line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
=head1 NAME |
2
|
|
|
|
|
|
|
|
3
|
|
|
|
|
|
|
Text::Record::Deduper - Separate complete, partial and near duplicate text records |
4
|
|
|
|
|
|
|
|
5
|
|
|
|
|
|
|
=head1 SYNOPSIS |
6
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
use Text::Record::Deduper; |
8
|
|
|
|
|
|
|
|
9
|
|
|
|
|
|
|
my $deduper = new Text::Record::Deduper; |
10
|
|
|
|
|
|
|
|
11
|
|
|
|
|
|
|
# Find and remove entire lines that are duplicated |
12
|
|
|
|
|
|
|
$deduper->dedupe_file("orig.txt"); |
13
|
|
|
|
|
|
|
|
14
|
|
|
|
|
|
|
# Dedupe comma separated records, duplicates defined by several fields |
15
|
|
|
|
|
|
|
$deduper->field_separator(','); |
16
|
|
|
|
|
|
|
$deduper->add_key(field_number => 1, ignore_case => 1 ); |
17
|
|
|
|
|
|
|
$deduper->add_key(field_number => 2, ignore_whitespace => 1); |
18
|
|
|
|
|
|
|
# unique records go to file names_uniqs.csv, dupes to names_dupes.csv |
19
|
|
|
|
|
|
|
$deduper->dedupe_file('names.csv'); |
20
|
|
|
|
|
|
|
|
21
|
|
|
|
|
|
|
# Find 'near' dupes by allowing for given name aliases |
22
|
|
|
|
|
|
|
my %nick_names = (Bob => 'Robert',Rob => 'Robert'); |
23
|
|
|
|
|
|
|
my $near_deduper = new Text::Record::Deduper(); |
24
|
|
|
|
|
|
|
$near_deduper->add_key(field_number => 2, alias => \%nick_names) or die; |
25
|
|
|
|
|
|
|
$near_deduper->dedupe_file('names.txt'); |
26
|
|
|
|
|
|
|
|
27
|
|
|
|
|
|
|
# Create a text report, names_report.txt to identify all duplicates |
28
|
|
|
|
|
|
|
$near_deduper->report_file('names.txt',all_records => 1); |
29
|
|
|
|
|
|
|
|
30
|
|
|
|
|
|
|
# Find 'near' dupes in an array of records, returning references |
31
|
|
|
|
|
|
|
# to a unique and a duplicate array |
32
|
|
|
|
|
|
|
my ($uniqs,$dupes) = $near_deduper->dedupe_array(\@some_records); |
33
|
|
|
|
|
|
|
|
34
|
|
|
|
|
|
|
# Create a report on unique and duplicate records |
35
|
|
|
|
|
|
|
$deduper->report_file("orig.txt",all_records => 0); |
36
|
|
|
|
|
|
|
|
37
|
|
|
|
|
|
|
|
38
|
|
|
|
|
|
|
=head1 DESCRIPTION |
39
|
|
|
|
|
|
|
|
40
|
|
|
|
|
|
|
This module allows you to take a text file of records and split it into |
41
|
|
|
|
|
|
|
a file of unique and a file of duplicate records. Deduping of arrays is |
42
|
|
|
|
|
|
|
also possible. |
43
|
|
|
|
|
|
|
|
44
|
|
|
|
|
|
|
Records are defined as a set of fields. Fields may be separated by spaces, |
45
|
|
|
|
|
|
|
commas, tabs or any other delimiter. Records are separated by a new line. |
46
|
|
|
|
|
|
|
|
47
|
|
|
|
|
|
|
If no options are specifed, a duplicate will be created only when all the |
48
|
|
|
|
|
|
|
fields in a record (the entire line) are duplicated. |
49
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
|
By specifying options a duplicate record is defined by which fields or partial |
51
|
|
|
|
|
|
|
fields must not occur more than once per record. There are also options to |
52
|
|
|
|
|
|
|
ignore case sensitivity, leading and trailing white space. |
53
|
|
|
|
|
|
|
|
54
|
|
|
|
|
|
|
Additionally 'near' or 'fuzzy' duplicates can be defined. This is done by creating |
55
|
|
|
|
|
|
|
aliases, such as Bob => Robert. |
56
|
|
|
|
|
|
|
|
57
|
|
|
|
|
|
|
This module is useful for finding duplicates that have been created by |
58
|
|
|
|
|
|
|
multiple data entry, or merging of similar records |
59
|
|
|
|
|
|
|
|
60
|
|
|
|
|
|
|
|
61
|
|
|
|
|
|
|
=head1 METHODS |
62
|
|
|
|
|
|
|
|
63
|
|
|
|
|
|
|
=head2 new |
64
|
|
|
|
|
|
|
|
65
|
|
|
|
|
|
|
The C method creates an instance of a deduping object. This must be |
66
|
|
|
|
|
|
|
called before any of the following methods are invoked. |
67
|
|
|
|
|
|
|
|
68
|
|
|
|
|
|
|
=head2 field_separator |
69
|
|
|
|
|
|
|
|
70
|
|
|
|
|
|
|
Sets the token to use as the field delimiter. Accepts any character as well as |
71
|
|
|
|
|
|
|
Perl escaped characters such as "\t" etc. If this method ins not called the |
72
|
|
|
|
|
|
|
deduper assumes you have fixed width fields . |
73
|
|
|
|
|
|
|
|
74
|
|
|
|
|
|
|
$deduper->field_separator(','); |
75
|
|
|
|
|
|
|
|
76
|
|
|
|
|
|
|
|
77
|
|
|
|
|
|
|
=head2 add_key |
78
|
|
|
|
|
|
|
|
79
|
|
|
|
|
|
|
Lets you add a field to the definition of a duplicate record. If no keys |
80
|
|
|
|
|
|
|
have been added, the entire record will become the key, so that only records |
81
|
|
|
|
|
|
|
duplicated in their entirity are removed. |
82
|
|
|
|
|
|
|
|
83
|
|
|
|
|
|
|
$deduper->add_key |
84
|
|
|
|
|
|
|
( |
85
|
|
|
|
|
|
|
field_number => 1, |
86
|
|
|
|
|
|
|
key_length => 5, |
87
|
|
|
|
|
|
|
ignore_case => 1, |
88
|
|
|
|
|
|
|
ignore_whitespace => 1, |
89
|
|
|
|
|
|
|
alias => \%nick_names |
90
|
|
|
|
|
|
|
); |
91
|
|
|
|
|
|
|
|
92
|
|
|
|
|
|
|
=over 4 |
93
|
|
|
|
|
|
|
|
94
|
|
|
|
|
|
|
=item field_number |
95
|
|
|
|
|
|
|
|
96
|
|
|
|
|
|
|
Specifies the number of the field in the record to add to the key (1,2 ...). |
97
|
|
|
|
|
|
|
Note that this option only applies to character separated data. You will get a |
98
|
|
|
|
|
|
|
warning if you try to specify a field_number for fixed width data. |
99
|
|
|
|
|
|
|
|
100
|
|
|
|
|
|
|
=item start_pos |
101
|
|
|
|
|
|
|
|
102
|
|
|
|
|
|
|
Specifies the position of the field in characters to add to the key. Note that |
103
|
|
|
|
|
|
|
this option only applies to fixed width data. You will get a warning if you |
104
|
|
|
|
|
|
|
try to specify a start_pos for character separated data. You must also specify |
105
|
|
|
|
|
|
|
a key_length. |
106
|
|
|
|
|
|
|
|
107
|
|
|
|
|
|
|
Note that the first column is numbered 1, not 0. |
108
|
|
|
|
|
|
|
|
109
|
|
|
|
|
|
|
|
110
|
|
|
|
|
|
|
=item key_length |
111
|
|
|
|
|
|
|
|
112
|
|
|
|
|
|
|
The length of a key field. This must be specifed if you are using fixed width |
113
|
|
|
|
|
|
|
data (along with a start_pos). It is optional for character separated data. |
114
|
|
|
|
|
|
|
|
115
|
|
|
|
|
|
|
=item ignore_case |
116
|
|
|
|
|
|
|
|
117
|
|
|
|
|
|
|
When defining a duplicate, ignore the case of characters, so Robert and ROBERT |
118
|
|
|
|
|
|
|
are equivalent. |
119
|
|
|
|
|
|
|
|
120
|
|
|
|
|
|
|
=item ignore_whitespace |
121
|
|
|
|
|
|
|
|
122
|
|
|
|
|
|
|
When defining a duplicate, ignore white space that leasd or trails a field's data. |
123
|
|
|
|
|
|
|
|
124
|
|
|
|
|
|
|
=item alias |
125
|
|
|
|
|
|
|
|
126
|
|
|
|
|
|
|
When defining a duplicate, allow for aliases substitution. For example |
127
|
|
|
|
|
|
|
|
128
|
|
|
|
|
|
|
my %nick_names = (Bob => 'Robert',Rob => 'Robert'); |
129
|
|
|
|
|
|
|
$near_deduper->add_key(field_number => 2, alias => \%nick_names) or die; |
130
|
|
|
|
|
|
|
|
131
|
|
|
|
|
|
|
Whenever field 2 contains 'Bob', it will be treated as a duplicate of a record |
132
|
|
|
|
|
|
|
where field 2 contains 'Robert'. |
133
|
|
|
|
|
|
|
|
134
|
|
|
|
|
|
|
=back |
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
|
137
|
|
|
|
|
|
|
=head2 dedupe_file |
138
|
|
|
|
|
|
|
|
139
|
|
|
|
|
|
|
This method takes a file name F as it's only argument. The file is |
140
|
|
|
|
|
|
|
processed to detect duplicates, as defined by the methods above. Unique records |
141
|
|
|
|
|
|
|
are place in a file named F and duplicates in a file named |
142
|
|
|
|
|
|
|
F. Note that If either of this output files exist, they are |
143
|
|
|
|
|
|
|
over written The orignal file is left intact. |
144
|
|
|
|
|
|
|
|
145
|
|
|
|
|
|
|
$deduper->dedupe_file("orig.txt"); |
146
|
|
|
|
|
|
|
|
147
|
|
|
|
|
|
|
|
148
|
|
|
|
|
|
|
=head2 dedupe_array |
149
|
|
|
|
|
|
|
|
150
|
|
|
|
|
|
|
This method takes an array reference as it's only argument. The array is |
151
|
|
|
|
|
|
|
processed to detect duplicates, as defined by the methods above. Two array |
152
|
|
|
|
|
|
|
references are retuned, the first to the set of unique records and the second |
153
|
|
|
|
|
|
|
to the set of duplicates. |
154
|
|
|
|
|
|
|
|
155
|
|
|
|
|
|
|
Note that the memory constraints of your system may prevent you from processing |
156
|
|
|
|
|
|
|
very large arrays. |
157
|
|
|
|
|
|
|
|
158
|
|
|
|
|
|
|
my ($unique_records,duplicate_records) = $deduper->dedupe_array(\@some_records); |
159
|
|
|
|
|
|
|
|
160
|
|
|
|
|
|
|
|
161
|
|
|
|
|
|
|
=head2 report_file |
162
|
|
|
|
|
|
|
|
163
|
|
|
|
|
|
|
This method takes a file name F as it's initial argument. |
164
|
|
|
|
|
|
|
|
165
|
|
|
|
|
|
|
A text report is produced with the following columns |
166
|
|
|
|
|
|
|
|
167
|
|
|
|
|
|
|
record number : the line number of the record |
168
|
|
|
|
|
|
|
|
169
|
|
|
|
|
|
|
key : the key values that define record uniqueness |
170
|
|
|
|
|
|
|
|
171
|
|
|
|
|
|
|
type: the type of record |
172
|
|
|
|
|
|
|
unique : record only occurs once |
173
|
|
|
|
|
|
|
identical : record occurs more than once, first occurence has parent record number of 0 |
174
|
|
|
|
|
|
|
alias : record occurs more than once, after alias substitutions have been applied |
175
|
|
|
|
|
|
|
|
176
|
|
|
|
|
|
|
parent record number : the line number of the record that THIS record is a duplicate of. |
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
By default, the report file name is F. |
179
|
|
|
|
|
|
|
|
180
|
|
|
|
|
|
|
Various setup options may be defined in a hash that is passed as an optional argument to |
181
|
|
|
|
|
|
|
the C method. Note that all the arguments are optional. They include |
182
|
|
|
|
|
|
|
|
183
|
|
|
|
|
|
|
=over 4 |
184
|
|
|
|
|
|
|
|
185
|
|
|
|
|
|
|
=item all_records |
186
|
|
|
|
|
|
|
|
187
|
|
|
|
|
|
|
When this option is set to a positive value, all records will be included in |
188
|
|
|
|
|
|
|
the report. If this value is not set, only the duplicate records will be included |
189
|
|
|
|
|
|
|
in the report |
190
|
|
|
|
|
|
|
|
191
|
|
|
|
|
|
|
=back |
192
|
|
|
|
|
|
|
|
193
|
|
|
|
|
|
|
|
194
|
|
|
|
|
|
|
$deduper->report_file("orig.txt",all_records => 0) |
195
|
|
|
|
|
|
|
|
196
|
|
|
|
|
|
|
=head2 report_array |
197
|
|
|
|
|
|
|
|
198
|
|
|
|
|
|
|
This method takes an array as it's initial argument. The behaviour is the same as |
199
|
|
|
|
|
|
|
C above except that the report file is named F |
200
|
|
|
|
|
|
|
|
201
|
|
|
|
|
|
|
=head1 EXAMPLES |
202
|
|
|
|
|
|
|
|
203
|
|
|
|
|
|
|
=head2 Dedupe an array of single records |
204
|
|
|
|
|
|
|
|
205
|
|
|
|
|
|
|
Given an array of strings: |
206
|
|
|
|
|
|
|
|
207
|
|
|
|
|
|
|
my @emails = |
208
|
|
|
|
|
|
|
( |
209
|
|
|
|
|
|
|
'John.Smith@xyz.com', |
210
|
|
|
|
|
|
|
'Bob.Smith@xyz.com', |
211
|
|
|
|
|
|
|
'John.Brown@xyz.com.au, |
212
|
|
|
|
|
|
|
'John.Smith@xyz.com' |
213
|
|
|
|
|
|
|
); |
214
|
|
|
|
|
|
|
|
215
|
|
|
|
|
|
|
use Text::Record::Deduper; |
216
|
|
|
|
|
|
|
|
217
|
|
|
|
|
|
|
my $deduper = new Text::Record::Deduper(); |
218
|
|
|
|
|
|
|
my ($uniq,$dupe); |
219
|
|
|
|
|
|
|
($uniq,$dupe) = $deduper->dedupe_array(\@emails); |
220
|
|
|
|
|
|
|
|
221
|
|
|
|
|
|
|
The array reference $uniq now contains |
222
|
|
|
|
|
|
|
|
223
|
|
|
|
|
|
|
'John.Smith@xyz.com', |
224
|
|
|
|
|
|
|
'Bob.Smith@xyz.com', |
225
|
|
|
|
|
|
|
'John.Brown@xyz.com.au' |
226
|
|
|
|
|
|
|
|
227
|
|
|
|
|
|
|
The array reference $dupe now contains |
228
|
|
|
|
|
|
|
|
229
|
|
|
|
|
|
|
'John.Smith@xyz.com' |
230
|
|
|
|
|
|
|
|
231
|
|
|
|
|
|
|
|
232
|
|
|
|
|
|
|
=head2 Dedupe a file of fixed width records |
233
|
|
|
|
|
|
|
|
234
|
|
|
|
|
|
|
Given a text file F with space separated values and duplicates defined |
235
|
|
|
|
|
|
|
by the second and third columns: |
236
|
|
|
|
|
|
|
|
237
|
|
|
|
|
|
|
100 Bob Smith |
238
|
|
|
|
|
|
|
101 Robert Smith |
239
|
|
|
|
|
|
|
102 John Brown |
240
|
|
|
|
|
|
|
103 Jack White |
241
|
|
|
|
|
|
|
104 Bob Smythe |
242
|
|
|
|
|
|
|
105 Robert Smith |
243
|
|
|
|
|
|
|
|
244
|
|
|
|
|
|
|
|
245
|
|
|
|
|
|
|
use Text::Record::Deduper; |
246
|
|
|
|
|
|
|
|
247
|
|
|
|
|
|
|
my %nick_names = (Bob => 'Robert',Rob => 'Robert'); |
248
|
|
|
|
|
|
|
my $near_deduper = new Text::Record::Deduper(); |
249
|
|
|
|
|
|
|
$near_deduper->add_key(start_pos => 5, key_length => 9, ignore_whitespace => 1, alias => \%nick_names) or die; |
250
|
|
|
|
|
|
|
$near_deduper->add_key(start_pos => 14, key_length => 9,) or die; |
251
|
|
|
|
|
|
|
$near_deduper->dedupe_file("names.txt"); |
252
|
|
|
|
|
|
|
$near_deduper->report_file("names.txt"); |
253
|
|
|
|
|
|
|
|
254
|
|
|
|
|
|
|
|
255
|
|
|
|
|
|
|
Text::Record::Deduper will produce a file of unique records, F |
256
|
|
|
|
|
|
|
in the same directory as F. |
257
|
|
|
|
|
|
|
|
258
|
|
|
|
|
|
|
101 Robert Smith |
259
|
|
|
|
|
|
|
102 John Brown |
260
|
|
|
|
|
|
|
103 Jack White |
261
|
|
|
|
|
|
|
104 Bob Smythe |
262
|
|
|
|
|
|
|
|
263
|
|
|
|
|
|
|
|
264
|
|
|
|
|
|
|
and a file of duplicates, F in the same directory as F |
265
|
|
|
|
|
|
|
|
266
|
|
|
|
|
|
|
100 Bob Smith |
267
|
|
|
|
|
|
|
105 Robert Smith |
268
|
|
|
|
|
|
|
|
269
|
|
|
|
|
|
|
The original file, F is left intact. |
270
|
|
|
|
|
|
|
|
271
|
|
|
|
|
|
|
A report file F, is created in the same directory as F |
272
|
|
|
|
|
|
|
|
273
|
|
|
|
|
|
|
Number Key Type Parent Parent Key |
274
|
|
|
|
|
|
|
-------------------------------------------------------------------------------- |
275
|
|
|
|
|
|
|
1 Bob_Smith alias 2 Robert_Smith |
276
|
|
|
|
|
|
|
2 Robert_Smith identical 0 |
277
|
|
|
|
|
|
|
3 John_Brown unique 0 |
278
|
|
|
|
|
|
|
4 Jack_White unique 0 |
279
|
|
|
|
|
|
|
5 Bob_Smythe unique 0 |
280
|
|
|
|
|
|
|
6 Robert_Smith identical 2 Robert_Smith |
281
|
|
|
|
|
|
|
|
282
|
|
|
|
|
|
|
|
283
|
|
|
|
|
|
|
=head1 TO DO |
284
|
|
|
|
|
|
|
|
285
|
|
|
|
|
|
|
Allow for multi line records |
286
|
|
|
|
|
|
|
Add batch mode driven by config file or command line options |
287
|
|
|
|
|
|
|
Allow option to warn user when over writing output files |
288
|
|
|
|
|
|
|
Allow user to customise suffix for uniq and dupe output files |
289
|
|
|
|
|
|
|
|
290
|
|
|
|
|
|
|
|
291
|
|
|
|
|
|
|
=head1 SEE ALSO |
292
|
|
|
|
|
|
|
|
293
|
|
|
|
|
|
|
sort(3), uniq(3), L, L, L |
294
|
|
|
|
|
|
|
|
295
|
|
|
|
|
|
|
|
296
|
|
|
|
|
|
|
=head1 AUTHOR |
297
|
|
|
|
|
|
|
|
298
|
|
|
|
|
|
|
Text::Record::Deduper was written by Kim Ryan |
299
|
|
|
|
|
|
|
|
300
|
|
|
|
|
|
|
|
301
|
|
|
|
|
|
|
=head1 COPYRIGHT AND LICENSE |
302
|
|
|
|
|
|
|
|
303
|
|
|
|
|
|
|
Copyright (C) 2011 Kim Ryan. |
304
|
|
|
|
|
|
|
|
305
|
|
|
|
|
|
|
|
306
|
|
|
|
|
|
|
This library is free software; you can redistribute it and/or modify |
307
|
|
|
|
|
|
|
it under the same terms as Perl itself. |
308
|
|
|
|
|
|
|
|
309
|
|
|
|
|
|
|
|
310
|
|
|
|
|
|
|
=cut |
311
|
|
|
|
|
|
|
|
312
|
|
|
|
|
|
|
package Text::Record::Deduper; |
313
|
1
|
|
|
1
|
|
25002
|
use FileHandle; |
|
1
|
|
|
|
|
13962
|
|
|
1
|
|
|
|
|
7
|
|
314
|
1
|
|
|
1
|
|
439
|
use File::Basename; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
102
|
|
315
|
1
|
|
|
1
|
|
966
|
use Text::ParseWords; |
|
1
|
|
|
|
|
1524
|
|
|
1
|
|
|
|
|
70
|
|
316
|
1
|
|
|
1
|
|
1258
|
use Data::Dumper; |
|
1
|
|
|
|
|
11616
|
|
|
1
|
|
|
|
|
84
|
|
317
|
|
|
|
|
|
|
|
318
|
|
|
|
|
|
|
|
319
|
|
|
|
|
|
|
|
320
|
1
|
|
|
1
|
|
9
|
use strict; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
36
|
|
321
|
1
|
|
|
1
|
|
6
|
use warnings; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
4009
|
|
322
|
|
|
|
|
|
|
|
323
|
|
|
|
|
|
|
require Exporter; |
324
|
|
|
|
|
|
|
|
325
|
|
|
|
|
|
|
our @ISA = qw(Exporter); |
326
|
|
|
|
|
|
|
our $VERSION = '0.07'; |
327
|
|
|
|
|
|
|
|
328
|
|
|
|
|
|
|
|
329
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
330
|
|
|
|
|
|
|
# Create a new instance of a deduping object. |
331
|
|
|
|
|
|
|
|
332
|
|
|
|
|
|
|
sub new |
333
|
|
|
|
|
|
|
{ |
334
|
3
|
|
|
3
|
1
|
2091
|
my $class = shift; |
335
|
3
|
|
|
|
|
8
|
my %args = @_; |
336
|
|
|
|
|
|
|
|
337
|
3
|
|
|
|
|
9
|
my $deduper = {}; |
338
|
3
|
|
|
|
|
8
|
bless($deduper,$class); |
339
|
|
|
|
|
|
|
|
340
|
|
|
|
|
|
|
# Default to no separator, until we find otherwise |
341
|
3
|
|
|
|
|
13
|
$deduper->{field_separator} = ''; |
342
|
|
|
|
|
|
|
|
343
|
3
|
|
|
|
|
9
|
return ($deduper); |
344
|
|
|
|
|
|
|
} |
345
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
346
|
|
|
|
|
|
|
# Create a new instance of a deduping object. |
347
|
|
|
|
|
|
|
|
348
|
|
|
|
|
|
|
sub field_separator |
349
|
|
|
|
|
|
|
{ |
350
|
1
|
|
|
1
|
1
|
7
|
my $deduper = shift; |
351
|
|
|
|
|
|
|
|
352
|
1
|
|
|
|
|
2
|
my ($field_separator) = @_; |
353
|
|
|
|
|
|
|
|
354
|
|
|
|
|
|
|
# Escape pipe symbol so it does get interpreted as alternation character |
355
|
|
|
|
|
|
|
# when splitting fields in _get_key_fields |
356
|
1
|
50
|
|
|
|
5
|
$field_separator eq '|' and $field_separator = '\|'; |
357
|
|
|
|
|
|
|
|
358
|
|
|
|
|
|
|
# add more error checking here |
359
|
|
|
|
|
|
|
|
360
|
1
|
|
|
|
|
3
|
$deduper->{field_separator} = $field_separator; |
361
|
1
|
|
|
|
|
2
|
return($deduper); |
362
|
|
|
|
|
|
|
} |
363
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
364
|
|
|
|
|
|
|
# Dewfine a key field in the record |
365
|
|
|
|
|
|
|
sub add_key |
366
|
|
|
|
|
|
|
{ |
367
|
4
|
|
|
4
|
1
|
25
|
my $deduper = shift; |
368
|
4
|
|
|
|
|
18
|
my %args = @_; |
369
|
|
|
|
|
|
|
|
370
|
|
|
|
|
|
|
|
371
|
4
|
|
|
|
|
9
|
$deduper->{key_counter}++; |
372
|
|
|
|
|
|
|
|
373
|
4
|
100
|
|
|
|
14
|
if ( $args{field_number} ) |
|
|
50
|
|
|
|
|
|
374
|
|
|
|
|
|
|
{ |
375
|
2
|
50
|
|
|
|
8
|
unless ( $deduper->{field_separator} ) |
376
|
|
|
|
|
|
|
{ |
377
|
0
|
|
|
|
|
0
|
warn "Cannot use field_number on fixed width lines"; |
378
|
0
|
|
|
|
|
0
|
return; |
379
|
|
|
|
|
|
|
} |
380
|
|
|
|
|
|
|
} |
381
|
|
|
|
|
|
|
elsif ( $args{start_pos} ) |
382
|
|
|
|
|
|
|
{ |
383
|
2
|
50
|
|
|
|
9
|
if ( $deduper->{field_separator} ) |
384
|
|
|
|
|
|
|
{ |
385
|
0
|
|
|
|
|
0
|
warn "Cannot use start_pos on character separated records"; |
386
|
0
|
|
|
|
|
0
|
return; |
387
|
|
|
|
|
|
|
} |
388
|
|
|
|
|
|
|
else |
389
|
|
|
|
|
|
|
{ |
390
|
2
|
50
|
|
|
|
9
|
unless ( $args{key_length} ) |
391
|
|
|
|
|
|
|
{ |
392
|
0
|
|
|
|
|
0
|
warn "No key_length defined for start_pos: $args{start_pos}"; |
393
|
0
|
|
|
|
|
0
|
return; |
394
|
|
|
|
|
|
|
} |
395
|
|
|
|
|
|
|
} |
396
|
|
|
|
|
|
|
} |
397
|
|
|
|
|
|
|
|
398
|
4
|
|
|
|
|
13
|
foreach my $current_key (keys %args) |
399
|
|
|
|
|
|
|
{ |
400
|
15
|
100
|
|
|
|
38
|
if ($current_key eq 'ignore_case' ) |
401
|
|
|
|
|
|
|
{ |
402
|
4
|
|
|
|
|
13
|
$deduper->{ignore_case}{$deduper->{key_counter}} = 1; |
403
|
|
|
|
|
|
|
} |
404
|
15
|
100
|
|
|
|
36
|
if ($current_key eq 'ignore_whitespace' ) |
405
|
|
|
|
|
|
|
{ |
406
|
3
|
|
|
|
|
11
|
$deduper->{ignore_whitespace}{$deduper->{key_counter}} = 1; |
407
|
|
|
|
|
|
|
} |
408
|
15
|
100
|
|
|
|
33
|
if ($current_key eq 'alias' ) |
409
|
|
|
|
|
|
|
{ |
410
|
2
|
50
|
|
|
|
6
|
if ( $args{ignore_case} ) |
411
|
|
|
|
|
|
|
{ |
412
|
|
|
|
|
|
|
# if ignore case, fold all of alias to upper case |
413
|
2
|
|
|
|
|
4
|
my %current_alias = %{ $args{$current_key} }; |
|
2
|
|
|
|
|
10
|
|
414
|
2
|
|
|
|
|
5
|
my %corrected_alias; |
415
|
2
|
|
|
|
|
7
|
foreach my $current_alias_key ( keys %current_alias ) |
416
|
|
|
|
|
|
|
{ |
417
|
6
|
|
|
|
|
30
|
$corrected_alias{uc($current_alias_key)} = uc($current_alias{$current_alias_key}); |
418
|
|
|
|
|
|
|
|
419
|
|
|
|
|
|
|
} |
420
|
2
|
|
|
|
|
10
|
$deduper->{alias}{$deduper->{key_counter}} = \%corrected_alias; |
421
|
|
|
|
|
|
|
} |
422
|
|
|
|
|
|
|
else |
423
|
|
|
|
|
|
|
{ |
424
|
0
|
|
|
|
|
0
|
$deduper->{alias}{$deduper->{key_counter}} = $args{$current_key}; |
425
|
|
|
|
|
|
|
} |
426
|
|
|
|
|
|
|
} |
427
|
15
|
100
|
|
|
|
69
|
if ($current_key =~ /field_number|start_pos|key_length/ ) |
428
|
|
|
|
|
|
|
{ |
429
|
6
|
|
|
|
|
26
|
$deduper->{key}{$deduper->{key_counter}}{$current_key} = $args{$current_key}; |
430
|
|
|
|
|
|
|
} |
431
|
|
|
|
|
|
|
} |
432
|
|
|
|
|
|
|
|
433
|
4
|
|
|
|
|
25
|
return ($deduper); |
434
|
|
|
|
|
|
|
} |
435
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
436
|
|
|
|
|
|
|
# |
437
|
|
|
|
|
|
|
sub dedupe_file |
438
|
|
|
|
|
|
|
{ |
439
|
0
|
|
|
0
|
1
|
0
|
my $deduper = shift; |
440
|
0
|
|
|
|
|
0
|
my ($input_file_name) = @_; |
441
|
|
|
|
|
|
|
|
442
|
0
|
0
|
0
|
|
|
0
|
unless ( -T $input_file_name and -s $input_file_name ) |
443
|
|
|
|
|
|
|
{ |
444
|
0
|
|
|
|
|
0
|
warn("Could not find input file: $input_file_name"); |
445
|
0
|
|
|
|
|
0
|
return(0); |
446
|
|
|
|
|
|
|
} |
447
|
|
|
|
|
|
|
|
448
|
0
|
|
|
|
|
0
|
my $input_fh = new FileHandle "<$input_file_name"; |
449
|
0
|
0
|
|
|
|
0
|
unless ($input_fh) |
450
|
|
|
|
|
|
|
{ |
451
|
0
|
|
|
|
|
0
|
warn "Could not open input file: $input_file_name"; |
452
|
0
|
|
|
|
|
0
|
return(0); |
453
|
|
|
|
|
|
|
} |
454
|
|
|
|
|
|
|
|
455
|
0
|
|
|
|
|
0
|
my ($file_name,$path,$suffix) = File::Basename::fileparse($input_file_name,qr{\..*}); |
456
|
0
|
|
|
|
|
0
|
my $file_name_unique_records = "$path$file_name\_uniqs$suffix"; |
457
|
0
|
|
|
|
|
0
|
my $file_name_duplicate_records = "$path$file_name\_dupes$suffix"; |
458
|
|
|
|
|
|
|
|
459
|
|
|
|
|
|
|
# TO DO!!! test for overwriting of previous Deduper output |
460
|
0
|
|
|
|
|
0
|
my $unique_fh = new FileHandle ">$file_name_unique_records"; |
461
|
0
|
0
|
|
|
|
0
|
unless($unique_fh) |
462
|
|
|
|
|
|
|
{ |
463
|
0
|
|
|
|
|
0
|
warn "Could not open file: $file_name_unique_records: $!"; |
464
|
0
|
|
|
|
|
0
|
return(0); |
465
|
|
|
|
|
|
|
} |
466
|
|
|
|
|
|
|
|
467
|
0
|
|
|
|
|
0
|
my $dupes_fh = new FileHandle ">$file_name_duplicate_records"; |
468
|
0
|
0
|
|
|
|
0
|
unless ( $dupes_fh ) |
469
|
|
|
|
|
|
|
{ |
470
|
0
|
|
|
|
|
0
|
warn "Could not open file: $file_name_duplicate_records: $!"; |
471
|
0
|
|
|
|
|
0
|
return(0); |
472
|
|
|
|
|
|
|
} |
473
|
|
|
|
|
|
|
|
474
|
0
|
|
|
|
|
0
|
my ($master_record_index) = $deduper->_analyse('file',undef,$input_fh); |
475
|
0
|
|
|
|
|
0
|
$deduper->_separate('file',$master_record_index,undef,$input_fh,$unique_fh,$dupes_fh); |
476
|
|
|
|
|
|
|
|
477
|
0
|
|
|
|
|
0
|
$input_fh->close; |
478
|
0
|
|
|
|
|
0
|
$unique_fh->close; |
479
|
0
|
|
|
|
|
0
|
$dupes_fh->close; |
480
|
|
|
|
|
|
|
} |
481
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
482
|
|
|
|
|
|
|
|
483
|
|
|
|
|
|
|
sub _rewind_file |
484
|
|
|
|
|
|
|
{ |
485
|
0
|
|
|
0
|
|
0
|
my ($input_fh) = @_; |
486
|
0
|
|
|
|
|
0
|
$input_fh->seek(0,0); # rewind file |
487
|
|
|
|
|
|
|
} |
488
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
489
|
|
|
|
|
|
|
# |
490
|
|
|
|
|
|
|
sub dedupe_array |
491
|
|
|
|
|
|
|
{ |
492
|
3
|
|
|
3
|
1
|
17
|
my $deduper = shift; |
493
|
|
|
|
|
|
|
|
494
|
3
|
|
|
|
|
5
|
my ($input_array_ref) = @_; |
495
|
3
|
|
|
|
|
10
|
my ($master_record_index) = $deduper->_analyse('array',$input_array_ref,undef); |
496
|
3
|
|
|
|
|
11
|
my ($uniq_array_ref,$dupe_array_ref) = $deduper->_separate('array',$master_record_index,$input_array_ref,undef); |
497
|
|
|
|
|
|
|
|
498
|
3
|
|
|
|
|
21
|
return($uniq_array_ref,$dupe_array_ref); |
499
|
|
|
|
|
|
|
} |
500
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
501
|
|
|
|
|
|
|
# Produce a text report on deduping statistics |
502
|
|
|
|
|
|
|
sub report_file |
503
|
|
|
|
|
|
|
{ |
504
|
|
|
|
|
|
|
|
505
|
0
|
|
|
0
|
1
|
0
|
my $deduper = shift; |
506
|
0
|
|
|
|
|
0
|
my ($input_file_name,%report_options) = @_; |
507
|
|
|
|
|
|
|
|
508
|
0
|
0
|
0
|
|
|
0
|
unless ( -T $input_file_name and -s $input_file_name ) |
509
|
|
|
|
|
|
|
{ |
510
|
0
|
|
|
|
|
0
|
warn("Could not find input file: $input_file_name"); |
511
|
0
|
|
|
|
|
0
|
return(0); |
512
|
|
|
|
|
|
|
} |
513
|
|
|
|
|
|
|
|
514
|
0
|
|
|
|
|
0
|
my $input_fh = new FileHandle "<$input_file_name"; |
515
|
0
|
0
|
|
|
|
0
|
unless ($input_fh) |
516
|
|
|
|
|
|
|
{ |
517
|
0
|
|
|
|
|
0
|
warn "Could not open input file: $input_file_name"; |
518
|
0
|
|
|
|
|
0
|
return(0); |
519
|
|
|
|
|
|
|
} |
520
|
|
|
|
|
|
|
|
521
|
0
|
|
|
|
|
0
|
my ($master_record_index_ref) = $deduper->_analyse('file',undef,$input_fh); |
522
|
|
|
|
|
|
|
|
523
|
0
|
|
|
|
|
0
|
my $report_file_name; |
524
|
0
|
0
|
|
|
|
0
|
if ( $report_options{file_name} ) |
525
|
|
|
|
|
|
|
{ |
526
|
|
|
|
|
|
|
# user has specified name and path to report file |
527
|
0
|
|
|
|
|
0
|
$report_file_name = $report_options{file_name}; |
528
|
|
|
|
|
|
|
} |
529
|
|
|
|
|
|
|
else |
530
|
|
|
|
|
|
|
{ |
531
|
|
|
|
|
|
|
# use base of input file, append _report.txt to report file name |
532
|
0
|
|
|
|
|
0
|
my ($file_name,$path,$suffix) = File::Basename::fileparse($input_file_name,qr{\..*}); |
533
|
0
|
|
|
|
|
0
|
$report_file_name = "$path$file_name\_report\.txt"; |
534
|
|
|
|
|
|
|
} |
535
|
|
|
|
|
|
|
|
536
|
|
|
|
|
|
|
|
537
|
0
|
|
|
|
|
0
|
$deduper->_report($master_record_index_ref,$report_file_name,%report_options); |
538
|
0
|
|
|
|
|
0
|
$input_fh->close; |
539
|
|
|
|
|
|
|
|
540
|
|
|
|
|
|
|
} |
541
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
542
|
|
|
|
|
|
|
# |
543
|
|
|
|
|
|
|
sub report_array |
544
|
|
|
|
|
|
|
{ |
545
|
|
|
|
|
|
|
|
546
|
0
|
|
|
0
|
1
|
0
|
my $deduper = shift; |
547
|
0
|
|
|
|
|
0
|
my ($input_array_ref,%report_options) = @_; |
548
|
|
|
|
|
|
|
|
549
|
0
|
|
|
|
|
0
|
my ($master_record_index_ref) = $deduper->_analyse('array',$input_array_ref,undef); |
550
|
|
|
|
|
|
|
|
551
|
0
|
|
|
|
|
0
|
my $report_file_name; |
552
|
0
|
0
|
|
|
|
0
|
if ( $report_options{file_name} ) |
553
|
|
|
|
|
|
|
{ |
554
|
|
|
|
|
|
|
# user has specified name and path to report file |
555
|
0
|
|
|
|
|
0
|
$report_file_name = $report_options{file_name}; |
556
|
|
|
|
|
|
|
} |
557
|
|
|
|
|
|
|
else |
558
|
|
|
|
|
|
|
{ |
559
|
|
|
|
|
|
|
# TO DO, make name more unique, eg add time stamp |
560
|
0
|
|
|
|
|
0
|
$report_file_name = "./deduper_array_report.txt"; |
561
|
|
|
|
|
|
|
} |
562
|
|
|
|
|
|
|
|
563
|
0
|
|
|
|
|
0
|
$deduper->_report($master_record_index_ref,$report_file_name,%report_options); |
564
|
|
|
|
|
|
|
|
565
|
|
|
|
|
|
|
} |
566
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
567
|
|
|
|
|
|
|
# |
568
|
|
|
|
|
|
|
sub _report |
569
|
|
|
|
|
|
|
{ |
570
|
0
|
|
|
0
|
|
0
|
my $deduper = shift; |
571
|
0
|
|
|
|
|
0
|
my ($master_record_index_ref,$report_file_name,%report_options) = @_; |
572
|
|
|
|
|
|
|
|
573
|
|
|
|
|
|
|
|
574
|
0
|
|
|
|
|
0
|
my $report_fh = new FileHandle ">$report_file_name"; |
575
|
0
|
0
|
|
|
|
0
|
unless($report_fh) |
576
|
|
|
|
|
|
|
{ |
577
|
0
|
|
|
|
|
0
|
warn "Could not open report file: $report_file_name: $!"; |
578
|
0
|
|
|
|
|
0
|
return(0); |
579
|
|
|
|
|
|
|
} |
580
|
|
|
|
|
|
|
|
581
|
|
|
|
|
|
|
# TO DO, report format, side be side or interleaved? |
582
|
|
|
|
|
|
|
# options, all records or dupes (default), group all dupes even first |
583
|
|
|
|
|
|
|
# full record dump, not just key |
584
|
|
|
|
|
|
|
|
585
|
0
|
|
|
|
|
0
|
my $current_line = sprintf("%6s %-30.30s %-10.10s %6s %-30.30s\n", |
586
|
|
|
|
|
|
|
'Number', 'Key','Type','Parent','Parent Key'); |
587
|
0
|
|
|
|
|
0
|
$report_fh->print($current_line); |
588
|
0
|
|
|
|
|
0
|
$report_fh->print('-' x 80,"\n"); |
589
|
|
|
|
|
|
|
|
590
|
0
|
|
|
|
|
0
|
foreach my $record_num ( sort { $a <=> $b } keys %$master_record_index_ref ) |
|
0
|
|
|
|
|
0
|
|
591
|
|
|
|
|
|
|
{ |
592
|
|
|
|
|
|
|
|
593
|
0
|
0
|
0
|
|
|
0
|
if ( $report_options{all_records} or |
|
|
|
0
|
|
|
|
|
594
|
|
|
|
|
|
|
($master_record_index_ref->{$record_num}->{type} ne 'unique' and |
595
|
|
|
|
|
|
|
$master_record_index_ref->{$record_num}->{parent} > 0 ) ) |
596
|
|
|
|
|
|
|
{ |
597
|
0
|
|
|
|
|
0
|
my $parent_record_key = ''; |
598
|
0
|
0
|
|
|
|
0
|
if ( $master_record_index_ref->{$record_num}->{parent} ) |
599
|
|
|
|
|
|
|
{ |
600
|
0
|
|
|
|
|
0
|
$parent_record_key = $master_record_index_ref->{$master_record_index_ref->{$record_num}->{parent}}->{key}; |
601
|
|
|
|
|
|
|
} |
602
|
0
|
|
|
|
|
0
|
my $current_line = sprintf("%6d %-30.30s %-10.10s %6d %-30.30s\n", |
603
|
|
|
|
|
|
|
$record_num, |
604
|
|
|
|
|
|
|
$master_record_index_ref->{$record_num}->{key}, |
605
|
|
|
|
|
|
|
$master_record_index_ref->{$record_num}->{type}, |
606
|
|
|
|
|
|
|
$master_record_index_ref->{$record_num}->{parent}, |
607
|
|
|
|
|
|
|
$parent_record_key); |
608
|
|
|
|
|
|
|
|
609
|
0
|
|
|
|
|
0
|
$report_fh->print($current_line); |
610
|
|
|
|
|
|
|
} |
611
|
|
|
|
|
|
|
} |
612
|
0
|
|
|
|
|
0
|
$report_fh->close; |
613
|
|
|
|
|
|
|
} |
614
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
615
|
|
|
|
|
|
|
# |
616
|
|
|
|
|
|
|
sub _analyse |
617
|
|
|
|
|
|
|
{ |
618
|
3
|
|
|
3
|
|
5
|
my $deduper = shift; |
619
|
3
|
|
|
|
|
13
|
my ($storage_type,$input_array_ref,$input_fh) = @_; |
620
|
|
|
|
|
|
|
|
621
|
3
|
|
|
|
|
5
|
my $current_record_number = 0; |
622
|
3
|
|
|
|
|
4
|
my $current_line; |
623
|
3
|
|
|
|
|
4
|
my $finished = 0; |
624
|
|
|
|
|
|
|
|
625
|
|
|
|
|
|
|
|
626
|
3
|
|
|
|
|
4
|
my %alias_candidates; |
627
|
3
|
50
|
|
|
|
11
|
if ( $deduper->{alias} ) |
628
|
|
|
|
|
|
|
{ |
629
|
3
|
|
|
|
|
12
|
my %all_alias_values = $deduper->_get_all_alias_values; |
630
|
3
|
|
|
|
|
8
|
while ( not $finished ) |
631
|
|
|
|
|
|
|
{ |
632
|
16
|
|
|
|
|
39
|
($current_line,$finished) = _read_one_record($storage_type,$current_record_number,$input_array_ref,$input_fh); |
633
|
16
|
|
|
|
|
25
|
$current_record_number++; |
634
|
16
|
|
|
|
|
43
|
my $alias_candidate_key = $deduper->_alias_candidate($current_line,%all_alias_values); |
635
|
16
|
100
|
100
|
|
|
81
|
if ( $alias_candidate_key and not $alias_candidates{$alias_candidate_key} ) |
636
|
|
|
|
|
|
|
{ |
637
|
2
|
|
|
|
|
9
|
$alias_candidates{$alias_candidate_key} = $current_record_number; |
638
|
|
|
|
|
|
|
} |
639
|
|
|
|
|
|
|
} |
640
|
|
|
|
|
|
|
} |
641
|
|
|
|
|
|
|
# print(Dumper(\%alias_candidates)); |
642
|
|
|
|
|
|
|
# die; |
643
|
|
|
|
|
|
|
|
644
|
3
|
|
|
|
|
4
|
my %seen_exact_dupes; |
645
|
3
|
|
|
|
|
5
|
my $unique_ref = []; |
646
|
3
|
|
|
|
|
6
|
my $dupe_ref = []; |
647
|
3
|
|
|
|
|
4
|
$current_record_number = 0; |
648
|
3
|
|
|
|
|
4
|
my %master_record_index; |
649
|
|
|
|
|
|
|
|
650
|
3
|
|
|
|
|
4
|
$finished = 0; |
651
|
3
|
50
|
33
|
|
|
12
|
if ( $storage_type eq 'file' and $deduper->{alias} ) |
652
|
|
|
|
|
|
|
{ |
653
|
0
|
|
|
|
|
0
|
_rewind_file($input_fh); |
654
|
|
|
|
|
|
|
} |
655
|
|
|
|
|
|
|
|
656
|
3
|
|
|
|
|
8
|
while ( not $finished ) |
657
|
|
|
|
|
|
|
{ |
658
|
16
|
|
|
|
|
32
|
($current_line,$finished) = _read_one_record($storage_type,$current_record_number,$input_array_ref,$input_fh); |
659
|
16
|
|
|
|
|
20
|
$current_record_number++; |
660
|
|
|
|
|
|
|
|
661
|
16
|
|
|
|
|
17
|
my $dupe_type; |
662
|
16
|
|
|
|
|
36
|
my %record_keys = $deduper->_get_key_fields($current_line); |
663
|
|
|
|
|
|
|
|
664
|
|
|
|
|
|
|
|
665
|
16
|
|
|
|
|
47
|
%record_keys = $deduper->_transform_key_fields(%record_keys); |
666
|
16
|
|
|
|
|
59
|
my $full_key = _assemble_full_key(%record_keys); |
667
|
|
|
|
|
|
|
|
668
|
16
|
|
|
|
|
23
|
my $parent_record_number; |
669
|
16
|
100
|
|
|
|
44
|
if ( $parent_record_number = $deduper->_alias_dupe(\%alias_candidates,%record_keys) ) |
|
|
100
|
|
|
|
|
|
670
|
|
|
|
|
|
|
{ |
671
|
4
|
|
|
|
|
5
|
$dupe_type = 'alias'; |
672
|
|
|
|
|
|
|
} |
673
|
|
|
|
|
|
|
# add soundex dupe |
674
|
|
|
|
|
|
|
# add string approx dupe |
675
|
|
|
|
|
|
|
elsif ( $parent_record_number = _exact_dupe($current_line,$full_key,%seen_exact_dupes) ) |
676
|
|
|
|
|
|
|
{ |
677
|
5
|
|
|
|
|
8
|
$dupe_type = 'identical'; |
678
|
|
|
|
|
|
|
} |
679
|
|
|
|
|
|
|
else |
680
|
|
|
|
|
|
|
{ |
681
|
7
|
|
|
|
|
10
|
$dupe_type = 'unique'; |
682
|
|
|
|
|
|
|
# retain the record number of dupe, useful for detailed reporting and grouping |
683
|
7
|
|
|
|
|
15
|
$seen_exact_dupes{$full_key} = $current_record_number; |
684
|
7
|
|
|
|
|
8
|
$parent_record_number = 0; |
685
|
|
|
|
|
|
|
} |
686
|
|
|
|
|
|
|
|
687
|
16
|
|
|
|
|
41
|
_classify_record($dupe_type,$parent_record_number,$current_record_number,$full_key,\%master_record_index); |
688
|
|
|
|
|
|
|
} |
689
|
3
|
|
|
|
|
14
|
return(\%master_record_index); |
690
|
|
|
|
|
|
|
|
691
|
|
|
|
|
|
|
} |
692
|
|
|
|
|
|
|
|
693
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
694
|
|
|
|
|
|
|
sub _alias_candidate |
695
|
|
|
|
|
|
|
{ |
696
|
16
|
|
|
16
|
|
19
|
my $deduper = shift; |
697
|
16
|
|
|
|
|
31
|
my ($current_line,%all_alias_values) = @_; |
698
|
|
|
|
|
|
|
|
699
|
16
|
|
|
|
|
39
|
my %record_keys = $deduper->_get_key_fields($current_line); |
700
|
16
|
|
|
|
|
54
|
%record_keys = $deduper->_transform_key_fields(%record_keys); |
701
|
|
|
|
|
|
|
|
702
|
16
|
|
|
|
|
31
|
my $alias_candidate_key = ''; |
703
|
16
|
|
|
|
|
41
|
foreach my $current_key ( sort keys %record_keys ) |
704
|
|
|
|
|
|
|
{ |
705
|
|
|
|
|
|
|
|
706
|
20
|
|
|
|
|
32
|
my $current_key_data = $record_keys{$current_key}; |
707
|
20
|
100
|
|
|
|
49
|
if ( $deduper->{alias}{$current_key} ) |
708
|
|
|
|
|
|
|
{ |
709
|
16
|
50
|
|
|
|
38
|
not $all_alias_values{$current_key} and next; |
710
|
16
|
100
|
|
|
|
19
|
if ( grep(/^$current_key_data$/,@{ $all_alias_values{$current_key} }) ) |
|
16
|
|
|
|
|
204
|
|
711
|
|
|
|
|
|
|
{ |
712
|
4
|
|
|
|
|
12
|
$alias_candidate_key .= $current_key_data . ':'; |
713
|
|
|
|
|
|
|
} |
714
|
|
|
|
|
|
|
else |
715
|
|
|
|
|
|
|
{ |
716
|
12
|
|
|
|
|
44
|
return(0); |
717
|
|
|
|
|
|
|
} |
718
|
|
|
|
|
|
|
} |
719
|
|
|
|
|
|
|
else |
720
|
|
|
|
|
|
|
{ |
721
|
4
|
|
|
|
|
13
|
$alias_candidate_key .= $current_key_data . ':'; |
722
|
|
|
|
|
|
|
} |
723
|
|
|
|
|
|
|
} |
724
|
4
|
|
|
|
|
16
|
return($alias_candidate_key); |
725
|
|
|
|
|
|
|
} |
726
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
727
|
|
|
|
|
|
|
# |
728
|
|
|
|
|
|
|
sub _get_all_alias_values |
729
|
|
|
|
|
|
|
{ |
730
|
3
|
|
|
3
|
|
4
|
my $deduper = shift; |
731
|
|
|
|
|
|
|
|
732
|
3
|
|
|
|
|
4
|
my %all_alias_values; |
733
|
|
|
|
|
|
|
|
734
|
3
|
|
|
|
|
5
|
foreach my $key_number ( sort keys %{$deduper->{alias}} ) |
|
3
|
|
|
|
|
13
|
|
735
|
|
|
|
|
|
|
{ |
736
|
3
|
|
|
|
|
4
|
my %current_alias = %{ $deduper->{alias}{$key_number} }; |
|
3
|
|
|
|
|
35
|
|
737
|
3
|
|
|
|
|
5
|
my (@current_alias_values,%seen_alias_values); |
738
|
3
|
|
|
|
|
8
|
foreach my $current_alias_value ( values %current_alias ) |
739
|
|
|
|
|
|
|
{ |
740
|
9
|
100
|
|
|
|
26
|
unless ( $seen_alias_values{$current_alias_value} ) |
741
|
|
|
|
|
|
|
{ |
742
|
6
|
|
|
|
|
8
|
push(@current_alias_values,$current_alias_value); |
743
|
6
|
|
|
|
|
18
|
$seen_alias_values{$current_alias_value}++; |
744
|
|
|
|
|
|
|
} |
745
|
|
|
|
|
|
|
} |
746
|
3
|
|
|
|
|
20
|
$all_alias_values{$key_number}= [ @current_alias_values ]; |
747
|
|
|
|
|
|
|
} |
748
|
3
|
|
|
|
|
12
|
return(%all_alias_values) |
749
|
|
|
|
|
|
|
} |
750
|
|
|
|
|
|
|
|
751
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
752
|
|
|
|
|
|
|
# |
753
|
|
|
|
|
|
|
|
754
|
|
|
|
|
|
|
sub _get_key_fields |
755
|
|
|
|
|
|
|
{ |
756
|
|
|
|
|
|
|
|
757
|
32
|
|
|
32
|
|
36
|
my $deduper = shift; |
758
|
32
|
|
|
|
|
45
|
my ($current_line) = @_; |
759
|
|
|
|
|
|
|
|
760
|
32
|
|
|
|
|
34
|
my %record_keys; |
761
|
|
|
|
|
|
|
|
762
|
|
|
|
|
|
|
|
763
|
32
|
50
|
|
|
|
69
|
if ( $deduper->{key} ) |
764
|
|
|
|
|
|
|
{ |
765
|
32
|
100
|
|
|
|
89
|
if ( $deduper->{field_separator} ) |
766
|
|
|
|
|
|
|
{ |
767
|
|
|
|
|
|
|
|
768
|
|
|
|
|
|
|
# The ParseWords module will not handle single quotes within fields, |
769
|
|
|
|
|
|
|
# so add an escape sequence between any apostrophe bounded by a |
770
|
|
|
|
|
|
|
# letter on each side. Note that this applies even if there are no |
771
|
|
|
|
|
|
|
# quotes in your data, the module needs balanced quotes. |
772
|
12
|
100
|
|
|
|
40
|
if ( $current_line =~ /\w'\w/ ) |
773
|
|
|
|
|
|
|
{ |
774
|
|
|
|
|
|
|
# check for names with apostrophes, like O'Reilly |
775
|
4
|
|
|
|
|
32
|
$current_line =~ s/(\w)'(\w)/$1\\'$2/g; |
776
|
|
|
|
|
|
|
} |
777
|
|
|
|
|
|
|
|
778
|
|
|
|
|
|
|
# Use ParseWords module to spearate delimited field. |
779
|
|
|
|
|
|
|
# '0' option means don't return any quotes enclosing a field |
780
|
12
|
|
|
|
|
45
|
my (@field_data) = &Text::ParseWords::parse_line($deduper->{field_separator},0,$current_line); |
781
|
|
|
|
|
|
|
|
782
|
|
|
|
|
|
|
|
783
|
12
|
|
|
|
|
1427
|
foreach my $key_number ( sort keys %{$deduper->{key}} ) |
|
12
|
|
|
|
|
49
|
|
784
|
|
|
|
|
|
|
{ |
785
|
24
|
|
|
|
|
55
|
my $current_field_data = $field_data[$deduper->{key}->{$key_number}->{field_number} - 1]; |
786
|
24
|
50
|
|
|
|
52
|
unless ( $current_field_data ) |
787
|
|
|
|
|
|
|
{ |
788
|
|
|
|
|
|
|
# A record has less fields then we were expecting, so no |
789
|
|
|
|
|
|
|
# point searching for anymore. |
790
|
0
|
|
|
|
|
0
|
print("Short record\n"); |
791
|
0
|
|
|
|
|
0
|
print("Current line : $current_line\n"); |
792
|
0
|
|
|
|
|
0
|
print("All fields :", @field_data,"\n"); |
793
|
0
|
|
|
|
|
0
|
last; |
794
|
|
|
|
|
|
|
# TO DO, add a warning if user specifies records that must have |
795
|
|
|
|
|
|
|
# a full set of fields?? |
796
|
|
|
|
|
|
|
} |
797
|
|
|
|
|
|
|
|
798
|
24
|
50
|
|
|
|
56
|
if ( $deduper->{key}->{$key_number}->{key_length} ) |
799
|
|
|
|
|
|
|
{ |
800
|
0
|
|
|
|
|
0
|
$current_field_data = substr($current_field_data,0,$deduper->{key}->{$key_number}->{key_length}); |
801
|
|
|
|
|
|
|
} |
802
|
24
|
|
|
|
|
81
|
$record_keys{$key_number} = $current_field_data; |
803
|
|
|
|
|
|
|
} |
804
|
|
|
|
|
|
|
} |
805
|
|
|
|
|
|
|
else |
806
|
|
|
|
|
|
|
{ |
807
|
20
|
|
|
|
|
25
|
foreach my $key_number ( sort keys %{$deduper->{key}} ) |
|
20
|
|
|
|
|
63
|
|
808
|
|
|
|
|
|
|
{ |
809
|
40
|
|
|
|
|
196
|
my $current_field_data = substr($current_line,$deduper->{key}->{$key_number}->{start_pos} - 1, |
810
|
|
|
|
|
|
|
$deduper->{key}->{$key_number}->{key_length}); |
811
|
40
|
50
|
|
|
|
61
|
if ( $current_field_data ) |
812
|
|
|
|
|
|
|
{ |
813
|
40
|
|
|
|
|
107
|
$record_keys{$key_number} = $current_field_data; |
814
|
|
|
|
|
|
|
} |
815
|
|
|
|
|
|
|
else |
816
|
|
|
|
|
|
|
{ |
817
|
0
|
|
|
|
|
0
|
print("Short record\n"); |
818
|
0
|
|
|
|
|
0
|
print("Current line : $current_line\n"); |
819
|
0
|
|
|
|
|
0
|
last; |
820
|
|
|
|
|
|
|
# TO DO, add a warning if user specifies records must have |
821
|
|
|
|
|
|
|
# a full set of fields?? |
822
|
|
|
|
|
|
|
} |
823
|
|
|
|
|
|
|
} |
824
|
|
|
|
|
|
|
} |
825
|
|
|
|
|
|
|
} |
826
|
|
|
|
|
|
|
else |
827
|
|
|
|
|
|
|
{ |
828
|
|
|
|
|
|
|
# no key fileds defined, use whole line as key |
829
|
0
|
|
|
|
|
0
|
$record_keys{1} = $current_line; |
830
|
|
|
|
|
|
|
} |
831
|
32
|
|
|
|
|
314
|
return(%record_keys); |
832
|
|
|
|
|
|
|
} |
833
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
834
|
|
|
|
|
|
|
# |
835
|
|
|
|
|
|
|
|
836
|
|
|
|
|
|
|
sub _transform_key_fields |
837
|
|
|
|
|
|
|
{ |
838
|
32
|
|
|
32
|
|
43
|
my $deduper = shift; |
839
|
32
|
|
|
|
|
64
|
my (%record_keys) = @_; |
840
|
|
|
|
|
|
|
|
841
|
32
|
50
|
|
|
|
84
|
if ( $deduper->{ignore_whitespace} ) |
842
|
|
|
|
|
|
|
{ |
843
|
32
|
|
|
|
|
35
|
foreach my $key_number ( keys %{$deduper->{ignore_whitespace}} ) |
|
32
|
|
|
|
|
75
|
|
844
|
|
|
|
|
|
|
{ |
845
|
|
|
|
|
|
|
# strip out leading and/or trailing whitespace |
846
|
44
|
|
|
|
|
91
|
$record_keys{$key_number} =~ s/^\s+//; |
847
|
44
|
|
|
|
|
162
|
$record_keys{$key_number} =~ s/\s+$//; |
848
|
|
|
|
|
|
|
} |
849
|
|
|
|
|
|
|
} |
850
|
|
|
|
|
|
|
|
851
|
32
|
50
|
|
|
|
96
|
if ( $deduper->{ignore_case} ) |
852
|
|
|
|
|
|
|
{ |
853
|
|
|
|
|
|
|
# Transform every field where ignore_case was specified |
854
|
|
|
|
|
|
|
|
855
|
32
|
|
|
|
|
34
|
foreach my $key_number ( keys %{$deduper->{ignore_case}} ) |
|
32
|
|
|
|
|
77
|
|
856
|
|
|
|
|
|
|
{ |
857
|
|
|
|
|
|
|
# If this key is case insensitive, fold data to upper case |
858
|
64
|
|
|
|
|
144
|
$record_keys{$key_number} = uc($record_keys{$key_number}); |
859
|
|
|
|
|
|
|
} |
860
|
|
|
|
|
|
|
} |
861
|
32
|
|
|
|
|
167
|
return(%record_keys); |
862
|
|
|
|
|
|
|
} |
863
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
864
|
|
|
|
|
|
|
# |
865
|
|
|
|
|
|
|
|
866
|
|
|
|
|
|
|
sub _assemble_full_key |
867
|
|
|
|
|
|
|
{ |
868
|
16
|
|
|
16
|
|
33
|
my (%record_keys) = @_; |
869
|
16
|
|
|
|
|
16
|
my $full_key; |
870
|
|
|
|
|
|
|
my @each_key; |
871
|
16
|
|
|
|
|
37
|
foreach my $current_key ( sort keys %record_keys ) |
872
|
|
|
|
|
|
|
{ |
873
|
32
|
|
|
|
|
59
|
push(@each_key,$record_keys{$current_key}); |
874
|
|
|
|
|
|
|
} |
875
|
16
|
|
|
|
|
38
|
$full_key = join('_',@each_key); |
876
|
16
|
|
|
|
|
46
|
return($full_key); |
877
|
|
|
|
|
|
|
|
878
|
|
|
|
|
|
|
} |
879
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
880
|
|
|
|
|
|
|
# |
881
|
|
|
|
|
|
|
|
882
|
|
|
|
|
|
|
sub _alias_dupe |
883
|
|
|
|
|
|
|
{ |
884
|
16
|
|
|
16
|
|
17
|
my $deduper = shift; |
885
|
16
|
|
|
|
|
34
|
my ($alias_candidates_ref,%record_keys) = @_; |
886
|
|
|
|
|
|
|
|
887
|
|
|
|
|
|
|
|
888
|
16
|
|
|
|
|
19
|
my $alias_dupe = 0; |
889
|
16
|
50
|
|
|
|
41
|
if ( $deduper->{alias} ) |
890
|
|
|
|
|
|
|
{ |
891
|
16
|
|
|
|
|
15
|
my $alias_was_substituted = 0; |
892
|
|
|
|
|
|
|
|
893
|
16
|
|
|
|
|
18
|
foreach my $key_number ( keys %{$deduper->{alias}} ) |
|
16
|
|
|
|
|
38
|
|
894
|
|
|
|
|
|
|
{ |
895
|
16
|
|
|
|
|
13
|
my %current_alias = %{ $deduper->{alias}{$key_number} }; |
|
16
|
|
|
|
|
66
|
|
896
|
16
|
|
|
|
|
35
|
foreach my $current_alias_key ( keys %current_alias ) |
897
|
|
|
|
|
|
|
{ |
898
|
39
|
100
|
|
|
|
123
|
if ( $record_keys{$key_number} eq $current_alias_key ) |
899
|
|
|
|
|
|
|
{ |
900
|
8
|
|
|
|
|
9
|
$alias_was_substituted = 1; |
901
|
8
|
|
|
|
|
12
|
$record_keys{$key_number} = $current_alias{$current_alias_key}; |
902
|
8
|
|
|
|
|
27
|
last; |
903
|
|
|
|
|
|
|
} |
904
|
|
|
|
|
|
|
} |
905
|
|
|
|
|
|
|
} |
906
|
16
|
100
|
|
|
|
43
|
if ( $alias_was_substituted ) |
907
|
|
|
|
|
|
|
{ |
908
|
8
|
|
|
|
|
10
|
my $full_key; |
909
|
8
|
|
|
|
|
18
|
foreach my $current_key ( sort keys %record_keys ) |
910
|
|
|
|
|
|
|
{ |
911
|
16
|
|
|
|
|
33
|
$full_key .= $record_keys{$current_key} . ':'; |
912
|
|
|
|
|
|
|
} |
913
|
8
|
100
|
|
|
|
22
|
if ( $alias_candidates_ref->{$full_key} ) |
914
|
|
|
|
|
|
|
{ |
915
|
4
|
|
|
|
|
8
|
$alias_dupe = $alias_candidates_ref->{$full_key}; |
916
|
|
|
|
|
|
|
} |
917
|
|
|
|
|
|
|
} |
918
|
|
|
|
|
|
|
} |
919
|
|
|
|
|
|
|
# returns the number of the orignal unique record for which this current record is an alias dupe of |
920
|
16
|
|
|
|
|
63
|
return($alias_dupe); |
921
|
|
|
|
|
|
|
} |
922
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
923
|
|
|
|
|
|
|
# |
924
|
|
|
|
|
|
|
|
925
|
|
|
|
|
|
|
sub _exact_dupe |
926
|
|
|
|
|
|
|
{ |
927
|
12
|
|
|
12
|
|
16
|
my $deduper = shift; |
928
|
12
|
|
|
|
|
27
|
my ($full_key,%seen_exact_dupes) = @_; |
929
|
|
|
|
|
|
|
# problem with unitialized value, set to undef?? |
930
|
12
|
100
|
|
|
|
23
|
if ( $seen_exact_dupes{$full_key} ) |
931
|
|
|
|
|
|
|
{ |
932
|
5
|
|
|
|
|
17
|
return($seen_exact_dupes{$full_key}); |
933
|
|
|
|
|
|
|
} |
934
|
|
|
|
|
|
|
else |
935
|
|
|
|
|
|
|
{ |
936
|
7
|
|
|
|
|
20
|
return(0); |
937
|
|
|
|
|
|
|
} |
938
|
|
|
|
|
|
|
} |
939
|
|
|
|
|
|
|
|
940
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
941
|
|
|
|
|
|
|
# |
942
|
|
|
|
|
|
|
|
943
|
|
|
|
|
|
|
sub _read_one_record |
944
|
|
|
|
|
|
|
{ |
945
|
48
|
|
|
48
|
|
78
|
my ($storage_type,$current_record_number,$input_array_ref,$input_fh) = @_; |
946
|
|
|
|
|
|
|
|
947
|
48
|
|
|
|
|
51
|
my $finished = 0; |
948
|
48
|
|
|
|
|
51
|
my $current_line; |
949
|
|
|
|
|
|
|
|
950
|
48
|
50
|
|
|
|
116
|
if ( $storage_type eq 'file' ) |
|
|
50
|
|
|
|
|
|
951
|
|
|
|
|
|
|
{ |
952
|
0
|
0
|
|
|
|
0
|
if ( $current_line = $input_fh->getline ) |
953
|
|
|
|
|
|
|
{ |
954
|
0
|
|
|
|
|
0
|
chomp($current_line); |
955
|
0
|
0
|
|
|
|
0
|
if ( $input_fh->eof ) |
956
|
|
|
|
|
|
|
{ |
957
|
0
|
|
|
|
|
0
|
$finished = 1; |
958
|
|
|
|
|
|
|
} |
959
|
|
|
|
|
|
|
} |
960
|
|
|
|
|
|
|
else |
961
|
|
|
|
|
|
|
{ |
962
|
0
|
|
|
|
|
0
|
warn "Could not read line from input file"; |
963
|
0
|
|
|
|
|
0
|
$finished = 1; |
964
|
|
|
|
|
|
|
} |
965
|
|
|
|
|
|
|
} |
966
|
|
|
|
|
|
|
elsif ( $storage_type eq 'array' ) |
967
|
|
|
|
|
|
|
{ |
968
|
48
|
|
|
|
|
73
|
$current_line = @$input_array_ref[$current_record_number]; |
969
|
48
|
|
|
|
|
70
|
my $last_element = @$input_array_ref - 1; |
970
|
48
|
100
|
|
|
|
128
|
if ( $current_record_number == $last_element ) |
|
|
50
|
|
|
|
|
|
971
|
|
|
|
|
|
|
{ |
972
|
9
|
|
|
|
|
13
|
$finished = 1; |
973
|
|
|
|
|
|
|
} |
974
|
|
|
|
|
|
|
elsif ( $current_record_number > $last_element ) |
975
|
|
|
|
|
|
|
{ |
976
|
0
|
|
|
|
|
0
|
warn "You are trying to access beyond the input array boundaries"; |
977
|
0
|
|
|
|
|
0
|
$finished = 1; |
978
|
|
|
|
|
|
|
} |
979
|
|
|
|
|
|
|
} |
980
|
|
|
|
|
|
|
else |
981
|
|
|
|
|
|
|
{ |
982
|
0
|
|
|
|
|
0
|
warn "Illegal storage type"; |
983
|
0
|
|
|
|
|
0
|
$finished = 1; |
984
|
|
|
|
|
|
|
} |
985
|
48
|
|
|
|
|
125
|
return($current_line,$finished); |
986
|
|
|
|
|
|
|
} |
987
|
|
|
|
|
|
|
|
988
|
|
|
|
|
|
|
|
989
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
990
|
|
|
|
|
|
|
# |
991
|
|
|
|
|
|
|
|
992
|
|
|
|
|
|
|
sub _classify_record |
993
|
|
|
|
|
|
|
{ |
994
|
16
|
|
|
16
|
|
27
|
my ($dupe_type,$parent_record_number,$current_record_number,$full_key,$master_record_index) = @_; |
995
|
16
|
|
|
|
|
50
|
$master_record_index->{$current_record_number}{key} = $full_key; |
996
|
16
|
|
|
|
|
29
|
$master_record_index->{$current_record_number}{parent} = $parent_record_number; |
997
|
16
|
|
|
|
|
28
|
$master_record_index->{$current_record_number}{type} = $dupe_type; |
998
|
|
|
|
|
|
|
|
999
|
|
|
|
|
|
|
# If there is a parent, update it now, so that is marked as the first |
1000
|
|
|
|
|
|
|
# dupe in a set (of current type alias, indentical etc). Note that |
1001
|
|
|
|
|
|
|
# a record can be parent to several record types, eg alias and indentical |
1002
|
|
|
|
|
|
|
# Currently only updating from last child, but may want to record all in future. |
1003
|
|
|
|
|
|
|
|
1004
|
16
|
100
|
|
|
|
50
|
if ( $parent_record_number ) |
1005
|
|
|
|
|
|
|
{ |
1006
|
9
|
|
|
|
|
17
|
$master_record_index->{$parent_record_number}{type} = $dupe_type; |
1007
|
9
|
|
|
|
|
102
|
$master_record_index->{$parent_record_number}{parent} = 0; |
1008
|
|
|
|
|
|
|
} |
1009
|
|
|
|
|
|
|
} |
1010
|
|
|
|
|
|
|
|
1011
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
1012
|
|
|
|
|
|
|
# |
1013
|
|
|
|
|
|
|
|
1014
|
|
|
|
|
|
|
sub _separate |
1015
|
|
|
|
|
|
|
{ |
1016
|
3
|
|
|
3
|
|
36
|
my $deduper = shift; |
1017
|
3
|
|
|
|
|
7
|
my ($storage_type,$master_record_index_ref,$input_array_ref,$input_fh,$unique_fh,$dupes_fh) = @_; |
1018
|
|
|
|
|
|
|
|
1019
|
3
|
50
|
|
|
|
10
|
if ( $storage_type eq 'file' ) |
1020
|
|
|
|
|
|
|
{ |
1021
|
0
|
|
|
|
|
0
|
_rewind_file($input_fh); |
1022
|
|
|
|
|
|
|
} |
1023
|
|
|
|
|
|
|
|
1024
|
3
|
|
|
|
|
5
|
my $unique_ref = []; |
1025
|
3
|
|
|
|
|
4
|
my $dupe_ref = []; |
1026
|
|
|
|
|
|
|
|
1027
|
|
|
|
|
|
|
|
1028
|
3
|
|
|
|
|
4
|
my $current_record_number = 0; |
1029
|
3
|
|
|
|
|
4
|
my $current_line; |
1030
|
3
|
|
|
|
|
4
|
my $finished = 0; |
1031
|
|
|
|
|
|
|
|
1032
|
|
|
|
|
|
|
|
1033
|
3
|
|
|
|
|
7
|
while ( not $finished ) |
1034
|
|
|
|
|
|
|
{ |
1035
|
16
|
|
|
|
|
20
|
my $current_line; |
1036
|
16
|
|
|
|
|
26
|
($current_line,$finished) = _read_one_record($storage_type,$current_record_number,$input_array_ref,$input_fh); |
1037
|
16
|
|
|
|
|
22
|
$current_record_number++; |
1038
|
16
|
|
|
|
|
31
|
my $dupe_type = $master_record_index_ref->{$current_record_number}{type}; |
1039
|
16
|
|
|
|
|
26
|
my $parent_record_number = $master_record_index_ref->{$current_record_number}{parent}; |
1040
|
|
|
|
|
|
|
|
1041
|
|
|
|
|
|
|
# The first duplicate in a set of 1 or more dupes (the parent), is treated as a unique record |
1042
|
|
|
|
|
|
|
# TO DO!!! allow user to define this initial dupe as not unique, and group with it's childeren dupes |
1043
|
|
|
|
|
|
|
# TO DO!!! separate out to alias, soundex dupes to their own file if needed |
1044
|
16
|
100
|
|
|
|
37
|
if ( $parent_record_number == 0 ) |
1045
|
|
|
|
|
|
|
{ |
1046
|
7
|
|
|
|
|
11
|
$dupe_type = 'unique'; |
1047
|
|
|
|
|
|
|
} |
1048
|
|
|
|
|
|
|
|
1049
|
16
|
|
|
|
|
25
|
_write_one_record($storage_type,$dupe_type,$current_line,$unique_ref,$dupe_ref,$input_fh,$unique_fh,$dupes_fh); |
1050
|
|
|
|
|
|
|
} |
1051
|
3
|
|
|
|
|
10
|
return($unique_ref,$dupe_ref); |
1052
|
|
|
|
|
|
|
|
1053
|
|
|
|
|
|
|
} |
1054
|
|
|
|
|
|
|
#------------------------------------------------------------------------------- |
1055
|
|
|
|
|
|
|
# |
1056
|
|
|
|
|
|
|
|
1057
|
|
|
|
|
|
|
sub _write_one_record |
1058
|
|
|
|
|
|
|
{ |
1059
|
16
|
|
|
16
|
|
31
|
my ($storage_type,$dupe_type,$current_line,$unique_ref,$dupe_ref,$input_fh,$unique_fh,$dupes_fh) = @_; |
1060
|
|
|
|
|
|
|
|
1061
|
16
|
50
|
|
|
|
45
|
if ( $storage_type eq 'file' ) |
|
|
50
|
|
|
|
|
|
1062
|
|
|
|
|
|
|
{ |
1063
|
|
|
|
|
|
|
|
1064
|
0
|
0
|
|
|
|
0
|
if ( $dupe_type eq 'unique' ) |
1065
|
|
|
|
|
|
|
{ |
1066
|
0
|
|
|
|
|
0
|
$unique_fh->print("$current_line\n"); |
1067
|
|
|
|
|
|
|
} |
1068
|
|
|
|
|
|
|
else |
1069
|
|
|
|
|
|
|
{ |
1070
|
0
|
|
|
|
|
0
|
$dupes_fh->print("$current_line\n"); |
1071
|
|
|
|
|
|
|
} |
1072
|
|
|
|
|
|
|
} |
1073
|
|
|
|
|
|
|
elsif ( $storage_type eq 'array' ) |
1074
|
|
|
|
|
|
|
{ |
1075
|
16
|
100
|
|
|
|
28
|
if ( $dupe_type eq 'unique' ) |
1076
|
|
|
|
|
|
|
{ |
1077
|
7
|
|
|
|
|
25
|
push(@$unique_ref,$current_line); |
1078
|
|
|
|
|
|
|
} |
1079
|
|
|
|
|
|
|
else |
1080
|
|
|
|
|
|
|
{ |
1081
|
|
|
|
|
|
|
# TO DO!!! separate out to alias, soundex dupes etc if needed |
1082
|
9
|
|
|
|
|
31
|
push(@$dupe_ref,$current_line); |
1083
|
|
|
|
|
|
|
} |
1084
|
|
|
|
|
|
|
} |
1085
|
|
|
|
|
|
|
} |
1086
|
|
|
|
|
|
|
|
1087
|
|
|
|
|
|
|
|
1088
|
|
|
|
|
|
|
1; |
1089
|
|
|
|
|
|
|
|