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 | |
757891ed |
43 | my $orig_aliastypes = |
44 | $attrs->{_precalculated_aliastypes} |
45 | || |
46 | $self->_resolve_aliastypes_from_select_args($attrs) |
47 | ; |
4b1b5ea3 |
48 | |
eb58c082 |
49 | my $new_aliastypes = { %$orig_aliastypes }; |
50 | |
51 | # we will be recreating this entirely |
52 | my @reclassify = 'joining'; |
97e130fa |
53 | |
4b1b5ea3 |
54 | # a grouped set will not be affected by amount of rows. Thus any |
eb58c082 |
55 | # purely multiplicator classifications can go |
56 | # (will be reintroduced below if needed by something else) |
57 | push @reclassify, qw(multiplying premultiplied) |
437a9cfa |
58 | if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by}; |
4b1b5ea3 |
59 | |
eb58c082 |
60 | # nuke what will be recalculated |
61 | delete @{$new_aliastypes}{@reclassify}; |
62 | |
e1861c2c |
63 | my @newfrom = $attrs->{from}[0]; # FROM head is always present |
052e8431 |
64 | |
eb58c082 |
65 | # recalculate what we need once the multipliers are potentially gone |
66 | # ignore premultiplies, since they do not add any value to anything |
a4812caa |
67 | my %need_joins; |
eb58c082 |
68 | for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) { |
a4812caa |
69 | # add all requested aliases |
70 | $need_joins{$_} = 1 for keys %$_; |
71 | |
72 | # add all their parents (as per joinpath which is an AoH { table => alias }) |
97e130fa |
73 | $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_; |
a4812caa |
74 | } |
97e130fa |
75 | |
e1861c2c |
76 | for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) { |
539ffe87 |
77 | push @newfrom, $j if ( |
a6ef93cb |
78 | (! defined $j->[0]{-alias}) # legacy crap |
539ffe87 |
79 | || |
80 | $need_joins{$j->[0]{-alias}} |
81 | ); |
052e8431 |
82 | } |
83 | |
eb58c082 |
84 | # we have a new set of joiners - for everything we nuked pull the classification |
85 | # off the original stack |
86 | for my $ctype (@reclassify) { |
87 | $new_aliastypes->{$ctype} = { map |
88 | { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () } |
89 | keys %{$orig_aliastypes->{$ctype}} |
90 | } |
91 | } |
92 | |
93 | return ( \@newfrom, $new_aliastypes ); |
052e8431 |
94 | } |
95 | |
052e8431 |
96 | # |
d28bb90d |
97 | # This is the code producing joined subqueries like: |
8273e845 |
98 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
d28bb90d |
99 | # |
100 | sub _adjust_select_args_for_complex_prefetch { |
e1861c2c |
101 | my ($self, $attrs) = @_; |
d28bb90d |
102 | |
e1861c2c |
103 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless ( |
104 | ref $attrs->{from} eq 'ARRAY' |
105 | and |
106 | @{$attrs->{from}} > 1 |
107 | and |
108 | ref $attrs->{from}[0] eq 'HASH' |
109 | and |
110 | ref $attrs->{from}[1] eq 'ARRAY' |
111 | ); |
d28bb90d |
112 | |
1e4f9fb3 |
113 | my $root_alias = $attrs->{alias}; |
114 | |
d28bb90d |
115 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
116 | my $outer_attrs = { %$attrs }; |
e1861c2c |
117 | delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)}; |
d28bb90d |
118 | |
6aa93928 |
119 | my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 }; |
120 | delete @{$inner_attrs}{qw(for collapse select as)}; |
d28bb90d |
121 | |
4df1400e |
122 | # there is no point of ordering the insides if there is no limit |
123 | delete $inner_attrs->{order_by} if ( |
124 | delete $inner_attrs->{_order_is_artificial} |
125 | or |
126 | ! $inner_attrs->{rows} |
127 | ); |
946f6260 |
128 | |
d28bb90d |
129 | # generate the inner/outer select lists |
130 | # for inside we consider only stuff *not* brought in by the prefetch |
131 | # on the outside we substitute any function for its alias |
e1861c2c |
132 | $outer_attrs->{select} = [ @{$attrs->{select}} ]; |
36fd7f07 |
133 | |
97e130fa |
134 | my ($root_node, $root_node_offset); |
27e0370d |
135 | |
e1861c2c |
136 | for my $i (0 .. $#{$inner_attrs->{from}}) { |
137 | my $node = $inner_attrs->{from}[$i]; |
27e0370d |
138 | my $h = (ref $node eq 'HASH') ? $node |
139 | : (ref $node eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0] |
140 | : next |
141 | ; |
142 | |
1e4f9fb3 |
143 | if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) { |
97e130fa |
144 | $root_node = $h; |
145 | $root_node_offset = $i; |
27e0370d |
146 | last; |
147 | } |
148 | } |
149 | |
150 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
97e130fa |
151 | unless $root_node; |
27e0370d |
152 | |
153 | # use the heavy duty resolver to take care of aliased/nonaliased naming |
e1861c2c |
154 | my $colinfo = $self->_resolve_column_info($inner_attrs->{from}); |
27e0370d |
155 | my $selected_root_columns; |
156 | |
e1861c2c |
157 | for my $i (0 .. $#{$outer_attrs->{select}}) { |
158 | my $sel = $outer_attrs->{select}->[$i]; |
d28bb90d |
159 | |
1e4f9fb3 |
160 | next if ( |
161 | $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias |
162 | ); |
163 | |
d28bb90d |
164 | if (ref $sel eq 'HASH' ) { |
165 | $sel->{-as} ||= $attrs->{as}[$i]; |
e1861c2c |
166 | $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") ); |
d28bb90d |
167 | } |
27e0370d |
168 | elsif (! ref $sel and my $ci = $colinfo->{$sel}) { |
169 | $selected_root_columns->{$ci->{-colname}} = 1; |
170 | } |
d28bb90d |
171 | |
e1861c2c |
172 | push @{$inner_attrs->{select}}, $sel; |
bb9bffea |
173 | |
174 | push @{$inner_attrs->{as}}, $attrs->{as}[$i]; |
d28bb90d |
175 | } |
176 | |
757891ed |
177 | my $inner_aliastypes = $self->_resolve_aliastypes_from_select_args($inner_attrs); |
178 | |
179 | # In the inner subq we will need to fetch *only* native columns which may |
97e130fa |
180 | # be a part of an *outer* join condition, or an order_by (which needs to be |
e1861c2c |
181 | # preserved outside), or wheres. In other words everything but the inner |
182 | # selector |
97e130fa |
183 | # We can not just fetch everything because a potential has_many restricting |
184 | # join collapse *will not work* on heavy data types. |
97e130fa |
185 | |
757891ed |
186 | # essentially a map of all non-selecting seen columns |
187 | # the sort is there for a nicer select list |
188 | for ( |
189 | sort |
190 | map |
191 | { keys %{$_->{-seen_columns}||{}} } |
192 | map |
193 | { values %{$inner_aliastypes->{$_}} } |
194 | grep |
195 | { $_ ne 'selecting' } |
196 | keys %$inner_aliastypes |
197 | ) { |
97e130fa |
198 | my $ci = $colinfo->{$_} or next; |
199 | if ( |
1e4f9fb3 |
200 | $ci->{-source_alias} eq $root_alias |
97e130fa |
201 | and |
202 | ! $selected_root_columns->{$ci->{-colname}}++ |
203 | ) { |
204 | # adding it to both to keep limits not supporting dark selectors happy |
e1861c2c |
205 | push @{$inner_attrs->{select}}, $ci->{-fq_colname}; |
97e130fa |
206 | push @{$inner_attrs->{as}}, $ci->{-fq_colname}; |
27e0370d |
207 | } |
208 | } |
209 | |
e1861c2c |
210 | # construct the inner {from} and lock it in a subquery |
48580715 |
211 | # we need to prune first, because this will determine if we need a group_by below |
97e130fa |
212 | # throw away all non-selecting, non-restricting multijoins |
eb58c082 |
213 | # (since we def. do not care about multiplication of the contents of the subquery) |
6395604e |
214 | my $inner_subq = do { |
ea95892e |
215 | |
eb58c082 |
216 | # must use it here regardless of user requests (vastly gentler on optimizer) |
ea95892e |
217 | local $self->{_use_join_optimizer} = 1; |
218 | |
97e130fa |
219 | # throw away multijoins since we def. do not care about those inside the subquery |
757891ed |
220 | # $inner_aliastypes *will* be redefined at this point |
221 | ($inner_attrs->{from}, $inner_aliastypes ) = $self->_prune_unused_joins ({ |
222 | %$inner_attrs, |
223 | _force_prune_multiplying_joins => 1, |
224 | _precalculated_aliastypes => $inner_aliastypes, |
437a9cfa |
225 | }); |
ea95892e |
226 | |
eb58c082 |
227 | # uh-oh a multiplier (which is not us) left in, this is a problem for limits |
228 | # we will need to add a group_by to collapse the resultset for proper counts |
0a3441ee |
229 | if ( |
eb58c082 |
230 | grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} } |
1e4f9fb3 |
231 | and |
560978e2 |
232 | # if there are user-supplied groups - assume user knows wtf they are up to |
233 | ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} ) |
0a3441ee |
234 | ) { |
1e4f9fb3 |
235 | |
eb58c082 |
236 | my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} }; |
1e4f9fb3 |
237 | |
eb58c082 |
238 | # *possibly* supplement the main selection with pks if not already |
239 | # there, as they will have to be a part of the group_by to collapse |
240 | # things properly |
241 | my $inner_select_with_extras; |
242 | my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns |
243 | or $self->throw_exception( sprintf |
244 | 'Unable to perform complex limited prefetch off %s without declared primary key', |
245 | $root_node->{-rsrc}->source_name, |
e1861c2c |
246 | ); |
eb58c082 |
247 | for my $col (@pks) { |
248 | push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col |
249 | unless $cur_sel->{$col}++; |
1e4f9fb3 |
250 | } |
eb58c082 |
251 | |
252 | ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({ |
253 | %$inner_attrs, |
254 | $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (), |
255 | _aliastypes => $inner_aliastypes, |
256 | }); |
0a3441ee |
257 | } |
d28bb90d |
258 | |
e1861c2c |
259 | # we already optimized $inner_attrs->{from} above |
97e130fa |
260 | # and already local()ized |
261 | $self->{_use_join_optimizer} = 0; |
d28bb90d |
262 | |
ea95892e |
263 | # generate the subquery |
6395604e |
264 | $self->_select_args_to_query ( |
e1861c2c |
265 | @{$inner_attrs}{qw(from select where)}, |
ea95892e |
266 | $inner_attrs, |
267 | ); |
d28bb90d |
268 | }; |
269 | |
270 | # Generate the outer from - this is relatively easy (really just replace |
271 | # the join slot with the subquery), with a major caveat - we can not |
272 | # join anything that is non-selecting (not part of the prefetch), but at |
273 | # the same time is a multi-type relationship, as it will explode the result. |
274 | # |
275 | # There are two possibilities here |
276 | # - either the join is non-restricting, in which case we simply throw it away |
277 | # - it is part of the restrictions, in which case we need to collapse the outer |
278 | # result by tackling yet another group_by to the outside of the query |
279 | |
27e0370d |
280 | # work on a shallow copy |
e1861c2c |
281 | my @orig_from = @{$attrs->{from}}; |
282 | |
052e8431 |
283 | |
e1861c2c |
284 | $outer_attrs->{from} = \ my @outer_from; |
53c29913 |
285 | |
27e0370d |
286 | # we may not be the head |
97e130fa |
287 | if ($root_node_offset) { |
e1861c2c |
288 | # first generate the outer_from, up to the substitution point |
289 | @outer_from = splice @orig_from, 0, $root_node_offset; |
27e0370d |
290 | |
e1861c2c |
291 | # substitute the subq at the right spot |
27e0370d |
292 | push @outer_from, [ |
293 | { |
1e4f9fb3 |
294 | -alias => $root_alias, |
97e130fa |
295 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
296 | $root_alias => $inner_subq, |
27e0370d |
297 | }, |
e1861c2c |
298 | # preserve attrs from what is now the head of the from after the splice |
299 | @{$orig_from[0]}[1 .. $#{$orig_from[0]}], |
27e0370d |
300 | ]; |
301 | } |
302 | else { |
27e0370d |
303 | @outer_from = { |
1e4f9fb3 |
304 | -alias => $root_alias, |
27e0370d |
305 | -rsrc => $root_node->{-rsrc}, |
1e4f9fb3 |
306 | $root_alias => $inner_subq, |
27e0370d |
307 | }; |
d28bb90d |
308 | } |
309 | |
e1861c2c |
310 | shift @orig_from; # what we just replaced above |
97e130fa |
311 | |
ea95892e |
312 | # scan the *remaining* from spec against different attributes, and see which joins are needed |
052e8431 |
313 | # in what role |
975b573a |
314 | my $outer_aliastypes = $outer_attrs->{_aliastypes} = |
e1861c2c |
315 | $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from }); |
052e8431 |
316 | |
a4812caa |
317 | # unroll parents |
1e4f9fb3 |
318 | my ($outer_select_chain, @outer_nonselecting_chains) = map { +{ |
319 | map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} } |
320 | } } qw/selecting restricting grouping ordering/; |
a4812caa |
321 | |
d28bb90d |
322 | # see what's left - throw away if not selecting/restricting |
eb58c082 |
323 | my $may_need_outer_group_by; |
e1861c2c |
324 | while (my $j = shift @orig_from) { |
d28bb90d |
325 | my $alias = $j->[0]{-alias}; |
326 | |
a4812caa |
327 | if ( |
328 | $outer_select_chain->{$alias} |
329 | ) { |
330 | push @outer_from, $j |
d28bb90d |
331 | } |
1e4f9fb3 |
332 | elsif (first { $_->{$alias} } @outer_nonselecting_chains ) { |
d28bb90d |
333 | push @outer_from, $j; |
eb58c082 |
334 | $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0; |
d28bb90d |
335 | } |
336 | } |
337 | |
eb58c082 |
338 | # also throw in a synthetic group_by if a non-selecting multiplier, |
339 | # to guard against cross-join explosions |
340 | # the logic is somewhat fragile, but relies on the idea that if a user supplied |
341 | # a group by on their own - they know what they were doing |
342 | if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) { |
343 | ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({ |
560978e2 |
344 | %$outer_attrs, |
345 | from => \@outer_from, |
560978e2 |
346 | }); |
36fd7f07 |
347 | } |
348 | |
e1861c2c |
349 | # This is totally horrific - the {where} ends up in both the inner and outer query |
d28bb90d |
350 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
351 | # then if where conditions apply to the *right* side of the prefetch, you may have |
352 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
4a0eed52 |
353 | # the outer select to exclude joins you didn't want in the first place |
d28bb90d |
354 | # |
355 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
e1861c2c |
356 | return $outer_attrs; |
d28bb90d |
357 | } |
358 | |
1a736efb |
359 | # |
360 | # I KNOW THIS SUCKS! GET SQLA2 OUT THE DOOR SO THIS CAN DIE! |
361 | # |
ad630f4b |
362 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
363 | # select/where/order/group attributes, in order to determine what |
4a0eed52 |
364 | # aliases are needed to fulfill the query. This information is used |
ad630f4b |
365 | # throughout the code to prune unnecessary JOINs from the queries |
366 | # in an attempt to reduce the execution time. |
367 | # Although the method is pretty horrific, the worst thing that can |
1a736efb |
368 | # happen is for it to fail due to some scalar SQL, which in turn will |
369 | # result in a vocal exception. |
539ffe87 |
370 | sub _resolve_aliastypes_from_select_args { |
e1861c2c |
371 | my ( $self, $attrs ) = @_; |
546f1cd9 |
372 | |
ad630f4b |
373 | $self->throw_exception ('Unable to analyze custom {from}') |
e1861c2c |
374 | if ref $attrs->{from} ne 'ARRAY'; |
546f1cd9 |
375 | |
ad630f4b |
376 | # what we will return |
964a3c71 |
377 | my $aliases_by_type; |
546f1cd9 |
378 | |
ad630f4b |
379 | # see what aliases are there to work with |
eb58c082 |
380 | # and record who is a multiplier and who is premultiplied |
ad630f4b |
381 | my $alias_list; |
e1861c2c |
382 | for my $node (@{$attrs->{from}}) { |
383 | |
384 | my $j = $node; |
ad630f4b |
385 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
386 | my $al = $j->{-alias} |
387 | or next; |
388 | |
389 | $alias_list->{$al} = $j; |
eb58c082 |
390 | |
391 | $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] } |
a4812caa |
392 | # not array == {from} head == can't be multiplying |
eb58c082 |
393 | if ref($node) eq 'ARRAY' and ! $j->{-is_single}; |
394 | |
395 | $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] } |
396 | # parts of the path that are not us but are multiplying |
397 | if grep { $aliases_by_type->{multiplying}{$_} } |
398 | grep { $_ ne $al } |
399 | map { values %$_ } |
400 | @{ $j->{-join_path}||[] } |
546f1cd9 |
401 | } |
546f1cd9 |
402 | |
318e3d94 |
403 | # get a column to source/alias map (including unambiguous unqualified ones) |
e1861c2c |
404 | my $colinfo = $self->_resolve_column_info ($attrs->{from}); |
1a736efb |
405 | |
ad630f4b |
406 | # set up a botched SQLA |
407 | my $sql_maker = $self->sql_maker; |
07f31d19 |
408 | |
4c2b30d6 |
409 | # these are throw away results, do not pollute the bind stack |
0542ec57 |
410 | local $sql_maker->{where_bind}; |
411 | local $sql_maker->{group_bind}; |
412 | local $sql_maker->{having_bind}; |
97e130fa |
413 | local $sql_maker->{from_bind}; |
3f5b99fe |
414 | |
415 | # we can't scan properly without any quoting (\b doesn't cut it |
416 | # everywhere), so unless there is proper quoting set - use our |
417 | # own weird impossible character. |
418 | # Also in the case of no quoting, we need to explicitly disable |
419 | # name_sep, otherwise sorry nasty legacy syntax like |
420 | # { 'count(foo.id)' => { '>' => 3 } } will stop working >:( |
421 | local $sql_maker->{quote_char} = $sql_maker->{quote_char}; |
422 | local $sql_maker->{name_sep} = $sql_maker->{name_sep}; |
423 | |
424 | unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) { |
e493ecb2 |
425 | $sql_maker->{quote_char} = ["\x00", "\xFF"]; |
426 | # if we don't unset it we screw up retarded but unfortunately working |
427 | # 'MAX(foo.bar)' => { '>', 3 } |
3f5b99fe |
428 | $sql_maker->{name_sep} = ''; |
429 | } |
430 | |
431 | my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep); |
07f31d19 |
432 | |
1a736efb |
433 | # generate sql chunks |
434 | my $to_scan = { |
435 | restricting => [ |
a9e985b7 |
436 | ($sql_maker->_recurse_where ($attrs->{where}))[0], |
1e4f9fb3 |
437 | $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }), |
438 | ], |
439 | grouping => [ |
440 | $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }), |
1a736efb |
441 | ], |
97e130fa |
442 | joining => [ |
443 | $sql_maker->_recurse_from ( |
e1861c2c |
444 | ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0], |
445 | @{$attrs->{from}}[1 .. $#{$attrs->{from}}], |
97e130fa |
446 | ), |
447 | ], |
1a736efb |
448 | selecting => [ |
fdd47fe8 |
449 | # kill all selectors which look like a proper subquery |
450 | # this is a sucky heuristic *BUT* - if we get it wrong the query will simply |
451 | # fail to run, so we are relatively safe |
452 | grep |
453 | { $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi } |
454 | map |
455 | { ($sql_maker->_recurse_fields($_))[0] } |
456 | @{$attrs->{select}} |
1e4f9fb3 |
457 | ], |
66bbb12c |
458 | ordering => [ map |
459 | { |
460 | ( my $sql = (ref $_ ? $_->[0] : $_) ) =~ s/ \s+ (?: ASC | DESC ) \s* \z //xi; |
461 | $sql; |
462 | } |
463 | $sql_maker->_order_by_chunks( $attrs->{order_by} ), |
1a736efb |
464 | ], |
465 | }; |
07f31d19 |
466 | |
89203568 |
467 | # we will be bulk-scanning anyway - pieces will not matter in that case, |
468 | # thus join everything up |
fdd47fe8 |
469 | # throw away empty-string chunks, and make sure no binds snuck in |
470 | # note that we operate over @{$to_scan->{$type}}, hence the |
471 | # semi-mindbending ... map ... for values ... |
89203568 |
472 | ( $_ = join ' ', map { |
0dadd60d |
473 | |
89203568 |
474 | ( ! defined $_ ) ? () |
475 | : ( length ref $_ ) ? (require Data::Dumper::Concise && $self->throw_exception( |
fdd47fe8 |
476 | "Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($_) |
477 | )) |
89203568 |
478 | : ( $_ =~ /^\s*$/ ) ? () |
479 | : $_ |
0dadd60d |
480 | |
89203568 |
481 | } @$_ ) for values %$to_scan; |
fdd47fe8 |
482 | |
483 | # throw away empty to-scan's |
484 | ( |
89203568 |
485 | length $to_scan->{$_} |
fdd47fe8 |
486 | or |
487 | delete $to_scan->{$_} |
488 | ) for keys %$to_scan; |
0dadd60d |
489 | |
07f31d19 |
490 | |
89203568 |
491 | |
90c9dd1d |
492 | # these will be used for matching in the loop below |
493 | my $all_aliases = join ' | ', map { quotemeta $_ } keys %$alias_list; |
494 | my $fq_col_re = qr/ |
495 | $lquote ( $all_aliases ) $rquote $sep (?: $lquote ([^$rquote]+) $rquote )? |
496 | | |
497 | \b ( $all_aliases ) \. ( [^\s\)\($rquote]+ )? |
498 | /x; |
499 | |
89203568 |
500 | |
90c9dd1d |
501 | my $all_unq_columns = join ' | ', |
502 | map |
503 | { quotemeta $_ } |
504 | grep |
505 | # using a regex here shows up on profiles, boggle |
506 | { index( $_, '.') < 0 } |
507 | keys %$colinfo |
508 | ; |
509 | my $unq_col_re = $all_unq_columns |
89203568 |
510 | ? qr/ |
511 | $lquote ( $all_unq_columns ) $rquote |
512 | | |
513 | (?: \A | \s ) ( $all_unq_columns ) (?: \s | \z ) |
514 | /x |
90c9dd1d |
515 | : undef |
516 | ; |
517 | |
518 | |
19955cdf |
519 | # the actual scan, per type |
318e3d94 |
520 | for my $type (keys %$to_scan) { |
19955cdf |
521 | |
90c9dd1d |
522 | |
19955cdf |
523 | # now loop through all fully qualified columns and get the corresponding |
524 | # alias (should work even if they are in scalarrefs) |
90c9dd1d |
525 | # |
89203568 |
526 | # The regex captures in multiples of 4, with one of the two pairs being |
90c9dd1d |
527 | # undef. There may be a *lot* of matches, hence the convoluted loop |
89203568 |
528 | my @matches = $to_scan->{$type} =~ /$fq_col_re/g; |
90c9dd1d |
529 | my $i = 0; |
530 | while( $i < $#matches ) { |
531 | |
532 | if ( |
533 | defined $matches[$i] |
534 | ) { |
535 | $aliases_by_type->{$type}{$matches[$i]} ||= { -parents => $alias_list->{$matches[$i]}{-join_path}||[] }; |
536 | |
537 | $aliases_by_type->{$type}{$matches[$i]}{-seen_columns}{"$matches[$i].$matches[$i+1]"} = "$matches[$i].$matches[$i+1]" |
538 | if defined $matches[$i+1]; |
539 | |
540 | $i += 2; |
1a736efb |
541 | } |
1a736efb |
542 | |
90c9dd1d |
543 | $i += 2; |
544 | } |
1a736efb |
545 | |
07f31d19 |
546 | |
90c9dd1d |
547 | # now loop through unqualified column names, and try to locate them within |
548 | # the chunks, if there are any unqualified columns in the 1st place |
549 | next unless $unq_col_re; |
89203568 |
550 | |
551 | # The regex captures in multiples of 2, one of the two being undef |
552 | for ( $to_scan->{$type} =~ /$unq_col_re/g ) { |
553 | defined $_ or next; |
90c9dd1d |
554 | my $alias = $colinfo->{$_}{-source_alias} or next; |
555 | $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] }; |
556 | $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_ |
07f31d19 |
557 | } |
558 | } |
559 | |
90c9dd1d |
560 | |
07f31d19 |
561 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
19955cdf |
562 | ( |
563 | $_->{-alias} |
564 | and |
565 | ! $aliases_by_type->{restricting}{ $_->{-alias} } |
566 | and |
567 | ( |
568 | not $_->{-join_type} |
07f31d19 |
569 | or |
19955cdf |
570 | $_->{-join_type} !~ /^left (?: \s+ outer)? $/xi |
571 | ) |
572 | and |
573 | $aliases_by_type->{restricting}{ $_->{-alias} } = { -parents => $_->{-join_path}||[] } |
574 | ) for values %$alias_list; |
07f31d19 |
575 | |
90c9dd1d |
576 | |
19955cdf |
577 | # final cleanup |
578 | ( |
579 | keys %{$aliases_by_type->{$_}} |
580 | or |
581 | delete $aliases_by_type->{$_} |
582 | ) for keys %$aliases_by_type; |
1e4f9fb3 |
583 | |
90c9dd1d |
584 | |
19955cdf |
585 | $aliases_by_type; |
07f31d19 |
586 | } |
587 | |
eb58c082 |
588 | # This is the engine behind { distinct => 1 } and the general |
589 | # complex prefetch grouper |
0a3441ee |
590 | sub _group_over_selection { |
560978e2 |
591 | my ($self, $attrs) = @_; |
0a3441ee |
592 | |
560978e2 |
593 | my $colinfos = $self->_resolve_column_info ($attrs->{from}); |
0a3441ee |
594 | |
595 | my (@group_by, %group_index); |
596 | |
36fd7f07 |
597 | # the logic is: if it is a { func => val } we assume an aggregate, |
598 | # otherwise if \'...' or \[...] we assume the user knows what is |
599 | # going on thus group over it |
560978e2 |
600 | for (@{$attrs->{select}}) { |
0a3441ee |
601 | if (! ref($_) or ref ($_) ne 'HASH' ) { |
602 | push @group_by, $_; |
603 | $group_index{$_}++; |
560978e2 |
604 | if ($colinfos->{$_} and $_ !~ /\./ ) { |
0a3441ee |
605 | # add a fully qualified version as well |
560978e2 |
606 | $group_index{"$colinfos->{$_}{-source_alias}.$_"}++; |
0a3441ee |
607 | } |
07f31d19 |
608 | } |
609 | } |
ad630f4b |
610 | |
eb58c082 |
611 | my @order_by = $self->_extract_order_criteria($attrs->{order_by}) |
612 | or return (\@group_by, $attrs->{order_by}); |
613 | |
614 | # add any order_by parts that are not already present in the group_by |
615 | # to maintain SQL cross-compatibility and general sanity |
616 | # |
617 | # also in case the original selection is *not* unique, or in case part |
618 | # of the ORDER BY refers to a multiplier - we will need to replace the |
619 | # skipped order_by elements with their MIN/MAX equivalents as to maintain |
620 | # the proper overall order without polluting the group criteria (and |
621 | # possibly changing the outcome entirely) |
622 | |
623 | my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes); |
624 | |
625 | my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by); |
626 | |
627 | for my $o_idx (0 .. $#order_by) { |
628 | |
629 | # if the chunk is already a min/max function - there is nothing left to touch |
630 | next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix; |
631 | |
0a3441ee |
632 | # only consider real columns (for functions the user got to do an explicit group_by) |
eb58c082 |
633 | my $chunk_ci; |
634 | if ( |
635 | @{$order_by[$o_idx]} != 1 |
636 | or |
637 | # only declare an unknown *plain* identifier as "leftover" if we are called with |
638 | # aliastypes to examine. If there are none - we are still in _resolve_attrs, and |
639 | # can just assume the user knows what they want |
640 | ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} ) |
641 | ) { |
642 | push @$leftovers, $order_by[$o_idx][0]; |
14e26c5f |
643 | } |
560978e2 |
644 | |
eb58c082 |
645 | next unless $chunk_ci; |
646 | |
647 | # no duplication of group criteria |
648 | next if $group_index{$chunk_ci->{-fq_colname}}; |
649 | |
650 | $aliastypes ||= ( |
651 | $attrs->{_aliastypes} |
560978e2 |
652 | or |
eb58c082 |
653 | $self->_resolve_aliastypes_from_select_args({ |
654 | from => $attrs->{from}, |
655 | order_by => $attrs->{order_by}, |
656 | }) |
657 | ) if $group_already_unique; |
658 | |
659 | # check that we are not ordering by a multiplier (if a check is requested at all) |
660 | if ( |
661 | $group_already_unique |
662 | and |
663 | ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}} |
664 | and |
665 | ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}} |
560978e2 |
666 | ) { |
eb58c082 |
667 | push @group_by, $chunk_ci->{-fq_colname}; |
668 | $group_index{$chunk_ci->{-fq_colname}}++ |
560978e2 |
669 | } |
eb58c082 |
670 | else { |
671 | # We need to order by external columns without adding them to the group |
672 | # (eiehter a non-unique selection, or a multi-external) |
673 | # |
674 | # This doesn't really make sense in SQL, however from DBICs point |
675 | # of view is rather valid (e.g. order the leftmost objects by whatever |
676 | # criteria and get the offset/rows many). There is a way around |
677 | # this however in SQL - we simply tae the direction of each piece |
678 | # of the external order and convert them to MIN(X) for ASC or MAX(X) |
679 | # for DESC, and group_by the root columns. The end result should be |
680 | # exactly what we expect |
681 | |
682 | # FIXME - this code is a joke, will need to be completely rewritten in |
683 | # the DQ branch. But I need to push a POC here, otherwise the |
684 | # pesky tests won't pass |
685 | # wrap any part of the order_by that "responds" to an ordering alias |
686 | # into a MIN/MAX |
687 | $sql_maker ||= $self->sql_maker; |
688 | $order_chunks ||= [ |
689 | map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by}) |
690 | ]; |
0a3441ee |
691 | |
eb58c082 |
692 | my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]); |
693 | |
694 | $new_order_by[$o_idx] = \[ |
695 | sprintf( '%s( %s )%s', |
696 | ($is_desc ? 'MAX' : 'MIN'), |
697 | $chunk, |
698 | ($is_desc ? ' DESC' : ''), |
699 | ), |
700 | @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ] |
701 | ]; |
702 | } |
0a3441ee |
703 | } |
704 | |
eb58c082 |
705 | $self->throw_exception ( sprintf |
9736be65 |
706 | 'Unable to programatically derive a required group_by from the supplied ' |
707 | . 'order_by criteria. To proceed either add an explicit group_by, or ' |
708 | . 'simplify your order_by to only include plain columns ' |
709 | . '(supplied order_by: %s)', |
eb58c082 |
710 | join ', ', map { "'$_'" } @$leftovers, |
711 | ) if $leftovers; |
712 | |
713 | # recreate the untouched order parts |
714 | if (@new_order_by) { |
715 | $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks ); |
716 | } |
717 | |
718 | return ( |
719 | \@group_by, |
720 | (@new_order_by ? \@new_order_by : $attrs->{order_by} ), # same ref as original == unchanged |
721 | ); |
07f31d19 |
722 | } |
723 | |
d28bb90d |
724 | sub _resolve_ident_sources { |
725 | my ($self, $ident) = @_; |
726 | |
727 | my $alias2source = {}; |
d28bb90d |
728 | |
729 | # the reason this is so contrived is that $ident may be a {from} |
730 | # structure, specifying multiple tables to join |
6298a324 |
731 | if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) { |
d28bb90d |
732 | # this is compat mode for insert/update/delete which do not deal with aliases |
733 | $alias2source->{me} = $ident; |
d28bb90d |
734 | } |
735 | elsif (ref $ident eq 'ARRAY') { |
736 | |
737 | for (@$ident) { |
738 | my $tabinfo; |
739 | if (ref $_ eq 'HASH') { |
740 | $tabinfo = $_; |
d28bb90d |
741 | } |
742 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
743 | $tabinfo = $_->[0]; |
744 | } |
745 | |
4376a157 |
746 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc} |
747 | if ($tabinfo->{-rsrc}); |
d28bb90d |
748 | } |
749 | } |
750 | |
90f10b5a |
751 | return $alias2source; |
d28bb90d |
752 | } |
753 | |
754 | # Takes $ident, \@column_names |
755 | # |
756 | # returns { $column_name => \%column_info, ... } |
757 | # also note: this adds -result_source => $rsrc to the column info |
758 | # |
09e14fdc |
759 | # If no columns_names are supplied returns info about *all* columns |
760 | # for all sources |
d28bb90d |
761 | sub _resolve_column_info { |
762 | my ($self, $ident, $colnames) = @_; |
8d005ad9 |
763 | |
764 | return {} if $colnames and ! @$colnames; |
765 | |
229401a0 |
766 | my $sources = $self->_resolve_ident_sources($ident); |
767 | |
768 | $_ = { rsrc => $_, colinfos => $_->columns_info } |
769 | for values %$sources; |
d28bb90d |
770 | |
52416317 |
771 | my (%seen_cols, @auto_colnames); |
d28bb90d |
772 | |
773 | # compile a global list of column names, to be able to properly |
774 | # disambiguate unqualified column names (if at all possible) |
229401a0 |
775 | for my $alias (keys %$sources) { |
776 | ( |
777 | ++$seen_cols{$_}{$alias} |
778 | and |
779 | ! $colnames |
780 | and |
781 | push @auto_colnames, "$alias.$_" |
782 | ) for keys %{ $sources->{$alias}{colinfos} }; |
d28bb90d |
783 | } |
784 | |
09e14fdc |
785 | $colnames ||= [ |
786 | @auto_colnames, |
229401a0 |
787 | ( grep { keys %{$seen_cols{$_}} == 1 } keys %seen_cols ), |
09e14fdc |
788 | ]; |
789 | |
229401a0 |
790 | my %return; |
791 | for (@$colnames) { |
792 | my ($colname, $source_alias) = reverse split /\./, $_; |
d28bb90d |
793 | |
229401a0 |
794 | my $assumed_alias = |
795 | $source_alias |
796 | || |
797 | # if the column was seen exactly once - we know which rsrc it came from |
798 | ( |
799 | $seen_cols{$colname} |
800 | and |
801 | keys %{$seen_cols{$colname}} == 1 |
802 | and |
803 | ( %{$seen_cols{$colname}} )[0] |
804 | ) |
805 | || |
806 | next |
807 | ; |
52416317 |
808 | |
229401a0 |
809 | $self->throw_exception( |
810 | "No such column '$colname' on source " . $sources->{$assumed_alias}{rsrc}->source_name |
811 | ) unless $seen_cols{$colname}{$assumed_alias}; |
52416317 |
812 | |
229401a0 |
813 | $return{$_} = { |
814 | %{ $sources->{$assumed_alias}{colinfos}{$colname} }, |
815 | -result_source => $sources->{$assumed_alias}{rsrc}, |
816 | -source_alias => $assumed_alias, |
817 | -fq_colname => "$assumed_alias.$colname", |
81bf295c |
818 | -colname => $colname, |
d28bb90d |
819 | }; |
81bf295c |
820 | |
229401a0 |
821 | $return{"$assumed_alias.$colname"} = $return{$_} |
822 | unless $source_alias; |
d28bb90d |
823 | } |
824 | |
825 | return \%return; |
826 | } |
827 | |
289ac713 |
828 | # The DBIC relationship chaining implementation is pretty simple - every |
829 | # new related_relationship is pushed onto the {from} stack, and the {select} |
830 | # window simply slides further in. This means that when we count somewhere |
831 | # in the middle, we got to make sure that everything in the join chain is an |
832 | # actual inner join, otherwise the count will come back with unpredictable |
833 | # results (a resultset may be generated with _some_ rows regardless of if |
834 | # the relation which the $rs currently selects has rows or not). E.g. |
835 | # $artist_rs->cds->count - normally generates: |
836 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
837 | # which actually returns the number of artists * (number of cds || 1) |
838 | # |
839 | # So what we do here is crawl {from}, determine if the current alias is at |
840 | # the top of the stack, and if not - make sure the chain is inner-joined down |
841 | # to the root. |
842 | # |
31a8aaaf |
843 | sub _inner_join_to_node { |
289ac713 |
844 | my ($self, $from, $alias) = @_; |
845 | |
302d35f8 |
846 | my $switch_branch = $self->_find_join_path_to_node($from, $alias); |
289ac713 |
847 | |
302d35f8 |
848 | return $from unless @{$switch_branch||[]}; |
289ac713 |
849 | |
850 | # So it looks like we will have to switch some stuff around. |
851 | # local() is useless here as we will be leaving the scope |
852 | # anyway, and deep cloning is just too fucking expensive |
8273e845 |
853 | # So replace the first hashref in the node arrayref manually |
289ac713 |
854 | my @new_from = ($from->[0]); |
faeb2407 |
855 | my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path |
289ac713 |
856 | |
857 | for my $j (@{$from}[1 .. $#$from]) { |
858 | my $jalias = $j->[0]{-alias}; |
859 | |
860 | if ($sw_idx->{$jalias}) { |
861 | my %attrs = %{$j->[0]}; |
862 | delete $attrs{-join_type}; |
863 | push @new_from, [ |
864 | \%attrs, |
865 | @{$j}[ 1 .. $#$j ], |
866 | ]; |
867 | } |
868 | else { |
869 | push @new_from, $j; |
870 | } |
871 | } |
872 | |
873 | return \@new_from; |
874 | } |
875 | |
302d35f8 |
876 | sub _find_join_path_to_node { |
877 | my ($self, $from, $target_alias) = @_; |
878 | |
879 | # subqueries and other oddness are naturally not supported |
880 | return undef if ( |
881 | ref $from ne 'ARRAY' |
882 | || |
883 | ref $from->[0] ne 'HASH' |
884 | || |
885 | ! defined $from->[0]{-alias} |
886 | ); |
887 | |
888 | # no path - the head is the alias |
889 | return [] if $from->[0]{-alias} eq $target_alias; |
890 | |
891 | for my $i (1 .. $#$from) { |
892 | return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias ); |
893 | } |
894 | |
895 | # something else went quite wrong |
896 | return undef; |
897 | } |
898 | |
bac358c9 |
899 | sub _extract_order_criteria { |
1a736efb |
900 | my ($self, $order_by, $sql_maker) = @_; |
c0748280 |
901 | |
1a736efb |
902 | my $parser = sub { |
e6977bbb |
903 | my ($sql_maker, $order_by, $orig_quote_chars) = @_; |
c0748280 |
904 | |
1a736efb |
905 | return scalar $sql_maker->_order_by_chunks ($order_by) |
906 | unless wantarray; |
c0748280 |
907 | |
e6977bbb |
908 | my ($lq, $rq, $sep) = map { quotemeta($_) } ( |
909 | ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars), |
910 | $sql_maker->name_sep |
911 | ); |
912 | |
1a736efb |
913 | my @chunks; |
bac358c9 |
914 | for ($sql_maker->_order_by_chunks ($order_by) ) { |
e6977bbb |
915 | my $chunk = ref $_ ? [ @$_ ] : [ $_ ]; |
cb3e87f5 |
916 | ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]); |
e6977bbb |
917 | |
918 | # order criteria may have come back pre-quoted (literals and whatnot) |
919 | # this is fragile, but the best we can currently do |
920 | $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe |
921 | or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x; |
922 | |
1a736efb |
923 | push @chunks, $chunk; |
bac6c4fb |
924 | } |
1a736efb |
925 | |
926 | return @chunks; |
927 | }; |
928 | |
929 | if ($sql_maker) { |
930 | return $parser->($sql_maker, $order_by); |
bac6c4fb |
931 | } |
932 | else { |
1a736efb |
933 | $sql_maker = $self->sql_maker; |
e6977bbb |
934 | |
935 | # pass these in to deal with literals coming from |
936 | # the user or the deep guts of prefetch |
937 | my $orig_quote_chars = [$sql_maker->_quote_chars]; |
938 | |
1a736efb |
939 | local $sql_maker->{quote_char}; |
e6977bbb |
940 | return $parser->($sql_maker, $order_by, $orig_quote_chars); |
bac6c4fb |
941 | } |
bac6c4fb |
942 | } |
943 | |
7cec4356 |
944 | sub _order_by_is_stable { |
5f11e54f |
945 | my ($self, $ident, $order_by, $where) = @_; |
c0748280 |
946 | |
eb58c082 |
947 | my @cols = ( |
8d005ad9 |
948 | ( map { $_->[0] } $self->_extract_order_criteria($order_by) ), |
8e40a627 |
949 | ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ), |
df4312bc |
950 | ) or return 0; |
eb58c082 |
951 | |
952 | my $colinfo = $self->_resolve_column_info($ident, \@cols); |
953 | |
954 | return keys %$colinfo |
955 | ? $self->_columns_comprise_identifying_set( $colinfo, \@cols ) |
df4312bc |
956 | : 0 |
eb58c082 |
957 | ; |
958 | } |
c0748280 |
959 | |
eb58c082 |
960 | sub _columns_comprise_identifying_set { |
961 | my ($self, $colinfo, $columns) = @_; |
7cec4356 |
962 | |
963 | my $cols_per_src; |
eb58c082 |
964 | $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_ |
965 | for grep { defined $_ } @{$colinfo}{@$columns}; |
7cec4356 |
966 | |
967 | for (values %$cols_per_src) { |
968 | my $src = (values %$_)[0]->{-result_source}; |
969 | return 1 if $src->_identifying_column_set($_); |
c0748280 |
970 | } |
971 | |
df4312bc |
972 | return 0; |
7cec4356 |
973 | } |
974 | |
df4312bc |
975 | # this is almost similar to _order_by_is_stable, except it takes |
0e81e691 |
976 | # a single rsrc, and will succeed only if the first portion of the order |
977 | # by is stable. |
978 | # returns that portion as a colinfo hashref on success |
df4312bc |
979 | sub _extract_colinfo_of_stable_main_source_order_by_portion { |
302d35f8 |
980 | my ($self, $attrs) = @_; |
0e81e691 |
981 | |
302d35f8 |
982 | my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias}); |
983 | |
984 | return unless defined $nodes; |
0e81e691 |
985 | |
986 | my @ord_cols = map |
987 | { $_->[0] } |
302d35f8 |
988 | ( $self->_extract_order_criteria($attrs->{order_by}) ) |
0e81e691 |
989 | ; |
990 | return unless @ord_cols; |
991 | |
302d35f8 |
992 | my $valid_aliases = { map { $_ => 1 } ( |
993 | $attrs->{from}[0]{-alias}, |
994 | map { values %$_ } @$nodes, |
995 | ) }; |
318e3d94 |
996 | |
302d35f8 |
997 | my $colinfos = $self->_resolve_column_info($attrs->{from}); |
998 | |
999 | my ($colinfos_to_return, $seen_main_src_cols); |
1000 | |
1001 | for my $col (@ord_cols) { |
1002 | # if order criteria is unresolvable - there is nothing we can do |
1003 | my $colinfo = $colinfos->{$col} or last; |
1004 | |
1005 | # if we reached the end of the allowed aliases - also nothing we can do |
1006 | last unless $valid_aliases->{$colinfo->{-source_alias}}; |
1007 | |
1008 | $colinfos_to_return->{$col} = $colinfo; |
1009 | |
1010 | $seen_main_src_cols->{$colinfo->{-colname}} = 1 |
1011 | if $colinfo->{-source_alias} eq $attrs->{alias}; |
0e81e691 |
1012 | } |
1013 | |
302d35f8 |
1014 | # FIXME the condition may be singling out things on its own, so we |
1015 | # conceivable could come back wi "stable-ordered by nothing" |
1016 | # not confient enough in the parser yet, so punt for the time being |
1017 | return unless $seen_main_src_cols; |
0e81e691 |
1018 | |
302d35f8 |
1019 | my $main_src_fixed_cols_from_cond = [ $attrs->{where} |
1020 | ? ( |
1021 | map |
1022 | { |
1023 | ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} ) |
1024 | ? $colinfos->{$_}{-colname} |
1025 | : () |
1026 | } |
8e40a627 |
1027 | keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) } |
302d35f8 |
1028 | ) |
1029 | : () |
1030 | ]; |
0e81e691 |
1031 | |
302d35f8 |
1032 | return $attrs->{result_source}->_identifying_column_set([ |
1033 | keys %$seen_main_src_cols, |
1034 | @$main_src_fixed_cols_from_cond, |
1035 | ]) ? $colinfos_to_return : (); |
0e81e691 |
1036 | } |
1037 | |
8d005ad9 |
1038 | # Attempts to flatten a passed in SQLA condition as much as possible towards |
1039 | # a plain hashref, *without* altering its semantics. Required by |
1040 | # create/populate being able to extract definitive conditions from preexisting |
1041 | # resultset {where} stacks |
1042 | # |
1043 | # FIXME - while relatively robust, this is still imperfect, one of the first |
1044 | # things to tackle with DQ |
1045 | sub _collapse_cond { |
1046 | my ($self, $where, $where_is_anded_array) = @_; |
1047 | |
135ac69d |
1048 | my $fin; |
1049 | |
8d005ad9 |
1050 | if (! $where) { |
1051 | return; |
1052 | } |
1053 | elsif ($where_is_anded_array or ref $where eq 'HASH') { |
1054 | |
1055 | my @pairs; |
1056 | |
1057 | my @pieces = $where_is_anded_array ? @$where : $where; |
1058 | while (@pieces) { |
1059 | my $chunk = shift @pieces; |
1060 | |
1061 | if (ref $chunk eq 'HASH') { |
e466c62b |
1062 | for (sort keys %$chunk) { |
1063 | |
1064 | # Match SQLA 1.79 behavior |
1065 | if ($_ eq '') { |
1066 | is_literal_value($chunk->{$_}) |
1067 | ? carp 'Hash-pairs consisting of an empty string with a literal are deprecated, use -and => [ $literal ] instead' |
1068 | : $self->throw_exception("Supplying an empty left hand side argument is not supported in hash-pairs") |
1069 | ; |
1070 | } |
1071 | |
1072 | push @pairs, $_ => $chunk->{$_}; |
1073 | } |
8d005ad9 |
1074 | } |
1075 | elsif (ref $chunk eq 'ARRAY') { |
6565d2c3 |
1076 | push @pairs, -or => $chunk |
8d005ad9 |
1077 | if @$chunk; |
1078 | } |
b34d9331 |
1079 | elsif ( ! length ref $chunk) { |
e466c62b |
1080 | |
1081 | # Match SQLA 1.79 behavior |
1082 | $self->throw_exception("Supplying an empty left hand side argument is not supported in array-pairs") |
1083 | if $where_is_anded_array and (! defined $chunk or $chunk eq ''); |
1084 | |
6565d2c3 |
1085 | push @pairs, $chunk, shift @pieces; |
8d005ad9 |
1086 | } |
1087 | else { |
6565d2c3 |
1088 | push @pairs, '', $chunk; |
8d005ad9 |
1089 | } |
1090 | } |
1091 | |
1092 | return unless @pairs; |
1093 | |
1094 | my @conds = $self->_collapse_cond_unroll_pairs(\@pairs) |
1095 | or return; |
1096 | |
1097 | # Consolidate various @conds back into something more compact |
8d005ad9 |
1098 | for my $c (@conds) { |
1099 | if (ref $c ne 'HASH') { |
1100 | push @{$fin->{-and}}, $c; |
1101 | } |
1102 | else { |
1103 | for my $col (sort keys %$c) { |
8d005ad9 |
1104 | |
135ac69d |
1105 | # consolidate all -and nodes |
1106 | if ($col =~ /^\-and$/i) { |
1107 | push @{$fin->{-and}}, |
1108 | ref $c->{$col} eq 'ARRAY' ? @{$c->{$col}} |
1109 | : ref $c->{$col} eq 'HASH' ? %{$c->{$col}} |
1110 | : { $col => $c->{$col} } |
1111 | ; |
1112 | } |
1113 | elsif ($col =~ /^\-/) { |
1114 | push @{$fin->{-and}}, { $col => $c->{$col} }; |
1115 | } |
1116 | elsif (exists $fin->{$col}) { |
1117 | $fin->{$col} = [ -and => map { |
1118 | (ref $_ eq 'ARRAY' and ($_->[0]||'') =~ /^\-and$/i ) |
1119 | ? @{$_}[1..$#$_] |
1120 | : $_ |
1121 | ; |
1122 | } ($fin->{$col}, $c->{$col}) ]; |
8d005ad9 |
1123 | } |
1124 | else { |
1125 | $fin->{$col} = $c->{$col}; |
1126 | } |
1127 | } |
1128 | } |
1129 | } |
8d005ad9 |
1130 | } |
1131 | elsif (ref $where eq 'ARRAY') { |
22485a7e |
1132 | # we are always at top-level here, it is safe to dump empty *standalone* pieces |
1133 | my $fin_idx; |
8d005ad9 |
1134 | |
22485a7e |
1135 | for (my $i = 0; $i <= $#$where; $i++ ) { |
8d005ad9 |
1136 | |
e466c62b |
1137 | # Match SQLA 1.79 behavior |
1138 | $self->throw_exception( |
1139 | "Supplying an empty left hand side argument is not supported in array-pairs" |
1140 | ) if (! defined $where->[$i] or ! length $where->[$i]); |
1141 | |
22485a7e |
1142 | my $logic_mod = lc ( ($where->[$i] =~ /^(\-(?:and|or))$/i)[0] || '' ); |
1143 | |
1144 | if ($logic_mod) { |
1145 | $i++; |
1146 | $self->throw_exception("Unsupported top-level op/arg pair: [ $logic_mod => $where->[$i] ]") |
1147 | unless ref $where->[$i] eq 'HASH' or ref $where->[$i] eq 'ARRAY'; |
1148 | |
1149 | my $sub_elt = $self->_collapse_cond({ $logic_mod => $where->[$i] }) |
1150 | or next; |
1151 | |
e466c62b |
1152 | my @keys = keys %$sub_elt; |
1153 | if ( @keys == 1 and $keys[0] !~ /^\-/ ) { |
1154 | $fin_idx->{ "COL_$keys[0]_" . serialize $sub_elt } = $sub_elt; |
1155 | } |
1156 | else { |
1157 | $fin_idx->{ "SER_" . serialize $sub_elt } = $sub_elt; |
1158 | } |
22485a7e |
1159 | } |
1160 | elsif (! length ref $where->[$i] ) { |
135ac69d |
1161 | my $sub_elt = $self->_collapse_cond({ @{$where}[$i, $i+1] }) |
1162 | or next; |
1163 | |
1164 | $fin_idx->{ "COL_$where->[$i]_" . serialize $sub_elt } = $sub_elt; |
22485a7e |
1165 | $i++; |
8d005ad9 |
1166 | } |
1167 | else { |
135ac69d |
1168 | $fin_idx->{ "SER_" . serialize $where->[$i] } = $self->_collapse_cond( $where->[$i] ) || next; |
8d005ad9 |
1169 | } |
1170 | } |
22485a7e |
1171 | |
07add744 |
1172 | if (! $fin_idx) { |
1173 | return; |
1174 | } |
1175 | elsif ( keys %$fin_idx == 1 ) { |
1176 | $fin = (values %$fin_idx)[0]; |
1177 | } |
1178 | else { |
1179 | my @or; |
1180 | |
1181 | # at this point everything is at most one level deep - unroll if needed |
1182 | for (sort keys %$fin_idx) { |
1183 | if ( ref $fin_idx->{$_} eq 'HASH' and keys %{$fin_idx->{$_}} == 1 ) { |
1184 | my ($l, $r) = %{$fin_idx->{$_}}; |
1185 | |
1186 | if ( |
1187 | ref $r eq 'ARRAY' |
1188 | and |
1189 | ( |
1190 | ( @$r == 1 and $l =~ /^\-and$/i ) |
1191 | or |
1192 | $l =~ /^\-or$/i |
1193 | ) |
1194 | ) { |
1195 | push @or, @$r |
1196 | } |
1197 | |
1198 | elsif ( |
1199 | ref $r eq 'HASH' |
1200 | and |
1201 | keys %$r == 1 |
1202 | and |
1203 | $l =~ /^\-(?:and|or)$/i |
1204 | ) { |
1205 | push @or, %$r; |
1206 | } |
1207 | |
1208 | else { |
1209 | push @or, $l, $r; |
1210 | } |
1211 | } |
1212 | else { |
1213 | push @or, $fin_idx->{$_}; |
1214 | } |
1215 | } |
1216 | |
1217 | $fin->{-or} = \@or; |
1218 | } |
8d005ad9 |
1219 | } |
1220 | else { |
1221 | # not a hash not an array |
07add744 |
1222 | $fin = { -and => [ $where ] }; |
135ac69d |
1223 | } |
1224 | |
1225 | # unroll single-element -and's |
1226 | while ( |
1227 | $fin->{-and} |
1228 | and |
1229 | @{$fin->{-and}} < 2 |
1230 | ) { |
1231 | my $and = delete $fin->{-and}; |
1232 | last if @$and == 0; |
1233 | |
1234 | # at this point we have @$and == 1 |
1235 | if ( |
1236 | ref $and->[0] eq 'HASH' |
1237 | and |
1238 | ! grep { exists $fin->{$_} } keys %{$and->[0]} |
1239 | ) { |
1240 | $fin = { |
1241 | %$fin, %{$and->[0]} |
1242 | }; |
1243 | } |
07add744 |
1244 | else { |
1245 | $fin->{-and} = $and; |
1246 | last; |
1247 | } |
135ac69d |
1248 | } |
1249 | |
1250 | # compress same-column conds found in $fin |
1251 | for my $col ( grep { $_ !~ /^\-/ } keys %$fin ) { |
1252 | next unless ref $fin->{$col} eq 'ARRAY' and ($fin->{$col}[0]||'') =~ /^\-and$/i; |
1253 | my $val_bag = { map { |
5379386e |
1254 | (! defined $_ ) ? ( UNDEF => undef ) |
1255 | : ( ! length ref $_ or is_plain_value $_ ) ? ( "VAL_$_" => $_ ) |
135ac69d |
1256 | : ( ( 'SER_' . serialize $_ ) => $_ ) |
1257 | } @{$fin->{$col}}[1 .. $#{$fin->{$col}}] }; |
1258 | |
1259 | if (keys %$val_bag == 1 ) { |
1260 | ($fin->{$col}) = values %$val_bag; |
1261 | } |
1262 | else { |
1263 | $fin->{$col} = [ -and => map { $val_bag->{$_} } sort keys %$val_bag ]; |
1264 | } |
8d005ad9 |
1265 | } |
1266 | |
135ac69d |
1267 | return keys %$fin ? $fin : (); |
8d005ad9 |
1268 | } |
1269 | |
1270 | sub _collapse_cond_unroll_pairs { |
1271 | my ($self, $pairs) = @_; |
1272 | |
1273 | my @conds; |
1274 | |
1275 | while (@$pairs) { |
6565d2c3 |
1276 | my ($lhs, $rhs) = splice @$pairs, 0, 2; |
8d005ad9 |
1277 | |
1278 | if ($lhs eq '') { |
1279 | push @conds, $self->_collapse_cond($rhs); |
1280 | } |
1281 | elsif ( $lhs =~ /^\-and$/i ) { |
1282 | push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY')); |
1283 | } |
1284 | elsif ( $lhs =~ /^\-or$/i ) { |
1285 | push @conds, $self->_collapse_cond( |
1286 | (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs |
1287 | ); |
1288 | } |
1289 | else { |
1290 | if (ref $rhs eq 'HASH' and ! keys %$rhs) { |
1291 | # FIXME - SQLA seems to be doing... nothing...? |
1292 | } |
f6fff270 |
1293 | # normalize top level -ident, for saner extract_fixed_condition_columns code |
5f35ba0f |
1294 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) { |
1295 | push @conds, { $lhs => { '=', $rhs } }; |
1296 | } |
1297 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-value} and is_plain_value $rhs->{-value}) { |
1298 | push @conds, { $lhs => $rhs->{-value} }; |
1299 | } |
8d005ad9 |
1300 | elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) { |
f6fff270 |
1301 | if ( length ref $rhs->{'='} and is_literal_value $rhs->{'='} ) { |
5f35ba0f |
1302 | push @conds, { $lhs => $rhs }; |
1303 | } |
1304 | else { |
6565d2c3 |
1305 | for my $p ($self->_collapse_cond_unroll_pairs([ $lhs => $rhs->{'='} ])) { |
5f35ba0f |
1306 | |
1307 | # extra sanity check |
1308 | if (keys %$p > 1) { |
1309 | require Data::Dumper::Concise; |
1310 | local $Data::Dumper::Deepcopy = 1; |
1311 | $self->throw_exception( |
1312 | "Internal error: unexpected collapse unroll:" |
1313 | . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p } |
1314 | ); |
1315 | } |
8d005ad9 |
1316 | |
5f35ba0f |
1317 | my ($l, $r) = %$p; |
8d005ad9 |
1318 | |
f6fff270 |
1319 | push @conds, ( |
1320 | ! length ref $r |
1321 | or |
1322 | # the unroller recursion may return a '=' prepended value already |
1323 | ref $r eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='} |
1324 | or |
1325 | is_plain_value($r) |
1326 | ) |
5f35ba0f |
1327 | ? { $l => $r } |
1328 | : { $l => { '=' => $r } } |
1329 | ; |
1330 | } |
8d005ad9 |
1331 | } |
1332 | } |
1333 | elsif (ref $rhs eq 'ARRAY') { |
1334 | # some of these conditionals encounter multi-values - roll them out using |
1335 | # an unshift, which will cause extra looping in the while{} above |
1336 | if (! @$rhs ) { |
1337 | push @conds, { $lhs => [] }; |
1338 | } |
1339 | elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) { |
1340 | $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ") |
1341 | if @$rhs == 1; |
1342 | |
1343 | if( $rhs->[0] =~ /^\-and$/i ) { |
6565d2c3 |
1344 | unshift @$pairs, map { $lhs => $_ } @{$rhs}[1..$#$rhs]; |
8d005ad9 |
1345 | } |
1346 | # if not an AND then it's an OR |
1347 | elsif(@$rhs == 2) { |
6565d2c3 |
1348 | unshift @$pairs, $lhs => $rhs->[1]; |
8d005ad9 |
1349 | } |
1350 | else { |
953d5b7d |
1351 | push @conds, { $lhs => [ @{$rhs}[1..$#$rhs] ] }; |
8d005ad9 |
1352 | } |
1353 | } |
1354 | elsif (@$rhs == 1) { |
6565d2c3 |
1355 | unshift @$pairs, $lhs => $rhs->[0]; |
8d005ad9 |
1356 | } |
1357 | else { |
1358 | push @conds, { $lhs => $rhs }; |
1359 | } |
1360 | } |
c1f3f2e8 |
1361 | # unroll func + { -value => ... } |
1362 | elsif ( |
1363 | ref $rhs eq 'HASH' |
1364 | and |
1365 | ( my ($subop) = keys %$rhs ) == 1 |
1366 | and |
1367 | length ref ((values %$rhs)[0]) |
1368 | and |
1369 | my $vref = is_plain_value( (values %$rhs)[0] ) |
1370 | ) { |
5379386e |
1371 | push @conds, { $lhs => { $subop => $$vref } } |
c1f3f2e8 |
1372 | } |
8d005ad9 |
1373 | else { |
1374 | push @conds, { $lhs => $rhs }; |
1375 | } |
1376 | } |
1377 | } |
1378 | |
1379 | return @conds; |
1380 | } |
1381 | |
8e40a627 |
1382 | # Analyzes a given condition and attempts to extract all columns |
1383 | # with a definitive fixed-condition criteria. Returns a hashref |
1384 | # of k/v pairs suitable to be passed to set_columns(), with a |
1385 | # MAJOR CAVEAT - multi-value (contradictory) equalities are still |
1386 | # represented as a reference to the UNRESOVABLE_CONDITION constant |
1387 | # The reason we do this is that some codepaths only care about the |
1388 | # codition being stable, as opposed to actually making sense |
5f11e54f |
1389 | # |
8e40a627 |
1390 | # The normal mode is used to figure out if a resultset is constrained |
1391 | # to a column which is part of a unique constraint, which in turn |
1392 | # allows us to better predict how ordering will behave etc. |
1393 | # |
1394 | # With the optional "consider_nulls" boolean argument, the function |
1395 | # is instead used to infer inambiguous values from conditions |
1396 | # (e.g. the inheritance of resultset conditions on new_result) |
1397 | # |
5f11e54f |
1398 | sub _extract_fixed_condition_columns { |
8e40a627 |
1399 | my ($self, $where, $consider_nulls) = @_; |
1400 | my $where_hash = $self->_collapse_cond($_[1]); |
1401 | |
1402 | my $res = {}; |
1403 | my ($c, $v); |
1404 | for $c (keys %$where_hash) { |
1405 | my $vals; |
1406 | |
1407 | if (!defined ($v = $where_hash->{$c}) ) { |
b34d9331 |
1408 | $vals->{UNDEF} = $v if $consider_nulls |
8e40a627 |
1409 | } |
1410 | elsif ( |
8e40a627 |
1411 | ref $v eq 'HASH' |
1412 | and |
1413 | keys %$v == 1 |
5f35ba0f |
1414 | ) { |
1415 | if (exists $v->{-value}) { |
1416 | if (defined $v->{-value}) { |
b34d9331 |
1417 | $vals->{"VAL_$v->{-value}"} = $v->{-value} |
5f35ba0f |
1418 | } |
1419 | elsif( $consider_nulls ) { |
b34d9331 |
1420 | $vals->{UNDEF} = $v->{-value}; |
5f35ba0f |
1421 | } |
1422 | } |
8e40a627 |
1423 | # do not need to check for plain values - _collapse_cond did it for us |
f6fff270 |
1424 | elsif( |
1425 | length ref $v->{'='} |
1426 | and |
1427 | ( |
1428 | ( ref $v->{'='} eq 'HASH' and keys %{$v->{'='}} == 1 and exists $v->{'='}{-ident} ) |
1429 | or |
1430 | is_literal_value($v->{'='}) |
1431 | ) |
1432 | ) { |
b34d9331 |
1433 | $vals->{ 'SER_' . serialize $v->{'='} } = $v->{'='}; |
5f35ba0f |
1434 | } |
1435 | } |
1436 | elsif ( |
1437 | ! length ref $v |
1438 | or |
1439 | is_plain_value ($v) |
8e40a627 |
1440 | ) { |
b34d9331 |
1441 | $vals->{"VAL_$v"} = $v; |
8e40a627 |
1442 | } |
1443 | elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') { |
1444 | for ( @{$v}[1..$#$v] ) { |
1445 | my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls'); # always fish nulls out on recursion |
1446 | next unless exists $subval->{$c}; # didn't find anything |
b34d9331 |
1447 | $vals->{ |
1448 | ! defined $subval->{$c} ? 'UNDEF' |
1449 | : ( ! length ref $subval->{$c} or is_plain_value $subval->{$c} ) ? "VAL_$subval->{$c}" |
1450 | : ( 'SER_' . serialize $subval->{$c} ) |
1451 | } = $subval->{$c}; |
8d005ad9 |
1452 | } |
5f11e54f |
1453 | } |
8e40a627 |
1454 | |
1455 | if (keys %$vals == 1) { |
1456 | ($res->{$c}) = (values %$vals) |
b34d9331 |
1457 | unless !$consider_nulls and exists $vals->{UNDEF}; |
8e40a627 |
1458 | } |
1459 | elsif (keys %$vals > 1) { |
1460 | $res->{$c} = UNRESOLVABLE_CONDITION; |
1461 | } |
5f11e54f |
1462 | } |
8d005ad9 |
1463 | |
8e40a627 |
1464 | $res; |
c0748280 |
1465 | } |
bac6c4fb |
1466 | |
d28bb90d |
1467 | 1; |