 | }; |
81bf295c |
776 | |
229401a0 |
777 | $return{"$assumed_alias.$colname"} = $return{$_} |
778 | unless $source_alias; |
d28bb90d |
779 | } |
780 | |
781 | return \%return; |
782 | } |
783 | |
289ac713 |
784 | # The DBIC relationship chaining implementation is pretty simple - every |
785 | # new related_relationship is pushed onto the {from} stack, and the {select} |
786 | # window simply slides further in. This means that when we count somewhere |
787 | # in the middle, we got to make sure that everything in the join chain is an |
788 | # actual inner join, otherwise the count will come back with unpredictable |
789 | # results (a resultset may be generated with _some_ rows regardless of if |
790 | # the relation which the $rs currently selects has rows or not). E.g. |
791 | # $artist_rs->cds->count - normally generates: |
792 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
793 | # which actually returns the number of artists * (number of cds || 1) |
794 | # |
795 | # So what we do here is crawl {from}, determine if the current alias is at |
796 | # the top of the stack, and if not - make sure the chain is inner-joined down |
797 | # to the root. |
798 | # |
31a8aaaf |
799 | sub _inner_join_to_node { |
289ac713 |
800 | my ($self, $from, $alias) = @_; |
801 | |
302d35f8 |
802 | my $switch_branch = $self->_find_join_path_to_node($from, $alias); |
289ac713 |
803 | |
302d35f8 |
804 | return $from unless @{$switch_branch||[]}; |
289ac713 |
805 | |
806 | # So it looks like we will have to switch some stuff around. |
807 | # local() is useless here as we will be leaving the scope |
808 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
809 | # So replace the first hashref in the node arrayref manually |
289ac713 |
810 | my @new_from = ($from->[0]); |
faeb2407 |
811 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
812 | |
813 | for my $j (@{$from}[1 .. $#$from]) { |
814 | my $jalias = $j->[0]{-alias}; |
815 | |
816 | if ($sw_idx->{$jalias}) { |
817 | my %attrs = %{$j->[0]}; |
818 | delete $attrs{-join_type}; |
819 | push @new_from, [ |
820 | \%attrs, |
821 | @{$j}[ 1 .. $#$j ], |
822 | ]; |
823 | } |
824 | else { |
825 | push @new_from, $j; |
826 | } |
827 | } |
828 | |
829 | return \@new_from; |
830 | } |
831 | |
302d35f8 |
832 | sub _find_join_path_to_node { |
833 | my ($self, $from, $target_alias) = @_; |
834 | |
835 | # subqueries and other oddness are naturally not supported |
836 | return undef if ( |
837 | ref $from ne 'ARRAY' |
838 | || |
839 | ref $from->[0] ne 'HASH' |
840 | || |
841 | ! defined $from->[0]{-alias} |
842 | ); |
843 | |
844 | # no path - the head is the alias |
845 | return [] if $from->[0]{-alias} eq $target_alias; |
846 | |
847 | for my $i (1 .. $#$from) { |
848 | return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias ); |
849 | } |
850 | |
851 | # something else went quite wrong |
852 | return undef; |
853 | } |
854 | |
bac358c9 |
855 | sub _extract_order_criteria { |
1a736efb |
856 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
857 | |
1a736efb |
858 | my $parser = sub { |
e6977bbb |
859 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
860 | |
1a736efb |
861 | return scalar $sql_maker->_order_by_chunks ($order_by) |
862 | unless wantarray; |
c0748280 |
863 | |
e6977bbb |
864 | my ($lq, $rq, $sep) = map { quotemeta($_) } ( |
865 | ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars), |
866 | $sql_maker->name_sep |
867 | ); |
868 | |
1a736efb |
869 | my @chunks; |
bac358c9 |
870 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
871 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
872 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
873 | |
874 | # order criteria may have come back pre-quoted (literals and whatnot) |
875 | # this is fragile, but the best we can currently do |
876 | $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe |
877 | or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x; |
878 | |
1a736efb |
879 | push @chunks, $chunk; |
bac6c4fb |
880 | } |
1a736efb |
881 | |
882 | return @chunks; |
883 | }; |
884 | |
885 | if ($sql_maker) { |
886 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
887 | } |
888 | else { |
1a736efb |
889 | $sql_maker = $self->sql_maker; |
e6977bbb |
890 | |
891 | # pass these in to deal with literals coming from |
892 | # the user or the deep guts of prefetch |
893 | my $orig_quote_chars = [$sql_maker->_quote_chars]; |
894 | |
1a736efb |
895 | local $sql_maker->{quote_char}; |
e6977bbb |
896 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
897 | } |
bac6c4fb |
898 | } |
899 | |
7cec4356 |
900 | sub _order_by_is_stable { |
5f11e54f |
901 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
902 | |
eb58c082 |
903 | my @cols = ( |
8d005ad9 |
904 | ( map { $_->[0] } $self->_extract_order_criteria($order_by) ), |
8e40a627 |
905 | ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ), |
df4312bc |
906 | ) or return 0; |
eb58c082 |
907 | |
908 | my $colinfo = $self->_resolve_column_info($ident, \@cols); |
909 | |
910 | return keys %$colinfo |
911 | ? $self->_columns_comprise_identifying_set( $colinfo, \@cols ) |
df4312bc |
912 | : 0 |
eb58c082 |
913 | ; |
914 | } |
c0748280 |
915 | |
eb58c082 |
916 | sub _columns_comprise_identifying_set { |
917 | my ($self, $colinfo, $columns) = @_; |
7cec4356 |
918 | |
919 | my $cols_per_src; |
eb58c082 |
920 | $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_ |
921 | for grep { defined $_ } @{$colinfo}{@$columns}; |
7cec4356 |
922 | |
923 | for (values %$cols_per_src) { |
924 | my $src = (values %$_)[0]->{-result_source}; |
925 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
926 | } |
927 | |
df4312bc |
928 | return 0; |
7cec4356 |
929 | } |
930 | |
df4312bc |
931 | # this is almost similar to _order_by_is_stable, except it takes |
0e81e691 |
932 | # a single rsrc, and will succeed only if the first portion of the order |
933 | # by is stable. |
934 | # returns that portion as a colinfo hashref on success |
df4312bc |
935 | sub _extract_colinfo_of_stable_main_source_order_by_portion { |
302d35f8 |
936 | my ($self, $attrs) = @_; |
0e81e691 |
937 | |
302d35f8 |
938 | my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias}); |
939 | |
940 | return unless defined $nodes; |
0e81e691 |
941 | |
942 | my @ord_cols = map |
943 | { $_->[0] } |
302d35f8 |
944 | ( $self->_extract_order_criteria($attrs->{order_by}) ) |
0e81e691 |
945 | ; |
946 | return unless @ord_cols; |
947 | |
302d35f8 |
948 | my $valid_aliases = { map { $_ => 1 } ( |
949 | $attrs->{from}[0]{-alias}, |
950 | map { values %$_ } @$nodes, |
951 | ) }; |
318e3d94 |
952 | |
302d35f8 |
953 | my $colinfos = $self->_resolve_column_info($attrs->{from}); |
954 | |
955 | my ($colinfos_to_return, $seen_main_src_cols); |
956 | |
957 | for my $col (@ord_cols) { |
958 | # if order criteria is unresolvable - there is nothing we can do |
959 | my $colinfo = $colinfos->{$col} or last; |
960 | |
961 | # if we reached the end of the allowed aliases - also nothing we can do |
962 | last unless $valid_aliases->{$colinfo->{-source_alias}}; |
963 | |
964 | $colinfos_to_return->{$col} = $colinfo; |
965 | |
966 | $seen_main_src_cols->{$colinfo->{-colname}} = 1 |
967 | if $colinfo->{-source_alias} eq $attrs->{alias}; |
0e81e691 |
968 | } |
969 | |
302d35f8 |
970 | # FIXME the condition may be singling out things on its own, so we |
971 | # conceivable could come back wi "stable-ordered by nothing" |
972 | # not confient enough in the parser yet, so punt for the time being |
973 | return unless $seen_main_src_cols; |
0e81e691 |
974 | |
302d35f8 |
975 | my $main_src_fixed_cols_from_cond = [ $attrs->{where} |
976 | ? ( |
977 | map |
978 | { |
979 | ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} ) |
980 | ? $colinfos->{$_}{-colname} |
981 | : () |
982 | } |
8e40a627 |
983 | keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) } |
302d35f8 |
984 | ) |
985 | : () |
986 | ]; |
0e81e691 |
987 | |
302d35f8 |
988 | return $attrs->{result_source}->_identifying_column_set([ |
989 | keys %$seen_main_src_cols, |
990 | @$main_src_fixed_cols_from_cond, |
991 | ]) ? $colinfos_to_return : (); |
0e81e691 |
992 | } |
993 | |
8d005ad9 |
994 | # Attempts to flatten a passed in SQLA condition as much as possible towards |
995 | # a plain hashref, *without* altering its semantics. Required by |
996 | # create/populate being able to extract definitive conditions from preexisting |
997 | # resultset {where} stacks |
998 | # |
999 | # FIXME - while relatively robust, this is still imperfect, one of the first |
1000 | # things to tackle with DQ |
1001 | sub _collapse_cond { |
1002 | my ($self, $where, $where_is_anded_array) = @_; |
1003 | |
135ac69d |
1004 | my $fin; |
1005 | |
8d005ad9 |
1006 | if (! $where) { |
1007 | return; |
1008 | } |
1009 | elsif ($where_is_anded_array or ref $where eq 'HASH') { |
1010 | |
1011 | my @pairs; |
1012 | |
1013 | my @pieces = $where_is_anded_array ? @$where : $where; |
1014 | while (@pieces) { |
1015 | my $chunk = shift @pieces; |
1016 | |
1017 | if (ref $chunk eq 'HASH') { |
e466c62b |
1018 | for (sort keys %$chunk) { |
1019 | |
1020 | # Match SQLA 1.79 behavior |
1021 | if ($_ eq '') { |
1022 | is_literal_value($chunk->{$_}) |
1023 | ? carp 'Hash-pairs consisting of an empty string with a literal are deprecated, use -and => [ $literal ] instead' |
1024 | : $self->throw_exception("Supplying an empty left hand side argument is not supported in hash-pairs") |
1025 | ; |
1026 | } |
1027 | |
1028 | push @pairs, $_ => $chunk->{$_}; |
1029 | } |
8d005ad9 |
1030 | } |
1031 | elsif (ref $chunk eq 'ARRAY') { |
6565d2c3 |
1032 | push @pairs, -or => $chunk |
8d005ad9 |
1033 | if @$chunk; |
1034 | } |
b34d9331 |
1035 | elsif ( ! length ref $chunk) { |
e466c62b |
1036 | |
1037 | # Match SQLA 1.79 behavior |
1038 | $self->throw_exception("Supplying an empty left hand side argument is not supported in array-pairs") |
1039 | if $where_is_anded_array and (! defined $chunk or $chunk eq ''); |
1040 | |
6565d2c3 |
1041 | push @pairs, $chunk, shift @pieces; |
8d005ad9 |
1042 | } |
1043 | else { |
6565d2c3 |
1044 | push @pairs, '', $chunk; |
8d005ad9 |
1045 | } |
1046 | } |
1047 | |
1048 | return unless @pairs; |
1049 | |
1050 | my @conds = $self->_collapse_cond_unroll_pairs(\@pairs) |
1051 | or return; |
1052 | |
1053 | # Consolidate various @conds back into something more compact |
8d005ad9 |
1054 | for my $c (@conds) { |
1055 | if (ref $c ne 'HASH') { |
1056 | push @{$fin->{-and}}, $c; |
1057 | } |
1058 | else { |
1059 | for my $col (sort keys %$c) { |
8d005ad9 |
1060 | |
135ac69d |
1061 | # consolidate all -and nodes |
1062 | if ($col =~ /^\-and$/i) { |
1063 | push @{$fin->{-and}}, |
1064 | ref $c->{$col} eq 'ARRAY' ? @{$c->{$col}} |
1065 | : ref $c->{$col} eq 'HASH' ? %{$c->{$col}} |
1066 | : { $col => $c->{$col} } |
1067 | ; |
1068 | } |
1069 | elsif ($col =~ /^\-/) { |
1070 | push @{$fin->{-and}}, { $col => $c->{$col} }; |
1071 | } |
1072 | elsif (exists $fin->{$col}) { |
1073 | $fin->{$col} = [ -and => map { |
1074 | (ref $_ eq 'ARRAY' and ($_->[0]||'') =~ /^\-and$/i ) |
1075 | ? @{$_}[1..$#$_] |
1076 | : $_ |
1077 | ; |
1078 | } ($fin->{$col}, $c->{$col}) ]; |
8d005ad9 |
1079 | } |
1080 | else { |
1081 | $fin->{$col} = $c->{$col}; |
1082 | } |
1083 | } |
1084 | } |
1085 | } |
8d005ad9 |
1086 | } |
1087 | elsif (ref $where eq 'ARRAY') { |
22485a7e |
1088 | # we are always at top-level here, it is safe to dump empty *standalone* pieces |
1089 | my $fin_idx; |
8d005ad9 |
1090 | |
22485a7e |
1091 | for (my $i = 0; $i <= $#$where; $i++ ) { |
8d005ad9 |
1092 | |
e466c62b |
1093 | # Match SQLA 1.79 behavior |
1094 | $self->throw_exception( |
1095 | "Supplying an empty left hand side argument is not supported in array-pairs" |
1096 | ) if (! defined $where->[$i] or ! length $where->[$i]); |
1097 | |
22485a7e |
1098 | my $logic_mod = lc ( ($where->[$i] =~ /^(\-(?:and|or))$/i)[0] || '' ); |
1099 | |
1100 | if ($logic_mod) { |
1101 | $i++; |
1102 | $self->throw_exception("Unsupported top-level op/arg pair: [ $logic_mod => $where->[$i] ]") |
1103 | unless ref $where->[$i] eq 'HASH' or ref $where->[$i] eq 'ARRAY'; |
1104 | |
1105 | my $sub_elt = $self->_collapse_cond({ $logic_mod => $where->[$i] }) |
1106 | or next; |
1107 | |
e466c62b |
1108 | my @keys = keys %$sub_elt; |
1109 | if ( @keys == 1 and $keys[0] !~ /^\-/ ) { |
1110 | $fin_idx->{ "COL_$keys[0]_" . serialize $sub_elt } = $sub_elt; |
1111 | } |
1112 | else { |
1113 | $fin_idx->{ "SER_" . serialize $sub_elt } = $sub_elt; |
1114 | } |
22485a7e |
1115 | } |
1116 | elsif (! length ref $where->[$i] ) { |
135ac69d |
1117 | my $sub_elt = $self->_collapse_cond({ @{$where}[$i, $i+1] }) |
1118 | or next; |
1119 | |
1120 | $fin_idx->{ "COL_$where->[$i]_" . serialize $sub_elt } = $sub_elt; |
22485a7e |
1121 | $i++; |
8d005ad9 |
1122 | } |
1123 | else { |
135ac69d |
1124 | $fin_idx->{ "SER_" . serialize $where->[$i] } = $self->_collapse_cond( $where->[$i] ) || next; |
8d005ad9 |
1125 | } |
1126 | } |
22485a7e |
1127 | |
07add744 |
1128 | if (! $fin_idx) { |
1129 | return; |
1130 | } |
1131 | elsif ( keys %$fin_idx == 1 ) { |
1132 | $fin = (values %$fin_idx)[0]; |
1133 | } |
1134 | else { |
1135 | my @or; |
1136 | |
1137 | # at this point everything is at most one level deep - unroll if needed |
1138 | for (sort keys %$fin_idx) { |
1139 | if ( ref $fin_idx->{$_} eq 'HASH' and keys %{$fin_idx->{$_}} == 1 ) { |
1140 | my ($l, $r) = %{$fin_idx->{$_}}; |
1141 | |
1142 | if ( |
1143 | ref $r eq 'ARRAY' |
1144 | and |
1145 | ( |
1146 | ( @$r == 1 and $l =~ /^\-and$/i ) |
1147 | or |
1148 | $l =~ /^\-or$/i |
1149 | ) |
1150 | ) { |
1151 | push @or, @$r |
1152 | } |
1153 | |
1154 | elsif ( |
1155 | ref $r eq 'HASH' |
1156 | and |
1157 | keys %$r == 1 |
1158 | and |
1159 | $l =~ /^\-(?:and|or)$/i |
1160 | ) { |
1161 | push @or, %$r; |
1162 | } |
1163 | |
1164 | else { |
1165 | push @or, $l, $r; |
1166 | } |
1167 | } |
1168 | else { |
1169 | push @or, $fin_idx->{$_}; |
1170 | } |
1171 | } |
1172 | |
1173 | $fin->{-or} = \@or; |
1174 | } |
8d005ad9 |
1175 | } |
1176 | else { |
1177 | # not a hash not an array |
07add744 |
1178 | $fin = { -and => [ $where ] }; |
135ac69d |
1179 | } |
1180 | |
1181 | # unroll single-element -and's |
1182 | while ( |
1183 | $fin->{-and} |
1184 | and |
1185 | @{$fin->{-and}} < 2 |
1186 | ) { |
1187 | my $and = delete $fin->{-and}; |
1188 | last if @$and == 0; |
1189 | |
1190 | # at this point we have @$and == 1 |
1191 | if ( |
1192 | ref $and->[0] eq 'HASH' |
1193 | and |
1194 | ! grep { exists $fin->{$_} } keys %{$and->[0]} |
1195 | ) { |
1196 | $fin = { |
1197 | %$fin, %{$and->[0]} |
1198 | }; |
1199 | } |
07add744 |
1200 | else { |
1201 | $fin->{-and} = $and; |
1202 | last; |
1203 | } |
135ac69d |
1204 | } |
1205 | |
1206 | # compress same-column conds found in $fin |
1207 | for my $col ( grep { $_ !~ /^\-/ } keys %$fin ) { |
1208 | next unless ref $fin->{$col} eq 'ARRAY' and ($fin->{$col}[0]||'') =~ /^\-and$/i; |
1209 | my $val_bag = { map { |
5379386e |
1210 | (! defined $_ ) ? ( UNDEF => undef ) |
1211 | : ( ! length ref $_ or is_plain_value $_ ) ? ( "VAL_$_" => $_ ) |
135ac69d |
1212 | : ( ( 'SER_' . serialize $_ ) => $_ ) |
1213 | } @{$fin->{$col}}[1 .. $#{$fin->{$col}}] }; |
1214 | |
1215 | if (keys %$val_bag == 1 ) { |
1216 | ($fin->{$col}) = values %$val_bag; |
1217 | } |
1218 | else { |
1219 | $fin->{$col} = [ -and => map { $val_bag->{$_} } sort keys %$val_bag ]; |
1220 | } |
8d005ad9 |
1221 | } |
1222 | |
135ac69d |
1223 | return keys %$fin ? $fin : (); |
8d005ad9 |
1224 | } |
1225 | |
1226 | sub _collapse_cond_unroll_pairs { |
1227 | my ($self, $pairs) = @_; |
1228 | |
1229 | my @conds; |
1230 | |
1231 | while (@$pairs) { |
6565d2c3 |
1232 | my ($lhs, $rhs) = splice @$pairs, 0, 2; |
8d005ad9 |
1233 | |
1234 | if ($lhs eq '') { |
1235 | push @conds, $self->_collapse_cond($rhs); |
1236 | } |
1237 | elsif ( $lhs =~ /^\-and$/i ) { |
1238 | push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY')); |
1239 | } |
1240 | elsif ( $lhs =~ /^\-or$/i ) { |
1241 | push @conds, $self->_collapse_cond( |
1242 | (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs |
1243 | ); |
1244 | } |
1245 | else { |
1246 | if (ref $rhs eq 'HASH' and ! keys %$rhs) { |
1247 | # FIXME - SQLA seems to be doing... nothing...? |
1248 | } |
f6fff270 |
1249 | # normalize top level -ident, for saner extract_fixed_condition_columns code |
5f35ba0f |
1250 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) { |
1251 | push @conds, { $lhs => { '=', $rhs } }; |
1252 | } |
1253 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-value} and is_plain_value $rhs->{-value}) { |
1254 | push @conds, { $lhs => $rhs->{-value} }; |
1255 | } |
8d005ad9 |
1256 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) { |
f6fff270 |
1257 | if ( length ref $rhs->{'='} and is_literal_value $rhs->{'='} ) { |
5f35ba0f |
1258 | push @conds, { $lhs => $rhs }; |
1259 | } |
1260 | else { |
6565d2c3 |
1261 | for my $p ($self->_collapse_cond_unroll_pairs([ $lhs => $rhs->{'='} ])) { |
5f35ba0f |
1262 | |
1263 | # extra sanity check |
1264 | if (keys %$p > 1) { |
1265 | require Data::Dumper::Concise; |
1266 | local $Data::Dumper::Deepcopy = 1; |
1267 | $self->throw_exception( |
1268 | "Internal error: unexpected collapse unroll:" |
1269 | . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p } |
1270 | ); |
1271 | } |
8d005ad9 |
1272 | |
5f35ba0f |
1273 | my ($l, $r) = %$p; |
8d005ad9 |
1274 | |
f6fff270 |
1275 | push @conds, ( |
1276 | ! length ref $r |
1277 | or |
1278 | # the unroller recursion may return a '=' prepended value already |
1279 | ref $r eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='} |
1280 | or |
1281 | is_plain_value($r) |
1282 | ) |
5f35ba0f |
1283 | ? { $l => $r } |
1284 | : { $l => { '=' => $r } } |
1285 | ; |
1286 | } |
8d005ad9 |
1287 | } |
1288 | } |
1289 | elsif (ref $rhs eq 'ARRAY') { |
1290 | # some of these conditionals encounter multi-values - roll them out using |
1291 | # an unshift, which will cause extra looping in the while{} above |
1292 | if (! @$rhs ) { |
1293 | push @conds, { $lhs => [] }; |
1294 | } |
1295 | elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) { |
1296 | $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ") |
1297 | if @$rhs == 1; |
1298 | |
1299 | if( $rhs->[0] =~ /^\-and$/i ) { |
6565d2c3 |
1300 | unshift @$pairs, map { $lhs => $_ } @{$rhs}[1..$#$rhs]; |
8d005ad9 |
1301 | } |
1302 | # if not an AND then it's an OR |
1303 | elsif(@$rhs == 2) { |
6565d2c3 |
1304 | unshift @$pairs, $lhs => $rhs->[1]; |
8d005ad9 |
1305 | } |
1306 | else { |
953d5b7d |
1307 | push @conds, { $lhs => [ @{$rhs}[1..$#$rhs] ] }; |
8d005ad9 |
1308 | } |
1309 | } |
1310 | elsif (@$rhs == 1) { |
6565d2c3 |
1311 | unshift @$pairs, $lhs => $rhs->[0]; |
8d005ad9 |
1312 | } |
1313 | else { |
1314 | push @conds, { $lhs => $rhs }; |
1315 | } |
1316 | } |
c1f3f2e8 |
1317 | # unroll func + { -value => ... } |
1318 | elsif ( |
1319 | ref $rhs eq 'HASH' |
1320 | and |
1321 | ( my ($subop) = keys %$rhs ) == 1 |
1322 | and |
1323 | length ref ((values %$rhs)[0]) |
1324 | and |
1325 | my $vref = is_plain_value( (values %$rhs)[0] ) |
1326 | ) { |
5379386e |
1327 | push @conds, { $lhs => { $subop => $$vref } } |
c1f3f2e8 |
1328 | } |
8d005ad9 |
1329 | else { |
1330 | push @conds, { $lhs => $rhs }; |
1331 | } |
1332 | } |
1333 | } |
1334 | |
1335 | return @conds; |
1336 | } |
1337 | |
8e40a627 |
1338 | # Analyzes a given condition and attempts to extract all columns |
1339 | # with a definitive fixed-condition criteria. Returns a hashref |
1340 | # of k/v pairs suitable to be passed to set_columns(), with a |
1341 | # MAJOR CAVEAT - multi-value (contradictory) equalities are still |
1342 | # represented as a reference to the UNRESOVABLE_CONDITION constant |
1343 | # The reason we do this is that some codepaths only care about the |
1344 | # codition being stable, as opposed to actually making sense |
5f11e54f |
1345 | # |
8e40a627 |
1346 | # The normal mode is used to figure out if a resultset is constrained |
1347 | # to a column which is part of a unique constraint, which in turn |
1348 | # allows us to better predict how ordering will behave etc. |
1349 | # |
1350 | # With the optional "consider_nulls" boolean argument, the function |
1351 | # is instead used to infer inambiguous values from conditions |
1352 | # (e.g. the inheritance of resultset conditions on new_result) |
1353 | # |
5f11e54f |
1354 | sub _extract_fixed_condition_columns { |
8e40a627 |
1355 | my ($self, $where, $consider_nulls) = @_; |
1356 | my $where_hash = $self->_collapse_cond($_[1]); |
1357 | |
1358 | my $res = {}; |
1359 | my ($c, $v); |
1360 | for $c (keys %$where_hash) { |
1361 | my $vals; |
1362 | |
1363 | if (!defined ($v = $where_hash->{$c}) ) { |
b34d9331 |
1364 | $vals->{UNDEF} = $v if $consider_nulls |
8e40a627 |
1365 | } |
1366 | elsif ( |
8e40a627 |
1367 | ref $v eq 'HASH' |
1368 | and |
1369 | keys %$v == 1 |
5f35ba0f |
1370 | ) { |
1371 | if (exists $v->{-value}) { |
1372 | if (defined $v->{-value}) { |
b34d9331 |
1373 | $vals->{"VAL_$v->{-value}"} = $v->{-value} |
5f35ba0f |
1374 | } |
1375 | elsif( $consider_nulls ) { |
b34d9331 |
1376 | $vals->{UNDEF} = $v->{-value}; |
5f35ba0f |
1377 | } |
1378 | } |
8e40a627 |
1379 | # do not need to check for plain values - _collapse_cond did it for us |
f6fff270 |
1380 | elsif( |
1381 | length ref $v->{'='} |
1382 | and |
1383 | ( |
1384 | ( ref $v->{'='} eq 'HASH' and keys %{$v->{'='}} == 1 and exists $v->{'='}{-ident} ) |
1385 | or |
1386 | is_literal_value($v->{'='}) |
1387 | ) |
1388 | ) { |
b34d9331 |
1389 | $vals->{ 'SER_' . serialize $v->{'='} } = $v->{'='}; |
5f35ba0f |
1390 | } |
1391 | } |
1392 | elsif ( |
1393 | ! length ref $v |
1394 | or |
1395 | is_plain_value ($v) |
8e40a627 |
1396 | ) { |
b34d9331 |
1397 | $vals->{"VAL_$v"} = $v; |
8e40a627 |
1398 | } |
1399 | elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') { |
1400 | for ( @{$v}[1..$#$v] ) { |
1401 | my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls'); # always fish nulls out on recursion |
1402 | next unless exists $subval->{$c}; # didn't find anything |
b34d9331 |
1403 | $vals->{ |
1404 | ! defined $subval->{$c} ? 'UNDEF' |
1405 | : ( ! length ref $subval->{$c} or is_plain_value $subval->{$c} ) ? "VAL_$subval->{$c}" |
1406 | : ( 'SER_' . serialize $subval->{$c} ) |
1407 | } = $subval->{$c}; |
8d005ad9 |
1408 | } |
5f11e54f |
1409 | } |
8e40a627 |
1410 | |
1411 | if (keys %$vals == 1) { |
1412 | ($res->{$c}) = (values %$vals) |
b34d9331 |
1413 | unless !$consider_nulls and exists $vals->{UNDEF}; |
8e40a627 |
1414 | } |
1415 | elsif (keys %$vals > 1) { |
1416 | $res->{$c} = UNRESOLVABLE_CONDITION; |
1417 | } |
5f11e54f |
1418 | } |
8d005ad9 |
1419 | |
8e40a627 |
1420 | $res; |
c0748280 |
1421 | } |
bac6c4fb |
1422 | |
d28bb90d |
1423 | 1; |