line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
#=============================================================================== |
2
|
|
|
|
|
|
|
# |
3
|
|
|
|
|
|
|
# FILE: String.pm |
4
|
|
|
|
|
|
|
# |
5
|
|
|
|
|
|
|
# DESCRIPTION: Utilities for easy string processing |
6
|
|
|
|
|
|
|
# |
7
|
|
|
|
|
|
|
# NOTE: This module ported from Wono framework |
8
|
|
|
|
|
|
|
# AUTHOR: Michael Bochkaryov (Rattler), <misha@rattler.kiev.ua> |
9
|
|
|
|
|
|
|
# COMPANY: Net.Style |
10
|
|
|
|
|
|
|
# VERSION: 1.044 |
11
|
|
|
|
|
|
|
# CREATED: 03.08.2008 15:04:22 EEST |
12
|
|
|
|
|
|
|
#=============================================================================== |
13
|
|
|
|
|
|
|
|
14
|
|
|
|
|
|
|
=head1 NAME |
15
|
|
|
|
|
|
|
|
16
|
|
|
|
|
|
|
NetSDS::Util::String - string prcessing routines |
17
|
|
|
|
|
|
|
|
18
|
|
|
|
|
|
|
=head1 SYNOPSIS |
19
|
|
|
|
|
|
|
|
20
|
|
|
|
|
|
|
use NetSDS::Util::String qw(); |
21
|
|
|
|
|
|
|
|
22
|
|
|
|
|
|
|
# Read from standard input |
23
|
|
|
|
|
|
|
my $string = <STDIN>; |
24
|
|
|
|
|
|
|
|
25
|
|
|
|
|
|
|
# Encode string to internal structure |
26
|
|
|
|
|
|
|
$string = string_encode($tring); |
27
|
|
|
|
|
|
|
|
28
|
|
|
|
|
|
|
|
29
|
|
|
|
|
|
|
=head1 DESCRIPTION |
30
|
|
|
|
|
|
|
|
31
|
|
|
|
|
|
|
C<NetSDS::Util::String> module contains functions may be used to quickly solve |
32
|
|
|
|
|
|
|
string processing tasks like parsing, recoding, formatting. |
33
|
|
|
|
|
|
|
|
34
|
|
|
|
|
|
|
As in other NetSDS modules standard encoding is UTF-8. |
35
|
|
|
|
|
|
|
|
36
|
|
|
|
|
|
|
=cut |
37
|
|
|
|
|
|
|
|
38
|
|
|
|
|
|
|
package NetSDS::Util::String; |
39
|
|
|
|
|
|
|
|
40
|
2
|
|
|
2
|
|
13325
|
use 5.8.0; |
|
2
|
|
|
|
|
9
|
|
|
2
|
|
|
|
|
115
|
|
41
|
2
|
|
|
2
|
|
16
|
use warnings 'all'; |
|
2
|
|
|
|
|
5
|
|
|
2
|
|
|
|
|
86
|
|
42
|
2
|
|
|
2
|
|
13
|
use strict; |
|
2
|
|
|
|
|
13
|
|
|
2
|
|
|
|
|
89
|
|
43
|
|
|
|
|
|
|
|
44
|
2
|
|
|
2
|
|
13
|
use base 'Exporter'; |
|
2
|
|
|
|
|
4
|
|
|
2
|
|
|
|
|
287
|
|
45
|
|
|
|
|
|
|
|
46
|
2
|
|
|
2
|
|
14
|
use version; our $VERSION = '1.044'; |
|
2
|
|
|
|
|
5
|
|
|
2
|
|
|
|
|
21
|
|
47
|
|
|
|
|
|
|
|
48
|
|
|
|
|
|
|
our @EXPORT = qw( |
49
|
|
|
|
|
|
|
str_encode |
50
|
|
|
|
|
|
|
str_decode |
51
|
|
|
|
|
|
|
str_recode |
52
|
|
|
|
|
|
|
str_trim |
53
|
|
|
|
|
|
|
str_trim_left |
54
|
|
|
|
|
|
|
str_trim_right |
55
|
|
|
|
|
|
|
str_clean |
56
|
|
|
|
|
|
|
str_camelize |
57
|
|
|
|
|
|
|
str_decamelize |
58
|
|
|
|
|
|
|
); |
59
|
|
|
|
|
|
|
|
60
|
2
|
|
|
2
|
|
219
|
use POSIX; |
|
2
|
|
|
|
|
4
|
|
|
2
|
|
|
|
|
19
|
|
61
|
2
|
|
|
|
|
258
|
use Encode qw( |
62
|
|
|
|
|
|
|
encode |
63
|
|
|
|
|
|
|
decode |
64
|
|
|
|
|
|
|
encode_utf8 |
65
|
|
|
|
|
|
|
decode_utf8 |
66
|
|
|
|
|
|
|
from_to |
67
|
|
|
|
|
|
|
is_utf8 |
68
|
2
|
|
|
2
|
|
8336
|
); |
|
2
|
|
|
|
|
12376
|
|
69
|
|
|
|
|
|
|
|
70
|
|
|
|
|
|
|
my $BLANK = "[:blank:][:space:][:cntrl:]"; |
71
|
|
|
|
|
|
|
|
72
|
2
|
|
|
2
|
|
14
|
use constant DEFAULT_ENCODING => 'UTF-8'; |
|
2
|
|
|
|
|
6
|
|
|
2
|
|
|
|
|
1764
|
|
73
|
|
|
|
|
|
|
|
74
|
|
|
|
|
|
|
#*********************************************************************** |
75
|
|
|
|
|
|
|
# |
76
|
|
|
|
|
|
|
# ENCODING/DECODING/RECODING FUNCTIONS |
77
|
|
|
|
|
|
|
# |
78
|
|
|
|
|
|
|
#*********************************************************************** |
79
|
|
|
|
|
|
|
|
80
|
|
|
|
|
|
|
=head1 EXPORTED FUNCTIONS |
81
|
|
|
|
|
|
|
|
82
|
|
|
|
|
|
|
=over |
83
|
|
|
|
|
|
|
|
84
|
|
|
|
|
|
|
=item B<str_encode($str[, $encoding])> - encode string to internal UTF-8 |
85
|
|
|
|
|
|
|
|
86
|
|
|
|
|
|
|
By default this function treat first argument as byte string in UTF-8 |
87
|
|
|
|
|
|
|
and return it's internal Unicode representation. |
88
|
|
|
|
|
|
|
|
89
|
|
|
|
|
|
|
In case of external character set isn't UTF-8 it should be added as second |
90
|
|
|
|
|
|
|
argument of function. |
91
|
|
|
|
|
|
|
|
92
|
|
|
|
|
|
|
|
93
|
|
|
|
|
|
|
# Convert UTF-8 byte string to internal Unicode representation |
94
|
|
|
|
|
|
|
$uni_string = str_encode($byte_string); |
95
|
|
|
|
|
|
|
|
96
|
|
|
|
|
|
|
# Convert KOI8-U byte string to internal |
97
|
|
|
|
|
|
|
$uni_string = str_encode($koi8_string, 'KOI8-U'); |
98
|
|
|
|
|
|
|
|
99
|
|
|
|
|
|
|
After C<str_encode()> it's possible to process this string correctly |
100
|
|
|
|
|
|
|
including regular expressions. All characters will be understood |
101
|
|
|
|
|
|
|
as UTF-8 symbols instead of byte sequences. |
102
|
|
|
|
|
|
|
|
103
|
|
|
|
|
|
|
=cut |
104
|
|
|
|
|
|
|
|
105
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
106
|
|
|
|
|
|
|
sub str_encode { |
107
|
0
|
|
|
0
|
1
|
|
my ( $txt, $enc ) = @_; |
108
|
|
|
|
|
|
|
|
109
|
0
|
0
|
0
|
|
|
|
if ( defined($txt) and ( $txt ne '' ) ) { |
110
|
0
|
0
|
|
|
|
|
unless ( is_utf8($txt) ) { |
111
|
0
|
|
0
|
|
|
|
$txt = decode( $enc || DEFAULT_ENCODING, $txt ); |
112
|
|
|
|
|
|
|
} |
113
|
|
|
|
|
|
|
} |
114
|
|
|
|
|
|
|
|
115
|
0
|
|
|
|
|
|
return $txt; |
116
|
|
|
|
|
|
|
} |
117
|
|
|
|
|
|
|
|
118
|
|
|
|
|
|
|
#*********************************************************************** |
119
|
|
|
|
|
|
|
|
120
|
|
|
|
|
|
|
=item B<str_decode($str[, $encoding])> - decode internal UTF-8 to byte string |
121
|
|
|
|
|
|
|
|
122
|
|
|
|
|
|
|
By default this function treat first argument as string in internal UTF-8 |
123
|
|
|
|
|
|
|
and return it in byte string (external) representation. |
124
|
|
|
|
|
|
|
|
125
|
|
|
|
|
|
|
In case of external character set isn't UTF-8 it should be added as second |
126
|
|
|
|
|
|
|
argument of function. |
127
|
|
|
|
|
|
|
|
128
|
|
|
|
|
|
|
|
129
|
|
|
|
|
|
|
# Get UTF-8 byte string from internal Unicode representation |
130
|
|
|
|
|
|
|
$byte_string = str_decode($uni_string); |
131
|
|
|
|
|
|
|
|
132
|
|
|
|
|
|
|
# Convert to KOI8-U byte string from internal Unicode |
133
|
|
|
|
|
|
|
$koi8_string = str_encode($uni_string, 'KOI8-U'); |
134
|
|
|
|
|
|
|
|
135
|
|
|
|
|
|
|
It's recommended to use C<str_encode()> when preparing data for |
136
|
|
|
|
|
|
|
communication with external systems (especially networking). |
137
|
|
|
|
|
|
|
|
138
|
|
|
|
|
|
|
=cut |
139
|
|
|
|
|
|
|
|
140
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
141
|
|
|
|
|
|
|
sub str_decode { |
142
|
0
|
|
|
0
|
1
|
|
my ( $txt, $enc ) = @_; |
143
|
|
|
|
|
|
|
|
144
|
0
|
0
|
0
|
|
|
|
if ( defined($txt) and ( $txt ne '' ) ) { |
145
|
0
|
0
|
|
|
|
|
if ( is_utf8($txt) ) { |
146
|
0
|
|
0
|
|
|
|
$txt = encode( $enc || DEFAULT_ENCODING, $txt ); |
147
|
|
|
|
|
|
|
} |
148
|
|
|
|
|
|
|
} |
149
|
|
|
|
|
|
|
|
150
|
0
|
|
|
|
|
|
return $txt; |
151
|
|
|
|
|
|
|
} |
152
|
|
|
|
|
|
|
|
153
|
|
|
|
|
|
|
#*********************************************************************** |
154
|
|
|
|
|
|
|
|
155
|
|
|
|
|
|
|
=item B<str_recode($str, $FROM_ENC[, $TO_ENC])> - recode string |
156
|
|
|
|
|
|
|
|
157
|
|
|
|
|
|
|
Translate string between different encodings. |
158
|
|
|
|
|
|
|
If target encoding is not set UTF-8 used as default one. |
159
|
|
|
|
|
|
|
|
160
|
|
|
|
|
|
|
=cut |
161
|
|
|
|
|
|
|
|
162
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
163
|
|
|
|
|
|
|
sub str_recode { |
164
|
0
|
|
|
0
|
1
|
|
my ( $txt, $enc, $trg ) = @_; |
165
|
|
|
|
|
|
|
|
166
|
0
|
0
|
0
|
|
|
|
if ( defined($txt) and ( $txt ne '' ) ) { |
167
|
0
|
0
|
|
|
|
|
if ($enc) { |
168
|
0
|
|
0
|
|
|
|
my $len = from_to( $txt, $enc, $trg || DEFAULT_ENCODING ); |
169
|
0
|
0
|
|
|
|
|
unless ( defined($len) ) { |
170
|
0
|
|
|
|
|
|
$txt = undef; |
171
|
|
|
|
|
|
|
} |
172
|
|
|
|
|
|
|
} |
173
|
|
|
|
|
|
|
} |
174
|
|
|
|
|
|
|
|
175
|
0
|
|
|
|
|
|
return $txt; |
176
|
|
|
|
|
|
|
} |
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
#*********************************************************************** |
179
|
|
|
|
|
|
|
# |
180
|
|
|
|
|
|
|
# CLEANING STRINGS |
181
|
|
|
|
|
|
|
# |
182
|
|
|
|
|
|
|
#*********************************************************************** |
183
|
|
|
|
|
|
|
|
184
|
|
|
|
|
|
|
=item B<str_trim($str)> - remove leading/trailing space characters |
185
|
|
|
|
|
|
|
|
186
|
|
|
|
|
|
|
$orig_str = " string with spaces "; |
187
|
|
|
|
|
|
|
$new_str = str_trim($orig_str); |
188
|
|
|
|
|
|
|
|
189
|
|
|
|
|
|
|
# Output: "string with spaces" |
190
|
|
|
|
|
|
|
print $new_str; |
191
|
|
|
|
|
|
|
|
192
|
|
|
|
|
|
|
=cut |
193
|
|
|
|
|
|
|
|
194
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
195
|
|
|
|
|
|
|
sub str_trim { |
196
|
0
|
|
|
0
|
1
|
|
my ($s) = @_; |
197
|
|
|
|
|
|
|
|
198
|
0
|
0
|
0
|
|
|
|
if ( defined($s) and ( $s ne '' ) ) { |
199
|
0
|
|
|
|
|
|
$s =~ s/^[$BLANK]+//s; |
200
|
0
|
|
|
|
|
|
$s =~ s/[$BLANK]+$//s; |
201
|
|
|
|
|
|
|
} |
202
|
|
|
|
|
|
|
|
203
|
0
|
|
|
|
|
|
return $s; |
204
|
|
|
|
|
|
|
} |
205
|
|
|
|
|
|
|
|
206
|
|
|
|
|
|
|
#*********************************************************************** |
207
|
|
|
|
|
|
|
|
208
|
|
|
|
|
|
|
=item B<str_trim_left($str)> - removes leading whitespaces |
209
|
|
|
|
|
|
|
|
210
|
|
|
|
|
|
|
This function is similar to C<str_trim()> except of it removes only |
211
|
|
|
|
|
|
|
leading space characters and leave trailing ones. |
212
|
|
|
|
|
|
|
|
213
|
|
|
|
|
|
|
=cut |
214
|
|
|
|
|
|
|
|
215
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
216
|
|
|
|
|
|
|
sub str_trim_left { |
217
|
0
|
0
|
|
0
|
1
|
|
my ($s) = @_ ? @_ : $_; |
218
|
|
|
|
|
|
|
|
219
|
0
|
0
|
0
|
|
|
|
if ( defined($s) and ( $s ne '' ) ) { |
220
|
0
|
|
|
|
|
|
$s =~ s/^[$BLANK]+//s; |
221
|
|
|
|
|
|
|
} |
222
|
|
|
|
|
|
|
|
223
|
0
|
|
|
|
|
|
return $s; |
224
|
|
|
|
|
|
|
} |
225
|
|
|
|
|
|
|
|
226
|
|
|
|
|
|
|
#*********************************************************************** |
227
|
|
|
|
|
|
|
|
228
|
|
|
|
|
|
|
=item B<str_trim_right($str)> - removes trailing whitespaces |
229
|
|
|
|
|
|
|
|
230
|
|
|
|
|
|
|
This function is similar to C<str_trim()> except of it removes only |
231
|
|
|
|
|
|
|
trailing space characters and leave leading ones. |
232
|
|
|
|
|
|
|
|
233
|
|
|
|
|
|
|
=cut |
234
|
|
|
|
|
|
|
|
235
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
236
|
|
|
|
|
|
|
sub str_trim_right { |
237
|
0
|
0
|
|
0
|
1
|
|
my ($s) = @_ ? @_ : $_; |
238
|
|
|
|
|
|
|
|
239
|
0
|
0
|
0
|
|
|
|
if ( defined($s) and ( $s ne '' ) ) { |
240
|
0
|
|
|
|
|
|
$s =~ s/[$BLANK]+$//s; |
241
|
|
|
|
|
|
|
} |
242
|
|
|
|
|
|
|
|
243
|
0
|
|
|
|
|
|
return $s; |
244
|
|
|
|
|
|
|
} |
245
|
|
|
|
|
|
|
|
246
|
|
|
|
|
|
|
#*********************************************************************** |
247
|
|
|
|
|
|
|
|
248
|
|
|
|
|
|
|
=item B<str_clean($str)> - clean string from extra spaces |
249
|
|
|
|
|
|
|
|
250
|
|
|
|
|
|
|
Function is similar to C<str_trim()> but also changes all spacing chains |
251
|
|
|
|
|
|
|
inside string to single spaces. |
252
|
|
|
|
|
|
|
|
253
|
|
|
|
|
|
|
=cut |
254
|
|
|
|
|
|
|
|
255
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
256
|
|
|
|
|
|
|
sub str_clean { |
257
|
|
|
|
|
|
|
|
258
|
0
|
|
|
0
|
1
|
|
my ($txt) = @_; |
259
|
|
|
|
|
|
|
|
260
|
0
|
0
|
0
|
|
|
|
if ( defined($txt) and ( $txt ne '' ) ) { |
261
|
0
|
|
|
|
|
|
$txt =~ s/^[$BLANK]+//s; |
262
|
0
|
|
|
|
|
|
$txt =~ s/[$BLANK]+$//s; |
263
|
0
|
|
|
|
|
|
$txt =~ s/[$BLANK]+/ /gs; |
264
|
|
|
|
|
|
|
} |
265
|
|
|
|
|
|
|
|
266
|
0
|
|
|
|
|
|
return $txt; |
267
|
|
|
|
|
|
|
} |
268
|
|
|
|
|
|
|
|
269
|
|
|
|
|
|
|
#************************************************************************** |
270
|
|
|
|
|
|
|
|
271
|
|
|
|
|
|
|
=item B<str_camelize($strin)> |
272
|
|
|
|
|
|
|
|
273
|
|
|
|
|
|
|
If pass undef - return undef. |
274
|
|
|
|
|
|
|
If pass '' - return ''. |
275
|
|
|
|
|
|
|
|
276
|
|
|
|
|
|
|
Examples: |
277
|
|
|
|
|
|
|
|
278
|
|
|
|
|
|
|
# returns 'getValue' |
279
|
|
|
|
|
|
|
str_camelize( 'get_value' ) |
280
|
|
|
|
|
|
|
|
281
|
|
|
|
|
|
|
# returns 'addUserAction' |
282
|
|
|
|
|
|
|
str_camelize( 'ADD_User_actION' ) |
283
|
|
|
|
|
|
|
|
284
|
|
|
|
|
|
|
=cut |
285
|
|
|
|
|
|
|
|
286
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
287
|
|
|
|
|
|
|
sub str_camelize { |
288
|
|
|
|
|
|
|
|
289
|
0
|
|
|
0
|
1
|
|
my $s = shift; |
290
|
|
|
|
|
|
|
|
291
|
0
|
0
|
0
|
|
|
|
if ( defined($s) and ( $s ne '' ) ) { |
292
|
0
|
|
|
|
|
|
$s = lc($s); |
293
|
0
|
|
|
|
|
|
$s =~ s/_([0-9a-z])/\U$1/g; |
294
|
|
|
|
|
|
|
} |
295
|
|
|
|
|
|
|
|
296
|
0
|
|
|
|
|
|
return $s; |
297
|
|
|
|
|
|
|
} |
298
|
|
|
|
|
|
|
|
299
|
|
|
|
|
|
|
#************************************************************************** |
300
|
|
|
|
|
|
|
|
301
|
|
|
|
|
|
|
=item B<str_decamelize(...)> |
302
|
|
|
|
|
|
|
|
303
|
|
|
|
|
|
|
If pass undef - return undef. |
304
|
|
|
|
|
|
|
If pass '' - return ''. |
305
|
|
|
|
|
|
|
|
306
|
|
|
|
|
|
|
Examples: |
307
|
|
|
|
|
|
|
|
308
|
|
|
|
|
|
|
# returns 'get_value' |
309
|
|
|
|
|
|
|
str_decamelize( 'getValue' ) |
310
|
|
|
|
|
|
|
|
311
|
|
|
|
|
|
|
=cut |
312
|
|
|
|
|
|
|
|
313
|
|
|
|
|
|
|
#----------------------------------------------------------------------- |
314
|
|
|
|
|
|
|
sub str_decamelize { |
315
|
|
|
|
|
|
|
|
316
|
0
|
|
|
0
|
1
|
|
my $s = shift; |
317
|
|
|
|
|
|
|
|
318
|
0
|
|
|
|
|
|
$s =~ s/([A-Z])/_\L$1/g; |
319
|
|
|
|
|
|
|
|
320
|
0
|
|
|
|
|
|
return lc($s); |
321
|
|
|
|
|
|
|
} |
322
|
|
|
|
|
|
|
|
323
|
|
|
|
|
|
|
1; |
324
|
|
|
|
|
|
|
__END__ |
325
|
|
|
|
|
|
|
|
326
|
|
|
|
|
|
|
=back |
327
|
|
|
|
|
|
|
|
328
|
|
|
|
|
|
|
=head1 EXAMPLES |
329
|
|
|
|
|
|
|
|
330
|
|
|
|
|
|
|
None yet |
331
|
|
|
|
|
|
|
|
332
|
|
|
|
|
|
|
=head1 BUGS |
333
|
|
|
|
|
|
|
|
334
|
|
|
|
|
|
|
Unknown yet |
335
|
|
|
|
|
|
|
|
336
|
|
|
|
|
|
|
=head1 TODO |
337
|
|
|
|
|
|
|
|
338
|
|
|
|
|
|
|
Implement examples and tests. |
339
|
|
|
|
|
|
|
|
340
|
|
|
|
|
|
|
=head1 SEE ALSO |
341
|
|
|
|
|
|
|
|
342
|
|
|
|
|
|
|
L<Encode>, L<perlunicode> |
343
|
|
|
|
|
|
|
|
344
|
|
|
|
|
|
|
=head1 AUTHORS |
345
|
|
|
|
|
|
|
|
346
|
|
|
|
|
|
|
Valentyn Solomko <pere@pere.org.ua> |
347
|
|
|
|
|
|
|
|
348
|
|
|
|
|
|
|
Michael Bochkaryov <misha@rattler.kiev.ua> |
349
|
|
|
|
|
|
|
|
350
|
|
|
|
|
|
|
=cut |