Commit | Line | Data |
218aa968 |
1 | use strict; |
c4af4c4e |
2 | use warnings; |
218aa968 |
3 | |
4 | use Test::More; |
5 | use Test::Exception; |
6 | use lib qw(t/lib); |
a5a7bb73 |
7 | use DBICTest ':DiffSQL'; |
218aa968 |
8 | |
9 | my $schema = DBICTest->init_schema(); |
10 | |
218aa968 |
11 | my $rs = $schema->resultset('CD')->search({}, |
12 | { |
13 | '+select' => \ 'COUNT(*)', |
14 | '+as' => 'count' |
15 | } |
16 | ); |
17 | lives_ok(sub { $rs->first->get_column('count') }, 'additional count rscolumn present'); |
18 | dies_ok(sub { $rs->first->get_column('nonexistent_column') }, 'nonexistant column requests still throw exceptions'); |
19 | |
20 | $rs = $schema->resultset('CD')->search({}, |
21 | { |
22 | '+select' => [ \ 'COUNT(*)', 'title' ], |
23 | '+as' => [ 'count', 'addedtitle' ] |
24 | } |
25 | ); |
26 | lives_ok(sub { $rs->first->get_column('count') }, 'multiple +select/+as columns, 1st rscolumn present'); |
27 | lives_ok(sub { $rs->first->get_column('addedtitle') }, 'multiple +select/+as columns, 2nd rscolumn present'); |
28 | |
29 | $rs = $schema->resultset('CD')->search({}, |
30 | { |
31 | '+select' => [ \ 'COUNT(*)', 'title' ], |
32 | '+as' => [ 'count', 'addedtitle' ] |
33 | } |
34 | )->search({}, |
35 | { |
36 | '+select' => 'title', |
37 | '+as' => 'addedtitle2' |
38 | } |
39 | ); |
40 | lives_ok(sub { $rs->first->get_column('count') }, '+select/+as chained search 1st rscolumn present'); |
41 | lives_ok(sub { $rs->first->get_column('addedtitle') }, '+select/+as chained search 1st rscolumn present'); |
42 | lives_ok(sub { $rs->first->get_column('addedtitle2') }, '+select/+as chained search 3rd rscolumn present'); |
0eb27426 |
43 | |
44 | |
45 | # test the from search attribute (gets between the FROM and WHERE keywords, allows arbitrary subselects) |
46 | # also shows that outer select attributes are ok (i.e. order_by) |
47 | # |
48 | # from doesn't seem to be useful without using a scalarref - there were no initial tests >:( |
49 | # |
0eb27426 |
50 | my $cds = $schema->resultset ('CD')->search ({}, { order_by => 'me.cdid'}); # make sure order is consistent |
51 | cmp_ok ($cds->count, '>', 2, 'Initially populated with more than 2 CDs'); |
52 | |
53 | my $table = $cds->result_source->name; |
80625830 |
54 | $table = $$table if ref $table eq 'SCALAR'; |
0eb27426 |
55 | my $subsel = $cds->search ({}, { |
56 | columns => [qw/cdid title/], |
57 | from => \ "(SELECT cdid, title FROM $table LIMIT 2) me", |
58 | }); |
59 | |
60 | is ($subsel->count, 2, 'Subselect correctly limited the rs to 2 cds'); |
61 | is ($subsel->next->title, $cds->next->title, 'First CD title match'); |
62 | is ($subsel->next->title, $cds->next->title, 'Second CD title match'); |
bbdff861 |
63 | |
64 | is($schema->resultset('CD')->current_source_alias, "me", '$rs->current_source_alias returns "me"'); |
61d26fae |
65 | |
faf79003 |
66 | |
67 | |
68 | $rs = $schema->resultset('CD')->search({}, |
61d26fae |
69 | { |
70 | 'join' => 'artist', |
faf79003 |
71 | 'columns' => ['cdid', 'title', 'artist.name'], |
61d26fae |
72 | } |
73 | ); |
faf79003 |
74 | |
faf79003 |
75 | is_same_sql_bind ( |
af6aac2d |
76 | $rs->as_query, |
faf79003 |
77 | '(SELECT me.cdid, me.title, artist.name FROM cd me JOIN artist artist ON artist.artistid = me.artist)', |
78 | [], |
79 | 'Use of columns attribute results in proper sql' |
80 | ); |
81 | |
82 | lives_ok(sub { |
83 | $rs->first->get_column('cdid') |
84 | }, 'columns 1st rscolumn present'); |
85 | |
86 | lives_ok(sub { |
87 | $rs->first->get_column('title') |
88 | }, 'columns 2nd rscolumn present'); |
89 | |
0ca2f0d1 |
90 | lives_ok(sub { |
bbd828a6 |
91 | $rs->first->artist->get_column('name') |
92 | }, 'columns 3rd rscolumn present'); |
4c253752 |
93 | |
94 | |
faf79003 |
95 | |
96 | $rs = $schema->resultset('CD')->search({}, |
bbd828a6 |
97 | { |
faf79003 |
98 | 'join' => 'artist', |
99 | '+columns' => ['cdid', 'title', 'artist.name'], |
100 | } |
101 | ); |
102 | |
faf79003 |
103 | is_same_sql_bind ( |
af6aac2d |
104 | $rs->as_query, |
37aafa2e |
105 | '(SELECT me.cdid, me.artist, me.title, me.year, me.genreid, me.single_track, artist.name FROM cd me JOIN artist artist ON artist.artistid = me.artist)', |
faf79003 |
106 | [], |
107 | 'Use of columns attribute results in proper sql' |
108 | ); |
109 | |
110 | lives_ok(sub { |
bbd828a6 |
111 | $rs->first->get_column('cdid') |
faf79003 |
112 | }, 'columns 1st rscolumn present'); |
113 | |
114 | lives_ok(sub { |
115 | $rs->first->get_column('title') |
116 | }, 'columns 2nd rscolumn present'); |
4c253752 |
117 | |
0ca2f0d1 |
118 | lives_ok(sub { |
119 | $rs->first->artist->get_column('name') |
120 | }, 'columns 3rd rscolumn present'); |
b3b13ec9 |
121 | |
122 | |
123 | $rs = $schema->resultset('CD')->search({'tracks.position' => { -in => [2] } }, |
124 | { |
125 | join => 'tracks', |
126 | columns => [qw/me.cdid me.title/], |
127 | '+select' => ['tracks.position'], |
128 | '+as' => ['track_position'], |
129 | |
130 | # get a hashref of CD1 only (the first with a second track) |
131 | result_class => 'DBIx::Class::ResultClass::HashRefInflator', |
132 | order_by => 'cdid', |
133 | rows => 1, |
134 | } |
135 | ); |
136 | |
137 | is_deeply ( |
138 | $rs->single, |
139 | { |
140 | cdid => 1, |
141 | track_position => 2, |
142 | title => 'Spoonful of bees', |
143 | }, |
144 | 'limited prefetch via column works on a multi-relationship', |
145 | ); |
146 | |
18850595 |
147 | my $sub_rs = $rs->search ({}, |
b3b13ec9 |
148 | { |
149 | columns => [qw/artist tracks.trackid/], # columns should not be merged but override $rs columns |
150 | '+select' => ['tracks.title'], |
151 | '+as' => ['tracks.title'], |
152 | } |
153 | ); |
154 | |
c9d29bb2 |
155 | is_deeply( |
b3b13ec9 |
156 | $sub_rs->single, |
157 | { |
c9d29bb2 |
158 | artist => 1, |
0d5e6c6c |
159 | tracks => { |
160 | title => 'Apiary', |
1dfb11d4 |
161 | trackid => 17, |
c9d29bb2 |
162 | }, |
b3b13ec9 |
163 | }, |
164 | 'columns/select/as fold properly on sub-searches', |
165 | ); |
166 | |
fd4b0742 |
167 | # *very* esoteric use-case, yet valid (the "empty" object should not be undef): |
168 | $rs = $schema->resultset('Artist'); |
169 | $rs->create({ artistid => 69, name => 'Ranetki' }); |
170 | |
171 | my $relations_or_1_count = |
172 | $rs->search_related('cds')->count |
173 | + |
174 | $rs->search({ 'cds.cdid' => undef }, { join => 'cds' })->count |
175 | ; |
176 | |
177 | my $weird_rs = $rs->search({}, { |
178 | order_by => { -desc => [ 'me.artistid', 'cds.cdid' ] }, |
179 | columns => [{ cd_title => 'cds.title', cd_year => 'cds.year' }], |
180 | join => 'cds', |
181 | }); |
182 | |
183 | my $weird_rs_hri = $weird_rs->search({}, { result_class => 'DBIx::Class::ResultClass::HashRefInflator' }); |
184 | |
185 | for my $rs ($weird_rs, $weird_rs_hri) { |
186 | is ($rs->count, $relations_or_1_count, 'count on rhs data injection matches'); |
187 | |
188 | my @all; |
189 | while (my $r = $rs->next) { |
190 | push @all, $r; |
191 | } |
192 | |
193 | is (scalar @all, $relations_or_1_count, 'object count on rhs data injection matches'); |
194 | is_deeply ( |
195 | ( $rs->result_class eq 'DBIx::Class::ResultClass::HashRefInflator' |
196 | ? \@all |
197 | : [ map { +{$_->get_columns} } @all ] |
198 | ), |
199 | [ |
200 | { |
201 | cd_title => undef, |
202 | cd_year => undef, |
203 | }, |
204 | { |
205 | cd_title => "Come Be Depressed With Us", |
206 | cd_year => 1998, |
207 | }, |
208 | { |
209 | cd_title => "Generic Manufactured Singles", |
210 | cd_year => 2001, |
211 | }, |
212 | { |
213 | cd_title => "Caterwaulin' Blues", |
214 | cd_year => 1997, |
215 | }, |
216 | { |
217 | cd_title => "Forkful of bees", |
218 | cd_year => 2001, |
219 | }, |
220 | { |
221 | cd_title => "Spoonful of bees", |
222 | cd_year => 1999, |
223 | }, |
224 | ], |
225 | 'Correct data retrieved' |
226 | ); |
227 | |
228 | is_deeply( [ $rs->all ], \@all, '->all matches' ); |
229 | } |
230 | |
c4af4c4e |
231 | done_testing; |