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 |
20 | # {from} specs, aiding the RDBMS query optimizer. It will leave any |
21 | # unused type-multi joins, if the amount of returned rows is |
22 | # important (i.e. count without collapse) |
23 | # |
24 | sub _prune_unused_joins { |
25 | my $self = shift; |
26 | |
27 | my $from = shift; |
28 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY') { |
29 | return $from; # only standard {from} specs are supported |
30 | } |
31 | |
32 | my $aliastypes = $self->_resolve_aliases_from_select_args($from, @_); |
33 | |
34 | my @newfrom = $from->[0]; # FROM head is always present |
35 | |
36 | my %need_joins = (map { %{$_||{}} } (values %$aliastypes) ); |
37 | for my $j (@{$from}[1..$#$from]) { |
38 | push @newfrom, $j if $need_joins{$j->[0]{-alias}}; |
39 | } |
40 | |
41 | return \@newfrom; |
42 | } |
43 | |
44 | |
45 | # |
d28bb90d |
46 | # This is the code producing joined subqueries like: |
47 | # SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ... |
48 | # |
49 | sub _adjust_select_args_for_complex_prefetch { |
50 | my ($self, $from, $select, $where, $attrs) = @_; |
51 | |
52 | $self->throw_exception ('Nothing to prefetch... how did we get here?!') |
53 | if not @{$attrs->{_prefetch_select}}; |
54 | |
55 | $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') |
56 | if (ref $from ne 'ARRAY' || ref $from->[0] ne 'HASH' || ref $from->[1] ne 'ARRAY'); |
57 | |
58 | |
59 | # generate inner/outer attribute lists, remove stuff that doesn't apply |
60 | my $outer_attrs = { %$attrs }; |
61 | delete $outer_attrs->{$_} for qw/where bind rows offset group_by having/; |
62 | |
63 | my $inner_attrs = { %$attrs }; |
64 | delete $inner_attrs->{$_} for qw/for collapse _prefetch_select _collapse_order_by select as/; |
65 | |
66 | |
67 | # bring over all non-collapse-induced order_by into the inner query (if any) |
68 | # the outer one will have to keep them all |
69 | delete $inner_attrs->{order_by}; |
70 | if (my $ord_cnt = @{$outer_attrs->{order_by}} - @{$outer_attrs->{_collapse_order_by}} ) { |
71 | $inner_attrs->{order_by} = [ |
72 | @{$outer_attrs->{order_by}}[ 0 .. $ord_cnt - 1] |
73 | ]; |
74 | } |
75 | |
d28bb90d |
76 | # generate the inner/outer select lists |
77 | # for inside we consider only stuff *not* brought in by the prefetch |
78 | # on the outside we substitute any function for its alias |
79 | my $outer_select = [ @$select ]; |
80 | my $inner_select = []; |
81 | for my $i (0 .. ( @$outer_select - @{$outer_attrs->{_prefetch_select}} - 1) ) { |
82 | my $sel = $outer_select->[$i]; |
83 | |
84 | if (ref $sel eq 'HASH' ) { |
85 | $sel->{-as} ||= $attrs->{as}[$i]; |
86 | $outer_select->[$i] = join ('.', $attrs->{alias}, ($sel->{-as} || "inner_column_$i") ); |
87 | } |
88 | |
89 | push @$inner_select, $sel; |
90 | } |
91 | |
d28bb90d |
92 | # construct the inner $from for the subquery |
052e8431 |
93 | my $inner_from = $self->_prune_unused_joins ($from, $inner_select, $where, $inner_attrs); |
ad630f4b |
94 | |
d28bb90d |
95 | # if a multi-type join was needed in the subquery ("multi" is indicated by |
96 | # presence in {collapse}) - add a group_by to simulate the collapse in the subq |
97 | unless ($inner_attrs->{group_by}) { |
052e8431 |
98 | for my $alias (map { $_->[0]{-alias} } (@{$inner_from}[1 .. $#$inner_from]) ) { |
d28bb90d |
99 | |
100 | # the dot comes from some weirdness in collapse |
101 | # remove after the rewrite |
102 | if ($attrs->{collapse}{".$alias"}) { |
103 | $inner_attrs->{group_by} ||= $inner_select; |
104 | last; |
105 | } |
106 | } |
107 | } |
108 | |
d28bb90d |
109 | # generate the subquery |
110 | my $subq = $self->_select_args_to_query ( |
052e8431 |
111 | $inner_from, |
d28bb90d |
112 | $inner_select, |
113 | $where, |
114 | $inner_attrs, |
115 | ); |
116 | |
117 | my $subq_joinspec = { |
118 | -alias => $attrs->{alias}, |
052e8431 |
119 | -source_handle => $inner_from->[0]{-source_handle}, |
d28bb90d |
120 | $attrs->{alias} => $subq, |
121 | }; |
122 | |
123 | # Generate the outer from - this is relatively easy (really just replace |
124 | # the join slot with the subquery), with a major caveat - we can not |
125 | # join anything that is non-selecting (not part of the prefetch), but at |
126 | # the same time is a multi-type relationship, as it will explode the result. |
127 | # |
128 | # There are two possibilities here |
129 | # - either the join is non-restricting, in which case we simply throw it away |
130 | # - it is part of the restrictions, in which case we need to collapse the outer |
131 | # result by tackling yet another group_by to the outside of the query |
132 | |
052e8431 |
133 | # normalize a copy of $from, so it will be easier to work with further |
134 | # down (i.e. promote the initial hashref to an AoH) |
135 | $from = [ @$from ]; |
136 | $from->[0] = [ $from->[0] ]; |
137 | |
d28bb90d |
138 | # so first generate the outer_from, up to the substitution point |
139 | my @outer_from; |
140 | while (my $j = shift @$from) { |
141 | if ($j->[0]{-alias} eq $attrs->{alias}) { # time to swap |
142 | push @outer_from, [ |
143 | $subq_joinspec, |
144 | @{$j}[1 .. $#$j], |
145 | ]; |
146 | last; # we'll take care of what's left in $from below |
147 | } |
148 | else { |
149 | push @outer_from, $j; |
150 | } |
151 | } |
152 | |
052e8431 |
153 | # scan the from spec against different attributes, and see which joins are needed |
154 | # in what role |
155 | my $outer_aliastypes = |
156 | $self->_resolve_aliases_from_select_args( $from, $outer_select, $where, $outer_attrs ); |
157 | |
d28bb90d |
158 | # see what's left - throw away if not selecting/restricting |
159 | # also throw in a group_by if restricting to guard against |
160 | # cross-join explosions |
161 | # |
162 | while (my $j = shift @$from) { |
163 | my $alias = $j->[0]{-alias}; |
164 | |
964a3c71 |
165 | if ($outer_aliastypes->{select}{$alias}) { |
d28bb90d |
166 | push @outer_from, $j; |
167 | } |
964a3c71 |
168 | elsif ($outer_aliastypes->{restrict}{$alias}) { |
d28bb90d |
169 | push @outer_from, $j; |
170 | |
171 | # FIXME - this should be obviated by SQLA2, as I'll be able to |
172 | # have restrict_inner and restrict_outer... or something to that |
173 | # effect... I think... |
174 | |
175 | # FIXME2 - I can't find a clean way to determine if a particular join |
176 | # is a multi - instead I am just treating everything as a potential |
177 | # explosive join (ribasushi) |
178 | # |
179 | # if (my $handle = $j->[0]{-source_handle}) { |
180 | # my $rsrc = $handle->resolve; |
181 | # ... need to bail out of the following if this is not a multi, |
182 | # as it will be much easier on the db ... |
183 | |
184 | $outer_attrs->{group_by} ||= $outer_select; |
185 | # } |
186 | } |
187 | } |
188 | |
189 | # demote the outer_from head |
190 | $outer_from[0] = $outer_from[0][0]; |
191 | |
192 | # This is totally horrific - the $where ends up in both the inner and outer query |
193 | # Unfortunately not much can be done until SQLA2 introspection arrives, and even |
194 | # then if where conditions apply to the *right* side of the prefetch, you may have |
195 | # to both filter the inner select (e.g. to apply a limit) and then have to re-filter |
196 | # the outer select to exclude joins you didin't want in the first place |
197 | # |
198 | # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;) |
199 | return (\@outer_from, $outer_select, $where, $outer_attrs); |
200 | } |
201 | |
ad630f4b |
202 | # Due to a lack of SQLA2 we fall back to crude scans of all the |
203 | # select/where/order/group attributes, in order to determine what |
204 | # aliases are neded to fulfill the query. This information is used |
205 | # throughout the code to prune unnecessary JOINs from the queries |
206 | # in an attempt to reduce the execution time. |
207 | # Although the method is pretty horrific, the worst thing that can |
208 | # happen is for it to fail due to an unqualified column, which in |
209 | # turn will result in a vocal exception. Qualifying the column will |
210 | # invariably solve the problem. |
546f1cd9 |
211 | sub _resolve_aliases_from_select_args { |
052e8431 |
212 | my ( $self, $from, $select, $where, $attrs ) = @_; |
546f1cd9 |
213 | |
ad630f4b |
214 | $self->throw_exception ('Unable to analyze custom {from}') |
215 | if ref $from ne 'ARRAY'; |
546f1cd9 |
216 | |
ad630f4b |
217 | # what we will return |
964a3c71 |
218 | my $aliases_by_type; |
546f1cd9 |
219 | |
ad630f4b |
220 | # see what aliases are there to work with |
221 | my $alias_list; |
222 | my @from = @$from; # if I don't copy weird shit happens |
223 | for my $j (@from) { |
224 | $j = $j->[0] if ref $j eq 'ARRAY'; |
225 | $alias_list->{$j->{-alias}} = $j; |
546f1cd9 |
226 | } |
546f1cd9 |
227 | |
ad630f4b |
228 | # set up a botched SQLA |
229 | my $sql_maker = $self->sql_maker; |
230 | my $sep = quotemeta ($self->_sql_maker_opts->{name_sep} || '.'); |
231 | local $sql_maker->{quote_char}; # so that we can regex away |
07f31d19 |
232 | |
233 | |
ad630f4b |
234 | my $select_sql = $sql_maker->_recurse_fields ($select); |
235 | my $where_sql = $sql_maker->where ($where); |
236 | my $group_by_sql = $sql_maker->_order_by({ |
237 | map { $_ => $attrs->{$_} } qw/group_by having/ |
238 | }); |
239 | my @order_by_chunks = (map |
240 | { ref $_ ? $_->[0] : $_ } |
241 | $sql_maker->_order_by_chunks ($attrs->{order_by}) |
242 | ); |
07f31d19 |
243 | |
ad630f4b |
244 | # match every alias to the sql chunks above |
245 | for my $alias (keys %$alias_list) { |
246 | my $al_re = qr/\b $alias $sep/x; |
07f31d19 |
247 | |
ad630f4b |
248 | for my $piece ($where_sql, $group_by_sql) { |
964a3c71 |
249 | $aliases_by_type->{restrict}{$alias} = 1 if ($piece =~ $al_re); |
ad630f4b |
250 | } |
07f31d19 |
251 | |
ad630f4b |
252 | for my $piece ($select_sql, @order_by_chunks ) { |
964a3c71 |
253 | $aliases_by_type->{select}{$alias} = 1 if ($piece =~ $al_re); |
07f31d19 |
254 | } |
255 | } |
256 | |
257 | # Add any non-left joins to the restriction list (such joins are indeed restrictions) |
ad630f4b |
258 | for my $j (values %$alias_list) { |
07f31d19 |
259 | my $alias = $j->{-alias} or next; |
964a3c71 |
260 | $aliases_by_type->{restrict}{$alias} = 1 if ( |
07f31d19 |
261 | (not $j->{-join_type}) |
262 | or |
263 | ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi) |
264 | ); |
265 | } |
266 | |
267 | # mark all join parents as mentioned |
268 | # (e.g. join => { cds => 'tracks' } - tracks will need to bring cds too ) |
964a3c71 |
269 | for my $type (keys %$aliases_by_type) { |
270 | for my $alias (keys %{$aliases_by_type->{$type}}) { |
271 | $aliases_by_type->{$type}{$_} = 1 |
ad630f4b |
272 | for (@{ $alias_list->{$alias}{-join_path} || [] }); |
07f31d19 |
273 | } |
274 | } |
ad630f4b |
275 | |
964a3c71 |
276 | return $aliases_by_type; |
07f31d19 |
277 | } |
278 | |
d28bb90d |
279 | sub _resolve_ident_sources { |
280 | my ($self, $ident) = @_; |
281 | |
282 | my $alias2source = {}; |
283 | my $rs_alias; |
284 | |
285 | # the reason this is so contrived is that $ident may be a {from} |
286 | # structure, specifying multiple tables to join |
287 | if ( Scalar::Util::blessed($ident) && $ident->isa("DBIx::Class::ResultSource") ) { |
288 | # this is compat mode for insert/update/delete which do not deal with aliases |
289 | $alias2source->{me} = $ident; |
290 | $rs_alias = 'me'; |
291 | } |
292 | elsif (ref $ident eq 'ARRAY') { |
293 | |
294 | for (@$ident) { |
295 | my $tabinfo; |
296 | if (ref $_ eq 'HASH') { |
297 | $tabinfo = $_; |
298 | $rs_alias = $tabinfo->{-alias}; |
299 | } |
300 | if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') { |
301 | $tabinfo = $_->[0]; |
302 | } |
303 | |
304 | $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-source_handle}->resolve |
305 | if ($tabinfo->{-source_handle}); |
306 | } |
307 | } |
308 | |
309 | return ($alias2source, $rs_alias); |
310 | } |
311 | |
312 | # Takes $ident, \@column_names |
313 | # |
314 | # returns { $column_name => \%column_info, ... } |
315 | # also note: this adds -result_source => $rsrc to the column info |
316 | # |
09e14fdc |
317 | # If no columns_names are supplied returns info about *all* columns |
318 | # for all sources |
d28bb90d |
319 | sub _resolve_column_info { |
320 | my ($self, $ident, $colnames) = @_; |
321 | my ($alias2src, $root_alias) = $self->_resolve_ident_sources($ident); |
322 | |
323 | my $sep = $self->_sql_maker_opts->{name_sep} || '.'; |
09e14fdc |
324 | my $qsep = quotemeta $sep; |
d28bb90d |
325 | |
09e14fdc |
326 | my (%return, %seen_cols, @auto_colnames); |
d28bb90d |
327 | |
328 | # compile a global list of column names, to be able to properly |
329 | # disambiguate unqualified column names (if at all possible) |
330 | for my $alias (keys %$alias2src) { |
331 | my $rsrc = $alias2src->{$alias}; |
332 | for my $colname ($rsrc->columns) { |
333 | push @{$seen_cols{$colname}}, $alias; |
09e14fdc |
334 | push @auto_colnames, "$alias$sep$colname" unless $colnames; |
d28bb90d |
335 | } |
336 | } |
337 | |
09e14fdc |
338 | $colnames ||= [ |
339 | @auto_colnames, |
340 | grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols), |
341 | ]; |
342 | |
d28bb90d |
343 | COLUMN: |
344 | foreach my $col (@$colnames) { |
09e14fdc |
345 | my ($alias, $colname) = $col =~ m/^ (?: ([^$qsep]+) $qsep)? (.+) $/x; |
d28bb90d |
346 | |
347 | unless ($alias) { |
348 | # see if the column was seen exactly once (so we know which rsrc it came from) |
349 | if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1) { |
350 | $alias = $seen_cols{$colname}[0]; |
351 | } |
352 | else { |
353 | next COLUMN; |
354 | } |
355 | } |
356 | |
357 | my $rsrc = $alias2src->{$alias}; |
358 | $return{$col} = $rsrc && { |
359 | %{$rsrc->column_info($colname)}, |
360 | -result_source => $rsrc, |
361 | -source_alias => $alias, |
362 | }; |
363 | } |
364 | |
365 | return \%return; |
366 | } |
367 | |
289ac713 |
368 | # The DBIC relationship chaining implementation is pretty simple - every |
369 | # new related_relationship is pushed onto the {from} stack, and the {select} |
370 | # window simply slides further in. This means that when we count somewhere |
371 | # in the middle, we got to make sure that everything in the join chain is an |
372 | # actual inner join, otherwise the count will come back with unpredictable |
373 | # results (a resultset may be generated with _some_ rows regardless of if |
374 | # the relation which the $rs currently selects has rows or not). E.g. |
375 | # $artist_rs->cds->count - normally generates: |
376 | # SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid |
377 | # which actually returns the number of artists * (number of cds || 1) |
378 | # |
379 | # So what we do here is crawl {from}, determine if the current alias is at |
380 | # the top of the stack, and if not - make sure the chain is inner-joined down |
381 | # to the root. |
382 | # |
383 | sub _straight_join_to_node { |
384 | my ($self, $from, $alias) = @_; |
385 | |
386 | # subqueries and other oddness are naturally not supported |
387 | return $from if ( |
388 | ref $from ne 'ARRAY' |
389 | || |
390 | @$from <= 1 |
391 | || |
392 | ref $from->[0] ne 'HASH' |
393 | || |
394 | ! $from->[0]{-alias} |
395 | || |
7eb76996 |
396 | $from->[0]{-alias} eq $alias # this last bit means $alias is the head of $from - nothing to do |
289ac713 |
397 | ); |
398 | |
399 | # find the current $alias in the $from structure |
400 | my $switch_branch; |
401 | JOINSCAN: |
402 | for my $j (@{$from}[1 .. $#$from]) { |
403 | if ($j->[0]{-alias} eq $alias) { |
404 | $switch_branch = $j->[0]{-join_path}; |
405 | last JOINSCAN; |
406 | } |
407 | } |
408 | |
7eb76996 |
409 | # something else went quite wrong |
289ac713 |
410 | return $from unless $switch_branch; |
411 | |
412 | # So it looks like we will have to switch some stuff around. |
413 | # local() is useless here as we will be leaving the scope |
414 | # anyway, and deep cloning is just too fucking expensive |
7eb76996 |
415 | # So replace the first hashref in the node arrayref manually |
289ac713 |
416 | my @new_from = ($from->[0]); |
417 | my $sw_idx = { map { $_ => 1 } @$switch_branch }; |
418 | |
419 | for my $j (@{$from}[1 .. $#$from]) { |
420 | my $jalias = $j->[0]{-alias}; |
421 | |
422 | if ($sw_idx->{$jalias}) { |
423 | my %attrs = %{$j->[0]}; |
424 | delete $attrs{-join_type}; |
425 | push @new_from, [ |
426 | \%attrs, |
427 | @{$j}[ 1 .. $#$j ], |
428 | ]; |
429 | } |
430 | else { |
431 | push @new_from, $j; |
432 | } |
433 | } |
434 | |
435 | return \@new_from; |
436 | } |
437 | |
bac6c4fb |
438 | # Most databases do not allow aliasing of tables in UPDATE/DELETE. Thus |
439 | # a condition containing 'me' or other table prefixes will not work |
440 | # at all. What this code tries to do (badly) is introspect the condition |
441 | # and remove all column qualifiers. If it bails out early (returns undef) |
442 | # the calling code should try another approach (e.g. a subquery) |
443 | sub _strip_cond_qualifiers { |
444 | my ($self, $where) = @_; |
445 | |
446 | my $cond = {}; |
447 | |
448 | # No-op. No condition, we're updating/deleting everything |
449 | return $cond unless $where; |
450 | |
451 | if (ref $where eq 'ARRAY') { |
452 | $cond = [ |
453 | map { |
454 | my %hash; |
455 | foreach my $key (keys %{$_}) { |
456 | $key =~ /([^.]+)$/; |
457 | $hash{$1} = $_->{$key}; |
458 | } |
459 | \%hash; |
460 | } @$where |
461 | ]; |
462 | } |
463 | elsif (ref $where eq 'HASH') { |
464 | if ( (keys %$where) == 1 && ( (keys %{$where})[0] eq '-and' )) { |
465 | $cond->{-and} = []; |
466 | my @cond = @{$where->{-and}}; |
467 | for (my $i = 0; $i < @cond; $i++) { |
468 | my $entry = $cond[$i]; |
469 | my $hash; |
470 | if (ref $entry eq 'HASH') { |
471 | $hash = $self->_strip_cond_qualifiers($entry); |
472 | } |
473 | else { |
474 | $entry =~ /([^.]+)$/; |
475 | $hash->{$1} = $cond[++$i]; |
476 | } |
477 | push @{$cond->{-and}}, $hash; |
478 | } |
479 | } |
480 | else { |
481 | foreach my $key (keys %$where) { |
482 | $key =~ /([^.]+)$/; |
483 | $cond->{$1} = $where->{$key}; |
484 | } |
485 | } |
486 | } |
487 | else { |
488 | return undef; |
489 | } |
490 | |
491 | return $cond; |
492 | } |
493 | |
494 | |
d28bb90d |
495 | 1; |