line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
package App::CSVUtils::csv_add_fields; |
2
|
|
|
|
|
|
|
|
3
|
1
|
|
|
1
|
|
5104
|
use 5.010001; |
|
1
|
|
|
|
|
4
|
|
4
|
1
|
|
|
1
|
|
5
|
use strict; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
23
|
|
5
|
1
|
|
|
1
|
|
4
|
use warnings; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
71
|
|
6
|
|
|
|
|
|
|
|
7
|
|
|
|
|
|
|
our $AUTHORITY = 'cpan:PERLANCAR'; # AUTHORITY |
8
|
|
|
|
|
|
|
our $DATE = '2023-04-01'; # DATE |
9
|
|
|
|
|
|
|
our $DIST = 'App-CSVUtils'; # DIST |
10
|
|
|
|
|
|
|
our $VERSION = '1.024'; # VERSION |
11
|
|
|
|
|
|
|
|
12
|
1
|
|
|
|
|
860
|
use App::CSVUtils qw( |
13
|
|
|
|
|
|
|
gen_csv_util |
14
|
|
|
|
|
|
|
compile_eval_code |
15
|
|
|
|
|
|
|
eval_code |
16
|
1
|
|
|
1
|
|
5
|
); |
|
1
|
|
|
|
|
3
|
|
17
|
|
|
|
|
|
|
|
18
|
|
|
|
|
|
|
gen_csv_util( |
19
|
|
|
|
|
|
|
name => 'csv_add_fields', |
20
|
|
|
|
|
|
|
summary => 'Add one or more fields to CSV file', |
21
|
|
|
|
|
|
|
description => <<'_', |
22
|
|
|
|
|
|
|
|
23
|
|
|
|
|
|
|
The new fields by default will be added at the end, unless you specify one of |
24
|
|
|
|
|
|
|
`--after` (to put after a certain field), `--before` (to put before a certain |
25
|
|
|
|
|
|
|
field), or `--at` (to put at specific position, 1 means the first field). The |
26
|
|
|
|
|
|
|
new fields will be clustered together though, you currently cannot set the |
27
|
|
|
|
|
|
|
position of each new field. But you can later reorder fields using |
28
|
|
|
|
|
|
|
<prog:csv-sort-fields>. |
29
|
|
|
|
|
|
|
|
30
|
|
|
|
|
|
|
If supplied, your Perl code (`-e`) will be called for each row (excluding the |
31
|
|
|
|
|
|
|
header row) and should return the value for the new fields (either as a list or |
32
|
|
|
|
|
|
|
as an arrayref). `$_` contains the current row (as arrayref, or if you specify |
33
|
|
|
|
|
|
|
`-H`, as a hashref). `$main::row` is available and contains the current row |
34
|
|
|
|
|
|
|
(always as an arrayref). `$main::rownum` contains the row number (2 means the |
35
|
|
|
|
|
|
|
first data row). `$csv` is the <pm:Text::CSV_XS> object. `$main::fields_idx` is |
36
|
|
|
|
|
|
|
also available for additional information. |
37
|
|
|
|
|
|
|
|
38
|
|
|
|
|
|
|
If `-e` is not supplied, the new fields will be getting the default value of |
39
|
|
|
|
|
|
|
empty string (`''`). |
40
|
|
|
|
|
|
|
|
41
|
|
|
|
|
|
|
_ |
42
|
|
|
|
|
|
|
add_args => { |
43
|
|
|
|
|
|
|
%App::CSVUtils::argspec_fields_1plus_nocomp, |
44
|
|
|
|
|
|
|
%App::CSVUtils::argspecopt_eval, |
45
|
|
|
|
|
|
|
%App::CSVUtils::argspecopt_hash, |
46
|
|
|
|
|
|
|
after => { |
47
|
|
|
|
|
|
|
summary => 'Put the new field(s) after specified field', |
48
|
|
|
|
|
|
|
schema => 'str*', |
49
|
|
|
|
|
|
|
completion => \&_complete_field, |
50
|
|
|
|
|
|
|
}, |
51
|
|
|
|
|
|
|
before => { |
52
|
|
|
|
|
|
|
summary => 'Put the new field(s) before specified field', |
53
|
|
|
|
|
|
|
schema => 'str*', |
54
|
|
|
|
|
|
|
completion => \&_complete_field, |
55
|
|
|
|
|
|
|
}, |
56
|
|
|
|
|
|
|
at => { |
57
|
|
|
|
|
|
|
summary => 'Put the new field(s) at specific position '. |
58
|
|
|
|
|
|
|
'(1 means at the front of all others)', |
59
|
|
|
|
|
|
|
schema => 'posint*', |
60
|
|
|
|
|
|
|
}, |
61
|
|
|
|
|
|
|
}, |
62
|
|
|
|
|
|
|
add_args_rels => { |
63
|
|
|
|
|
|
|
choose_one => [qw/after before at/], |
64
|
|
|
|
|
|
|
}, |
65
|
|
|
|
|
|
|
tags => ['category:munging','adds-fields'], |
66
|
|
|
|
|
|
|
|
67
|
|
|
|
|
|
|
examples => [ |
68
|
|
|
|
|
|
|
{ |
69
|
|
|
|
|
|
|
summary => 'Add a few new blank fields at the end', |
70
|
|
|
|
|
|
|
argv => ['file.csv', 'field4', 'field6', 'field5'], |
71
|
|
|
|
|
|
|
test => 0, |
72
|
|
|
|
|
|
|
'x.doc.show_result' => 0, |
73
|
|
|
|
|
|
|
}, |
74
|
|
|
|
|
|
|
{ |
75
|
|
|
|
|
|
|
summary => 'Add a few new blank fields after a certain field', |
76
|
|
|
|
|
|
|
argv => ['file.csv', 'field4', 'field6', 'field5', '--after', 'field2'], |
77
|
|
|
|
|
|
|
test => 0, |
78
|
|
|
|
|
|
|
'x.doc.show_result' => 0, |
79
|
|
|
|
|
|
|
}, |
80
|
|
|
|
|
|
|
{ |
81
|
|
|
|
|
|
|
summary => 'Add a new field and set its value', |
82
|
|
|
|
|
|
|
argv => ['file.csv', 'after_tax', '-e', '$main::row->[5] * 1.11'], |
83
|
|
|
|
|
|
|
test => 0, |
84
|
|
|
|
|
|
|
'x.doc.show_result' => 0, |
85
|
|
|
|
|
|
|
}, |
86
|
|
|
|
|
|
|
{ |
87
|
|
|
|
|
|
|
summary => 'Add a couple new fields and set their values', |
88
|
|
|
|
|
|
|
argv => ['file.csv', 'tax_rate', 'after_tax', '-e', '(0.11, $main::row->[5] * 1.11)'], |
89
|
|
|
|
|
|
|
test => 0, |
90
|
|
|
|
|
|
|
'x.doc.show_result' => 0, |
91
|
|
|
|
|
|
|
}, |
92
|
|
|
|
|
|
|
], |
93
|
|
|
|
|
|
|
|
94
|
|
|
|
|
|
|
on_begin => sub { |
95
|
|
|
|
|
|
|
my $r = shift; |
96
|
|
|
|
|
|
|
|
97
|
|
|
|
|
|
|
# check arguments |
98
|
|
|
|
|
|
|
if (!defined($r->{util_args}{fields}) || !@{ $r->{util_args}{fields} }) { |
99
|
|
|
|
|
|
|
die [400, "Please specify one or more fields (-f)"]; |
100
|
|
|
|
|
|
|
} |
101
|
|
|
|
|
|
|
}, |
102
|
|
|
|
|
|
|
|
103
|
|
|
|
|
|
|
on_input_header_row => sub { |
104
|
|
|
|
|
|
|
my $r = shift; |
105
|
|
|
|
|
|
|
|
106
|
|
|
|
|
|
|
# check that the new fields are not duplicate (against existing fields |
107
|
|
|
|
|
|
|
# and against itself) |
108
|
|
|
|
|
|
|
my %seen; |
109
|
|
|
|
|
|
|
for (@{ $r->{util_args}{fields} }) { |
110
|
|
|
|
|
|
|
unless (length $_) { |
111
|
|
|
|
|
|
|
die [400, "New field name cannot be empty"]; |
112
|
|
|
|
|
|
|
} |
113
|
|
|
|
|
|
|
if (defined $r->{input_fields_idx}{$_}) { |
114
|
|
|
|
|
|
|
die [412, "Field '$_' already exists"]; |
115
|
|
|
|
|
|
|
} |
116
|
|
|
|
|
|
|
if ($seen{$_}++) { |
117
|
|
|
|
|
|
|
die [412, "Duplicate new field '$_'"]; |
118
|
|
|
|
|
|
|
} |
119
|
|
|
|
|
|
|
} |
120
|
|
|
|
|
|
|
|
121
|
|
|
|
|
|
|
# determine the position at which to insert the new fields |
122
|
|
|
|
|
|
|
my $new_fields_idx; |
123
|
|
|
|
|
|
|
if (defined $r->{util_args}{at}) { |
124
|
|
|
|
|
|
|
$new_fields_idx = $r->{util_args}{at}-1; |
125
|
|
|
|
|
|
|
} elsif (defined $r->{util_args}{before}) { |
126
|
|
|
|
|
|
|
for (0..$#{ $r->{input_fields} }) { |
127
|
|
|
|
|
|
|
if ($r->{input_fields}[$_] eq $r->{util_args}{before}) { |
128
|
|
|
|
|
|
|
$new_fields_idx = $_; |
129
|
|
|
|
|
|
|
last; |
130
|
|
|
|
|
|
|
} |
131
|
|
|
|
|
|
|
} |
132
|
|
|
|
|
|
|
die [400, "Field '$r->{util_args}{before}' (to add new fields before) not found"] |
133
|
|
|
|
|
|
|
unless defined $new_fields_idx; |
134
|
|
|
|
|
|
|
} elsif (defined $r->{util_args}{after}) { |
135
|
|
|
|
|
|
|
for (0..$#{ $r->{input_fields} }) { |
136
|
|
|
|
|
|
|
if ($r->{input_fields}[$_] eq $r->{util_args}{after}) { |
137
|
|
|
|
|
|
|
$new_fields_idx = $_+1; |
138
|
|
|
|
|
|
|
last; |
139
|
|
|
|
|
|
|
} |
140
|
|
|
|
|
|
|
} |
141
|
|
|
|
|
|
|
die [400, "Field '$r->{util_args}{after}' (to add new fields after) not found"] |
142
|
|
|
|
|
|
|
unless defined $new_fields_idx; |
143
|
|
|
|
|
|
|
} else { |
144
|
|
|
|
|
|
|
$new_fields_idx = @{ $r->{input_fields} }; |
145
|
|
|
|
|
|
|
} |
146
|
|
|
|
|
|
|
|
147
|
|
|
|
|
|
|
# for printing the header |
148
|
|
|
|
|
|
|
$r->{output_fields} = [@{ $r->{input_fields} }]; |
149
|
|
|
|
|
|
|
splice @{ $r->{output_fields} }, $new_fields_idx, 0, @{ $r->{util_args}{fields} }; |
150
|
|
|
|
|
|
|
|
151
|
|
|
|
|
|
|
$r->{wants_input_row_as_hashref} = 1 if $r->{util_args}{hash}; |
152
|
|
|
|
|
|
|
|
153
|
|
|
|
|
|
|
# we add the following keys to the stash |
154
|
|
|
|
|
|
|
$r->{code} = compile_eval_code($r->{util_args}{eval} // 'return', 'eval'); |
155
|
|
|
|
|
|
|
$r->{new_fields_idx} = $new_fields_idx; |
156
|
|
|
|
|
|
|
}, |
157
|
|
|
|
|
|
|
|
158
|
|
|
|
|
|
|
on_input_data_row => sub { |
159
|
|
|
|
|
|
|
my $r = shift; |
160
|
|
|
|
|
|
|
|
161
|
|
|
|
|
|
|
my @vals; |
162
|
|
|
|
|
|
|
eval { @vals = eval_code($r->{code}, $r, $r->{wants_input_row_as_hashref} ? $r->{input_row_as_hashref} : $r->{input_row}) }; |
163
|
|
|
|
|
|
|
die [500, "Error while adding field(s) '".join(",", @{$r->{util_args}{fields}})."' for row #$r->{input_rownum}: $@"] |
164
|
|
|
|
|
|
|
if $@; |
165
|
|
|
|
|
|
|
if (ref $vals[0] eq 'ARRAY') { @vals = @{ $vals[0] } } |
166
|
|
|
|
|
|
|
splice @{ $r->{input_row} }, $r->{new_fields_idx}, 0, |
167
|
|
|
|
|
|
|
(map { $_ // '' } @vals[0 .. $#{$r->{util_args}{fields}}]); |
168
|
|
|
|
|
|
|
$r->{code_print_row}->($r->{input_row}); |
169
|
|
|
|
|
|
|
}, |
170
|
|
|
|
|
|
|
); |
171
|
|
|
|
|
|
|
|
172
|
|
|
|
|
|
|
1; |
173
|
|
|
|
|
|
|
# ABSTRACT: Add one or more fields to CSV file |
174
|
|
|
|
|
|
|
|
175
|
|
|
|
|
|
|
__END__ |
176
|
|
|
|
|
|
|
|
177
|
|
|
|
|
|
|
=pod |
178
|
|
|
|
|
|
|
|
179
|
|
|
|
|
|
|
=encoding UTF-8 |
180
|
|
|
|
|
|
|
|
181
|
|
|
|
|
|
|
=head1 NAME |
182
|
|
|
|
|
|
|
|
183
|
|
|
|
|
|
|
App::CSVUtils::csv_add_fields - Add one or more fields to CSV file |
184
|
|
|
|
|
|
|
|
185
|
|
|
|
|
|
|
=head1 VERSION |
186
|
|
|
|
|
|
|
|
187
|
|
|
|
|
|
|
This document describes version 1.024 of App::CSVUtils::csv_add_fields (from Perl distribution App-CSVUtils), released on 2023-04-01. |
188
|
|
|
|
|
|
|
|
189
|
|
|
|
|
|
|
=head1 FUNCTIONS |
190
|
|
|
|
|
|
|
|
191
|
|
|
|
|
|
|
|
192
|
|
|
|
|
|
|
=head2 csv_add_fields |
193
|
|
|
|
|
|
|
|
194
|
|
|
|
|
|
|
Usage: |
195
|
|
|
|
|
|
|
|
196
|
|
|
|
|
|
|
csv_add_fields(%args) -> [$status_code, $reason, $payload, \%result_meta] |
197
|
|
|
|
|
|
|
|
198
|
|
|
|
|
|
|
Add one or more fields to CSV file. |
199
|
|
|
|
|
|
|
|
200
|
|
|
|
|
|
|
Examples: |
201
|
|
|
|
|
|
|
|
202
|
|
|
|
|
|
|
=over |
203
|
|
|
|
|
|
|
|
204
|
|
|
|
|
|
|
=item * Add a few new blank fields at the end: |
205
|
|
|
|
|
|
|
|
206
|
|
|
|
|
|
|
csv_add_fields( |
207
|
|
|
|
|
|
|
input_filename => "file.csv", |
208
|
|
|
|
|
|
|
fields => ["field4", "field6", "field5"] |
209
|
|
|
|
|
|
|
); |
210
|
|
|
|
|
|
|
|
211
|
|
|
|
|
|
|
=item * Add a few new blank fields after a certain field: |
212
|
|
|
|
|
|
|
|
213
|
|
|
|
|
|
|
csv_add_fields( |
214
|
|
|
|
|
|
|
input_filename => "file.csv", |
215
|
|
|
|
|
|
|
fields => ["field4", "field6", "field5"], |
216
|
|
|
|
|
|
|
after => "field2" |
217
|
|
|
|
|
|
|
); |
218
|
|
|
|
|
|
|
|
219
|
|
|
|
|
|
|
=item * Add a new field and set its value: |
220
|
|
|
|
|
|
|
|
221
|
|
|
|
|
|
|
csv_add_fields( |
222
|
|
|
|
|
|
|
input_filename => "file.csv", |
223
|
|
|
|
|
|
|
fields => ["after_tax"], |
224
|
|
|
|
|
|
|
eval => "\$main::row->[5] * 1.11" |
225
|
|
|
|
|
|
|
); |
226
|
|
|
|
|
|
|
|
227
|
|
|
|
|
|
|
=item * Add a couple new fields and set their values: |
228
|
|
|
|
|
|
|
|
229
|
|
|
|
|
|
|
csv_add_fields( |
230
|
|
|
|
|
|
|
input_filename => "file.csv", |
231
|
|
|
|
|
|
|
fields => ["tax_rate", "after_tax"], |
232
|
|
|
|
|
|
|
eval => "(0.11, \$main::row->[5] * 1.11)" |
233
|
|
|
|
|
|
|
); |
234
|
|
|
|
|
|
|
|
235
|
|
|
|
|
|
|
=back |
236
|
|
|
|
|
|
|
|
237
|
|
|
|
|
|
|
The new fields by default will be added at the end, unless you specify one of |
238
|
|
|
|
|
|
|
C<--after> (to put after a certain field), C<--before> (to put before a certain |
239
|
|
|
|
|
|
|
field), or C<--at> (to put at specific position, 1 means the first field). The |
240
|
|
|
|
|
|
|
new fields will be clustered together though, you currently cannot set the |
241
|
|
|
|
|
|
|
position of each new field. But you can later reorder fields using |
242
|
|
|
|
|
|
|
L<csv-sort-fields>. |
243
|
|
|
|
|
|
|
|
244
|
|
|
|
|
|
|
If supplied, your Perl code (C<-e>) will be called for each row (excluding the |
245
|
|
|
|
|
|
|
header row) and should return the value for the new fields (either as a list or |
246
|
|
|
|
|
|
|
as an arrayref). C<$_> contains the current row (as arrayref, or if you specify |
247
|
|
|
|
|
|
|
C<-H>, as a hashref). C<$main::row> is available and contains the current row |
248
|
|
|
|
|
|
|
(always as an arrayref). C<$main::rownum> contains the row number (2 means the |
249
|
|
|
|
|
|
|
first data row). C<$csv> is the L<Text::CSV_XS> object. C<$main::fields_idx> is |
250
|
|
|
|
|
|
|
also available for additional information. |
251
|
|
|
|
|
|
|
|
252
|
|
|
|
|
|
|
If C<-e> is not supplied, the new fields will be getting the default value of |
253
|
|
|
|
|
|
|
empty string (C<''>). |
254
|
|
|
|
|
|
|
|
255
|
|
|
|
|
|
|
This function is not exported. |
256
|
|
|
|
|
|
|
|
257
|
|
|
|
|
|
|
Arguments ('*' denotes required arguments): |
258
|
|
|
|
|
|
|
|
259
|
|
|
|
|
|
|
=over 4 |
260
|
|
|
|
|
|
|
|
261
|
|
|
|
|
|
|
=item * B<after> => I<str> |
262
|
|
|
|
|
|
|
|
263
|
|
|
|
|
|
|
Put the new field(s) after specified field. |
264
|
|
|
|
|
|
|
|
265
|
|
|
|
|
|
|
=item * B<at> => I<posint> |
266
|
|
|
|
|
|
|
|
267
|
|
|
|
|
|
|
Put the new field(s) at specific position (1 means at the front of all others). |
268
|
|
|
|
|
|
|
|
269
|
|
|
|
|
|
|
=item * B<before> => I<str> |
270
|
|
|
|
|
|
|
|
271
|
|
|
|
|
|
|
Put the new field(s) before specified field. |
272
|
|
|
|
|
|
|
|
273
|
|
|
|
|
|
|
=item * B<eval> => I<str|code> |
274
|
|
|
|
|
|
|
|
275
|
|
|
|
|
|
|
Perl code. |
276
|
|
|
|
|
|
|
|
277
|
|
|
|
|
|
|
=item * B<fields>* => I<array[str]> |
278
|
|
|
|
|
|
|
|
279
|
|
|
|
|
|
|
Field names. |
280
|
|
|
|
|
|
|
|
281
|
|
|
|
|
|
|
=item * B<hash> => I<bool> |
282
|
|
|
|
|
|
|
|
283
|
|
|
|
|
|
|
Provide row in $_ as hashref instead of arrayref. |
284
|
|
|
|
|
|
|
|
285
|
|
|
|
|
|
|
=item * B<inplace> => I<true> |
286
|
|
|
|
|
|
|
|
287
|
|
|
|
|
|
|
Output to the same file as input. |
288
|
|
|
|
|
|
|
|
289
|
|
|
|
|
|
|
Normally, you output to a different file than input. If you try to output to the |
290
|
|
|
|
|
|
|
same file (C<-o INPUT.csv -O>) you will clobber the input file; thus the utility |
291
|
|
|
|
|
|
|
prevents you from doing it. However, with this C<--inplace> option, you can |
292
|
|
|
|
|
|
|
output to the same file. Like perl's C<-i> option, this will first output to a |
293
|
|
|
|
|
|
|
temporary file in the same directory as the input file then rename to the final |
294
|
|
|
|
|
|
|
file at the end. You cannot specify output file (C<-o>) when using this option, |
295
|
|
|
|
|
|
|
but you can specify backup extension with C<-b> option. |
296
|
|
|
|
|
|
|
|
297
|
|
|
|
|
|
|
Some caveats: |
298
|
|
|
|
|
|
|
|
299
|
|
|
|
|
|
|
=over |
300
|
|
|
|
|
|
|
|
301
|
|
|
|
|
|
|
=item * if input file is a symbolic link, it will be replaced with a regular file; |
302
|
|
|
|
|
|
|
|
303
|
|
|
|
|
|
|
=item * renaming (implemented using C<rename()>) can fail if input filename is too long; |
304
|
|
|
|
|
|
|
|
305
|
|
|
|
|
|
|
=item * value specified in C<-b> is currently not checked for acceptable characters; |
306
|
|
|
|
|
|
|
|
307
|
|
|
|
|
|
|
=item * things can also fail if permissions are restrictive; |
308
|
|
|
|
|
|
|
|
309
|
|
|
|
|
|
|
=back |
310
|
|
|
|
|
|
|
|
311
|
|
|
|
|
|
|
=item * B<inplace_backup_ext> => I<str> (default: "") |
312
|
|
|
|
|
|
|
|
313
|
|
|
|
|
|
|
Extension to add for backup of input file. |
314
|
|
|
|
|
|
|
|
315
|
|
|
|
|
|
|
In inplace mode (C<--inplace>), if this option is set to a non-empty string, will |
316
|
|
|
|
|
|
|
rename the input file using this extension as a backup. The old existing backup |
317
|
|
|
|
|
|
|
will be overwritten, if any. |
318
|
|
|
|
|
|
|
|
319
|
|
|
|
|
|
|
=item * B<input_escape_char> => I<str> |
320
|
|
|
|
|
|
|
|
321
|
|
|
|
|
|
|
Specify character to escape value in field in input CSV, will be passed to Text::CSV_XS. |
322
|
|
|
|
|
|
|
|
323
|
|
|
|
|
|
|
Defaults to C<\\> (backslash). Overrides C<--input-tsv> option. |
324
|
|
|
|
|
|
|
|
325
|
|
|
|
|
|
|
=item * B<input_filename> => I<filename> (default: "-") |
326
|
|
|
|
|
|
|
|
327
|
|
|
|
|
|
|
Input CSV file. |
328
|
|
|
|
|
|
|
|
329
|
|
|
|
|
|
|
Use C<-> to read from stdin. |
330
|
|
|
|
|
|
|
|
331
|
|
|
|
|
|
|
Encoding of input file is assumed to be UTF-8. |
332
|
|
|
|
|
|
|
|
333
|
|
|
|
|
|
|
=item * B<input_header> => I<bool> (default: 1) |
334
|
|
|
|
|
|
|
|
335
|
|
|
|
|
|
|
Specify whether input CSV has a header row. |
336
|
|
|
|
|
|
|
|
337
|
|
|
|
|
|
|
By default, the first row of the input CSV will be assumed to contain field |
338
|
|
|
|
|
|
|
names (and the second row contains the first data row). When you declare that |
339
|
|
|
|
|
|
|
input CSV does not have header row (C<--no-input-header>), the first row of the |
340
|
|
|
|
|
|
|
CSV is assumed to contain the first data row. Fields will be named C<field1>, |
341
|
|
|
|
|
|
|
C<field2>, and so on. |
342
|
|
|
|
|
|
|
|
343
|
|
|
|
|
|
|
=item * B<input_quote_char> => I<str> |
344
|
|
|
|
|
|
|
|
345
|
|
|
|
|
|
|
Specify field quote character in input CSV, will be passed to Text::CSV_XS. |
346
|
|
|
|
|
|
|
|
347
|
|
|
|
|
|
|
Defaults to C<"> (double quote). Overrides C<--input-tsv> option. |
348
|
|
|
|
|
|
|
|
349
|
|
|
|
|
|
|
=item * B<input_sep_char> => I<str> |
350
|
|
|
|
|
|
|
|
351
|
|
|
|
|
|
|
Specify field separator character in input CSV, will be passed to Text::CSV_XS. |
352
|
|
|
|
|
|
|
|
353
|
|
|
|
|
|
|
Defaults to C<,> (comma). Overrides C<--input-tsv> option. |
354
|
|
|
|
|
|
|
|
355
|
|
|
|
|
|
|
=item * B<input_tsv> => I<true> |
356
|
|
|
|
|
|
|
|
357
|
|
|
|
|
|
|
Inform that input file is in TSV (tab-separated) format instead of CSV. |
358
|
|
|
|
|
|
|
|
359
|
|
|
|
|
|
|
Overriden by C<--input-sep-char>, C<--input-quote-char>, C<--input-escape-char> |
360
|
|
|
|
|
|
|
options. If one of those options is specified, then C<--input-tsv> will be |
361
|
|
|
|
|
|
|
ignored. |
362
|
|
|
|
|
|
|
|
363
|
|
|
|
|
|
|
=item * B<output_always_quote> => I<bool> (default: 0) |
364
|
|
|
|
|
|
|
|
365
|
|
|
|
|
|
|
Whether to always quote values. |
366
|
|
|
|
|
|
|
|
367
|
|
|
|
|
|
|
When set to false (the default), values are quoted only when necessary: |
368
|
|
|
|
|
|
|
|
369
|
|
|
|
|
|
|
field1,field2,"field three contains comma (,)",field4 |
370
|
|
|
|
|
|
|
|
371
|
|
|
|
|
|
|
When set to true, then all values will be quoted: |
372
|
|
|
|
|
|
|
|
373
|
|
|
|
|
|
|
"field1","field2","field three contains comma (,)","field4" |
374
|
|
|
|
|
|
|
|
375
|
|
|
|
|
|
|
=item * B<output_escape_char> => I<str> |
376
|
|
|
|
|
|
|
|
377
|
|
|
|
|
|
|
Specify character to escape value in field in output CSV, will be passed to Text::CSV_XS. |
378
|
|
|
|
|
|
|
|
379
|
|
|
|
|
|
|
This is like C<--input-escape-char> option but for output instead of input. |
380
|
|
|
|
|
|
|
|
381
|
|
|
|
|
|
|
Defaults to C<\\> (backslash). Overrides C<--output-tsv> option. |
382
|
|
|
|
|
|
|
|
383
|
|
|
|
|
|
|
=item * B<output_filename> => I<filename> |
384
|
|
|
|
|
|
|
|
385
|
|
|
|
|
|
|
Output filename. |
386
|
|
|
|
|
|
|
|
387
|
|
|
|
|
|
|
Use C<-> to output to stdout (the default if you don't specify this option). |
388
|
|
|
|
|
|
|
|
389
|
|
|
|
|
|
|
Encoding of output file is assumed to be UTF-8. |
390
|
|
|
|
|
|
|
|
391
|
|
|
|
|
|
|
=item * B<output_header> => I<bool> |
392
|
|
|
|
|
|
|
|
393
|
|
|
|
|
|
|
Whether output CSV should have a header row. |
394
|
|
|
|
|
|
|
|
395
|
|
|
|
|
|
|
By default, a header row will be output I<if> input CSV has header row. Under |
396
|
|
|
|
|
|
|
C<--output-header>, a header row will be output even if input CSV does not have |
397
|
|
|
|
|
|
|
header row (value will be something like "col0,col1,..."). Under |
398
|
|
|
|
|
|
|
C<--no-output-header>, header row will I<not> be printed even if input CSV has |
399
|
|
|
|
|
|
|
header row. So this option can be used to unconditionally add or remove header |
400
|
|
|
|
|
|
|
row. |
401
|
|
|
|
|
|
|
|
402
|
|
|
|
|
|
|
=item * B<output_quote_char> => I<str> |
403
|
|
|
|
|
|
|
|
404
|
|
|
|
|
|
|
Specify field quote character in output CSV, will be passed to Text::CSV_XS. |
405
|
|
|
|
|
|
|
|
406
|
|
|
|
|
|
|
This is like C<--input-quote-char> option but for output instead of input. |
407
|
|
|
|
|
|
|
|
408
|
|
|
|
|
|
|
Defaults to C<"> (double quote). Overrides C<--output-tsv> option. |
409
|
|
|
|
|
|
|
|
410
|
|
|
|
|
|
|
=item * B<output_quote_empty> => I<bool> (default: 0) |
411
|
|
|
|
|
|
|
|
412
|
|
|
|
|
|
|
Whether to quote empty values. |
413
|
|
|
|
|
|
|
|
414
|
|
|
|
|
|
|
When set to false (the default), empty values are not quoted: |
415
|
|
|
|
|
|
|
|
416
|
|
|
|
|
|
|
field1,field2,,field4 |
417
|
|
|
|
|
|
|
|
418
|
|
|
|
|
|
|
When set to true, then empty values will be quoted: |
419
|
|
|
|
|
|
|
|
420
|
|
|
|
|
|
|
field1,field2,"",field4 |
421
|
|
|
|
|
|
|
|
422
|
|
|
|
|
|
|
=item * B<output_sep_char> => I<str> |
423
|
|
|
|
|
|
|
|
424
|
|
|
|
|
|
|
Specify field separator character in output CSV, will be passed to Text::CSV_XS. |
425
|
|
|
|
|
|
|
|
426
|
|
|
|
|
|
|
This is like C<--input-sep-char> option but for output instead of input. |
427
|
|
|
|
|
|
|
|
428
|
|
|
|
|
|
|
Defaults to C<,> (comma). Overrides C<--output-tsv> option. |
429
|
|
|
|
|
|
|
|
430
|
|
|
|
|
|
|
=item * B<output_tsv> => I<bool> |
431
|
|
|
|
|
|
|
|
432
|
|
|
|
|
|
|
Inform that output file is TSV (tab-separated) format instead of CSV. |
433
|
|
|
|
|
|
|
|
434
|
|
|
|
|
|
|
This is like C<--input-tsv> option but for output instead of input. |
435
|
|
|
|
|
|
|
|
436
|
|
|
|
|
|
|
Overriden by C<--output-sep-char>, C<--output-quote-char>, C<--output-escape-char> |
437
|
|
|
|
|
|
|
options. If one of those options is specified, then C<--output-tsv> will be |
438
|
|
|
|
|
|
|
ignored. |
439
|
|
|
|
|
|
|
|
440
|
|
|
|
|
|
|
=item * B<overwrite> => I<bool> |
441
|
|
|
|
|
|
|
|
442
|
|
|
|
|
|
|
Whether to override existing output file. |
443
|
|
|
|
|
|
|
|
444
|
|
|
|
|
|
|
|
445
|
|
|
|
|
|
|
=back |
446
|
|
|
|
|
|
|
|
447
|
|
|
|
|
|
|
Returns an enveloped result (an array). |
448
|
|
|
|
|
|
|
|
449
|
|
|
|
|
|
|
First element ($status_code) is an integer containing HTTP-like status code |
450
|
|
|
|
|
|
|
(200 means OK, 4xx caller error, 5xx function error). Second element |
451
|
|
|
|
|
|
|
($reason) is a string containing error message, or something like "OK" if status is |
452
|
|
|
|
|
|
|
200. Third element ($payload) is the actual result, but usually not present when enveloped result is an error response ($status_code is not 2xx). Fourth |
453
|
|
|
|
|
|
|
element (%result_meta) is called result metadata and is optional, a hash |
454
|
|
|
|
|
|
|
that contains extra information, much like how HTTP response headers provide additional metadata. |
455
|
|
|
|
|
|
|
|
456
|
|
|
|
|
|
|
Return value: (any) |
457
|
|
|
|
|
|
|
|
458
|
|
|
|
|
|
|
=head1 HOMEPAGE |
459
|
|
|
|
|
|
|
|
460
|
|
|
|
|
|
|
Please visit the project's homepage at L<https://metacpan.org/release/App-CSVUtils>. |
461
|
|
|
|
|
|
|
|
462
|
|
|
|
|
|
|
=head1 SOURCE |
463
|
|
|
|
|
|
|
|
464
|
|
|
|
|
|
|
Source repository is at L<https://github.com/perlancar/perl-App-CSVUtils>. |
465
|
|
|
|
|
|
|
|
466
|
|
|
|
|
|
|
=head1 AUTHOR |
467
|
|
|
|
|
|
|
|
468
|
|
|
|
|
|
|
perlancar <perlancar@cpan.org> |
469
|
|
|
|
|
|
|
|
470
|
|
|
|
|
|
|
=head1 CONTRIBUTING |
471
|
|
|
|
|
|
|
|
472
|
|
|
|
|
|
|
|
473
|
|
|
|
|
|
|
To contribute, you can send patches by email/via RT, or send pull requests on |
474
|
|
|
|
|
|
|
GitHub. |
475
|
|
|
|
|
|
|
|
476
|
|
|
|
|
|
|
Most of the time, you don't need to build the distribution yourself. You can |
477
|
|
|
|
|
|
|
simply modify the code, then test via: |
478
|
|
|
|
|
|
|
|
479
|
|
|
|
|
|
|
% prove -l |
480
|
|
|
|
|
|
|
|
481
|
|
|
|
|
|
|
If you want to build the distribution (e.g. to try to install it locally on your |
482
|
|
|
|
|
|
|
system), you can install L<Dist::Zilla>, |
483
|
|
|
|
|
|
|
L<Dist::Zilla::PluginBundle::Author::PERLANCAR>, |
484
|
|
|
|
|
|
|
L<Pod::Weaver::PluginBundle::Author::PERLANCAR>, and sometimes one or two other |
485
|
|
|
|
|
|
|
Dist::Zilla- and/or Pod::Weaver plugins. Any additional steps required beyond |
486
|
|
|
|
|
|
|
that are considered a bug and can be reported to me. |
487
|
|
|
|
|
|
|
|
488
|
|
|
|
|
|
|
=head1 COPYRIGHT AND LICENSE |
489
|
|
|
|
|
|
|
|
490
|
|
|
|
|
|
|
This software is copyright (c) 2023, 2022, 2021, 2020, 2019, 2018, 2017, 2016 by perlancar <perlancar@cpan.org>. |
491
|
|
|
|
|
|
|
|
492
|
|
|
|
|
|
|
This is free software; you can redistribute it and/or modify it under |
493
|
|
|
|
|
|
|
the same terms as the Perl 5 programming language system itself. |
494
|
|
|
|
|
|
|
|
495
|
|
|
|
|
|
|
=head1 BUGS |
496
|
|
|
|
|
|
|
|
497
|
|
|
|
|
|
|
Please report any bugs or feature requests on the bugtracker website L<https://rt.cpan.org/Public/Dist/Display.html?Name=App-CSVUtils> |
498
|
|
|
|
|
|
|
|
499
|
|
|
|
|
|
|
When submitting a bug or request, please include a test-file or a |
500
|
|
|
|
|
|
|
patch to an existing test-file that illustrates the bug or desired |
501
|
|
|
|
|
|
|
feature. |
502
|
|
|
|
|
|
|
|
503
|
|
|
|
|
|
|
=cut |