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 |
7eb76996 |
7 | # display. The arrival of SQLA2 should immediately oboslete 90% of this |
d28bb90d |
8 | # |
9 | |
10 | use strict; |
11 | use warnings; |
12 | |
13 | use base 'DBIx::Class::Storage'; |
14 | use mro 'c3'; |
15 | |
16 | use Carp::Clan qw/^DBIx::Class/; |
17 | |
18 | # |
052e8431 |
19 | # This code will remove non-selecting/non-restricting joins from |
539ffe87 |
20 | # {from} specs, aiding the RDBMS query optimizer. |
052e8431 |
21 | # |
22 | sub _prune_unused_joins { |
23 | my $self = shift; |
24 | |
25 | my $from = shift; |
26 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY') { |
27 | return $from; # only standard {from} specs are supported |
28 | } |
29 | |
539ffe87 |
30 | my $aliastypes = $self->_resolve_aliastypes_from_select_args($from, @_); |
052e8431 |
31 | |
32 | my @newfrom = $from->[0]; # FROM head is always present |
33 | |
34 | my %need_joins = (map { %{$_||{}} } (values %$aliastypes) ); |
35 | for my $j (@{$from}[1..$#$from]) { |
539ffe87 |
36 | push @newfrom, $j if ( |
37 | ! $j->[0]{-alias} # legacy crap |
38 | || |
39 | $need_joins{$j->[0]{-alias}} |
40 | ); |
052e8431 |
41 | } |
42 | |
43 | return \@newfrom; |
44 | } |
45 | |
46 | |
47 | # |
d28bb90d |
48 | # This is the code producing joined subqueries like: |
49 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
50 | # |
51 | sub _adjust_select_args_for_complex_prefetch { |
52 | my ($self, $from, $select, $where, $attrs) = @_; |
53 | |
54 | $self->throw_exception ('Nothing to prefetch... how did we get here?!') |
55 | if not @{$attrs->{_prefetch_select}}; |
56 | |
57 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
58 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY'); |
59 | |
60 | |
61 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
62 | my $outer_attrs = { %$attrs }; |
63 | delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/; |
64 | |
65 | my $inner_attrs = { %$attrs }; |
66 | delete $inner_attrs->{$_} for qw/for collapse _prefetch_select _collapse_order_by select as/; |
67 | |
68 | |
69 | # bring over all non-collapse-induced order_by into the inner query (if any) |
70 | # the outer one will have to keep them all |
71 | delete $inner_attrs->{order_by}; |
72 | if (my $ord_cnt = @{$outer_attrs->{order_by}} - @{$outer_attrs->{_collapse_order_by}} ) { |
73 | $inner_attrs->{order_by} = [ |
74 | @{$outer_attrs->{order_by}}[ 0 .. $ord_cnt - 1] |
75 | ]; |
76 | } |
77 | |
d28bb90d |
78 | # generate the inner/outer select lists |
79 | # for inside we consider only stuff *not* brought in by the prefetch |
80 | # on the outside we substitute any function for its alias |
81 | my $outer_select = [ @$select ]; |
82 | my $inner_select = []; |
83 | for my $i (0 .. ( @$outer_select - @{$outer_attrs->{_prefetch_select}} - 1) ) { |
84 | my $sel = $outer_select->[$i]; |
85 | |
86 | if (ref $sel eq 'HASH' ) { |
87 | $sel->{-as} ||= $attrs->{as}[$i]; |
88 | $outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") ); |
89 | } |
90 | |
91 | push @$inner_select, $sel; |
92 | } |
93 | |
d28bb90d |
94 | # construct the inner $from for the subquery |
052e8431 |
95 | my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, $inner_attrs); |
ad630f4b |
96 | |
539ffe87 |
97 | # if a multi-type join was needed in the subquery - add a group_by to simulate the |
98 | # collapse in the subq |
99 | $inner_attrs->{group_by} ||= $inner_select |
100 | if List::Util::first |
101 | { ! $_->[0]{-is_single} } |
102 | (@{$inner_from}[1 .. $#$inner_from]) |
103 | ; |
d28bb90d |
104 | |
d28bb90d |
105 | # generate the subquery |
106 | my $subq = $self->_select_args_to_query ( |
052e8431 |
107 | $inner_from, |
d28bb90d |
108 | $inner_select, |
109 | $where, |
110 | $inner_attrs, |
111 | ); |
112 | |
113 | my $subq_joinspec = { |
114 | -alias => $attrs->{alias}, |
052e8431 |
115 | -source_handle => $inner_from->[0]{-source_handle}, |
d28bb90d |
116 | $attrs->{alias} => $subq, |
117 | }; |
118 | |
119 | # Generate the outer from - this is relatively easy (really just replace |
120 | # the join slot with the subquery), with a major caveat - we can not |
121 | # join anything that is non-selecting (not part of the prefetch), but at |
122 | # the same time is a multi-type relationship, as it will explode the result. |
123 | # |
124 | # There are two possibilities here |
125 | # - either the join is non-restricting, in which case we simply throw it away |
126 | # - it is part of the restrictions, in which case we need to collapse the outer |
127 | # result by tackling yet another group_by to the outside of the query |
128 | |
052e8431 |
129 | # normalize a copy of $from, so it will be easier to work with further |
130 | # down (i.e. promote the initial hashref to an AoH) |
131 | $from = [ @$from ]; |
132 | $from->[0] = [ $from->[0] ]; |
133 | |
d28bb90d |
134 | # so first generate the outer_from, up to the substitution point |
135 | my @outer_from; |
136 | while (my $j = shift @$from) { |
137 | if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap |
138 | push @outer_from, [ |
139 | $subq_joinspec, |
140 | @{$j}[1 .. $#$j], |
141 | ]; |
142 | last; # we'll take care of what's left in $from below |
143 | } |
144 | else { |
145 | push @outer_from, $j; |
146 | } |
147 | } |
148 | |
052e8431 |
149 | # scan the from spec against different attributes, and see which joins are needed |
150 | # in what role |
151 | my $outer_aliastypes = |
539ffe87 |
152 | $self->_resolve_aliastypes_from_select_args( $from, $outer_select, $where, $outer_attrs ); |
052e8431 |
153 | |
d28bb90d |
154 | # see what's left - throw away if not selecting/restricting |
155 | # also throw in a group_by if restricting to guard against |
156 | # cross-join explosions |
157 | # |
158 | while (my $j = shift @$from) { |
159 | my $alias = $j->[0]{-alias}; |
160 | |
964a3c71 |
161 | if ($outer_aliastypes->{select}{$alias}) { |
d28bb90d |
162 | push @outer_from, $j; |
163 | } |
964a3c71 |
164 | elsif ($outer_aliastypes->{restrict}{$alias}) { |
d28bb90d |
165 | push @outer_from, $j; |
539ffe87 |
166 | $outer_attrs->{group_by} ||= $outer_select unless $j->[0]{-is_single}; |
d28bb90d |
167 | } |
168 | } |
169 | |
170 | # demote the outer_from head |
171 | $outer_from[0] = $outer_from[0][0]; |
172 | |
173 | # This is totally horrific - the $where ends up in both the inner and outer query |
174 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
175 | # then if where conditions apply to the *right* side of the prefetch, you may have |
176 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
177 | # the outer select to exclude joins you didin't want in the first place |
178 | # |
179 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
180 | return (\@outer_from, $outer_select, $where, $outer_attrs); |
181 | } |
182 | |
ad630f4b |
183 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
184 | # select/where/order/group attributes, in order to determine what |
185 | # aliases are neded to fulfill the query. This information is used |
186 | # throughout the code to prune unnecessary JOINs from the queries |
187 | # in an attempt to reduce the execution time. |
188 | # Although the method is pretty horrific, the worst thing that can |
189 | # happen is for it to fail due to an unqualified column, which in |
190 | # turn will result in a vocal exception. Qualifying the column will |
191 | # invariably solve the problem. |
539ffe87 |
192 | sub _resolve_aliastypes_from_select_args { |
052e8431 |
193 | my ( $self, $from, $select, $where, $attrs ) = @_; |
546f1cd9 |
194 | |
ad630f4b |
195 | $self->throw_exception ('Unable to analyze custom {from}') |
196 | if ref $from ne 'ARRAY'; |
546f1cd9 |
197 | |
ad630f4b |
198 | # what we will return |
964a3c71 |
199 | my $aliases_by_type; |
546f1cd9 |
200 | |
ad630f4b |
201 | # see what aliases are there to work with |
202 | my $alias_list; |
539ffe87 |
203 | for (@$from) { |
204 | my $j = $_; |
ad630f4b |
205 | $j = $j->[0] if ref $j eq 'ARRAY'; |
539ffe87 |
206 | my $al = $j->{-alias} |
207 | or next; |
208 | |
209 | $alias_list->{$al} = $j; |
210 | $aliases_by_type->{multiplying}{$al} = 1 |
211 | unless $j->{-is_single}; |
546f1cd9 |
212 | } |
546f1cd9 |
213 | |
ad630f4b |
214 | # set up a botched SQLA |
215 | my $sql_maker = $self->sql_maker; |
216 | my $sep = quotemeta ($self->_sql_maker_opts->{name_sep} || '.'); |
217 | local $sql_maker->{quote_char}; # so that we can regex away |
07f31d19 |
218 | |
219 | |
ad630f4b |
220 | my $select_sql = $sql_maker->_recurse_fields ($select); |
221 | my $where_sql = $sql_maker->where ($where); |
222 | my $group_by_sql = $sql_maker->_order_by({ |
223 | map { $_ => $attrs->{$_} } qw/group_by having/ |
224 | }); |
225 | my @order_by_chunks = (map |
226 | { ref $_ ? $_->[0] : $_ } |
227 | $sql_maker->_order_by_chunks ($attrs->{order_by}) |
228 | ); |
07f31d19 |
229 | |
ad630f4b |
230 | # match every alias to the sql chunks above |
231 | for my $alias (keys %$alias_list) { |
232 | my $al_re = qr/\b $alias $sep/x; |
07f31d19 |
233 | |
ad630f4b |
234 | for my $piece ($where_sql, $group_by_sql) { |
964a3c71 |
235 | $aliases_by_type->{restrict}{$alias} = 1 if ($piece =~ $al_re); |
ad630f4b |
236 | } |
07f31d19 |
237 | |
ad630f4b |
238 | for my $piece ($select_sql, @order_by_chunks ) { |
964a3c71 |
239 | $aliases_by_type->{select}{$alias} = 1 if ($piece =~ $al_re); |
07f31d19 |
240 | } |
241 | } |
242 | |
243 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
244 | for my $j (values %$alias_list) { |
07f31d19 |
245 | my $alias = $j->{-alias} or next; |
964a3c71 |
246 | $aliases_by_type->{restrict}{$alias} = 1 if ( |
07f31d19 |
247 | (not $j->{-join_type}) |
248 | or |
249 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
250 | ); |
251 | } |
252 | |
253 | # mark all join parents as mentioned |
254 | # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too ) |
964a3c71 |
255 | for my $type (keys %$aliases_by_type) { |
256 | for my $alias (keys %{$aliases_by_type->{$type}}) { |
257 | $aliases_by_type->{$type}{$_} = 1 |
ad630f4b |
258 | for (@{ $alias_list->{$alias}{-join_path} || [] }); |
07f31d19 |
259 | } |
260 | } |
ad630f4b |
261 | |
964a3c71 |
262 | return $aliases_by_type; |
07f31d19 |
263 | } |
264 | |
d28bb90d |
265 | sub _resolve_ident_sources { |
266 | my ($self, $ident) = @_; |
267 | |
268 | my $alias2source = {}; |
269 | my $rs_alias; |
270 | |
271 | # the reason this is so contrived is that $ident may be a {from} |
272 | # structure, specifying multiple tables to join |
273 | if ( Scalar::Util::blessed($ident) && $ident->isa("DBIx::Class::ResultSource") ) { |
274 | # this is compat mode for insert/update/delete which do not deal with aliases |
275 | $alias2source->{me} = $ident; |
276 | $rs_alias = 'me'; |
277 | } |
278 | elsif (ref $ident eq 'ARRAY') { |
279 | |
280 | for (@$ident) { |
281 | my $tabinfo; |
282 | if (ref $_ eq 'HASH') { |
283 | $tabinfo = $_; |
284 | $rs_alias = $tabinfo->{-alias}; |
285 | } |
286 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
287 | $tabinfo = $_->[0]; |
288 | } |
289 | |
290 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-source_handle}->resolve |
291 | if ($tabinfo->{-source_handle}); |
292 | } |
293 | } |
294 | |
295 | return ($alias2source, $rs_alias); |
296 | } |
297 | |
298 | # Takes $ident, \@column_names |
299 | # |
300 | # returns { $column_name => \%column_info, ... } |
301 | # also note: this adds -result_source => $rsrc to the column info |
302 | # |
09e14fdc |
303 | # If no columns_names are supplied returns info about *all* columns |
304 | # for all sources |
d28bb90d |
305 | sub _resolve_column_info { |
306 | my ($self, $ident, $colnames) = @_; |
307 | my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); |
308 | |
309 | my $sep = $self->_sql_maker_opts->{name_sep} || '.'; |
09e14fdc |
310 | my $qsep = quotemeta $sep; |
d28bb90d |
311 | |
09e14fdc |
312 | my (%return, %seen_cols, @auto_colnames); |
d28bb90d |
313 | |
314 | # compile a global list of column names, to be able to properly |
315 | # disambiguate unqualified column names (if at all possible) |
316 | for my $alias (keys %$alias2src) { |
317 | my $rsrc = $alias2src->{$alias}; |
318 | for my $colname ($rsrc->columns) { |
319 | push @{$seen_cols{$colname}}, $alias; |
09e14fdc |
320 | push @auto_colnames, "$alias$sep$colname" unless $colnames; |
d28bb90d |
321 | } |
322 | } |
323 | |
09e14fdc |
324 | $colnames ||= [ |
325 | @auto_colnames, |
326 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
327 | ]; |
328 | |
d28bb90d |
329 | COLUMN: |
330 | foreach my $col (@$colnames) { |
09e14fdc |
331 | my ($alias, $colname) = $col =~ m/^ (?: ([^$qsep]+) $qsep)? (.+) $/x; |
d28bb90d |
332 | |
333 | unless ($alias) { |
334 | # see if the column was seen exactly once (so we know which rsrc it came from) |
335 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1) { |
336 | $alias = $seen_cols{$colname}[0]; |
337 | } |
338 | else { |
339 | next COLUMN; |
340 | } |
341 | } |
342 | |
343 | my $rsrc = $alias2src->{$alias}; |
344 | $return{$col} = $rsrc && { |
345 | %{$rsrc->column_info($colname)}, |
346 | -result_source => $rsrc, |
347 | -source_alias => $alias, |
348 | }; |
349 | } |
350 | |
351 | return \%return; |
352 | } |
353 | |
289ac713 |
354 | # The DBIC relationship chaining implementation is pretty simple - every |
355 | # new related_relationship is pushed onto the {from} stack, and the {select} |
356 | # window simply slides further in. This means that when we count somewhere |
357 | # in the middle, we got to make sure that everything in the join chain is an |
358 | # actual inner join, otherwise the count will come back with unpredictable |
359 | # results (a resultset may be generated with _some_ rows regardless of if |
360 | # the relation which the $rs currently selects has rows or not). E.g. |
361 | # $artist_rs->cds->count - normally generates: |
362 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
363 | # which actually returns the number of artists * (number of cds || 1) |
364 | # |
365 | # So what we do here is crawl {from}, determine if the current alias is at |
366 | # the top of the stack, and if not - make sure the chain is inner-joined down |
367 | # to the root. |
368 | # |
369 | sub _straight_join_to_node { |
370 | my ($self, $from, $alias) = @_; |
371 | |
372 | # subqueries and other oddness are naturally not supported |
373 | return $from if ( |
374 | ref $from ne 'ARRAY' |
375 | || |
376 | @$from <= 1 |
377 | || |
378 | ref $from->[0] ne 'HASH' |
379 | || |
380 | ! $from->[0]{-alias} |
381 | || |
7eb76996 |
382 | $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do |
289ac713 |
383 | ); |
384 | |
385 | # find the current $alias in the $from structure |
386 | my $switch_branch; |
387 | JOINSCAN: |
388 | for my $j (@{$from}[1 .. $#$from]) { |
389 | if ($j->[0]{-alias} eq $alias) { |
390 | $switch_branch = $j->[0]{-join_path}; |
391 | last JOINSCAN; |
392 | } |
393 | } |
394 | |
7eb76996 |
395 | # something else went quite wrong |
289ac713 |
396 | return $from unless $switch_branch; |
397 | |
398 | # So it looks like we will have to switch some stuff around. |
399 | # local() is useless here as we will be leaving the scope |
400 | # anyway, and deep cloning is just too fucking expensive |
7eb76996 |
401 | # So replace the first hashref in the node arrayref manually |
289ac713 |
402 | my @new_from = ($from->[0]); |
403 | my $sw_idx = { map { $_ => 1 } @$switch_branch }; |
404 | |
405 | for my $j (@{$from}[1 .. $#$from]) { |
406 | my $jalias = $j->[0]{-alias}; |
407 | |
408 | if ($sw_idx->{$jalias}) { |
409 | my %attrs = %{$j->[0]}; |
410 | delete $attrs{-join_type}; |
411 | push @new_from, [ |
412 | \%attrs, |
413 | @{$j}[ 1 .. $#$j ], |
414 | ]; |
415 | } |
416 | else { |
417 | push @new_from, $j; |
418 | } |
419 | } |
420 | |
421 | return \@new_from; |
422 | } |
423 | |
bac6c4fb |
424 | # Most databases do not allow aliasing of tables in UPDATE/DELETE. Thus |
425 | # a condition containing 'me' or other table prefixes will not work |
426 | # at all. What this code tries to do (badly) is introspect the condition |
427 | # and remove all column qualifiers. If it bails out early (returns undef) |
428 | # the calling code should try another approach (e.g. a subquery) |
429 | sub _strip_cond_qualifiers { |
430 | my ($self, $where) = @_; |
431 | |
432 | my $cond = {}; |
433 | |
434 | # No-op. No condition, we're updating/deleting everything |
435 | return $cond unless $where; |
436 | |
437 | if (ref $where eq 'ARRAY') { |
438 | $cond = [ |
439 | map { |
440 | my %hash; |
441 | foreach my $key (keys %{$_}) { |
442 | $key =~ /([^.]+)$/; |
443 | $hash{$1} = $_->{$key}; |
444 | } |
445 | \%hash; |
446 | } @$where |
447 | ]; |
448 | } |
449 | elsif (ref $where eq 'HASH') { |
450 | if ( (keys %$where) == 1 && ( (keys %{$where})[0] eq '-and' )) { |
451 | $cond->{-and} = []; |
452 | my @cond = @{$where->{-and}}; |
453 | for (my $i = 0; $i < @cond; $i++) { |
454 | my $entry = $cond[$i]; |
455 | my $hash; |
456 | if (ref $entry eq 'HASH') { |
457 | $hash = $self->_strip_cond_qualifiers($entry); |
458 | } |
459 | else { |
460 | $entry =~ /([^.]+)$/; |
461 | $hash->{$1} = $cond[++$i]; |
462 | } |
463 | push @{$cond->{-and}}, $hash; |
464 | } |
465 | } |
466 | else { |
467 | foreach my $key (keys %$where) { |
468 | $key =~ /([^.]+)$/; |
469 | $cond->{$1} = $where->{$key}; |
470 | } |
471 | } |
472 | } |
473 | else { |
474 | return undef; |
475 | } |
476 | |
477 | return $cond; |
478 | } |
479 | |
480 | |
d28bb90d |
481 | 1; |