line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
# License: Public Domain or CC0 |
2
|
|
|
|
|
|
|
# See https://creativecommons.org/publicdomain/zero/1.0/ |
3
|
|
|
|
|
|
|
# The author, Jim Avera (jim.avera at gmail) has waived all copyright and |
4
|
|
|
|
|
|
|
# related or neighboring rights. Attribution is requested but is not required. |
5
|
|
|
|
|
|
|
|
6
|
|
|
|
|
|
|
# Pod documentation is below (use perldoc to view) |
7
|
|
|
|
|
|
|
|
8
|
3
|
|
|
3
|
|
3458641
|
use 5.18.0; # lexical subs with bug we have worked around |
|
3
|
|
|
|
|
33
|
|
9
|
3
|
|
|
3
|
|
18
|
use strict; use warnings FATAL => 'all'; use utf8; |
|
3
|
|
|
3
|
|
6
|
|
|
3
|
|
|
3
|
|
86
|
|
|
3
|
|
|
|
|
31
|
|
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
122
|
|
|
3
|
|
|
|
|
16
|
|
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
66
|
|
10
|
3
|
|
|
3
|
|
100
|
no warnings qw(experimental::lexical_subs); |
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
127
|
|
11
|
3
|
|
|
3
|
|
34
|
use feature qw(state lexical_subs current_sub); |
|
3
|
|
|
|
|
14
|
|
|
3
|
|
|
|
|
387
|
|
12
|
|
|
|
|
|
|
|
13
|
|
|
|
|
|
|
package Spreadsheet::Edit; |
14
|
|
|
|
|
|
|
|
15
|
|
|
|
|
|
|
# Allow "use <thismodule> <someversion>;" in development sandbox to not bomb |
16
|
3
|
|
|
3
|
|
20
|
{ no strict 'refs'; ${__PACKAGE__."::VER"."SION"} = 1999.999; } |
|
3
|
|
|
|
|
7
|
|
|
3
|
|
|
|
|
496
|
|
17
|
|
|
|
|
|
|
|
18
|
|
|
|
|
|
|
our $VERSION = '1000.008'; # VERSION from Dist::Zilla::Plugin::OurPkgVersion |
19
|
|
|
|
|
|
|
our $DATE = '2023-09-19'; # DATE from Dist::Zilla::Plugin::OurDate |
20
|
|
|
|
|
|
|
|
21
|
|
|
|
|
|
|
# FIXME: cmd_nesting does nothing except prefix >s to log messages. |
22
|
|
|
|
|
|
|
# Shouldn't it skip that many "public" call frames??? |
23
|
|
|
|
|
|
|
|
24
|
|
|
|
|
|
|
# TODO FIXME: Integrate with Spreadsheet::Read and provide a formatting API |
25
|
|
|
|
|
|
|
|
26
|
|
|
|
|
|
|
# TODO: Allow & support undef cell values (see Text::CSV_XS), used to |
27
|
|
|
|
|
|
|
# represent "NULL" when interfacing with database systems. |
28
|
|
|
|
|
|
|
# OTOH, this conflicts with failed optional alias keys |
29
|
|
|
|
|
|
|
|
30
|
|
|
|
|
|
|
# TODO: Add some way to exit an apply() early, e.g. return value? |
31
|
|
|
|
|
|
|
# or maybe provide an abort_apply(resultval) function |
32
|
|
|
|
|
|
|
# which throws an exception we can catch in apply? |
33
|
|
|
|
|
|
|
|
34
|
|
|
|
|
|
|
# TODO: Use Tie::File to avoid storing entire sheet in memory |
35
|
|
|
|
|
|
|
# (requires seekable, so must depend on OpenAsCsv |
36
|
|
|
|
|
|
|
# copying "-" to a temp file if it is not a plain file). |
37
|
|
|
|
|
|
|
|
38
|
|
|
|
|
|
|
######################################### |
39
|
|
|
|
|
|
|
# sub NAMING CONVENTION: |
40
|
|
|
|
|
|
|
# sub public_method_or_function |
41
|
|
|
|
|
|
|
# sub _internal_method |
42
|
|
|
|
|
|
|
# sub __internal_function |
43
|
|
|
|
|
|
|
######################################### |
44
|
|
|
|
|
|
|
|
45
|
|
|
|
|
|
|
# If these globals are *defined* then the corresponding |
46
|
|
|
|
|
|
|
# (downcased) options default accordingly when new sheets are created. |
47
|
|
|
|
|
|
|
our ($Debug, $Verbose, $Silent); |
48
|
|
|
|
|
|
|
|
49
|
|
|
|
|
|
|
############################## Exporting stuff ############################ |
50
|
|
|
|
|
|
|
# |
51
|
3
|
|
|
3
|
|
1599
|
use Exporter::Tiny 1.001_000 (); # just to require version with non-sub generator support |
|
3
|
|
|
|
|
15810
|
|
|
3
|
|
|
|
|
104
|
|
52
|
3
|
|
|
3
|
|
23
|
use parent "Exporter::Tiny"; # make us be a derived class |
|
3
|
|
|
|
|
7
|
|
|
3
|
|
|
|
|
35
|
|
53
|
|
|
|
|
|
|
|
54
|
|
|
|
|
|
|
require mro; # makes next::can available |
55
|
|
|
|
|
|
|
|
56
|
|
|
|
|
|
|
sub import { |
57
|
|
|
|
|
|
|
# copied from List::Util |
58
|
|
|
|
|
|
|
# (RT88848) Touch the caller's $a and $b, to avoid the warning of |
59
|
|
|
|
|
|
|
# 'Name "main::a" used only once: possible typo' |
60
|
5
|
|
|
5
|
|
544
|
my $pkg = caller; |
61
|
3
|
|
|
3
|
|
323
|
no strict 'refs'; |
|
3
|
|
|
|
|
7
|
|
|
3
|
|
|
|
|
507
|
|
62
|
5
|
|
|
|
|
10
|
${"${pkg}::a"} = ${"${pkg}::a"}; |
|
5
|
|
|
|
|
20
|
|
|
5
|
|
|
|
|
30
|
|
63
|
5
|
|
|
|
|
9
|
${"${pkg}::b"} = ${"${pkg}::b"}; |
|
5
|
|
|
|
|
12
|
|
|
5
|
|
|
|
|
18
|
|
64
|
|
|
|
|
|
|
|
65
|
5
|
|
|
|
|
11
|
my $this = $_[0]; |
66
|
5
|
|
|
|
|
9
|
goto &{ $this->next::can }; # see 'perldoc mro' |
|
5
|
|
|
|
|
47
|
|
67
|
|
|
|
|
|
|
#goto &maybe::next::method ; ??? |
68
|
|
|
|
|
|
|
} |
69
|
|
|
|
|
|
|
|
70
|
3
|
|
|
3
|
|
23
|
use Symbol qw/gensym/; |
|
3
|
|
|
|
|
7
|
|
|
3
|
|
|
|
|
4856
|
|
71
|
|
|
|
|
|
|
|
72
|
|
|
|
|
|
|
our @EXPORT = qw( |
73
|
|
|
|
|
|
|
alias apply apply_all apply_exceptrx apply_torx attributes |
74
|
|
|
|
|
|
|
spectocx data_source delete_col delete_cols delete_row delete_rows |
75
|
|
|
|
|
|
|
first_data_rx transpose join_cols join_cols_sep last_data_rx move_col |
76
|
|
|
|
|
|
|
move_cols insert_col insert_cols insert_row insert_rows new_sheet only_cols |
77
|
|
|
|
|
|
|
options read_spreadsheet rename_cols reverse_cols |
78
|
|
|
|
|
|
|
sheet sheetname sort_rows split_col tie_column_vars title2ident title_row |
79
|
|
|
|
|
|
|
title_rx unalias write_csv write_spreadsheet ); |
80
|
|
|
|
|
|
|
|
81
|
|
|
|
|
|
|
my @stdvars = qw( $title_rx $first_data_rx $last_data_rx $num_cols |
82
|
|
|
|
|
|
|
@rows @linenums @meta_info %colx %colx_desc $title_row |
83
|
|
|
|
|
|
|
$rx $linenum @crow %crow ); |
84
|
|
|
|
|
|
|
|
85
|
|
|
|
|
|
|
our @EXPORT_OK = (@stdvars, qw/logmsg cx2let let2cx fmt_sheet/); |
86
|
|
|
|
|
|
|
|
87
|
|
|
|
|
|
|
our %EXPORT_TAGS = ( |
88
|
|
|
|
|
|
|
STDVARS => [@stdvars], |
89
|
|
|
|
|
|
|
FUNCS => [@EXPORT], |
90
|
|
|
|
|
|
|
default => [':FUNCS'], |
91
|
|
|
|
|
|
|
DEFAULT => [':default'], |
92
|
|
|
|
|
|
|
#all => [qw(:STDVARS :FUNCS cx2let let2cx logmsg)], |
93
|
|
|
|
|
|
|
all => [qw(:STDVARS :FUNCS)], |
94
|
|
|
|
|
|
|
); |
95
|
|
|
|
|
|
|
|
96
|
|
|
|
|
|
|
# Although the magic globals $row, $rx and friends are "imported" by 'use', |
97
|
|
|
|
|
|
|
# they can not simply alias variables in Spreadsheet::Edit becuase they |
98
|
|
|
|
|
|
|
# need to be tied with parameters specific to the user's package |
99
|
|
|
|
|
|
|
# so that each package has its own 'current sheet'. |
100
|
|
|
|
|
|
|
# |
101
|
|
|
|
|
|
|
# To accomplish this, we use Exporter:Tiny's "generator" mechanism to |
102
|
|
|
|
|
|
|
# create a new, unique variables tied appropriately for each import. |
103
|
|
|
|
|
|
|
# See perldoc Exporter::Tiny::Manual::QuickStart . |
104
|
|
|
|
|
|
|
# |
105
|
|
|
|
|
|
|
# If "methods" named _generate_SUBNAME, _generateScalar_SCALARNAME, |
106
|
|
|
|
|
|
|
# _generateArray_ARRAYNAME or _generateHash_HASHNAME exist in the |
107
|
|
|
|
|
|
|
# exporting package (that's us), then they are called to obtain |
108
|
|
|
|
|
|
|
# a ref to an object to export as SUBNAME, SCALARNAME, ARRAYNAME or HASHNAME. |
109
|
|
|
|
|
|
|
# |
110
|
|
|
|
|
|
|
# For example _generateScalar_rx() is called to get a ref to an $rx variable. |
111
|
|
|
|
|
|
|
|
112
|
|
|
|
|
|
|
sub __gen_x { |
113
|
36
|
|
|
36
|
|
100
|
my ($myclass, $sigilname, $args, $globals, # supplied by Exporter::Tiny |
114
|
|
|
|
|
|
|
$Type, $helpersub, @args) = @_; |
115
|
36
|
100
|
|
|
|
96
|
my $sigl = $Type eq 'Hash' ? '%' : $Type eq 'Array' ? '@' : '$'; |
|
|
100
|
|
|
|
|
|
116
|
36
|
|
|
|
|
65
|
my $tieclassname = 'Tie::Indirect::'.$Type; |
117
|
36
|
|
|
|
|
51
|
my $ref = *{gensym()}{uc($Type)}; # e.g. "ARARY" |
|
36
|
|
|
|
|
111
|
|
118
|
|
|
|
|
|
|
# e.g. tie $$ref, 'Tie::Indirect::Scalar', \&_scal_tiehelper, ... |
119
|
36
|
|
|
|
|
3752
|
eval "tie ${sigl}\$ref, \$tieclassname, \$helpersub, |
120
|
|
|
|
|
|
|
\$globals->{into}, \$sigilname, \@args"; |
121
|
36
|
50
|
|
|
|
717
|
die $@ if $@; |
122
|
36
|
|
|
|
|
232
|
$ref |
123
|
|
|
|
|
|
|
} |
124
|
|
|
|
|
|
|
sub __gen_scalar { |
125
|
12
|
|
|
12
|
|
31
|
my ($myclass, $sigilname, $args, $globals, |
126
|
|
|
|
|
|
|
$canon_ident, $onlyinapply) = @_; |
127
|
12
|
|
|
|
|
35
|
__gen_x($myclass, $sigilname, $args, $globals, |
128
|
|
|
|
|
|
|
"Scalar", \&Spreadsheet::Edit::_scal_tiehelper, |
129
|
|
|
|
|
|
|
$canon_ident, $onlyinapply); |
130
|
|
|
|
|
|
|
} |
131
|
3
|
|
|
3
|
|
276
|
sub _generateScalar_num_cols { __gen_scalar(@_, "num_cols") } |
132
|
3
|
|
|
3
|
|
301
|
sub _generateScalar_first_data_rx { __gen_scalar(@_, "first_data_rx") } |
133
|
3
|
|
|
3
|
|
275
|
sub _generateScalar_last_data_rx { __gen_scalar(@_, "last_data_rx") } |
134
|
3
|
|
|
3
|
|
261
|
sub _generateScalar_rx { __gen_scalar(@_, "current_rx", 1) } |
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
#sub _generateScalar_title_rx { __gen_scalar(@_, "title_rx") } |
137
|
|
|
|
|
|
|
sub _generateScalar_title_rx { |
138
|
|
|
|
|
|
|
__gen_x(@_, "Scalar", |
139
|
|
|
|
|
|
|
sub{ |
140
|
0
|
|
|
0
|
|
0
|
my ($mutating, $pkg, $uvar, $onlyinapply)=@_; |
141
|
0
|
|
|
|
|
0
|
my $sheet = &Spreadsheet::Edit::__getsheet(@_); |
142
|
|
|
|
|
|
|
\$$sheet->{title_rx} |
143
|
0
|
|
|
3
|
|
0
|
}, 0 # onlyinapply |
|
3
|
|
|
|
|
5515
|
|
144
|
|
|
|
|
|
|
) |
145
|
|
|
|
|
|
|
} |
146
|
|
|
|
|
|
|
|
147
|
|
|
|
|
|
|
sub __gen_aryelem { |
148
|
6
|
|
|
6
|
|
22
|
my ($myclass, $sigilname, $args, $globals, |
149
|
|
|
|
|
|
|
$index_ident, $array_ident, $onlyinapply) = @_; |
150
|
|
|
|
|
|
|
# N.B. _aryelem_tiehelper has special logic for 'current_rx' and 'title_rx' |
151
|
6
|
|
|
|
|
19
|
__gen_x($myclass, $sigilname, $args, $globals, |
152
|
|
|
|
|
|
|
"Scalar", \&Spreadsheet::Edit::_aryelem_tiehelper, |
153
|
|
|
|
|
|
|
$index_ident, $array_ident, $onlyinapply); |
154
|
|
|
|
|
|
|
} |
155
|
3
|
|
|
3
|
|
264
|
sub _generateScalar_title_row { __gen_aryelem(@_, "title_rx", "rows") } |
156
|
3
|
|
|
3
|
|
281
|
sub _generateScalar_linenum { __gen_aryelem(@_, "current_rx", "linenums", 1) } |
157
|
|
|
|
|
|
|
|
158
|
|
|
|
|
|
|
sub __gen_hash { |
159
|
6
|
|
|
6
|
|
22
|
my ($myclass, $sigilname, $args, $globals, |
160
|
|
|
|
|
|
|
$field_ident, $onlyinapply) = @_; |
161
|
6
|
|
|
|
|
17
|
__gen_x($myclass, $sigilname, $args, $globals, |
162
|
|
|
|
|
|
|
"Hash", \&Spreadsheet::Edit::_refval_tiehelper, |
163
|
|
|
|
|
|
|
$field_ident, $onlyinapply, 0); # mutable => 0 |
164
|
|
|
|
|
|
|
} |
165
|
3
|
|
|
3
|
|
276
|
sub _generateHash_colx { __gen_hash(@_, "colx", 0) } |
166
|
3
|
|
|
3
|
|
270
|
sub _generateHash_colx_desc { __gen_hash(@_, "colx_desc", 0) } |
167
|
|
|
|
|
|
|
|
168
|
|
|
|
|
|
|
sub __gen_array{ |
169
|
9
|
|
|
9
|
|
26
|
my ($myclass, $sigilname, $args, $globals, |
170
|
|
|
|
|
|
|
$field_ident, $onlyinapply, $mutable) = @_; |
171
|
9
|
|
|
|
|
28
|
__gen_x($myclass, $sigilname, $args, $globals, |
172
|
|
|
|
|
|
|
"Array", \&Spreadsheet::Edit::_refval_tiehelper, |
173
|
|
|
|
|
|
|
$field_ident, $onlyinapply, $mutable); |
174
|
|
|
|
|
|
|
} |
175
|
3
|
|
|
3
|
|
271
|
sub _generateArray_rows { __gen_array(@_, "rows", 0, 1) } |
176
|
|
|
|
|
|
|
|
177
|
|
|
|
|
|
|
# Currently @linenums is not mutable but maybe it should be? |
178
|
3
|
|
|
3
|
|
273
|
sub _generateArray_linenums { __gen_array(@_, "linenums", 0) } |
179
|
|
|
|
|
|
|
|
180
|
|
|
|
|
|
|
## FIXME: is meta_info still valid? |
181
|
3
|
|
|
3
|
|
319
|
sub _generateArray_meta_info { __gen_array(@_, "meta_info", 0) } |
182
|
|
|
|
|
|
|
|
183
|
|
|
|
|
|
|
sub __get_currentrow { |
184
|
0
|
|
|
0
|
|
0
|
my ($mutating, $pkg, $uvar) = @_; |
185
|
|
|
|
|
|
|
# Ignore mutating, as it applies to the element not the container |
186
|
0
|
|
|
|
|
0
|
my $sheet = &Spreadsheet::Edit::__getsheet(0, $pkg, $uvar, 1); |
187
|
|
|
|
|
|
|
# Returns the dual-typed ref (Magicrow) for the current row |
188
|
0
|
|
|
|
|
0
|
$$sheet->{rows}->[$$sheet->{current_rx}] |
189
|
|
|
|
|
|
|
} |
190
|
|
|
|
|
|
|
sub _generateArray_crow { # @crow aliases the current row during apply |
191
|
3
|
|
|
3
|
|
273
|
my ($myclass, $sigilname, $args, $globals) = @_; |
192
|
3
|
|
|
|
|
5
|
my $aref = *{gensym()}{ARRAY}; |
|
3
|
|
|
|
|
12
|
|
193
|
|
|
|
|
|
|
tie @$aref, 'Tie::Indirect::Array', |
194
|
3
|
|
|
|
|
56
|
\&__get_currentrow, $globals->{into}, $sigilname ; |
195
|
3
|
|
|
|
|
34
|
$aref |
196
|
|
|
|
|
|
|
} |
197
|
|
|
|
|
|
|
sub _generateHash_crow { # %crow indexes cells in the current row during apply |
198
|
3
|
|
|
3
|
|
251
|
my ($myclass, $sigilname, $args, $globals) = @_; |
199
|
3
|
|
|
|
|
5
|
my $href = *{gensym()}{HASH}; |
|
3
|
|
|
|
|
10
|
|
200
|
|
|
|
|
|
|
tie %$href, 'Tie::Indirect::Hash', |
201
|
3
|
|
|
|
|
61
|
\&__get_currentrow, $globals->{into}, $sigilname ; |
202
|
3
|
|
|
|
|
89
|
$href |
203
|
|
|
|
|
|
|
} |
204
|
|
|
|
|
|
|
# |
205
|
|
|
|
|
|
|
########################### End of Exporting stuff ########################## |
206
|
|
|
|
|
|
|
|
207
|
3
|
|
|
|
|
14
|
use Spreadsheet::Edit::Log qw/log_call fmt_call log_methcall fmt_methcall oops/, |
208
|
3
|
|
|
3
|
|
1594
|
':btw=SE${lno}:' ; |
|
3
|
|
|
|
|
8
|
|
209
|
|
|
|
|
|
|
|
210
|
3
|
|
|
3
|
|
28
|
use Data::Dumper (); |
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
102
|
|
211
|
3
|
|
|
3
|
|
26
|
use Data::Dumper::Interp 6.007 qw/:all/; |
|
3
|
|
|
|
|
67
|
|
|
3
|
|
|
|
|
56
|
|
212
|
|
|
|
|
|
|
|
213
|
3
|
|
|
3
|
|
78281
|
use Carp; |
|
3
|
|
|
|
|
8
|
|
|
3
|
|
|
|
|
317
|
|
214
|
|
|
|
|
|
|
our @CARP_NOT = qw(Spreadsheet::Edit |
215
|
|
|
|
|
|
|
Spreadsheet::Edit::IO |
216
|
|
|
|
|
|
|
Tie::Indirect::Array Tie::Indirect::Hash |
217
|
|
|
|
|
|
|
Tie::Indirect::Scalar |
218
|
|
|
|
|
|
|
); |
219
|
|
|
|
|
|
|
|
220
|
3
|
|
|
3
|
|
25
|
use Scalar::Util qw(looks_like_number openhandle reftype refaddr blessed); |
|
3
|
|
|
|
|
5
|
|
|
3
|
|
|
|
|
191
|
|
221
|
3
|
|
|
3
|
|
28
|
use List::Util qw(min max sum0 first any all pairs pairgrep); |
|
3
|
|
|
|
|
8
|
|
|
3
|
|
|
|
|
259
|
|
222
|
3
|
|
|
3
|
|
852
|
use File::Temp qw(tempfile tempdir); |
|
3
|
|
|
|
|
20374
|
|
|
3
|
|
|
|
|
236
|
|
223
|
3
|
|
|
3
|
|
24
|
use File::Basename qw(basename dirname fileparse); |
|
3
|
|
|
|
|
7
|
|
|
3
|
|
|
|
|
207
|
|
224
|
3
|
|
|
3
|
|
24
|
use File::Find (); |
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
99
|
|
225
|
3
|
|
|
3
|
|
17
|
use Symbol qw(gensym); |
|
3
|
|
|
|
|
7
|
|
|
3
|
|
|
|
|
159
|
|
226
|
3
|
|
|
3
|
|
28
|
use POSIX qw(INT_MAX); |
|
3
|
|
|
|
|
10
|
|
|
3
|
|
|
|
|
43
|
|
227
|
3
|
|
|
3
|
|
768
|
use Guard qw(scope_guard); |
|
3
|
|
|
|
|
572
|
|
|
3
|
|
|
|
|
248
|
|
228
|
|
|
|
|
|
|
|
229
|
|
|
|
|
|
|
require Tie::Indirect; # OUR CUSTOM STUFF: stop using it? |
230
|
|
|
|
|
|
|
|
231
|
3
|
|
|
3
|
|
2673
|
use Text::CSV 1.90_01; # 1st version with known_attributes() |
|
3
|
|
|
|
|
55981
|
|
|
3
|
|
|
|
|
243
|
|
232
|
3
|
|
|
|
|
624
|
use Spreadsheet::Edit::IO qw( |
233
|
|
|
|
|
|
|
OpenAsCsv @sane_CSV_read_options @sane_CSV_write_options |
234
|
|
|
|
|
|
|
convert_spreadsheet |
235
|
3
|
|
|
3
|
|
2181
|
sheetname_from_spec filepath_from_spec form_spec_with_sheetname); |
|
3
|
|
|
|
|
13
|
|
236
|
|
|
|
|
|
|
|
237
|
3
|
|
33
|
|
|
232
|
use constant _CALLER_OVERRIDE_CHECK_OK => |
238
|
|
|
|
|
|
|
(defined(&Carp::CALLER_OVERRIDE_CHECK_OK) |
239
|
3
|
|
|
3
|
|
28
|
&& &Carp::CALLER_OVERRIDE_CHECK_OK); |
|
3
|
|
|
|
|
11
|
|
240
|
|
|
|
|
|
|
|
241
|
|
|
|
|
|
|
# sub __mytraceback() { ... } archived in commit 482e09b6009 |
242
|
|
|
|
|
|
|
|
243
|
3
|
|
|
3
|
|
21
|
use constant DEFAULT_WRITE_ENCODING => 'UTF-8'; |
|
3
|
|
|
|
|
11
|
|
|
3
|
|
|
|
|
29546
|
|
244
|
|
|
|
|
|
|
#use constant DEFAULT_READ_ENCODINGS => 'UTF-8,windows-1252'; |
245
|
|
|
|
|
|
|
|
246
|
|
|
|
|
|
|
# This global is used by logmsg() to infer the current sheet if an apply is |
247
|
|
|
|
|
|
|
# active, even if logmsg is called indirectly via another pkg |
248
|
|
|
|
|
|
|
our $_inner_apply_sheet; # see &_apply_to_rows |
249
|
|
|
|
|
|
|
|
250
|
|
|
|
|
|
|
# The "current sheet", to which tied globals refer in any given package. |
251
|
|
|
|
|
|
|
our %pkg2currsheet; |
252
|
|
|
|
|
|
|
|
253
|
24
|
|
|
24
|
|
41
|
sub __looks_like_aref($) { eval{ 1+scalar(@{$_[0]}) } } #actual or overloaded |
|
24
|
|
|
|
|
37
|
|
|
24
|
|
|
|
|
61
|
|
254
|
|
|
|
|
|
|
|
255
|
|
|
|
|
|
|
# Utility FUNCTIONS |
256
|
|
|
|
|
|
|
# |
257
|
|
|
|
|
|
|
sub to_array(@) { @_ != 1 ? @_ # 0 or multiple values |
258
|
0
|
|
|
|
|
0
|
: ref($_[0]) eq "ARRAY" ? @{$_[0]} |
259
|
4
|
0
|
|
4
|
0
|
24
|
: ref($_[0]) eq "HASH" ? @{ %{$_[0]} } # (key, value, ...) |
|
0
|
0
|
|
|
|
0
|
|
|
0
|
50
|
|
|
|
0
|
|
260
|
|
|
|
|
|
|
: $_[0] # just 1 value |
261
|
|
|
|
|
|
|
} |
262
|
0
|
|
|
0
|
0
|
0
|
sub to_aref(@) { [ to_array(@_) ] } |
263
|
0
|
0
|
|
0
|
0
|
0
|
sub to_wanted(@) { goto &to_array if wantarray; goto &to_aref } |
|
0
|
|
|
|
|
0
|
|
264
|
|
|
|
|
|
|
|
265
|
|
|
|
|
|
|
sub to_hash(@) { |
266
|
4
|
50
|
33
|
4
|
0
|
24
|
@_==1 && ref($_[0]) eq "HASH" ? $_[0] : |
|
|
50
|
|
|
|
|
|
267
|
|
|
|
|
|
|
(@_ % 2)!=0 ? croak("odd arg count, expecting key => value pairs") : |
268
|
|
|
|
|
|
|
{ to_array(@_) } |
269
|
|
|
|
|
|
|
} |
270
|
|
|
|
|
|
|
|
271
|
|
|
|
|
|
|
sub cx2let(_) { # default arg is $_ |
272
|
52
|
|
|
52
|
0
|
77
|
my $cx = shift; |
273
|
52
|
|
|
|
|
72
|
my $ABC="A"; ++$ABC for (1..$cx); |
|
52
|
|
|
|
|
109
|
|
274
|
52
|
|
|
|
|
140
|
return $ABC |
275
|
|
|
|
|
|
|
} |
276
|
|
|
|
|
|
|
sub let2cx(_) { |
277
|
0
|
|
|
0
|
0
|
0
|
my $ABC = shift; |
278
|
0
|
|
|
|
|
0
|
my $n = ord(substr($ABC,0,1,"")) - ord('A'); |
279
|
0
|
|
|
|
|
0
|
while (length $ABC) { |
280
|
0
|
|
|
|
|
0
|
my $letter = substr($ABC,0,1,""); |
281
|
0
|
|
|
|
|
0
|
$n = (($n+1) * 26) + (ord($letter) - ord('A')); |
282
|
|
|
|
|
|
|
} |
283
|
0
|
|
|
|
|
0
|
return $n; |
284
|
|
|
|
|
|
|
} |
285
|
|
|
|
|
|
|
|
286
|
|
|
|
|
|
|
# Produce the "automatic alias" identifier for an arbitrary title |
287
|
|
|
|
|
|
|
sub title2ident($) { |
288
|
32
|
|
|
32
|
0
|
45
|
local $_ = shift; |
289
|
32
|
|
|
|
|
64
|
s/^\s+//; s/\s+$//; s/\W/_/g; s/^(?=\d)/_/; |
|
32
|
|
|
|
|
56
|
|
|
32
|
|
|
|
|
60
|
|
|
32
|
|
|
|
|
68
|
|
290
|
32
|
|
|
|
|
126
|
$_ |
291
|
|
|
|
|
|
|
} |
292
|
|
|
|
|
|
|
|
293
|
|
|
|
|
|
|
# Format list as "word,word,..." without parens ; Non-barewords are "quoted". |
294
|
0
|
|
|
0
|
|
0
|
sub __fmt_uqlist(@) { join(",",map{quotekey} @_) } |
|
0
|
|
|
|
|
0
|
|
295
|
0
|
|
|
0
|
|
0
|
sub __fmt_uqarray(@) { "(" . &__fmt_uqlist . ")" } |
296
|
|
|
|
|
|
|
|
297
|
|
|
|
|
|
|
# Format list as without parens with barewords in qw/.../ |
298
|
|
|
|
|
|
|
sub __fmt_uqlistwithqw(@) { |
299
|
0
|
|
|
0
|
|
0
|
my $barewords; |
300
|
0
|
|
|
|
|
0
|
my $s = ""; |
301
|
0
|
|
|
|
|
0
|
foreach (map{quotekey} @_) { |
|
0
|
|
|
|
|
0
|
|
302
|
0
|
0
|
|
|
|
0
|
if (/^\w/) { |
303
|
0
|
0
|
|
|
|
0
|
if ($barewords++) { |
304
|
0
|
|
|
|
|
0
|
$s .= " "; |
305
|
|
|
|
|
|
|
} else { |
306
|
0
|
0
|
|
|
|
0
|
$s .= ", " if $s; |
307
|
0
|
|
|
|
|
0
|
$s .= "qw/"; |
308
|
|
|
|
|
|
|
} |
309
|
|
|
|
|
|
|
} else { |
310
|
0
|
0
|
|
|
|
0
|
if ($barewords) { |
311
|
0
|
|
|
|
|
0
|
$s .= "/"; |
312
|
0
|
|
|
|
|
0
|
$barewords = 0; |
313
|
|
|
|
|
|
|
} |
314
|
0
|
0
|
|
|
|
0
|
$s .= ", " if $s; |
315
|
|
|
|
|
|
|
} |
316
|
0
|
|
|
|
|
0
|
$s .= $_; |
317
|
|
|
|
|
|
|
} |
318
|
0
|
0
|
|
|
|
0
|
$s .= "/" if $barewords; |
319
|
0
|
|
|
|
|
0
|
$s |
320
|
|
|
|
|
|
|
} |
321
|
0
|
|
|
0
|
|
0
|
sub __fmt_uqarraywithqw(@) { "(" . &__fmt_uqlistwithqw . ")" } |
322
|
|
|
|
|
|
|
|
323
|
|
|
|
|
|
|
# Format list of pairs as "key1 => val1, key2 => val2, ..." without parens |
324
|
|
|
|
|
|
|
sub __fmt_pairlist(@) { |
325
|
0
|
|
|
0
|
|
0
|
my $result = ""; |
326
|
0
|
|
|
|
|
0
|
while (@_) { |
327
|
0
|
0
|
|
|
|
0
|
confess "Odd arg count, expecting key => value pairs" if @_==1; |
328
|
0
|
0
|
|
|
|
0
|
$result .= ", " if $result; |
329
|
0
|
|
|
|
|
0
|
my $key = shift @_; |
330
|
0
|
|
|
|
|
0
|
my $val = shift @_; |
331
|
0
|
|
|
|
|
0
|
$result .= quotekey($key)." => ".vis($val); |
332
|
|
|
|
|
|
|
} |
333
|
|
|
|
|
|
|
$result |
334
|
0
|
|
|
|
|
0
|
} |
335
|
|
|
|
|
|
|
sub __fmt_pairs(@) { |
336
|
0
|
|
|
0
|
|
0
|
__fmt_pairlist( map{ @$_ } @_ ); |
|
0
|
|
|
|
|
0
|
|
337
|
|
|
|
|
|
|
} |
338
|
|
|
|
|
|
|
|
339
|
|
|
|
|
|
|
# Concatenate strings separated by spaces, folding as necessary |
340
|
|
|
|
|
|
|
# (strings are never broken; internal newlines go unnoticed). |
341
|
|
|
|
|
|
|
# All lines (including the first) are indented the specified number of |
342
|
|
|
|
|
|
|
# spaces. Explicit line-breaks may be included as "\n". |
343
|
|
|
|
|
|
|
# A final newline is *not* included unless the last item ends with "\n". |
344
|
|
|
|
|
|
|
sub __fill($;$$) { |
345
|
0
|
|
|
0
|
|
0
|
my ($items, $indent, $foldwidth) = @_; |
346
|
0
|
|
0
|
|
|
0
|
$indent //= 4; |
347
|
0
|
|
0
|
|
|
0
|
$foldwidth //= 72; |
348
|
0
|
|
|
|
|
0
|
my $buf = ""; |
349
|
0
|
|
|
|
|
0
|
my $llen = 0; |
350
|
0
|
|
|
|
|
0
|
foreach (@$items) { |
351
|
0
|
0
|
0
|
|
|
0
|
if ($_ eq "\n" or |
|
|
|
0
|
|
|
|
|
352
|
|
|
|
|
|
|
($llen > $indent && ($llen + length($_)) > $foldwidth)) { |
353
|
0
|
|
|
|
|
0
|
$buf .= "\n"; |
354
|
0
|
|
|
|
|
0
|
$llen = 0; |
355
|
0
|
0
|
|
|
|
0
|
next if $_ eq "\n"; |
356
|
|
|
|
|
|
|
} |
357
|
0
|
0
|
|
|
|
0
|
if ($llen == 0) { |
358
|
0
|
|
|
|
|
0
|
$buf .= (" " x $indent); |
359
|
0
|
|
|
|
|
0
|
$llen = $indent; |
360
|
|
|
|
|
|
|
} else { |
361
|
0
|
0
|
|
|
|
0
|
if (substr($buf,-1) =~ /\S/) { |
362
|
0
|
|
|
|
|
0
|
$buf .= " "; |
363
|
0
|
|
|
|
|
0
|
++$llen; |
364
|
|
|
|
|
|
|
} |
365
|
|
|
|
|
|
|
} |
366
|
0
|
|
|
|
|
0
|
$buf .= $_; |
367
|
0
|
|
|
|
|
0
|
$llen += length(); |
368
|
|
|
|
|
|
|
} |
369
|
0
|
|
|
|
|
0
|
$buf; |
370
|
|
|
|
|
|
|
} |
371
|
|
|
|
|
|
|
|
372
|
|
|
|
|
|
|
sub __fmt_colspec_cx($$) { |
373
|
|
|
|
|
|
|
# "cx NN" or "COLSPEC [cx NN]" or "COLSPEC (NOT DEFINED)" if undef cx |
374
|
0
|
|
|
0
|
|
0
|
my ($colspec, $cx) = @_; |
375
|
0
|
0
|
|
|
|
0
|
if (ref($colspec) eq "Regexp") { |
376
|
0
|
|
|
|
|
0
|
state $delimsets = [ |
377
|
|
|
|
|
|
|
[qw(/ /)], [qw({ })], [qw([ ])], [qw<( )>], [qw(< >)], [qw(« »)] ]; |
378
|
0
|
|
|
|
|
0
|
for (@$delimsets) { |
379
|
0
|
|
|
|
|
0
|
my ($left, $right) = @$_; |
380
|
0
|
0
|
0
|
|
|
0
|
if (index($colspec,$left)<0 && index($colspec,$right)<0) { |
381
|
0
|
|
|
|
|
0
|
$colspec = "qr${left}${colspec}${right}"; |
382
|
0
|
|
|
|
|
0
|
last; |
383
|
|
|
|
|
|
|
} |
384
|
|
|
|
|
|
|
} |
385
|
|
|
|
|
|
|
} else { |
386
|
0
|
|
|
|
|
0
|
$colspec = visq($colspec); |
387
|
|
|
|
|
|
|
} |
388
|
0
|
0
|
|
|
|
0
|
return "$colspec (NOT DEFINED)" |
389
|
|
|
|
|
|
|
if ! defined $cx; |
390
|
0
|
0
|
|
|
|
0
|
$colspec eq "$cx" ? "cx $cx" : "$colspec [cx $cx]" |
391
|
|
|
|
|
|
|
} |
392
|
|
|
|
|
|
|
sub __fmt_cx($) { |
393
|
28
|
|
|
28
|
|
43
|
my ($cx) = @_; |
394
|
28
|
50
|
|
|
|
76
|
defined($cx) ? "cx $cx=".cx2let($cx) : "(undefined)" |
395
|
|
|
|
|
|
|
} |
396
|
|
|
|
|
|
|
|
397
|
|
|
|
|
|
|
# Format %colx keys "intelligently". cx values are not shown for keys which are |
398
|
|
|
|
|
|
|
# absolute column refs. Keys with undef values (from alias {optional => 1}) |
399
|
|
|
|
|
|
|
# are omitted since they are not currently valid. A final newline IS included. |
400
|
|
|
|
|
|
|
sub _fmt_colx(;$$) { |
401
|
|
|
|
|
|
|
my $self = shift; |
402
|
|
|
|
|
|
|
my ($indent, $foldwidth) = @_; |
403
|
|
|
|
|
|
|
my ($colx, $num_cols) = @$$self{qw{colx num_cols}}; |
404
|
|
|
|
|
|
|
# copy %$colx omitting keys with undef cx |
405
|
|
|
|
|
|
|
my %hash = map{ defined($colx->{$_}) ? ($_ => $colx->{$_}) : () } keys %$colx; |
406
|
|
|
|
|
|
|
my sub sortbycx(@) { sort { ($colx->{$a}//-1) <=> ($colx->{$b}//-1) } @_ } |
407
|
|
|
|
|
|
|
my sub subset($) { # format items, deleting from %hash |
408
|
|
|
|
|
|
|
my $specs = shift; |
409
|
|
|
|
|
|
|
my (@items, $curr, $curr_desc); |
410
|
|
|
|
|
|
|
my $curr_cx = -1; |
411
|
|
|
|
|
|
|
my sub flush() { |
412
|
|
|
|
|
|
|
return unless $curr_cx >= 0; |
413
|
|
|
|
|
|
|
push @items, $curr.$curr_desc; |
414
|
|
|
|
|
|
|
$curr = $curr_desc = undef; ##DEBUGGING |
415
|
|
|
|
|
|
|
$curr_cx = -1; |
416
|
|
|
|
|
|
|
} |
417
|
|
|
|
|
|
|
my sub additem($$) { |
418
|
|
|
|
|
|
|
(local $_, my $cx) = @_; |
419
|
|
|
|
|
|
|
flush() if $curr_cx != $cx; |
420
|
|
|
|
|
|
|
if ($curr_cx >= 0) { |
421
|
|
|
|
|
|
|
$curr .= ",".quotekey($_); |
422
|
|
|
|
|
|
|
} else { |
423
|
|
|
|
|
|
|
$curr_cx = $cx; |
424
|
|
|
|
|
|
|
$curr = quotekey($_); |
425
|
|
|
|
|
|
|
my $misfit = (/^[A-Z]{1,2}$/ && $colx->{$_} != let2cx($_)) |
426
|
|
|
|
|
|
|
|| (/^\d+$/ && $colx->{$_} != $_) |
427
|
|
|
|
|
|
|
|| (/^\D../) # show titles with cx too |
428
|
|
|
|
|
|
|
; |
429
|
|
|
|
|
|
|
$curr_desc = $misfit ? "(cx ".vis($hash{$_}).")" : ""; |
430
|
|
|
|
|
|
|
} |
431
|
|
|
|
|
|
|
} |
432
|
|
|
|
|
|
|
foreach (@$specs) { |
433
|
|
|
|
|
|
|
if (ref $_) { |
434
|
|
|
|
|
|
|
push @items, $$_; # \"string" means insert "string" literally |
435
|
|
|
|
|
|
|
} else { |
436
|
|
|
|
|
|
|
# Work around old Perl lexical sub limitations... |
437
|
|
|
|
|
|
|
my $item=$hash{$_}; |
438
|
|
|
|
|
|
|
oops(dvis '$_ $specs hash=',hvis(%hash)) unless defined $item; |
439
|
|
|
|
|
|
|
additem($_, $item); |
440
|
|
|
|
|
|
|
delete $hash{$_} // oops; |
441
|
|
|
|
|
|
|
} |
442
|
|
|
|
|
|
|
} |
443
|
|
|
|
|
|
|
flush(); |
444
|
|
|
|
|
|
|
push @items, "\n" if @items; # wrap before next subset, or at end |
445
|
|
|
|
|
|
|
@items |
446
|
|
|
|
|
|
|
} |
447
|
|
|
|
|
|
|
my @ABCs = subset [ map{ my $A = cx2let($_); |
448
|
|
|
|
|
|
|
u($hash{$A}) eq $_ ? $A : \" " |
449
|
|
|
|
|
|
|
} 0..$num_cols-1 ]; |
450
|
|
|
|
|
|
|
|
451
|
|
|
|
|
|
|
# More lexical sub bug work-arounds... |
452
|
|
|
|
|
|
|
my @ss1 = subset [sortbycx grep{ /^(=.*\D)\w+$/ } keys %hash]; # normal titles |
453
|
|
|
|
|
|
|
my @ss2 = subset [sortbycx grep{ /^\d+$/ } keys %hash]; # numeric titles |
454
|
|
|
|
|
|
|
my @ss3 = subset [sortbycx keys %hash]; # oddities |
455
|
|
|
|
|
|
|
__fill [ |
456
|
|
|
|
|
|
|
@ABCs, |
457
|
|
|
|
|
|
|
@ss1, |
458
|
|
|
|
|
|
|
@ss2, |
459
|
|
|
|
|
|
|
@ss3, |
460
|
|
|
|
|
|
|
], $indent, $foldwidth; |
461
|
|
|
|
|
|
|
} |
462
|
|
|
|
|
|
|
|
463
|
|
|
|
|
|
|
# Is a title a special symbol or looks like a cx number? |
464
|
|
|
|
|
|
|
sub __unindexed_title($$) { |
465
|
14
|
|
|
14
|
|
26
|
my ($title, $num_cols) = @_; |
466
|
14
|
50
|
|
|
|
35
|
oops unless defined $title; |
467
|
14
|
50
|
33
|
|
|
134
|
$title eq "" |
|
|
|
33
|
|
|
|
|
|
|
|
33
|
|
|
|
|
|
|
|
33
|
|
|
|
|
468
|
|
|
|
|
|
|
|| $title eq '^' |
469
|
|
|
|
|
|
|
|| $title eq '$' |
470
|
|
|
|
|
|
|
|| ( ($title =~ /^[1-9]\d*$/ || $title eq "0") # not with leading zeros |
471
|
|
|
|
|
|
|
&& $title <= $num_cols ) |
472
|
|
|
|
|
|
|
} |
473
|
|
|
|
|
|
|
sub _unindexed_title { #method for use by tests |
474
|
0
|
|
|
0
|
|
0
|
my $self = shift; |
475
|
0
|
|
|
|
|
0
|
__unindexed_title(shift(), $$self->{num_cols}); |
476
|
|
|
|
|
|
|
} |
477
|
|
|
|
|
|
|
|
478
|
|
|
|
|
|
|
# Return (normals, unindexed) where each is [title => cx, ...] sorted by cx |
479
|
|
|
|
|
|
|
sub _get_usable_titles { |
480
|
6
|
|
|
6
|
|
12
|
my $self = shift; |
481
|
6
|
|
|
|
|
18
|
my ($rows, $title_rx, $num_cols) = @$$self{qw{rows title_rx num_cols}}; |
482
|
6
|
|
33
|
|
|
38
|
my $title_row = $rows->[$title_rx // oops]; |
483
|
6
|
|
|
|
|
21
|
my @unindexed; |
484
|
|
|
|
|
|
|
my @normals; |
485
|
6
|
|
|
|
|
0
|
my %seen; |
486
|
6
|
|
|
|
|
17
|
for my $cx (0 .. $num_cols-1) { |
487
|
14
|
|
|
|
|
39
|
my $title = $title_row->[$cx]; |
488
|
14
|
50
|
|
|
|
41
|
next if $title eq ""; |
489
|
14
|
50
|
|
|
|
45
|
if ($seen{$title}++) { |
490
|
0
|
0
|
|
|
|
0
|
$self->_carponce("Warning: Non-unique title ", visq($title), " will not be usable for COLSPEC\n") unless $$self->{silent}; |
491
|
0
|
|
|
|
|
0
|
@normals = grep{ $_->[0] ne $title } @normals; |
|
0
|
|
|
|
|
0
|
|
492
|
0
|
|
|
|
|
0
|
@unindexed = grep{ $_->[0] ne $title } @unindexed; |
|
0
|
|
|
|
|
0
|
|
493
|
0
|
|
|
|
|
0
|
next; |
494
|
|
|
|
|
|
|
} |
495
|
14
|
50
|
|
|
|
32
|
if (__unindexed_title($title, $num_cols)) { |
496
|
0
|
|
|
|
|
0
|
push @unindexed, [$title, $cx]; |
497
|
|
|
|
|
|
|
} else { |
498
|
14
|
|
|
|
|
42
|
push @normals, [$title, $cx]; |
499
|
|
|
|
|
|
|
} |
500
|
|
|
|
|
|
|
} |
501
|
9
|
|
|
|
|
64
|
[sort { $a->[1] <=> $b->[1] } @normals], |
502
|
6
|
|
|
|
|
29
|
[sort { $a->[1] <=> $b->[1] } @unindexed] |
|
0
|
|
|
|
|
0
|
|
503
|
|
|
|
|
|
|
} |
504
|
|
|
|
|
|
|
|
505
|
|
|
|
|
|
|
# Non-OO api: Explicitly create a new sheet and make it the "current sheet". |
506
|
|
|
|
|
|
|
# Options (e.g. to specify initial content) may be specified in an |
507
|
|
|
|
|
|
|
# initial {OPTHASH} and/or as linear key => value pairs. |
508
|
|
|
|
|
|
|
# |
509
|
|
|
|
|
|
|
# Note: Most functions automatically create an empty sheet if no sheet |
510
|
|
|
|
|
|
|
# exists, so this is only really needed when using more than one sheet |
511
|
|
|
|
|
|
|
# or if you want to initialize a sheet from data in memory. |
512
|
|
|
|
|
|
|
# N.B. the corresponding OO interface is Spreadsheet::Edit->new(...) |
513
|
|
|
|
|
|
|
# |
514
|
|
|
|
|
|
|
sub new_sheet(@) { |
515
|
2
|
|
|
2
|
1
|
10449
|
my $opthash = &__opthash; |
516
|
2
|
|
|
|
|
7
|
my %opts = (%$opthash, %{to_hash(@_)}); # new() merges these anyway |
|
2
|
|
|
|
|
8
|
|
517
|
|
|
|
|
|
|
|
518
|
2
|
|
|
|
|
6
|
my ($userpkg, $fn, $lno, $subname) = @{ __filter_frame(__usercall_info()) }; |
|
2
|
|
|
|
|
6
|
|
519
|
2
|
50
|
|
|
|
12
|
$userpkg = delete $opts{package} if exists $opts{package}; |
520
|
|
|
|
|
|
|
|
521
|
2
|
50
|
|
|
|
8
|
my $pkgmsg = $opts{package} ? " [for pkg $userpkg]" : ""; |
522
|
2
|
50
|
33
|
|
|
16
|
croak "Invalid 'package' ",u($userpkg),"\n" |
523
|
|
|
|
|
|
|
unless defined($userpkg) && $userpkg =~ /^[a-zA-Z][:\w]*$/a; |
524
|
|
|
|
|
|
|
|
525
|
2
|
|
33
|
|
|
17
|
$opts{data_source} ||= "Created at ${fn}:$lno by $subname"; |
526
|
|
|
|
|
|
|
|
527
|
2
|
|
|
|
|
17
|
my $sheet = __silent_new(\%opts); |
528
|
|
|
|
|
|
|
|
529
|
|
|
|
|
|
|
log_call [$opthash, @_], [\(fmt_sheet($sheet), \$pkgmsg)] |
530
|
2
|
50
|
|
|
|
16
|
if $$sheet->{verbose}; |
531
|
|
|
|
|
|
|
|
532
|
2
|
|
|
|
|
13
|
$pkg2currsheet{$userpkg} = $sheet |
533
|
|
|
|
|
|
|
} |
534
|
|
|
|
|
|
|
|
535
|
|
|
|
|
|
|
# logmsg() - Concatenate strings to form a "log message", possibly |
536
|
|
|
|
|
|
|
# prefixed with a description of a "focus" sheet and optionally |
537
|
|
|
|
|
|
|
# a specific row. A final \n is appended if needed. |
538
|
|
|
|
|
|
|
# |
539
|
|
|
|
|
|
|
# The "focus" sheet and row, if any, are determined as follows: |
540
|
|
|
|
|
|
|
# |
541
|
|
|
|
|
|
|
# If the first argument is a sheet object, [sheet_object], |
542
|
|
|
|
|
|
|
# [sheet_object, rx], or [sheet_object, undef] then the indicated |
543
|
|
|
|
|
|
|
# sheet and (optionally) row are used. Note that if called as a method |
544
|
|
|
|
|
|
|
# the first arg will be the sheet object. |
545
|
|
|
|
|
|
|
# |
546
|
|
|
|
|
|
|
# Otherwise the first arg is not special and is included in the message. |
547
|
|
|
|
|
|
|
# |
548
|
|
|
|
|
|
|
# If no sheet is identified above, then the caller's package active |
549
|
|
|
|
|
|
|
# sheet is used, if any. |
550
|
|
|
|
|
|
|
# |
551
|
|
|
|
|
|
|
# If still no sheet is identified, then the sheet of the innermost apply |
552
|
|
|
|
|
|
|
# currently executing (anywhere up the stack) is used, if any; this sheet |
553
|
|
|
|
|
|
|
# is internally saved in a global by the apply* methods. |
554
|
|
|
|
|
|
|
# |
555
|
|
|
|
|
|
|
# If a sheet is identified but no specific rx, then the |
556
|
|
|
|
|
|
|
# "current row" of an active apply on that sheet is used, if any. |
557
|
|
|
|
|
|
|
# |
558
|
|
|
|
|
|
|
# If a focus sheet or sheet & row were identified, then the caller-supplied |
559
|
|
|
|
|
|
|
# message is prefixed by "(<description>):" or "(row <num> <description>):" |
560
|
|
|
|
|
|
|
# where <description> comes from: |
561
|
|
|
|
|
|
|
# |
562
|
|
|
|
|
|
|
# 1) If the sheet attribute {logmsg_pfx_gen} is defined to a subref, |
563
|
|
|
|
|
|
|
# the sub is called and all returned items other than undef are |
564
|
|
|
|
|
|
|
# concatenated (any undefs in the returned list are ignored); otherwise |
565
|
|
|
|
|
|
|
# |
566
|
|
|
|
|
|
|
# 2) The "sheetname" property is used, if defined; otherwise |
567
|
|
|
|
|
|
|
# |
568
|
|
|
|
|
|
|
# 3) the "data_source" property is used, which defaults to the name of the |
569
|
|
|
|
|
|
|
# spreadsheet read by read_spreadsheet(). |
570
|
|
|
|
|
|
|
# |
571
|
|
|
|
|
|
|
# FIXME: I should either rename logmsg_pfx_gen as logmsg_sheetdesc_gen |
572
|
|
|
|
|
|
|
# to reflect that it only generated the sheet-description part, |
573
|
|
|
|
|
|
|
# or else make prefix generators produce the entire message prefix |
574
|
|
|
|
|
|
|
# including any row number. |
575
|
|
|
|
|
|
|
# |
576
|
|
|
|
|
|
|
sub _default_pfx_gen($$) { |
577
|
274
|
|
|
274
|
|
561
|
my ($sheet, $rx) = @_; |
578
|
274
|
50
|
|
|
|
1266
|
confess "bug" unless ref($sheet) =~ /^Spreadsheet::Edit\b/; |
579
|
274
|
50
|
|
|
|
607
|
($sheet->sheetname() || $sheet->data_source()) |
580
|
|
|
|
|
|
|
} |
581
|
|
|
|
|
|
|
sub logmsg(@) { |
582
|
285
|
|
|
285
|
1
|
283140
|
my ($sheet, $rx); |
583
|
285
|
100
|
100
|
|
|
1297
|
if (@_ > 0 && ref($_[0])) { |
584
|
208
|
100
|
33
|
|
|
802
|
if (ref($_[0]) =~ /^Spreadsheet::Edit\b/) { |
|
|
50
|
33
|
|
|
|
|
585
|
24
|
|
|
|
|
53
|
$sheet = shift; |
586
|
|
|
|
|
|
|
} |
587
|
|
|
|
|
|
|
elsif (ref($_[0]) eq "ARRAY" |
588
|
184
|
|
|
|
|
1104
|
&& @{ $_[0] } <= 2 |
589
|
|
|
|
|
|
|
&& ref($_[0]->[0])."" =~ /^Spreadsheet::Edit\b/) { |
590
|
184
|
|
|
|
|
312
|
($sheet, $rx) = @{ shift @_ }; |
|
184
|
|
|
|
|
420
|
|
591
|
|
|
|
|
|
|
} |
592
|
|
|
|
|
|
|
} |
593
|
285
|
|
|
|
|
608
|
my $msgstr = join("", grep{defined} @_); |
|
320
|
|
|
|
|
981
|
|
594
|
285
|
100
|
|
|
|
666
|
if (! defined $sheet) { |
595
|
77
|
|
|
|
|
206
|
$sheet = $pkg2currsheet{scalar(caller)}; |
596
|
|
|
|
|
|
|
} |
597
|
285
|
100
|
|
|
|
568
|
if (! defined $sheet) { |
598
|
22
|
|
|
|
|
36
|
$sheet = $Spreadsheet::Edit::_inner_apply_sheet; |
599
|
|
|
|
|
|
|
} |
600
|
285
|
100
|
|
|
|
555
|
if (! defined $rx) { |
601
|
125
|
100
|
|
|
|
256
|
$rx = eval{ $sheet->rx() } if defined($sheet); |
|
114
|
|
|
|
|
313
|
|
602
|
|
|
|
|
|
|
} |
603
|
285
|
|
|
|
|
464
|
my @prefix; |
604
|
285
|
100
|
|
|
|
515
|
if (defined $sheet) { |
605
|
274
|
|
|
|
|
496
|
push @prefix, "("; |
606
|
274
|
100
|
100
|
|
|
766
|
if (defined($rx) && ($rx < 0 || $rx > $#{$sheet->rows})) { |
|
|
|
100
|
|
|
|
|
607
|
80
|
|
|
|
|
247
|
push @prefix, "Row ".($rx+1)."[INVALID RX $rx] "; |
608
|
80
|
|
|
|
|
157
|
$rx = undef; # avoid confusing user's prefix generator |
609
|
|
|
|
|
|
|
} else { |
610
|
194
|
100
|
|
|
|
550
|
push @prefix, "Row ".($rx+1)." " if defined($rx); |
611
|
|
|
|
|
|
|
} |
612
|
274
|
|
50
|
|
|
725
|
my $pfxgen = $sheet->attributes->{logmsg_pfx_gen} // \&_default_pfx_gen; |
613
|
274
|
|
|
|
|
613
|
push @prefix, &$pfxgen($sheet, $rx), "): "; |
614
|
|
|
|
|
|
|
} |
615
|
285
|
100
|
|
|
|
834
|
my $suffix = ($msgstr =~ /\n\z/s ? "" : "\n"); |
616
|
|
|
|
|
|
|
# If the message string starts with newlines, put them before the prefix |
617
|
285
|
50
|
|
|
|
715
|
if ($msgstr =~ s/\A(\n+)([^\n])/$2/s) { |
618
|
0
|
|
|
|
|
0
|
unshift @prefix, $1; |
619
|
|
|
|
|
|
|
} |
620
|
285
|
|
|
|
|
503
|
return join "", grep{defined} @prefix, $msgstr, $suffix; |
|
1593
|
|
|
|
|
4428
|
|
621
|
|
|
|
|
|
|
} |
622
|
|
|
|
|
|
|
|
623
|
|
|
|
|
|
|
##################################################################### |
624
|
|
|
|
|
|
|
# Locate the nearest call to a public sub in the call stack. |
625
|
|
|
|
|
|
|
# [Now *GENERIC*] |
626
|
|
|
|
|
|
|
# |
627
|
|
|
|
|
|
|
# A "public" sub means anything named starting with [a-z], irrespective |
628
|
|
|
|
|
|
|
# of package, and not mentioned in $USERCALL_NOT. |
629
|
|
|
|
|
|
|
# The assumption is that internal subs are named with an initial |
630
|
|
|
|
|
|
|
# underscore or are ALLCAPS (e.g. for constants). |
631
|
|
|
|
|
|
|
# |
632
|
|
|
|
|
|
|
# RETURNS |
633
|
|
|
|
|
|
|
# ([frame], [called args]) in array context |
634
|
|
|
|
|
|
|
# [frame] in scalar context |
635
|
|
|
|
|
|
|
# |
636
|
|
|
|
|
|
|
# "frame" means caller(n) results: |
637
|
|
|
|
|
|
|
# 0 1 2 3 |
638
|
|
|
|
|
|
|
# package filename linenum subname ... |
639
|
|
|
|
|
|
|
# |
640
|
|
|
|
|
|
|
sub __usercall_info(;$) { |
641
|
15
|
|
|
15
|
|
39
|
for (my $lvl=1 ; ; ++$lvl) { |
642
|
53
|
|
|
|
|
118
|
@DB::args = \$lvl if _CALLER_OVERRIDE_CHECK_OK; # see mytraceback() |
643
|
53
|
|
|
|
|
77
|
my @frame = do{ package |
644
|
53
|
|
|
|
|
568
|
DB; caller($lvl) }; |
645
|
53
|
50
|
|
|
|
139
|
oops dvis('$lvl @frame') unless defined($frame[0]); |
646
|
53
|
100
|
|
|
|
307
|
if ( $frame[3] =~ /::([a-z][^:]*)/ ) { |
647
|
15
|
100
|
|
|
|
159
|
return \@frame unless wantarray; |
648
|
2
|
|
|
|
|
4
|
my @args; |
649
|
2
|
|
|
|
|
5
|
my $hasargs = $frame[4]; |
650
|
2
|
50
|
|
|
|
5
|
if ($hasargs) { |
651
|
2
|
|
|
|
|
6
|
eval{ @args = @DB::args }; |
|
2
|
|
|
|
|
5
|
|
652
|
2
|
50
|
|
|
|
6
|
@args=("<perl bug(?) prevented getting args>") if $@; |
653
|
|
|
|
|
|
|
} |
654
|
2
|
|
|
|
|
8
|
return (\@frame, \@args) |
655
|
|
|
|
|
|
|
} |
656
|
|
|
|
|
|
|
} |
657
|
|
|
|
|
|
|
} |
658
|
|
|
|
|
|
|
|
659
|
|
|
|
|
|
|
sub __filter_frame($) { #clean-up/abbreviate for display purposes |
660
|
5
|
|
|
5
|
|
11
|
my @frame = @{shift @_}; |
|
5
|
|
|
|
|
20
|
|
661
|
5
|
|
|
|
|
186
|
$frame[1] = basename $frame[1]; # filename |
662
|
5
|
|
|
|
|
37
|
$frame[3] =~ s/.*:://; # subname |
663
|
|
|
|
|
|
|
\@frame |
664
|
5
|
|
|
|
|
44
|
} |
665
|
|
|
|
|
|
|
sub __fn_ln_methname() { |
666
|
3
|
|
|
3
|
|
7
|
@{ __filter_frame(__usercall_info()) }[1,2,3]; # (fn, lno, subname) |
|
3
|
|
|
|
|
9
|
|
667
|
|
|
|
|
|
|
} |
668
|
|
|
|
|
|
|
|
669
|
|
|
|
|
|
|
sub __methname() { |
670
|
0
|
|
|
0
|
|
0
|
(&__fn_ln_methname())[2] |
671
|
|
|
|
|
|
|
} |
672
|
|
|
|
|
|
|
|
673
|
|
|
|
|
|
|
sub __find_userpkg() { |
674
|
10
|
|
|
10
|
|
29
|
${ __usercall_info() }[0]; |
|
10
|
|
|
|
|
41
|
|
675
|
|
|
|
|
|
|
} |
676
|
|
|
|
|
|
|
|
677
|
|
|
|
|
|
|
# This always returns the caller's caller's package but also |
678
|
|
|
|
|
|
|
# checks that it is not an internal call, which should never happen |
679
|
|
|
|
|
|
|
sub __callerpkg() { |
680
|
1
|
|
|
1
|
|
20
|
my $pkg = (caller(1))[0]; |
681
|
1
|
50
|
|
|
|
7
|
oops if index($pkg,__PACKAGE__) == 0; |
682
|
1
|
|
|
|
|
4
|
$pkg |
683
|
|
|
|
|
|
|
} |
684
|
|
|
|
|
|
|
|
685
|
|
|
|
|
|
|
# Create a new object without allowing any logging. This is used when |
686
|
|
|
|
|
|
|
# new() is called implicitly by something else and new's log messages |
687
|
|
|
|
|
|
|
# might display an internal filename (namely Edit.pm). |
688
|
|
|
|
|
|
|
# |
689
|
|
|
|
|
|
|
# debug/verbose args are removed from the arguments passed to new() |
690
|
|
|
|
|
|
|
# and put back into the object after it is created. |
691
|
|
|
|
|
|
|
sub __silent_new(@) { |
692
|
2
|
|
|
2
|
|
4
|
my $opthash = &__opthash; |
693
|
2
|
|
|
|
|
8
|
my $new_args = to_hash(@_); |
694
|
|
|
|
|
|
|
|
695
|
2
|
|
|
|
|
4
|
my %saved; |
696
|
2
|
|
|
|
|
6
|
foreach my $key (qw/verbose debug/) { |
697
|
4
|
50
|
|
|
|
14
|
$saved{$key} = $opthash->{$key} if exists($opthash->{$key}); |
698
|
4
|
|
|
|
|
9
|
$opthash->{$key} = 0; # force off |
699
|
4
|
50
|
|
|
|
12
|
$saved{$key} = delete($new_args->{$key}) if exists($new_args->{$key}); |
700
|
|
|
|
|
|
|
} |
701
|
|
|
|
|
|
|
|
702
|
2
|
|
|
|
|
20
|
my $self = Spreadsheet::Edit->new($opthash, %$new_args); |
703
|
|
|
|
|
|
|
|
704
|
2
|
|
|
|
|
12
|
delete @$$self{qw/verbose debug/}; |
705
|
2
|
|
|
|
|
7
|
$self->_set_verbose_debug_silent(%saved); |
706
|
|
|
|
|
|
|
|
707
|
2
|
|
|
|
|
7
|
$self |
708
|
|
|
|
|
|
|
} |
709
|
|
|
|
|
|
|
|
710
|
|
|
|
|
|
|
##################################################################### |
711
|
|
|
|
|
|
|
# Get "self" for a function/method combo sub: |
712
|
|
|
|
|
|
|
# If the first arg is an object ref we shift it off and use that |
713
|
|
|
|
|
|
|
# (i.e. assume it is called as a method); otherwise we assume it's a |
714
|
|
|
|
|
|
|
# functional-API function call and use the caller's "current sheet" |
715
|
|
|
|
|
|
|
# (if none exists, __self creates one but __selfmust throws). |
716
|
|
|
|
|
|
|
# |
717
|
|
|
|
|
|
|
# This must be used with special syntax like |
718
|
|
|
|
|
|
|
# my $self = &__self; |
719
|
|
|
|
|
|
|
# which re-uses @_ so we can shift @_ as seen by our caller. |
720
|
|
|
|
|
|
|
|
721
|
|
|
|
|
|
|
sub __self_ifexists { |
722
|
|
|
|
|
|
|
|
723
|
|
|
|
|
|
|
# If the first arg is an object ref, shift it off and return it; |
724
|
|
|
|
|
|
|
# Otherwise, if the caller's "current sheet" exists, return that; |
725
|
|
|
|
|
|
|
# otherwise return undef. |
726
|
|
|
|
|
|
|
|
727
|
|
|
|
|
|
|
(defined(blessed($_[0])) && $_[0]->isa(__PACKAGE__) && shift(@_)) |
728
|
1120
|
100
|
66
|
1120
|
|
8668
|
|| $pkg2currsheet{__find_userpkg()}; |
|
|
|
66
|
|
|
|
|
729
|
|
|
|
|
|
|
} |
730
|
|
|
|
|
|
|
sub __selfmust { # sheet must exist, otherwise throw |
731
|
841
|
50
|
|
841
|
|
1293
|
&__self_ifexists || do{ |
732
|
0
|
|
|
|
|
0
|
my $pkg = caller(1); |
733
|
0
|
|
|
|
|
0
|
croak __methname()," : No sheet is defined in $pkg\n" |
734
|
|
|
|
|
|
|
}; |
735
|
|
|
|
|
|
|
} |
736
|
|
|
|
|
|
|
|
737
|
|
|
|
|
|
|
sub __self { # a new empty sheet is created if necessary |
738
|
279
|
50
|
|
279
|
|
487
|
&__self_ifexists || do{ |
739
|
|
|
|
|
|
|
# Create a new empty sheet and make it the caller's "current sheet". |
740
|
0
|
|
|
|
|
0
|
my %opts; |
741
|
0
|
|
|
|
|
0
|
my ($frame, $args) = __usercall_info(); |
742
|
|
|
|
|
|
|
|
743
|
0
|
|
|
|
|
0
|
my ($userpkg, $fn, $lno, $subname) = @{ __filter_frame($frame) }; |
|
0
|
|
|
|
|
0
|
|
744
|
0
|
|
|
|
|
0
|
$opts{data_source} = "(Created implicitly by '$subname' at ${fn}:$lno)"; |
745
|
|
|
|
|
|
|
|
746
|
0
|
|
|
|
|
0
|
my $self = $pkg2currsheet{$userpkg} = __silent_new(\%opts); |
747
|
0
|
|
|
|
|
0
|
$self |
748
|
|
|
|
|
|
|
} |
749
|
|
|
|
|
|
|
} |
750
|
|
|
|
|
|
|
|
751
|
|
|
|
|
|
|
|
752
|
|
|
|
|
|
|
## Helpers... |
753
|
|
|
|
|
|
|
|
754
|
|
|
|
|
|
|
sub __opthash { |
755
|
35
|
100
|
|
35
|
|
130
|
ref($_[0]) eq "HASH" ? shift(@_) : {} |
756
|
|
|
|
|
|
|
} |
757
|
|
|
|
|
|
|
sub __selfmust_opthash { |
758
|
8
|
|
|
8
|
|
24
|
my $self = &__selfmust; |
759
|
8
|
|
|
|
|
40
|
my $opthash = &__opthash; |
760
|
8
|
|
|
|
|
31
|
($self, $opthash) |
761
|
|
|
|
|
|
|
} |
762
|
|
|
|
|
|
|
sub __self_opthash { |
763
|
5
|
|
|
5
|
|
16
|
my $self = &__self; |
764
|
5
|
|
|
|
|
23
|
my $opthash = &__opthash; |
765
|
5
|
|
|
|
|
19
|
($self, $opthash) |
766
|
|
|
|
|
|
|
} |
767
|
|
|
|
|
|
|
sub __selfonly { |
768
|
274
|
|
|
274
|
|
501
|
my $self = &__self; |
769
|
274
|
50
|
|
|
|
634
|
confess __methname, " expects no arguments!\n" if @_; |
770
|
274
|
|
|
|
|
1176
|
$self |
771
|
|
|
|
|
|
|
} |
772
|
|
|
|
|
|
|
sub __selfmustonly { |
773
|
553
|
|
|
553
|
|
852
|
my $self = &__selfmust; |
774
|
553
|
50
|
|
|
|
1260
|
confess __methname, " expects no arguments!\n" if @_; |
775
|
553
|
|
|
|
|
2127
|
$self |
776
|
|
|
|
|
|
|
} |
777
|
|
|
|
|
|
|
|
778
|
|
|
|
|
|
|
sub __self_opthash_Nargs($@) { # (num_expected_args, @_) |
779
|
2
|
|
|
2
|
|
5
|
my $Nargs = shift; |
780
|
2
|
|
|
|
|
6
|
my ($self, $opthash) = &__self_opthash; |
781
|
|
|
|
|
|
|
#croak |
782
|
2
|
50
|
|
|
|
7
|
croak __methname, " expects $Nargs arguments, not ",scalar(@_),"\n" |
783
|
|
|
|
|
|
|
if $Nargs != @_; |
784
|
2
|
|
|
|
|
7
|
($self, $opthash, @_) |
785
|
|
|
|
|
|
|
} |
786
|
0
|
|
|
0
|
|
0
|
sub __self_opthash_0args { unshift @_,0; goto &__self_opthash_Nargs } |
|
0
|
|
|
|
|
0
|
|
787
|
2
|
|
|
2
|
|
7
|
sub __self_opthash_1arg { unshift @_,1; goto &__self_opthash_Nargs } |
|
2
|
|
|
|
|
6
|
|
788
|
0
|
|
|
0
|
|
0
|
sub __self_opthash_2args { unshift @_,2; goto &__self_opthash_Nargs } |
|
0
|
|
|
|
|
0
|
|
789
|
0
|
|
|
0
|
|
0
|
sub __self_opthash_3args { unshift @_,3; goto &__self_opthash_Nargs } |
|
0
|
|
|
|
|
0
|
|
790
|
|
|
|
|
|
|
|
791
|
|
|
|
|
|
|
# Check that an option hash has only valid keys, and values aren't undef |
792
|
|
|
|
|
|
|
sub __validate_opthash($$;@) { |
793
|
10
|
|
|
10
|
|
55
|
my ($opthash, $valid_keys, %opts) = @_; |
794
|
10
|
50
|
|
|
|
27
|
return unless defined $opthash; # silently accept undef |
795
|
10
|
|
|
|
|
36
|
foreach my $k (keys %$opthash) { |
796
|
|
|
|
|
|
|
croak "Unrecognized ",($opts{desc}//"option")," '$k'" |
797
|
1
|
50
|
0
|
1
|
|
11
|
unless first{$_ eq $k} @$valid_keys; |
|
1
|
|
|
|
|
5
|
|
798
|
|
|
|
|
|
|
confess "Option '$k' must be defined" |
799
|
|
|
|
|
|
|
if $opts{undef_ok} && !defined($opthash->{$k}) |
800
|
1
|
50
|
33
|
|
|
12
|
&& !grep{$_ eq $k} @{$opts{undef_ok}}; |
|
0
|
|
33
|
|
|
0
|
|
|
0
|
|
|
|
|
0
|
|
801
|
|
|
|
|
|
|
} |
802
|
|
|
|
|
|
|
$opthash |
803
|
10
|
|
|
|
|
25
|
} |
804
|
|
|
|
|
|
|
|
805
|
|
|
|
|
|
|
# Copy verbose/debug/silent options into $self, deleting them from |
806
|
|
|
|
|
|
|
# the provided options hash. |
807
|
|
|
|
|
|
|
# RETURNS: Hash of original values to pass to _restore_stdopts() |
808
|
|
|
|
|
|
|
# |
809
|
|
|
|
|
|
|
# This is used by methods which accept {verbose} etc. options |
810
|
|
|
|
|
|
|
# which override what is in the object for the duration of that method call. |
811
|
|
|
|
|
|
|
sub _set_stdopts { |
812
|
10
|
|
|
10
|
|
27
|
my ($self, $opthash) = @_; |
813
|
10
|
|
|
|
|
21
|
my $previous = {}; |
814
|
10
|
|
|
|
|
31
|
foreach my $key (qw/verbose debug silent/) { |
815
|
30
|
100
|
|
|
|
78
|
if (exists $opthash->{$key}) { |
816
|
2
|
|
|
|
|
6
|
$previous->{$key} = $$self->{$key}; |
817
|
2
|
|
|
|
|
7
|
$$self->{$key} = delete($opthash->{$key}); |
818
|
|
|
|
|
|
|
} |
819
|
|
|
|
|
|
|
} |
820
|
|
|
|
|
|
|
$previous |
821
|
10
|
|
|
|
|
27
|
} |
822
|
|
|
|
|
|
|
sub _restore_stdopts { |
823
|
10
|
|
|
10
|
|
17
|
my $self = shift; |
824
|
10
|
|
|
|
|
18
|
my $saved = shift; |
825
|
10
|
|
|
|
|
85
|
@$$self{keys %$saved} = values %$saved; |
826
|
|
|
|
|
|
|
} |
827
|
|
|
|
|
|
|
|
828
|
|
|
|
|
|
|
sub _validate_ident($) { |
829
|
0
|
0
|
|
0
|
|
0
|
croak "identifier is undef!" unless defined $_[0]; |
830
|
0
|
0
|
|
|
|
0
|
croak "identifier is empty" unless $_[0] ne ""; |
831
|
0
|
0
|
|
|
|
0
|
croak ivisq '"$_[0]" is not a valid identifier\n' |
832
|
|
|
|
|
|
|
unless $_[0] eq title2ident($_[0]); |
833
|
0
|
|
|
|
|
0
|
$_[0] |
834
|
|
|
|
|
|
|
} |
835
|
|
|
|
|
|
|
|
836
|
|
|
|
|
|
|
# Check that an option hash has only valid keys |
837
|
|
|
|
|
|
|
sub __validate_4pthash($$;$) { |
838
|
0
|
|
|
0
|
|
0
|
my ($opthash, $valid_keys, $optdesc) = @_; |
839
|
0
|
0
|
|
|
|
0
|
return unless defined $opthash; # silently accept undef |
840
|
0
|
|
|
|
|
0
|
foreach my $k (keys %$opthash) { |
841
|
|
|
|
|
|
|
croak "Unrecognized ",($optdesc//"option")," '$k'" |
842
|
0
|
0
|
0
|
0
|
|
0
|
unless first{$_ eq $k} @$valid_keys; |
|
0
|
|
|
|
|
0
|
|
843
|
|
|
|
|
|
|
} |
844
|
|
|
|
|
|
|
$opthash |
845
|
0
|
|
|
|
|
0
|
} |
846
|
|
|
|
|
|
|
|
847
|
|
|
|
|
|
|
sub __validate_nonnegi($;$) { |
848
|
4
|
50
|
0
|
4
|
|
27
|
croak(($_[1]//"argument")." must be a non-negative integer", |
|
|
|
33
|
|
|
|
|
849
|
|
|
|
|
|
|
" (not ".u($_[0]).")") |
850
|
|
|
|
|
|
|
unless defined($_[0]) && "$_[0]" =~ /^\d+$/; |
851
|
4
|
|
|
|
|
10
|
$_[0] |
852
|
|
|
|
|
|
|
} |
853
|
|
|
|
|
|
|
sub __validate_nonnegi_or_undef($;$) { |
854
|
0
|
0
|
0
|
0
|
|
0
|
croak(($_[1]//"argument")." must be a non-negative integer or undef", |
|
|
|
0
|
|
|
|
|
855
|
|
|
|
|
|
|
" (not ".u($_[0]).")") |
856
|
|
|
|
|
|
|
unless !defined($_[0]) || "$_[0]" =~ /^\d+$/; |
857
|
0
|
|
|
|
|
0
|
$_[0] |
858
|
|
|
|
|
|
|
} |
859
|
|
|
|
|
|
|
|
860
|
|
|
|
|
|
|
sub __validate_pairs(@) { |
861
|
4
|
50
|
|
4
|
|
29
|
unless ((scalar(@_) % 2) == 0) { |
862
|
0
|
0
|
|
|
|
0
|
croak __methname," does not accept an {OPTIONS} hash here" |
863
|
|
|
|
|
|
|
if (ref($_[0]) eq "HASH"); |
864
|
0
|
|
|
|
|
0
|
confess "In call to ",__methname, |
865
|
|
|
|
|
|
|
" : uneven arg count, expecting key => value pairs" |
866
|
|
|
|
|
|
|
} |
867
|
4
|
|
|
|
|
46
|
foreach (pairs @_) { |
868
|
1
|
|
|
|
|
4
|
my $key = $_->[0]; |
869
|
1
|
50
|
|
|
|
6
|
confess "In call to ",__methname," the key '$key' looks suspicious" |
870
|
|
|
|
|
|
|
unless $key =~ /^\w+$/; |
871
|
|
|
|
|
|
|
} |
872
|
|
|
|
|
|
|
@_ |
873
|
4
|
|
|
|
|
24
|
} |
874
|
|
|
|
|
|
|
|
875
|
|
|
|
|
|
|
sub _check_rx { |
876
|
0
|
|
|
0
|
|
0
|
my ($self, $rx, $one_past_end_ok) = @_; |
877
|
0
|
0
|
0
|
|
|
0
|
confess __methname.": Illegal rx ",vis($rx),"\n" |
878
|
|
|
|
|
|
|
unless ($rx//"") =~ /^\d+$/; # non-negative integer |
879
|
0
|
|
|
|
|
0
|
my $maxrx = $#{$$self->{rows}}; |
|
0
|
|
|
|
|
0
|
|
880
|
0
|
0
|
|
|
|
0
|
confess __methname.": rx ".vis($rx)." is beyond the last row\n" |
|
|
0
|
|
|
|
|
|
881
|
|
|
|
|
|
|
.dvis(' $$self') |
882
|
|
|
|
|
|
|
if $rx > ($one_past_end_ok ? ($maxrx+1) : $maxrx); |
883
|
|
|
|
|
|
|
} |
884
|
|
|
|
|
|
|
|
885
|
|
|
|
|
|
|
# Diagnose scalar context if there are no results. |
886
|
|
|
|
|
|
|
sub __first_ifnot_wantarray(@) { |
887
|
0
|
|
|
0
|
|
0
|
my $wantarray = (caller(1))[5]; |
888
|
0
|
0
|
|
|
|
0
|
return @_ if $wantarray; |
889
|
0
|
0
|
|
|
|
0
|
return $_[0] if @_; |
890
|
0
|
0
|
|
|
|
0
|
croak __methname, " called in scalar context but that method does not return a result.\n" |
891
|
|
|
|
|
|
|
if defined($wantarray); |
892
|
|
|
|
|
|
|
} |
893
|
|
|
|
|
|
|
sub __validate_not_scalar_context(@) { |
894
|
0
|
|
|
0
|
|
0
|
my $wantarray = (caller(1))[5]; |
895
|
0
|
0
|
0
|
|
|
0
|
croak __methname, " returns an array, not a scalar" |
896
|
|
|
|
|
|
|
unless $wantarray || !defined($wantarray); |
897
|
|
|
|
|
|
|
@_ |
898
|
0
|
|
|
|
|
0
|
} |
899
|
|
|
|
|
|
|
|
900
|
|
|
|
|
|
|
sub _carponce { # if not silent |
901
|
0
|
|
|
0
|
|
0
|
my $self = shift; |
902
|
0
|
|
|
|
|
0
|
my $msg = join "",@_; |
903
|
0
|
0
|
|
|
|
0
|
return if $$self->{_carponce}->{$msg}++; |
904
|
0
|
0
|
|
|
|
0
|
$msg .= "\n" unless $msg =~ /\n\z/s; |
905
|
|
|
|
|
|
|
carp($msg) |
906
|
0
|
0
|
|
|
|
0
|
unless $$self->{silent}; # never appears even if silent is later unset |
907
|
|
|
|
|
|
|
} |
908
|
|
|
|
|
|
|
|
909
|
|
|
|
|
|
|
###################### METHODS/FUNCTIONS ####################### |
910
|
|
|
|
|
|
|
|
911
|
|
|
|
|
|
|
# Unlike other methods, new() takes key => value pair arguments. |
912
|
|
|
|
|
|
|
# For consistency with other methods an initial {OPTIONS} hash is |
913
|
|
|
|
|
|
|
# also allowed, but it is not special in any way and is merged |
914
|
|
|
|
|
|
|
# with any linear args (linear args override {OPTIONS}). |
915
|
|
|
|
|
|
|
|
916
|
|
|
|
|
|
|
sub new { # Strictly OO, this does not affect caller's "current sheet". |
917
|
|
|
|
|
|
|
# The corresponding functional API is new_sheet() which explicitly |
918
|
|
|
|
|
|
|
# creates a new sheet and makes it the 'current sheet'. |
919
|
4
|
|
|
4
|
1
|
9
|
my $classname = shift; |
920
|
4
|
50
|
33
|
|
|
33
|
croak "Invalid/missing CLASSNAME (i.e. \"this\") arg" |
921
|
|
|
|
|
|
|
unless defined($classname) && $classname =~ /^[\w_:]+$/; |
922
|
|
|
|
|
|
|
|
923
|
4
|
|
|
|
|
17
|
my $opthash = &__opthash; |
924
|
|
|
|
|
|
|
# Special handling of {cmd_nesting) since there was no object to begin with: |
925
|
|
|
|
|
|
|
# Internal callers may pass this as a "user" option in {OPTARGS}; |
926
|
|
|
|
|
|
|
# we won't log it, but we plant it into the object below. |
927
|
|
|
|
|
|
|
# **THE CALLER MUST DECREMENT IT LATER IF NEEDED* |
928
|
4
|
|
50
|
|
|
23
|
my $cmd_nesting = delete($opthash->{cmd_nesting}) // 0; |
929
|
|
|
|
|
|
|
|
930
|
4
|
|
|
|
|
22
|
my %opts = (verbose => $Verbose, debug => $Debug, silent => $Silent, |
931
|
|
|
|
|
|
|
%$opthash, |
932
|
|
|
|
|
|
|
__validate_pairs(@_)); |
933
|
|
|
|
|
|
|
|
934
|
4
|
|
|
|
|
9
|
my $self; |
935
|
4
|
50
|
|
|
|
59
|
if (my $clonee = delete $opts{clone}) { # untested as of 2/12/14 |
936
|
0
|
|
|
|
|
0
|
delete @opts{qw/verbose debug silent/}; |
937
|
0
|
0
|
|
|
|
0
|
croak "Other options not allowed with 'clone': ",hvis(%opts) if %opts; |
938
|
0
|
|
|
|
|
0
|
require Clone; |
939
|
0
|
|
|
|
|
0
|
$self = Clone::clone($clonee); # in all its glory |
940
|
|
|
|
|
|
|
$$self->{data_source} = (delete $opts{data_source}) |
941
|
0
|
|
0
|
|
|
0
|
// "cloned from $$self->{data_source}"; |
942
|
|
|
|
|
|
|
} else { |
943
|
|
|
|
|
|
|
my $hash = { |
944
|
|
|
|
|
|
|
attributes => delete $opts{attributes} // {}, |
945
|
|
|
|
|
|
|
linenums => delete $opts{linenums} // [], |
946
|
|
|
|
|
|
|
meta_info => delete $opts{meta_info} // [], ##### ???? obsolete ??? |
947
|
|
|
|
|
|
|
data_source => delete $opts{data_source}, # // "(none)", |
948
|
|
|
|
|
|
|
num_cols => delete $opts{num_cols}, # possibly undef |
949
|
|
|
|
|
|
|
|
950
|
|
|
|
|
|
|
# %colx maps titles, aliases (automatic and user-defined), and |
951
|
|
|
|
|
|
|
# spreadsheet column lettercodes to the corresponding column indicies. |
952
|
4
|
|
50
|
|
|
113
|
colx => {}, |
|
|
|
50
|
|
|
|
|
|
|
|
50
|
|
|
|
|
953
|
|
|
|
|
|
|
colx_desc => {}, # for use in error messages |
954
|
|
|
|
|
|
|
useraliases => {}, # key exists for user-defined alias names |
955
|
|
|
|
|
|
|
|
956
|
|
|
|
|
|
|
title_rx => undef, |
957
|
|
|
|
|
|
|
first_data_rx => undef, |
958
|
|
|
|
|
|
|
last_data_rx => undef, |
959
|
|
|
|
|
|
|
current_rx => undef, # valid during apply() |
960
|
|
|
|
|
|
|
|
961
|
|
|
|
|
|
|
pkg2tiedvarnames => {}, |
962
|
|
|
|
|
|
|
pkg2tieall => {}, |
963
|
|
|
|
|
|
|
}; |
964
|
|
|
|
|
|
|
|
965
|
|
|
|
|
|
|
# We can not use $hash directly as the object representation because %{} |
966
|
|
|
|
|
|
|
# is overloaded, so we use a scalar ref (pointing to the hashref) |
967
|
|
|
|
|
|
|
# as the object. |
968
|
4
|
|
|
|
|
17
|
$self = bless \$hash, $classname; |
969
|
|
|
|
|
|
|
|
970
|
|
|
|
|
|
|
# Create a tied virtual array which creates Magicrows when assigned to. |
971
|
4
|
|
|
|
|
7
|
my @rows; tie @rows, 'Spreadsheet::Edit::RowsTie', $self; |
|
4
|
|
|
|
|
34
|
|
972
|
4
|
|
|
|
|
18
|
$hash->{rows} = \@rows; |
973
|
|
|
|
|
|
|
|
974
|
4
|
100
|
|
|
|
15
|
if (my $newdata = delete $opts{rows}) { |
975
|
2
|
|
|
|
|
6
|
foreach (@$newdata) { |
976
|
6
|
|
|
|
|
30
|
push @rows, $_; |
977
|
|
|
|
|
|
|
} |
978
|
|
|
|
|
|
|
} |
979
|
|
|
|
|
|
|
}# not cloning |
980
|
|
|
|
|
|
|
|
981
|
4
|
|
|
|
|
70
|
$$self->{cmd_nesting} = $cmd_nesting; |
982
|
|
|
|
|
|
|
|
983
|
4
|
|
|
|
|
27
|
$self->_set_verbose_debug_silent(%opts); # croaks if other keys remain |
984
|
|
|
|
|
|
|
|
985
|
|
|
|
|
|
|
# Validate data, default num_cols, pad rows, etc. |
986
|
4
|
|
|
|
|
53
|
$self->_rows_replaced(); |
987
|
|
|
|
|
|
|
|
988
|
|
|
|
|
|
|
log_call [$opthash,@_], [\fmt_sheet($self)] |
989
|
4
|
50
|
|
|
|
14
|
if $$self->{verbose}; |
990
|
|
|
|
|
|
|
|
991
|
4
|
|
|
|
|
23
|
$self |
992
|
|
|
|
|
|
|
}#new |
993
|
|
|
|
|
|
|
|
994
|
|
|
|
|
|
|
use overload |
995
|
|
|
|
|
|
|
# As an ARRAYref, a sheet acts like \@rows which is (a ref to) a |
996
|
|
|
|
|
|
|
# virtual array of Magicrow objects, each of which is a dual array/hash |
997
|
|
|
|
|
|
|
# ref to cells in a given row (via RowsTie). |
998
|
0
|
|
|
0
|
|
0
|
'@{}' => sub { my $hash = ${ shift() }; $hash->{rows}; }, |
|
0
|
|
|
|
|
0
|
|
|
0
|
|
|
|
|
0
|
|
999
|
|
|
|
|
|
|
|
1000
|
|
|
|
|
|
|
# As a HASHref, a sheet acts like \%crow which is (a ref to) |
1001
|
|
|
|
|
|
|
# the hash view of the current row during 'apply' |
1002
|
0
|
|
|
0
|
|
0
|
'%{}' => sub { my $self = shift; |
1003
|
|
|
|
|
|
|
# probably less efficient but avoids repeating code |
1004
|
0
|
|
|
|
|
0
|
\%{ $self->crow() }; |
|
0
|
|
|
|
|
0
|
|
1005
|
|
|
|
|
|
|
}, |
1006
|
|
|
|
|
|
|
#'""' => sub { shift }, |
1007
|
|
|
|
|
|
|
#'0+' => sub { shift }, |
1008
|
|
|
|
|
|
|
#'==' => sub { my ($self, $other, $swap) = @_; $self == $other }, |
1009
|
|
|
|
|
|
|
#'eq' => sub { my ($self, $other, $swap) = @_; "$self" eq "$other" }, |
1010
|
3
|
|
|
|
|
30
|
fallback => 1, |
1011
|
3
|
|
|
3
|
|
34
|
; |
|
3
|
|
|
|
|
6
|
|
1012
|
|
|
|
|
|
|
|
1013
|
|
|
|
|
|
|
sub _rows_replaced { # completely new or replaced rows, linenums, etc. |
1014
|
6
|
|
|
6
|
|
20
|
my ($self) = @_; |
1015
|
6
|
|
|
|
|
14
|
my $hash = $$self; |
1016
|
|
|
|
|
|
|
|
1017
|
|
|
|
|
|
|
my ($rows, $linenums, $num_cols, $current_rx) |
1018
|
6
|
|
|
|
|
43
|
= @$hash{qw/rows linenums num_cols current_rx/}; |
1019
|
|
|
|
|
|
|
|
1020
|
6
|
50
|
|
|
|
21
|
croak "Can not replace sheet content during an apply!\n" |
1021
|
|
|
|
|
|
|
if defined $current_rx; |
1022
|
6
|
|
|
|
|
22
|
for my $rx (0..$#$rows) { |
1023
|
12
|
|
|
|
|
45
|
my $row = $rows->[$rx]; |
1024
|
12
|
50
|
|
|
|
29
|
croak "rows must contain refs to arrays of cells (row $rx is $row)" |
1025
|
|
|
|
|
|
|
unless __looks_like_aref($row); |
1026
|
12
|
|
|
|
|
26
|
for my $cx (0..$#$row) { |
1027
|
30
|
50
|
|
|
|
54
|
croak "New cell at Row ",$rx+1," Column ",cx2let($cx)," contains a ref" |
1028
|
|
|
|
|
|
|
if ref($row->[$cx]); |
1029
|
|
|
|
|
|
|
} |
1030
|
|
|
|
|
|
|
} |
1031
|
|
|
|
|
|
|
croak '\'linenums\' if present must be (a ref to) an array of numbers, ', |
1032
|
|
|
|
|
|
|
' "" or "???"', ivis('\nnot $linenums\n') |
1033
|
|
|
|
|
|
|
unless ref($linenums) eq "ARRAY" |
1034
|
6
|
50
|
33
|
6
|
|
74
|
&& all{ defined() and looks_like_number($_) || !/[^\?]/ } @$linenums; |
|
6
|
50
|
33
|
|
|
25
|
|
1035
|
|
|
|
|
|
|
|
1036
|
6
|
100
|
|
|
|
31
|
if (@$rows) { # Determine num_cols and pad short rows |
1037
|
4
|
|
|
|
|
8
|
my $nc = 0; |
1038
|
4
|
100
|
|
|
|
13
|
foreach (@$rows) { $nc = @$_ if @$_ > $nc } |
|
12
|
|
|
|
|
24
|
|
1039
|
4
|
50
|
33
|
|
|
15
|
if ($num_cols && $num_cols != $nc) { |
1040
|
0
|
|
|
|
|
0
|
croak "num_cols=$num_cols was specified along with initial data, but\n", |
1041
|
|
|
|
|
|
|
"the value doesn't match the data (which has up to $nc columns)\n" |
1042
|
|
|
|
|
|
|
} else { |
1043
|
4
|
|
|
|
|
23
|
$hash->{num_cols} = $num_cols = $nc; |
1044
|
|
|
|
|
|
|
} |
1045
|
|
|
|
|
|
|
# Pad short rows with empty fields |
1046
|
4
|
|
|
|
|
15
|
foreach my $row (@$rows) { |
1047
|
12
|
|
|
|
|
26
|
push @$row, (("") x ($num_cols - @$row)); |
1048
|
|
|
|
|
|
|
} |
1049
|
4
|
|
|
|
|
10
|
$#$linenums = $#$rows; |
1050
|
4
|
|
100
|
|
|
11
|
foreach (@$linenums) { $_ //= '???' }; |
|
12
|
|
|
|
|
43
|
|
1051
|
|
|
|
|
|
|
} else { |
1052
|
|
|
|
|
|
|
# There is no data. Default num_cols to zero, but leave any |
1053
|
|
|
|
|
|
|
# user-supplied value so a subsequent insert_rows() will know how |
1054
|
|
|
|
|
|
|
# many columns to create. |
1055
|
2
|
|
50
|
|
|
11
|
$hash->{num_cols} //= 0; |
1056
|
|
|
|
|
|
|
} |
1057
|
|
|
|
|
|
|
#oops unless $hash->{data_source}; |
1058
|
6
|
50
|
|
|
|
16
|
croak "#linenums ($#$linenums) != #rows ($#$rows)\n", |
1059
|
|
|
|
|
|
|
dvis '$hash->{linenums}\n$hash->{rows}' |
1060
|
|
|
|
|
|
|
unless @$linenums == @$rows; |
1061
|
|
|
|
|
|
|
|
1062
|
6
|
|
|
|
|
11
|
$hash->{title_rx} = undef; |
1063
|
6
|
|
|
|
|
13
|
$hash->{first_data_rx} = undef; |
1064
|
6
|
|
|
|
|
10
|
$hash->{last_data_rx} = undef; |
1065
|
6
|
|
|
|
|
16
|
$hash->{useraliases} = {}; |
1066
|
6
|
|
|
|
|
46
|
$self->_rebuild_colx; # Set up colx colx_desc |
1067
|
6
|
|
|
|
|
12
|
$self |
1068
|
|
|
|
|
|
|
}#_rows_replaced |
1069
|
|
|
|
|
|
|
|
1070
|
|
|
|
|
|
|
######################################################### |
1071
|
|
|
|
|
|
|
# Combination FUNCTION/METHOD |
1072
|
|
|
|
|
|
|
# These are declared with signatures for use as functional-API FUNCTIONs |
1073
|
|
|
|
|
|
|
# which use the caller's "current sheet" as the implicit object. |
1074
|
|
|
|
|
|
|
# |
1075
|
|
|
|
|
|
|
# However they may also be called as METHODs with an explicit object. |
1076
|
|
|
|
|
|
|
######################################################### |
1077
|
|
|
|
|
|
|
|
1078
|
|
|
|
|
|
|
# Allow user to find out names of tied variables |
1079
|
|
|
|
|
|
|
sub tied_varnames(;@) { |
1080
|
0
|
|
|
0
|
0
|
0
|
my ($self, $opts) = &__selfmust_opthash; |
1081
|
0
|
|
0
|
|
|
0
|
my $pkg = $opts->{package} // __callerpkg(); |
1082
|
0
|
|
0
|
|
|
0
|
my $h = $$self->{pkg2tiedvarnames}->{$pkg} //= {}; |
1083
|
0
|
|
|
|
|
0
|
return keys %$h; |
1084
|
|
|
|
|
|
|
} |
1085
|
|
|
|
|
|
|
|
1086
|
|
|
|
|
|
|
# Internal: Tie specified variables into a package if not already tied. |
1087
|
|
|
|
|
|
|
# Returns: |
1088
|
|
|
|
|
|
|
sub __TCV_REDUNDANT() { 1 } # if all were already tied |
1089
|
|
|
|
|
|
|
sub __TCV_OK() { 2 } # otherwise (some tied, or no idents specified) |
1090
|
|
|
|
|
|
|
# |
1091
|
|
|
|
|
|
|
sub _tie_col_vars { |
1092
|
3
|
|
|
3
|
|
6
|
my $self = shift; |
1093
|
3
|
|
|
|
|
7
|
my $pkg = shift; |
1094
|
3
|
|
|
|
|
5
|
my $parms = shift; |
1095
|
|
|
|
|
|
|
# Remaining arguments are idents |
1096
|
|
|
|
|
|
|
|
1097
|
3
|
|
|
|
|
23
|
my ($safe, $file, $lno) = @$parms; |
1098
|
3
|
100
|
|
|
|
17
|
my @safecheck_pkgs = $pkg eq "main" ? ($pkg) : ($pkg, "main"); |
1099
|
|
|
|
|
|
|
|
1100
|
|
|
|
|
|
|
my ($colx, $colx_desc, $debug, $silent) |
1101
|
3
|
|
|
|
|
26
|
= @$$self{qw/colx colx_desc debug silent/}; |
1102
|
|
|
|
|
|
|
|
1103
|
|
|
|
|
|
|
# FIXME: BUG/ISSUE ... |
1104
|
|
|
|
|
|
|
# Why is it correct to keep tiedvarnames PER-SHEET ? |
1105
|
|
|
|
|
|
|
# Isn't this a global property of each package? |
1106
|
|
|
|
|
|
|
|
1107
|
3
|
|
50
|
|
|
25
|
my $tiedvarnames = ($$self->{pkg2tiedvarnames}->{$pkg} //= {}); |
1108
|
|
|
|
|
|
|
|
1109
|
3
|
50
|
33
|
|
|
27
|
if (@_ > 0 && %$tiedvarnames) { |
1110
|
|
|
|
|
|
|
SHORTCUT: { |
1111
|
0
|
|
|
|
|
0
|
foreach (@_) { |
|
0
|
|
|
|
|
0
|
|
1112
|
0
|
0
|
|
|
|
0
|
last SHORTCUT unless exists $tiedvarnames->{$_}; |
1113
|
|
|
|
|
|
|
} |
1114
|
0
|
|
|
|
|
0
|
return __TCV_REDUNDANT; |
1115
|
|
|
|
|
|
|
} |
1116
|
|
|
|
|
|
|
} |
1117
|
|
|
|
|
|
|
|
1118
|
|
|
|
|
|
|
VAR: |
1119
|
3
|
|
|
|
|
12
|
foreach (sort {$a->[0] <=> $b->[0]} # sort for ease of debugging |
|
26
|
|
|
|
|
51
|
|
1120
|
|
|
|
|
|
|
map { |
1121
|
16
|
|
|
|
|
31
|
my $cx = $colx->{$_}; |
1122
|
|
|
|
|
|
|
defined($cx) |
1123
|
16
|
50
|
|
|
|
67
|
? [ $cx, $_, $colx_desc->{$_} ] |
1124
|
|
|
|
|
|
|
: [ 999999, $_, "(currently NOT DEFINED)" ]; |
1125
|
|
|
|
|
|
|
} @_ |
1126
|
|
|
|
|
|
|
) |
1127
|
|
|
|
|
|
|
{ |
1128
|
16
|
|
|
|
|
162
|
my ($cx, $ident, $desc) = @$_; |
1129
|
16
|
50
|
|
|
|
71
|
oops unless $ident =~ /^\w+$/; |
1130
|
|
|
|
|
|
|
|
1131
|
16
|
50
|
|
|
|
35
|
if (exists $tiedvarnames->{$ident}) { |
1132
|
0
|
0
|
|
|
|
0
|
$self->_log(" Previously tied: \$${pkg}::${ident}\n") if $debug; |
1133
|
|
|
|
|
|
|
next |
1134
|
0
|
|
|
|
|
0
|
} |
1135
|
|
|
|
|
|
|
|
1136
|
16
|
50
|
|
|
|
28
|
btw dvis '##ZZ $ident $safe ${^GLOBAL_PHASE} @safecheck_pkgs\n' if $debug; |
1137
|
|
|
|
|
|
|
|
1138
|
3
|
|
|
3
|
|
4713
|
no strict 'refs'; |
|
3
|
|
|
|
|
8
|
|
|
3
|
|
|
|
|
526
|
|
1139
|
16
|
100
|
|
|
|
35
|
if ($safe) { |
1140
|
8
|
50
|
|
|
|
21
|
if (${^GLOBAL_PHASE} ne "START") { |
1141
|
0
|
0
|
|
|
|
0
|
$self->_carponce("Not tieing new variables because :safe was used and this is not (any longer) during compile time\n") unless $silent; |
1142
|
0
|
|
|
|
|
0
|
return __TCV_REDUNDANT; ### IMMEDIATE EXIT ### |
1143
|
|
|
|
|
|
|
} |
1144
|
8
|
|
|
|
|
12
|
foreach my $p (@safecheck_pkgs) { |
1145
|
|
|
|
|
|
|
# Per 'man perlref' we can not use *foo{SCALAR} to detect a never- |
1146
|
|
|
|
|
|
|
# declared SCALAR (it's indistinguishable from an existing undef var). |
1147
|
|
|
|
|
|
|
# So we must insist that the entire glob does not exist. |
1148
|
16
|
50
|
|
|
|
29
|
btw ivis ' Checking pkg $p ident $ident ...\n' if $debug; |
1149
|
3
|
|
|
3
|
|
25
|
no strict 'refs'; |
|
3
|
|
|
|
|
27
|
|
|
3
|
|
|
|
|
511
|
|
1150
|
16
|
50
|
|
|
|
23
|
if (exists ${$p.'::'}{$ident}) { |
|
16
|
|
|
|
|
71
|
|
1151
|
0
|
|
|
|
|
0
|
my $msg = <<EOF ; |
1152
|
|
|
|
|
|
|
COLSPEC '$ident' clashes with an existing object in package '$p' . |
1153
|
|
|
|
|
|
|
This is dis-allowed when tie_column_vars was called with option :safe, |
1154
|
|
|
|
|
|
|
in this case at ${file}:${lno} . In this situation you can not |
1155
|
|
|
|
|
|
|
explicitly declare the tied variables, and they must be tied and |
1156
|
|
|
|
|
|
|
imported before the compiler sees them. |
1157
|
|
|
|
|
|
|
|
1158
|
|
|
|
|
|
|
Note: The clash might not be with a scalar \$$ident, but something else |
1159
|
|
|
|
|
|
|
named '${ident}' in the same package (array, hash, sub, filehandle, |
1160
|
|
|
|
|
|
|
etc.) Unfortunately it is not possible to distinguish a non-existent |
1161
|
|
|
|
|
|
|
scalar from a declared scalar containing an undef value |
1162
|
|
|
|
|
|
|
(see *foo{SCALAR} in 'man perlref'). Therefore, to be safe, |
1163
|
|
|
|
|
|
|
nothing is allowed to pre-exist with the name '${ident}'. |
1164
|
|
|
|
|
|
|
EOF |
1165
|
|
|
|
|
|
|
# We can detect anything other than an undef scalar |
1166
|
0
|
|
|
|
|
0
|
local $Data::Dumper::Maxdepth = 2; |
1167
|
0
|
|
|
|
|
0
|
local $Data::Dumper::Interp::Maxdepth = 2; |
1168
|
0
|
|
|
|
|
0
|
my $fqname = $p."::".$ident; |
1169
|
3
|
|
|
3
|
|
21
|
no strict 'refs'; |
|
3
|
|
|
|
|
13
|
|
|
3
|
|
|
|
|
35656
|
|
1170
|
0
|
0
|
|
|
|
0
|
if (my $r = *$fqname{SCALAR}) { |
1171
|
0
|
0
|
|
|
|
0
|
if (defined($$r)) { |
1172
|
0
|
|
|
|
|
0
|
$msg .= "\nExisting *Defined* Scalar \$$ident = ".vis($$r)."\n"; |
1173
|
|
|
|
|
|
|
} |
1174
|
|
|
|
|
|
|
} |
1175
|
0
|
0
|
|
|
|
0
|
if (my $r = *$fqname{ARRAY}) { |
1176
|
0
|
|
|
|
|
0
|
$msg .= "\nExisting Array \@$fqname = ".avis(@$r)."\n"; |
1177
|
|
|
|
|
|
|
} |
1178
|
0
|
0
|
|
|
|
0
|
if (my $r = *$fqname{HASH}) { |
1179
|
0
|
|
|
|
|
0
|
$msg .= "\nExisting Hash \%$fqname = ".hvis(%$r)."\n"; |
1180
|
|
|
|
|
|
|
} |
1181
|
0
|
0
|
|
|
|
0
|
if (my $r = *$fqname{CODE}) { |
1182
|
0
|
|
|
|
|
0
|
$msg .= "\nExisting $fqname = ".Data::Dumper->new([$r])->Terse(1)->Indent(0)->Deparse(1)->Dump()."\n"; |
1183
|
|
|
|
|
|
|
} |
1184
|
0
|
0
|
|
|
|
0
|
if (my $r = *$fqname{IO}) { |
1185
|
0
|
|
|
|
|
0
|
$msg .= "\nExisting IO object $fqname (file/dir handle, etc.)\n"; |
1186
|
|
|
|
|
|
|
} |
1187
|
0
|
0
|
|
|
|
0
|
if (eval{ require Devel::Symdump; }) { |
|
0
|
|
|
|
|
0
|
|
1188
|
0
|
|
|
|
|
0
|
my $obj = Devel::Symdump->new($p); |
1189
|
0
|
|
|
|
|
0
|
foreach (qw/FILEHANDLE.s DIRHANDLE.s UNKNOWN.s PACKAGE.s/) { |
1190
|
0
|
0
|
|
|
|
0
|
/^(.+)\.(.*)$/ or die; my $kind = lc $1; my $methname = lc $1.$2; |
|
0
|
|
|
|
|
0
|
|
|
0
|
|
|
|
|
0
|
|
1191
|
0
|
0
|
|
|
|
0
|
if (grep{$_ eq $ident} $obj->$methname()) { |
|
0
|
|
|
|
|
0
|
|
1192
|
0
|
|
|
|
|
0
|
$msg .= "\n Found existing $kind object $fqname\n"; |
1193
|
|
|
|
|
|
|
} |
1194
|
|
|
|
|
|
|
} |
1195
|
|
|
|
|
|
|
} |
1196
|
0
|
|
|
|
|
0
|
croak "\n$msg\n"; |
1197
|
|
|
|
|
|
|
} |
1198
|
|
|
|
|
|
|
} |
1199
|
|
|
|
|
|
|
} |
1200
|
|
|
|
|
|
|
|
1201
|
16
|
50
|
|
|
|
29
|
$self->_log("tie \$${pkg}::${ident} to $desc\n") if $debug; |
1202
|
|
|
|
|
|
|
|
1203
|
16
|
|
|
|
|
33
|
$tiedvarnames->{$ident} = 1; |
1204
|
|
|
|
|
|
|
|
1205
|
16
|
|
|
|
|
30
|
*{"$pkg\::$ident"} = \${ *{gensym()} }; |
|
16
|
|
|
|
|
245
|
|
|
16
|
|
|
|
|
20
|
|
|
16
|
|
|
|
|
106
|
|
1206
|
|
|
|
|
|
|
|
1207
|
16
|
|
|
|
|
33
|
tie ${"${pkg}::$ident"}, 'Tie::Indirect::Scalar', |
|
16
|
|
|
|
|
78
|
|
1208
|
|
|
|
|
|
|
\&_tiecell_helper, $pkg, $ident; |
1209
|
|
|
|
|
|
|
} |
1210
|
3
|
|
|
|
|
35
|
return __TCV_OK; |
1211
|
|
|
|
|
|
|
} |
1212
|
|
|
|
|
|
|
sub _tiecell_helper { |
1213
|
2
|
|
|
2
|
|
116
|
my($mutating, $pkg, $ident) = @_; |
1214
|
2
|
|
33
|
|
|
19
|
my $sheet = $pkg2currsheet{$pkg} |
1215
|
|
|
|
|
|
|
// croak "No sheet is currently valid for package $pkg\n"; |
1216
|
2
|
|
|
|
|
19
|
$sheet->_onlyinapply("tied variable \$$ident"); |
1217
|
|
|
|
|
|
|
|
1218
|
|
|
|
|
|
|
# WRONG... it croaks bc sheet->{rows}->[rx] is a rowhash which doesn't like undef keys |
1219
|
|
|
|
|
|
|
# WRONG: This returns \undef if $ident is not currently valid |
1220
|
2
|
|
|
|
|
12
|
\( $$sheet->{rows}->[$$sheet->{current_rx}]->{$ident} ) |
1221
|
|
|
|
|
|
|
} |
1222
|
|
|
|
|
|
|
|
1223
|
|
|
|
|
|
|
sub _all_valid_idents { |
1224
|
3
|
|
|
3
|
|
10
|
my $self = shift; |
1225
|
3
|
|
|
|
|
6
|
my %valid_idents; |
1226
|
3
|
|
|
|
|
7
|
foreach (keys %{ $$self->{colx} }) { |
|
3
|
|
|
|
|
17
|
|
1227
|
18
|
50
|
|
|
|
55
|
if (/^(?:REGERROR|REGMARK|AUTOLOAD)$/) { |
1228
|
0
|
|
|
|
|
0
|
$self->_carponce("WARNING: Column key ",visq($_)," conflicts with a Perl built-in; variable will not be tied.\n"); |
1229
|
0
|
|
|
|
|
0
|
next; |
1230
|
|
|
|
|
|
|
} |
1231
|
18
|
|
|
|
|
34
|
$valid_idents{ title2ident($_) } = 1; |
1232
|
|
|
|
|
|
|
} |
1233
|
3
|
|
|
|
|
24
|
return keys %valid_idents; |
1234
|
|
|
|
|
|
|
} |
1235
|
|
|
|
|
|
|
|
1236
|
|
|
|
|
|
|
sub tie_column_vars(;@) { |
1237
|
3
|
|
|
3
|
1
|
16
|
my ($self, $opts) = &__self_opthash; |
1238
|
|
|
|
|
|
|
# Any remaining args specify variable names matching |
1239
|
|
|
|
|
|
|
# alias names, either user-defined or automatic. |
1240
|
|
|
|
|
|
|
|
1241
|
3
|
50
|
|
|
|
11
|
croak "tie_column_vars without arguments (did you intend to use ':all'?)" |
1242
|
|
|
|
|
|
|
unless @_; |
1243
|
|
|
|
|
|
|
|
1244
|
3
|
|
33
|
|
|
28
|
local $$self->{silent} = $opts->{silent} // $$self->{silent}; |
1245
|
3
|
|
33
|
|
|
32
|
local $$self->{verbose} = $opts->{verbose} // $$self->{verbose}; |
1246
|
3
|
|
33
|
|
|
19
|
local $$self->{debug} = $opts->{debug} // $$self->{debug}; |
1247
|
|
|
|
|
|
|
|
1248
|
3
|
|
66
|
|
|
16
|
my $pkg = $opts->{package} // __callerpkg(); |
1249
|
|
|
|
|
|
|
|
1250
|
3
|
|
|
|
|
6
|
my (%tokens, @varnames); |
1251
|
3
|
50
|
|
|
|
9
|
foreach (@_) { if (/:/) { $tokens{$_} = 1 } else { push @varnames, $_ } } |
|
5
|
|
|
|
|
22
|
|
|
5
|
|
|
|
|
20
|
|
|
0
|
|
|
|
|
0
|
|
1252
|
3
|
|
|
|
|
11
|
foreach (@varnames) { |
1253
|
0
|
0
|
|
|
|
0
|
croak "Invalid variable name '$_'\n" unless /^\$?\w+$/; |
1254
|
0
|
|
|
|
|
0
|
s/^\$//; |
1255
|
|
|
|
|
|
|
} |
1256
|
|
|
|
|
|
|
|
1257
|
|
|
|
|
|
|
# With ':all' tie all possible variables, now and in the future. |
1258
|
|
|
|
|
|
|
# |
1259
|
|
|
|
|
|
|
# CURRENTLY UNDOCUMENTED: With the ":safe" token, a check is made |
1260
|
|
|
|
|
|
|
# that variables do not already exist immediately before tying them; |
1261
|
|
|
|
|
|
|
# otherwise an exception is thrown. |
1262
|
|
|
|
|
|
|
# |
1263
|
|
|
|
|
|
|
# When ':safe' is combined with ':all', variables will not be checked & tied |
1264
|
|
|
|
|
|
|
# except during compile time, i.e. within BEGIN{...}. Therefore a |
1265
|
|
|
|
|
|
|
# malicious spreadsheet can not cause an exception after the compilation |
1266
|
|
|
|
|
|
|
# phase. |
1267
|
3
|
|
|
|
|
6
|
my $safe = delete $tokens{':safe'}; |
1268
|
3
|
|
|
|
|
11
|
my ($file, $lno) = __fn_ln_methname(); |
1269
|
3
|
|
|
|
|
12
|
my $parms = [$safe, $file, $lno]; |
1270
|
|
|
|
|
|
|
|
1271
|
|
|
|
|
|
|
# Why? Obsolete? Only for :all?? [note added Dec22] |
1272
|
|
|
|
|
|
|
# FIXME TODO: is this bogus? |
1273
|
3
|
50
|
|
|
|
14
|
$self->title_rx($opts->{title_rx}) if exists $opts->{title_rx}; |
1274
|
|
|
|
|
|
|
|
1275
|
3
|
50
|
|
|
|
13
|
if (delete $tokens{':all'}) { |
1276
|
|
|
|
|
|
|
# Remember parameters for tie operations which might occur later |
1277
|
3
|
|
|
|
|
14
|
$$self->{pkg2tieall}->{$pkg} = $parms; |
1278
|
|
|
|
|
|
|
|
1279
|
3
|
|
|
|
|
32
|
push @varnames, sort $self->_all_valid_idents; |
1280
|
|
|
|
|
|
|
} |
1281
|
3
|
50
|
|
|
|
10
|
croak "Unrecognized token in arguments: ",avis(keys %tokens) if %tokens; |
1282
|
|
|
|
|
|
|
|
1283
|
3
|
|
|
|
|
16
|
my $r = $self->_tie_col_vars($pkg, $parms, @varnames); |
1284
|
|
|
|
|
|
|
|
1285
|
3
|
50
|
|
|
|
12
|
my $pfx = ($r == __TCV_REDUNDANT ? "[ALL REDUNDANT] " : ""); |
1286
|
|
|
|
|
|
|
log_methcall $self, [\$pfx, \__fmt_uqarraywithqw(keys %tokens, @varnames), |
1287
|
|
|
|
|
|
|
\" in package $pkg"] |
1288
|
3
|
50
|
|
|
|
22
|
if $$self->{verbose}; |
1289
|
|
|
|
|
|
|
}#tie_column_vars |
1290
|
|
|
|
|
|
|
|
1291
|
|
|
|
|
|
|
# |
1292
|
|
|
|
|
|
|
# Accessors for misc. sheet data |
1293
|
|
|
|
|
|
|
# |
1294
|
|
|
|
|
|
|
|
1295
|
274
|
|
|
274
|
1
|
387
|
sub attributes(@) { ${&__selfonly}->{attributes} } |
|
274
|
|
|
|
|
498
|
|
1296
|
0
|
|
|
0
|
1
|
0
|
sub colx() { ${&__selfmustonly}->{colx} } |
|
0
|
|
|
|
|
0
|
|
1297
|
0
|
|
|
0
|
1
|
0
|
sub colx_desc() { ${&__selfmustonly}->{colx_desc} } |
|
0
|
|
|
|
|
0
|
|
1298
|
|
|
|
|
|
|
sub data_source(;$) { |
1299
|
274
|
|
|
274
|
1
|
459
|
my $self = &__selfmust; |
1300
|
274
|
50
|
|
|
|
659
|
if (@_ == 0) { # 'get' request |
1301
|
274
|
50
|
|
|
|
654
|
log_methcall $self, [], [$$self->{data_source}] if $$self->{verbose}; |
1302
|
|
|
|
|
|
|
return $$self->{data_source} |
1303
|
274
|
|
|
|
|
1053
|
} |
1304
|
0
|
0
|
|
|
|
0
|
log_methcall $self, [@_] if $$self->{verbose}; |
1305
|
0
|
0
|
|
|
|
0
|
croak "Too many args" unless @_ == 1; |
1306
|
0
|
|
|
|
|
0
|
$$self->{data_source} = $_[0]; |
1307
|
0
|
|
|
|
|
0
|
$self |
1308
|
|
|
|
|
|
|
} |
1309
|
0
|
|
|
0
|
1
|
0
|
sub linenums() { ${&__selfmustonly}->{linenums} } |
|
0
|
|
|
|
|
0
|
|
1310
|
0
|
|
|
0
|
1
|
0
|
sub num_cols() { ${&__selfmustonly}->{num_cols} } |
|
0
|
|
|
|
|
0
|
|
1311
|
161
|
|
|
161
|
1
|
238
|
sub rows() { ${&__selfmustonly}->{rows} } |
|
161
|
|
|
|
|
288
|
|
1312
|
274
|
|
|
274
|
1
|
359
|
sub sheetname() { ${&__selfmustonly}->{sheetname} } |
|
274
|
|
|
|
|
511
|
|
1313
|
|
|
|
|
|
|
|
1314
|
0
|
|
|
0
|
0
|
0
|
sub iolayers() { ${&__selfmustonly}->{iolayers} } |
|
0
|
|
|
|
|
0
|
|
1315
|
0
|
|
|
0
|
0
|
0
|
sub meta_info() {${&__selfmustonly}->{meta_info} } |
|
0
|
|
|
|
|
0
|
|
1316
|
|
|
|
|
|
|
sub input_encoding() { |
1317
|
|
|
|
|
|
|
# Emulate old API. We actually store input_iolayers instead now, |
1318
|
|
|
|
|
|
|
# so as to include :crlf if necessary. |
1319
|
0
|
|
|
0
|
0
|
0
|
my $self = &__selfmustonly; |
1320
|
0
|
|
|
|
|
0
|
local $_; |
1321
|
|
|
|
|
|
|
return undef unless |
1322
|
0
|
|
|
|
|
0
|
exists(${$self}->{input_iolayers}) |
1323
|
0
|
0
|
0
|
|
|
0
|
&& ${$self}->{input_iolayers} =~ /encoding\(([^()]*)\)/; |
|
0
|
|
|
|
|
0
|
|
1324
|
0
|
|
|
|
|
0
|
return $1; |
1325
|
|
|
|
|
|
|
} |
1326
|
|
|
|
|
|
|
|
1327
|
|
|
|
|
|
|
# See below for title_rx() |
1328
|
|
|
|
|
|
|
sub title_row() { |
1329
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmustonly; |
1330
|
0
|
|
|
|
|
0
|
my $title_rx = $$self->{title_rx}; |
1331
|
0
|
0
|
|
|
|
0
|
my $r = defined($title_rx) ? $$self->{rows}->[$title_rx] : undef; |
1332
|
0
|
0
|
|
|
|
0
|
log_methcall $self, [], [$r] if $$self->{verbose}; |
1333
|
0
|
|
|
|
|
0
|
$r |
1334
|
|
|
|
|
|
|
} |
1335
|
118
|
|
|
118
|
1
|
234
|
sub rx() { ${ &__selfmustonly }->{current_rx} } |
|
118
|
|
|
|
|
246
|
|
1336
|
|
|
|
|
|
|
sub crow() { |
1337
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmustonly; |
1338
|
0
|
|
|
|
|
0
|
${ $self->_onlyinapply("crow() method") }->{rows}->[$$self->{current_rx}] |
|
0
|
|
|
|
|
0
|
|
1339
|
|
|
|
|
|
|
} |
1340
|
|
|
|
|
|
|
sub linenum() { |
1341
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmustonly; |
1342
|
0
|
|
0
|
|
|
0
|
my $current_rx = $$self->{current_rx} // return(undef); |
1343
|
0
|
|
|
|
|
0
|
$$self->{linenums}->[$current_rx]; |
1344
|
|
|
|
|
|
|
} |
1345
|
|
|
|
|
|
|
sub _getref { |
1346
|
0
|
|
|
0
|
|
0
|
my ($self, $rx, $ident) = @_; |
1347
|
0
|
|
|
|
|
0
|
my ($rows, $colx) = @$$self{qw/rows colx/}; |
1348
|
0
|
0
|
0
|
|
|
0
|
croak "get/set: rx $rx is out of range" if $rx < 0 || $rx > $#$rows; |
1349
|
0
|
|
|
|
|
0
|
my $row = $$self->{rows}->[$rx]; |
1350
|
0
|
|
|
|
|
0
|
my $cx = $colx->{$ident}; |
1351
|
0
|
0
|
0
|
|
|
0
|
oops("Invalid cx ".vis($cx)) if ! defined($cx) || $cx < 0 || $cx > $#$row; |
|
|
|
0
|
|
|
|
|
1352
|
0
|
|
|
|
|
0
|
\$row->[$cx]; |
1353
|
|
|
|
|
|
|
} |
1354
|
|
|
|
|
|
|
# get/set a cell given by (rx,COLSPEC) |
1355
|
|
|
|
|
|
|
sub get($$) { |
1356
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
1357
|
0
|
|
|
|
|
0
|
my $ref = $self->_getref(@_); |
1358
|
0
|
|
|
|
|
0
|
$$ref; |
1359
|
|
|
|
|
|
|
} |
1360
|
|
|
|
|
|
|
sub set($$$) { |
1361
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
1362
|
0
|
|
|
|
|
0
|
my ($rx, $colspec, $newval) = @_; |
1363
|
0
|
|
|
|
|
0
|
my $ref = $self->_getref($rx, $colspec); |
1364
|
0
|
|
|
|
|
0
|
$$ref = $newval; |
1365
|
0
|
|
|
|
|
0
|
$self |
1366
|
|
|
|
|
|
|
} |
1367
|
|
|
|
|
|
|
|
1368
|
|
|
|
|
|
|
sub __validate_sheet_arg($) { |
1369
|
12
|
|
|
12
|
|
25
|
my $sheet = shift; |
1370
|
12
|
50
|
33
|
|
|
111
|
croak "Argument '${\u($sheet)}' is not a Spreadsheet::Edit sheet object" |
|
0
|
|
66
|
|
|
0
|
|
1371
|
|
|
|
|
|
|
if defined($sheet) and |
1372
|
|
|
|
|
|
|
!blessed($sheet) || !$sheet->isa("Spreadsheet::Edit"); |
1373
|
12
|
|
|
|
|
34
|
$sheet; |
1374
|
|
|
|
|
|
|
} |
1375
|
|
|
|
|
|
|
|
1376
|
|
|
|
|
|
|
#---- logging ---------------------------------- |
1377
|
|
|
|
|
|
|
|
1378
|
|
|
|
|
|
|
# Print segmented log messages: |
1379
|
|
|
|
|
|
|
# Join args together, prefixing with "> " or ">> " etc. |
1380
|
|
|
|
|
|
|
# unless the previous call did not end with newline. |
1381
|
|
|
|
|
|
|
# Maintains internal state. A final call with an ending \n must occur. |
1382
|
|
|
|
|
|
|
sub _log { |
1383
|
0
|
|
|
0
|
|
0
|
my $self = shift; |
1384
|
0
|
|
|
|
|
0
|
state $in_midst; |
1385
|
|
|
|
|
|
|
print STDERR join "", |
1386
|
|
|
|
|
|
|
($in_midst ? "" : (">" x ($$self->{cmd_nesting}||1))), |
1387
|
0
|
0
|
0
|
|
|
0
|
map{u} @_; |
|
0
|
|
|
|
|
0
|
|
1388
|
0
|
|
|
|
|
0
|
$in_midst = ($_[$#_] !~ /\n\z/s); |
1389
|
|
|
|
|
|
|
} |
1390
|
|
|
|
|
|
|
|
1391
|
|
|
|
|
|
|
my $trunclen = 200; |
1392
|
|
|
|
|
|
|
sub fmt_sheet($) { |
1393
|
0
|
|
|
0
|
0
|
0
|
my $sheet = shift; |
1394
|
0
|
0
|
|
|
|
0
|
return "undef" unless defined($sheet); |
1395
|
|
|
|
|
|
|
#oops unless ref($sheet) ne "" && $sheet->isa(__PACKAGE__); |
1396
|
0
|
|
|
|
|
0
|
local $$sheet->{verbose} = 0; |
1397
|
0
|
|
0
|
|
|
0
|
my $desc = $sheet->sheetname() || $sheet->data_source() || "no data_source"; |
1398
|
0
|
0
|
|
|
|
0
|
if (length($desc) > $trunclen) { $desc = substr($desc,($trunclen-3))."..." } |
|
0
|
|
|
|
|
0
|
|
1399
|
0
|
|
|
|
|
0
|
my $r = addrvis($sheet)."($desc)"; |
1400
|
0
|
|
|
|
|
0
|
$r |
1401
|
|
|
|
|
|
|
} |
1402
|
|
|
|
|
|
|
|
1403
|
|
|
|
|
|
|
sub _logmethifv { # args: just linearized INPUT items |
1404
|
0
|
0
|
|
0
|
|
0
|
return unless ${$_[0]}->{verbose}; |
|
0
|
|
|
|
|
0
|
|
1405
|
0
|
|
|
|
|
0
|
my $self = shift; |
1406
|
0
|
|
|
|
|
0
|
log_methcall $self, [@_]; |
1407
|
|
|
|
|
|
|
} |
1408
|
|
|
|
|
|
|
|
1409
|
|
|
|
|
|
|
sub _logmethretifv { # args: [INPUTS], [OUTPUTS] |
1410
|
8
|
50
|
|
8
|
|
27
|
oops if @_ > 3; # (self, items, retvals) |
1411
|
8
|
50
|
|
|
|
14
|
return unless ${$_[0]}->{verbose}; |
|
8
|
|
|
|
|
39
|
|
1412
|
0
|
|
|
|
|
0
|
goto &log_methcall |
1413
|
|
|
|
|
|
|
} |
1414
|
|
|
|
|
|
|
#----------------------------------------------- |
1415
|
|
|
|
|
|
|
|
1416
|
|
|
|
|
|
|
my $__FILE__ = __FILE__; |
1417
|
|
|
|
|
|
|
my $__PACKAGE__ = __PACKAGE__; |
1418
|
|
|
|
|
|
|
|
1419
|
|
|
|
|
|
|
sub _call_usercode($$$) { |
1420
|
6
|
|
|
6
|
|
32
|
my ($self, $code, $cxlist) = @_; |
1421
|
|
|
|
|
|
|
|
1422
|
6
|
50
|
|
|
|
29
|
if (@$cxlist) { |
1423
|
0
|
|
|
|
|
0
|
my $row = $self->crow(); |
1424
|
0
|
|
|
|
|
0
|
for ($row->[$cxlist->[0]]) { # bind $_ to the first-specified column |
1425
|
0
|
|
|
|
|
0
|
eval{ &$code(@$row[@$cxlist]) }; |
|
0
|
|
|
|
|
0
|
|
1426
|
|
|
|
|
|
|
} |
1427
|
|
|
|
|
|
|
} else { |
1428
|
6
|
|
|
|
|
13
|
eval{ $code->() }; |
|
6
|
|
|
|
|
23
|
|
1429
|
|
|
|
|
|
|
##Simplify backtraces |
1430
|
|
|
|
|
|
|
#@_ = (); |
1431
|
|
|
|
|
|
|
#goto &$code; |
1432
|
|
|
|
|
|
|
} |
1433
|
6
|
50
|
|
|
|
1426
|
if ($@) { # filter out our internal stuff to make easier to read |
1434
|
0
|
|
|
|
|
0
|
{ local $_; |
|
0
|
|
|
|
|
0
|
|
1435
|
|
|
|
|
|
|
$@ =~ s/^[^\n]* called \Kat $__FILE__ line .*?(?=\s*${__PACKAGE__}::\w*apply)/from [apply internals]/msg |
1436
|
0
|
0
|
|
|
|
0
|
unless $$self->{debug}; |
1437
|
|
|
|
|
|
|
} |
1438
|
0
|
|
|
|
|
0
|
die "$@\n"; |
1439
|
|
|
|
|
|
|
} |
1440
|
|
|
|
|
|
|
} |
1441
|
|
|
|
|
|
|
|
1442
|
|
|
|
|
|
|
# Do apply, handling COLSPEC args. |
1443
|
|
|
|
|
|
|
# If $rxlists or $rxfirst & $rxlast are undef, visit all rows. |
1444
|
|
|
|
|
|
|
sub _apply_to_rows($$$;$$$) { |
1445
|
6
|
|
|
6
|
|
28
|
my ($self, $code, $cxlist, $rxlist, $rxfirst, $rxlast) = @_; |
1446
|
6
|
|
|
|
|
21
|
my $hash = $$self; |
1447
|
6
|
|
|
|
|
42
|
my ($linenums,$rows,$num_cols,$cl) = @$hash{qw/linenums rows num_cols/}; |
1448
|
|
|
|
|
|
|
|
1449
|
6
|
50
|
|
|
|
33
|
croak $self->_methmsg("Missing or incorrect {code} argument") |
1450
|
|
|
|
|
|
|
unless ref($code) eq "CODE"; |
1451
|
6
|
|
|
|
|
30
|
foreach (@$cxlist) { |
1452
|
0
|
0
|
0
|
|
|
0
|
if ($_ < 0 || $_ >= $num_cols) { |
1453
|
0
|
|
|
|
|
0
|
croak $self->_methmsg("cx $_ is out of range") |
1454
|
|
|
|
|
|
|
} |
1455
|
|
|
|
|
|
|
} |
1456
|
|
|
|
|
|
|
|
1457
|
|
|
|
|
|
|
{ # Temp save "current_rx" from an enclosing apply |
1458
|
6
|
|
|
|
|
20
|
local $hash->{current_rx} = undef; |
|
6
|
|
|
|
|
26
|
|
1459
|
|
|
|
|
|
|
|
1460
|
|
|
|
|
|
|
# Temp update "current apply sheet" for logmsg() |
1461
|
6
|
|
|
|
|
25
|
local $_inner_apply_sheet = $self; |
1462
|
|
|
|
|
|
|
|
1463
|
6
|
100
|
|
|
|
25
|
if (defined $rxlist) { |
1464
|
5
|
|
|
|
|
24
|
foreach my $rx (@$rxlist) { |
1465
|
5
|
50
|
33
|
|
|
43
|
croak "rx $rx is out of range" |
1466
|
|
|
|
|
|
|
if $rx < 0 || $rx > $#$rows; |
1467
|
5
|
|
|
|
|
24
|
$hash->{current_rx} = $rx; |
1468
|
5
|
|
|
|
|
54
|
_call_usercode($self,$code,$cxlist); |
1469
|
|
|
|
|
|
|
} |
1470
|
|
|
|
|
|
|
} else { |
1471
|
|
|
|
|
|
|
# Do not cache $#$rows so user can call insert_rows() or delete_rows() |
1472
|
1
|
|
50
|
|
|
17
|
for (my $rx = $rxfirst // 0; |
|
|
|
33
|
|
|
|
|
|
|
|
66
|
|
|
|
|
1473
|
|
|
|
|
|
|
$rx <= $#$rows && (!defined($rxlast) || $rx <= $rxlast); |
1474
|
|
|
|
|
|
|
$rx++) |
1475
|
|
|
|
|
|
|
{ |
1476
|
1
|
|
|
|
|
4
|
$hash->{current_rx} = $rx; |
1477
|
1
|
|
|
|
|
9
|
_call_usercode($self,$code,$cxlist); |
1478
|
1
|
|
|
|
|
10
|
$rx = $hash->{current_rx}; # might have been changed by delete_rows() |
1479
|
|
|
|
|
|
|
} |
1480
|
|
|
|
|
|
|
} |
1481
|
|
|
|
|
|
|
} |
1482
|
|
|
|
|
|
|
|
1483
|
|
|
|
|
|
|
croak "After completing apply, an enclosing apply was resumed, but", |
1484
|
|
|
|
|
|
|
" current_rx=",$hash->{current_rx}," now points beyond the last row!\n" |
1485
|
6
|
50
|
33
|
|
|
28
|
if defined($hash->{current_rx}) && $hash->{current_rx} > $#$rows; |
1486
|
6
|
|
|
|
|
33
|
$self |
1487
|
|
|
|
|
|
|
}#_apply_to_rows |
1488
|
|
|
|
|
|
|
|
1489
|
|
|
|
|
|
|
# Rebuild %colx and %colx_desc, and tie any required new variables. |
1490
|
|
|
|
|
|
|
# |
1491
|
|
|
|
|
|
|
# User-defined column aliases must already be valid in %colx; |
1492
|
|
|
|
|
|
|
# all other entries are deleted and re-created. |
1493
|
|
|
|
|
|
|
# |
1494
|
|
|
|
|
|
|
# Note: the special '^', '$' and numieric cx values (if in 0..num_cols) |
1495
|
|
|
|
|
|
|
# are handled algorithmically in _specs2cxdesclist() before consulting %colx. |
1496
|
|
|
|
|
|
|
# |
1497
|
|
|
|
|
|
|
# When building %colx, conflicts are resolved using these priorities: |
1498
|
|
|
|
|
|
|
# |
1499
|
|
|
|
|
|
|
# User-defined aliases (ALWAYS valid) |
1500
|
|
|
|
|
|
|
# Titles (if unique) |
1501
|
|
|
|
|
|
|
# Trimmed titles (with leading & trailing spaces removed) |
1502
|
|
|
|
|
|
|
# Automatic aliases |
1503
|
|
|
|
|
|
|
# ABC letter-codes |
1504
|
|
|
|
|
|
|
# |
1505
|
|
|
|
|
|
|
# Warnings are issued once for each conflict. |
1506
|
|
|
|
|
|
|
sub _rebuild_colx { |
1507
|
|
|
|
|
|
|
my $self = shift; |
1508
|
|
|
|
|
|
|
my $notie = $_[0]; # true during autodetect probing |
1509
|
|
|
|
|
|
|
|
1510
|
|
|
|
|
|
|
my ($silent, $colx, $colx_desc, $useraliases, $num_cols, $title_rx, |
1511
|
|
|
|
|
|
|
$rows, $debug, $pkg2tieall) |
1512
|
|
|
|
|
|
|
= @$$self{qw/silent colx colx_desc useraliases num_cols title_rx |
1513
|
|
|
|
|
|
|
rows debug pkg2tieall/}; |
1514
|
|
|
|
|
|
|
|
1515
|
|
|
|
|
|
|
# Save user-defined Aliases before wiping %colx |
1516
|
|
|
|
|
|
|
my %useralias; |
1517
|
|
|
|
|
|
|
foreach my $alias (keys %$useraliases) { |
1518
|
|
|
|
|
|
|
my $cx = $colx->{$alias}; |
1519
|
|
|
|
|
|
|
# cx may be undef if referenced column was deleted and/or if |
1520
|
|
|
|
|
|
|
# an alias was created with {optional => TRUE} with non-matching regex. |
1521
|
|
|
|
|
|
|
#next if !defined($cx); # the referenced column was deleted |
1522
|
|
|
|
|
|
|
$useralias{$alias} = [$cx, $colx_desc->{$alias}]; |
1523
|
|
|
|
|
|
|
} |
1524
|
|
|
|
|
|
|
|
1525
|
|
|
|
|
|
|
# Now re-generate |
1526
|
|
|
|
|
|
|
%$colx = (); |
1527
|
|
|
|
|
|
|
%$colx_desc = (); |
1528
|
|
|
|
|
|
|
|
1529
|
|
|
|
|
|
|
my sub __putback($$$;$) { |
1530
|
|
|
|
|
|
|
my ($key, $cx, $desc, $nomasking) = @_; |
1531
|
|
|
|
|
|
|
if (defined (my $ocx = $colx->{$key})) { |
1532
|
|
|
|
|
|
|
if ($cx != $ocx) { |
1533
|
|
|
|
|
|
|
oops if $nomasking; # _get_usable_titles should have screen out |
1534
|
|
|
|
|
|
|
$self->_carponce("Warning: ", visq($key), " ($desc) is MASKED BY (", |
1535
|
|
|
|
|
|
|
$colx_desc->{$key}, ")") unless $silent; |
1536
|
|
|
|
|
|
|
} |
1537
|
|
|
|
|
|
|
} else { |
1538
|
|
|
|
|
|
|
oops if exists $colx->{$key}; |
1539
|
|
|
|
|
|
|
$colx->{$key} = $cx; |
1540
|
|
|
|
|
|
|
$colx_desc->{$key} = $desc; |
1541
|
|
|
|
|
|
|
} |
1542
|
|
|
|
|
|
|
} |
1543
|
|
|
|
|
|
|
|
1544
|
|
|
|
|
|
|
# Put back the user aliases |
1545
|
|
|
|
|
|
|
while (my ($alias,$aref) = each %useralias) { |
1546
|
|
|
|
|
|
|
my ($cx, $desc) = @$aref; |
1547
|
|
|
|
|
|
|
__putback($alias, $cx, $desc); |
1548
|
|
|
|
|
|
|
} |
1549
|
|
|
|
|
|
|
|
1550
|
|
|
|
|
|
|
if (defined $title_rx) { |
1551
|
|
|
|
|
|
|
# Add non-conflicting titles |
1552
|
|
|
|
|
|
|
my ($normal_titles, $unindexed_titles) = $self->_get_usable_titles; |
1553
|
|
|
|
|
|
|
foreach (@$normal_titles) { |
1554
|
|
|
|
|
|
|
my ($title, $cx) = @$_; |
1555
|
|
|
|
|
|
|
__putback($title, $cx, __fmt_cx($cx).": Title", 1); # nomasking==1 |
1556
|
|
|
|
|
|
|
} |
1557
|
|
|
|
|
|
|
# Titles with leading & trailing spaces trimmed off |
1558
|
|
|
|
|
|
|
foreach (@$normal_titles) { |
1559
|
|
|
|
|
|
|
my ($title, $cx) = @$_; |
1560
|
|
|
|
|
|
|
my $key = $title; |
1561
|
|
|
|
|
|
|
$key =~ s/\A\s+//s; $key =~ s/\s+\z//s; |
1562
|
|
|
|
|
|
|
if ($key ne $title) { |
1563
|
|
|
|
|
|
|
__putback($key, $cx, __fmt_cx($cx).": Title sans lead/trailing spaces",1); |
1564
|
|
|
|
|
|
|
} |
1565
|
|
|
|
|
|
|
} |
1566
|
|
|
|
|
|
|
# Automatic aliases |
1567
|
|
|
|
|
|
|
# N.B. These come from all titles, not just "normal" ones |
1568
|
|
|
|
|
|
|
foreach (@$normal_titles, @$unindexed_titles) { |
1569
|
|
|
|
|
|
|
my ($title, $cx) = @$_; |
1570
|
|
|
|
|
|
|
my $ident = title2ident($title); |
1571
|
|
|
|
|
|
|
__putback($ident, $cx, __fmt_cx($cx).": Automatic alias for title"); |
1572
|
|
|
|
|
|
|
} |
1573
|
|
|
|
|
|
|
} |
1574
|
|
|
|
|
|
|
my %abc; |
1575
|
|
|
|
|
|
|
foreach my $cx ( 0..$num_cols-1 ) { |
1576
|
|
|
|
|
|
|
my $ABC = cx2let($cx); |
1577
|
|
|
|
|
|
|
__putback($ABC, $cx, "cx $cx: Standard letter-code"); |
1578
|
|
|
|
|
|
|
} |
1579
|
|
|
|
|
|
|
|
1580
|
|
|
|
|
|
|
unless ($notie) { |
1581
|
|
|
|
|
|
|
# export and tie newly-defined magic variables to packages which want that. |
1582
|
|
|
|
|
|
|
if (my @pkglist = grep {defined $pkg2tieall->{$_}} keys %$pkg2tieall) { |
1583
|
|
|
|
|
|
|
my @idents = $self->_all_valid_idents; |
1584
|
|
|
|
|
|
|
foreach my $pkg (@pkglist) { |
1585
|
|
|
|
|
|
|
$self->_tie_col_vars($pkg, $pkg2tieall->{$pkg}, @idents); |
1586
|
|
|
|
|
|
|
} |
1587
|
|
|
|
|
|
|
} |
1588
|
|
|
|
|
|
|
} |
1589
|
|
|
|
|
|
|
} # _rebuild_colx |
1590
|
|
|
|
|
|
|
|
1591
|
|
|
|
|
|
|
# Move and/or delete column positions. The argument is a ref to an array |
1592
|
|
|
|
|
|
|
# containing the old column indicies of current (i.e. surviving) columns, |
1593
|
|
|
|
|
|
|
# or undefs for new columns which did not exist previously. |
1594
|
|
|
|
|
|
|
sub _adjust_colx { |
1595
|
0
|
|
|
0
|
|
0
|
my ($self, $old_colxs) = @_; |
1596
|
|
|
|
|
|
|
my ($colx, $colx_desc, $num_cols, $useraliases, $debug) |
1597
|
0
|
|
|
|
|
0
|
= @$$self{qw/colx colx_desc num_cols useraliases debug/}; |
1598
|
0
|
0
|
|
|
|
0
|
oops unless @$old_colxs == $num_cols; |
1599
|
0
|
|
|
|
|
0
|
my %old2new; |
1600
|
0
|
|
|
|
|
0
|
foreach my $new_cx (0..$#$old_colxs) { |
1601
|
0
|
|
|
|
|
0
|
my $old_cx = $old_colxs->[$new_cx]; |
1602
|
0
|
0
|
|
|
|
0
|
$old2new{$old_cx} = $new_cx if defined $old_cx; |
1603
|
|
|
|
|
|
|
} |
1604
|
|
|
|
|
|
|
# User-defined aliases are for arbitrary columns, so fix them manually |
1605
|
0
|
|
|
|
|
0
|
foreach my $alias (keys %$useraliases) { |
1606
|
0
|
|
|
|
|
0
|
my $cx = $colx->{$alias}; |
1607
|
0
|
0
|
|
|
|
0
|
next unless defined $cx; # e.g. non-unique title; see _rebuild_colx() |
1608
|
0
|
0
|
|
|
|
0
|
if (defined (my $new_cx = $old2new{$cx})) { |
1609
|
0
|
0
|
|
|
|
0
|
warn ">adjusting colx{$alias} : $colx->{$alias} -> $new_cx\n" if $debug; |
1610
|
0
|
|
|
|
|
0
|
$colx->{$alias} = $new_cx; |
1611
|
|
|
|
|
|
|
} else { |
1612
|
0
|
0
|
|
|
|
0
|
warn ">deleting colx{$alias} (was $colx->{$alias})\n" if $debug; |
1613
|
0
|
|
|
|
|
0
|
delete $colx->{$alias}; |
1614
|
0
|
|
|
|
|
0
|
delete $colx_desc->{$alias}; |
1615
|
0
|
|
|
|
|
0
|
delete $useraliases->{$alias}; |
1616
|
|
|
|
|
|
|
} |
1617
|
|
|
|
|
|
|
} |
1618
|
|
|
|
|
|
|
# Everything else is derived from actual titles |
1619
|
0
|
|
|
|
|
0
|
$self->_rebuild_colx(); |
1620
|
|
|
|
|
|
|
} |
1621
|
|
|
|
|
|
|
|
1622
|
|
|
|
|
|
|
# Translate list of COLSPECs to a list of [cx,desc]. |
1623
|
|
|
|
|
|
|
# Regexes may match multiple columns. |
1624
|
|
|
|
|
|
|
# THROWS if a spec does not indicate any existing column. |
1625
|
|
|
|
|
|
|
sub _specs2cxdesclist { |
1626
|
0
|
|
|
0
|
|
0
|
my $self = shift; |
1627
|
0
|
|
|
|
|
0
|
my ($colx, $colx_desc, $num_cols) = @$$self{qw/colx colx_desc num_cols/}; |
1628
|
0
|
|
|
|
|
0
|
my @results; |
1629
|
0
|
|
|
|
|
0
|
foreach my $spec (@_) { |
1630
|
0
|
0
|
|
|
|
0
|
croak "Column specifier is undef!" unless defined $spec; |
1631
|
0
|
0
|
|
|
|
0
|
if ($spec eq '^') { |
1632
|
0
|
|
|
|
|
0
|
push @results, [0, "Special '^' specifier for first col"]; |
1633
|
|
|
|
|
|
|
next |
1634
|
0
|
|
|
|
|
0
|
} |
1635
|
0
|
0
|
|
|
|
0
|
if ($spec eq '$') { |
1636
|
0
|
|
|
|
|
0
|
push @results, [$num_cols-1, "Special '\$' specifier for last col"]; |
1637
|
|
|
|
|
|
|
next |
1638
|
0
|
|
|
|
|
0
|
} |
1639
|
0
|
0
|
0
|
|
|
0
|
if (($spec =~ /^[1-9]\d*$/ || $spec eq "0") |
|
|
|
0
|
|
|
|
|
1640
|
|
|
|
|
|
|
&& $spec <= $num_cols) { # allow one-past-end |
1641
|
0
|
|
|
|
|
0
|
push @results, [$spec, "Numeric column-index"]; |
1642
|
|
|
|
|
|
|
next |
1643
|
0
|
|
|
|
|
0
|
} |
1644
|
0
|
0
|
|
|
|
0
|
if (defined (my $cx = $colx->{$spec})) { |
1645
|
0
|
|
|
|
|
0
|
push @results, [$cx, $colx_desc->{$spec}]; |
1646
|
|
|
|
|
|
|
next |
1647
|
0
|
|
|
|
|
0
|
} |
1648
|
0
|
0
|
|
|
|
0
|
if (ref($spec) eq 'Regexp') { |
1649
|
0
|
|
|
|
|
0
|
my ($title_rx, $rows) = @$$self{qw/title_rx rows/}; |
1650
|
0
|
0
|
|
|
|
0
|
croak "Can not use regex: No title-row is defined!\n" |
1651
|
|
|
|
|
|
|
unless defined $title_rx; |
1652
|
0
|
|
0
|
|
|
0
|
my $title_row = $rows->[$title_rx] // oops; |
1653
|
0
|
|
|
|
|
0
|
my $matched; |
1654
|
0
|
|
|
|
|
0
|
for my $cx (0..$#$title_row) { |
1655
|
0
|
|
|
|
|
0
|
my $title = $title_row->[$cx]; |
1656
|
|
|
|
|
|
|
# Note: We can't use /s here! The regex compiler has already |
1657
|
|
|
|
|
|
|
# encapsulated /s or lack thereof in the compiled regex |
1658
|
0
|
0
|
|
|
|
0
|
if ($title =~ /$spec/) { |
1659
|
0
|
|
|
|
|
0
|
push @results, [$cx, "cx $cx: regex matched title '$title'"]; |
1660
|
0
|
|
|
|
|
0
|
$matched++; |
1661
|
|
|
|
|
|
|
} |
1662
|
|
|
|
|
|
|
} |
1663
|
0
|
0
|
|
|
|
0
|
if (! $matched) { |
1664
|
0
|
|
|
|
|
0
|
croak "\n--- Title Row (rx $title_rx) ---\n", |
1665
|
|
|
|
|
|
|
vis($title_row),"\n-----------------\n", |
1666
|
|
|
|
|
|
|
"Regex $spec\n", |
1667
|
|
|
|
|
|
|
"does not match any of the titles (see above) in '", |
1668
|
|
|
|
|
|
|
fmt_sheet($self),"'\n" |
1669
|
|
|
|
|
|
|
# N.B. check for "does not match" in alias() |
1670
|
|
|
|
|
|
|
} |
1671
|
|
|
|
|
|
|
next |
1672
|
0
|
|
|
|
|
0
|
} |
1673
|
0
|
|
|
|
|
0
|
croak "Invalid column specifier '${spec}'\nnum_cols=$num_cols. Valid keys are:\n", |
1674
|
|
|
|
|
|
|
$self->_fmt_colx; |
1675
|
|
|
|
|
|
|
} |
1676
|
0
|
0
|
|
|
|
0
|
oops unless wantarray; |
1677
|
|
|
|
|
|
|
@results |
1678
|
0
|
|
|
|
|
0
|
}#_specs2cxdesclist |
1679
|
|
|
|
|
|
|
sub _spec2cx { # return $cx or ($cx, $desc); throws if spec is invalid |
1680
|
0
|
|
|
0
|
|
0
|
my ($self, $spec) = @_; |
1681
|
0
|
|
|
|
|
0
|
my @list = $self->_specs2cxdesclist($spec); |
1682
|
0
|
0
|
|
|
|
0
|
if (@list > 1) { |
1683
|
|
|
|
|
|
|
croak ivis("Regexpr $spec matches multiple titles:\n "), |
1684
|
0
|
|
|
|
|
0
|
join("\n ",map{ vis $_->[1] } @list), "\n"; |
|
0
|
|
|
|
|
0
|
|
1685
|
|
|
|
|
|
|
} |
1686
|
0
|
|
|
|
|
0
|
__first_ifnot_wantarray( @{$list[0]} ) # cx or (cx,desc) |
|
0
|
|
|
|
|
0
|
|
1687
|
|
|
|
|
|
|
} |
1688
|
|
|
|
|
|
|
|
1689
|
|
|
|
|
|
|
sub _colspec2cx { |
1690
|
0
|
|
|
0
|
|
0
|
my ($self, $colspec) = @_; |
1691
|
0
|
0
|
|
|
|
0
|
croak "COLSPEC may not be a regex" if ref($colspec) eq 'Regexp'; |
1692
|
0
|
|
|
|
|
0
|
goto &_spec2cx |
1693
|
|
|
|
|
|
|
} |
1694
|
|
|
|
|
|
|
|
1695
|
|
|
|
|
|
|
# The user-callable API |
1696
|
|
|
|
|
|
|
# THROWS if a spec does not indicate any existing column. |
1697
|
|
|
|
|
|
|
# Can return multiple results due to multple args and/or Regexp multimatch |
1698
|
|
|
|
|
|
|
# In scalar context returns the first result. |
1699
|
|
|
|
|
|
|
sub spectocx(@) { # the user-callable API |
1700
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
1701
|
0
|
|
|
|
|
0
|
my @list = $self->_specs2cxdesclist(@_); |
1702
|
0
|
|
|
|
|
0
|
__first_ifnot_wantarray( map{ $_->[0] } @list ) |
|
0
|
|
|
|
|
0
|
|
1703
|
|
|
|
|
|
|
} |
1704
|
|
|
|
|
|
|
|
1705
|
|
|
|
|
|
|
# Translate a possibly-relative column specification which |
1706
|
|
|
|
|
|
|
# indicate 1 off the end. |
1707
|
|
|
|
|
|
|
# |
1708
|
|
|
|
|
|
|
# The specification may be |
1709
|
|
|
|
|
|
|
# >something (the column after 'something') |
1710
|
|
|
|
|
|
|
# or |
1711
|
|
|
|
|
|
|
# an absolute column indicator (cx or ABC), possibly 1 off the end |
1712
|
|
|
|
|
|
|
# or |
1713
|
|
|
|
|
|
|
# refer to an existing column |
1714
|
|
|
|
|
|
|
# |
1715
|
|
|
|
|
|
|
sub _relspec2cx { |
1716
|
0
|
|
|
0
|
|
0
|
my ($self, $spec) = @_; |
1717
|
0
|
|
|
|
|
0
|
my $colx = $$self->{colx}; |
1718
|
0
|
0
|
|
|
|
0
|
if ($spec =~ /^>(.*)/) { |
1719
|
0
|
|
|
|
|
0
|
my $cx = $self->_colspec2cx($1); # croaks if not an existing column |
1720
|
0
|
|
|
|
|
0
|
return $cx + 1 |
1721
|
|
|
|
|
|
|
} |
1722
|
0
|
|
|
|
|
0
|
$self->_colspec2cx($spec); # croaks if not an existing column |
1723
|
|
|
|
|
|
|
} |
1724
|
|
|
|
|
|
|
|
1725
|
|
|
|
|
|
|
sub alias(@) { |
1726
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
1727
|
0
|
0
|
|
|
|
0
|
my $opthash = ref($_[0]) eq 'HASH' ? shift() : {}; |
1728
|
0
|
0
|
|
|
|
0
|
if ($opthash) { |
1729
|
0
|
|
|
|
|
0
|
__validate_opthash($opthash, [qw(optional)], |
1730
|
|
|
|
|
|
|
desc => "alias option"); |
1731
|
|
|
|
|
|
|
} |
1732
|
0
|
0
|
|
|
|
0
|
croak "'alias' expects an even number of arguments\n" |
1733
|
|
|
|
|
|
|
unless scalar(@_ % 2)==0; |
1734
|
|
|
|
|
|
|
|
1735
|
|
|
|
|
|
|
my ($colx, $colx_desc, $num_cols, $useraliases, $rows, $silent, $debug) |
1736
|
0
|
|
|
|
|
0
|
= @$$self{qw/colx colx_desc num_cols useraliases rows silent debug/}; |
1737
|
|
|
|
|
|
|
|
1738
|
0
|
|
|
|
|
0
|
my ($file, $lno) = __fn_ln_methname(); |
1739
|
|
|
|
|
|
|
|
1740
|
0
|
|
|
|
|
0
|
my @cxlist; |
1741
|
0
|
|
|
|
|
0
|
while (@_) { |
1742
|
0
|
|
|
|
|
0
|
my $ident = _validate_ident( shift @_ ); |
1743
|
0
|
|
|
|
|
0
|
my $spec = shift @_; |
1744
|
|
|
|
|
|
|
|
1745
|
0
|
0
|
|
|
|
0
|
if (my $wheredefined = $useraliases->{$ident}) { |
1746
|
0
|
|
|
|
|
0
|
croak "'$ident' is already a user-defined alias", |
1747
|
|
|
|
|
|
|
" for cx ", scalar($self->_spec2cx($ident)), |
1748
|
|
|
|
|
|
|
" defined at ", $wheredefined, "\n" |
1749
|
|
|
|
|
|
|
#" (for cx ", scalar($self->_spec2cx($ident)), ")" |
1750
|
|
|
|
|
|
|
} |
1751
|
|
|
|
|
|
|
croak "'$ident' is already a user-defined alias (for cx ", |
1752
|
|
|
|
|
|
|
scalar($self->_spec2cx($ident)), ")" |
1753
|
0
|
0
|
|
|
|
0
|
if $useraliases->{$ident}; |
1754
|
|
|
|
|
|
|
|
1755
|
0
|
|
|
|
|
0
|
my $cx = eval{ $self->_spec2cx($spec) }; |
|
0
|
|
|
|
|
0
|
|
1756
|
0
|
0
|
|
|
|
0
|
unless(defined $cx) { |
1757
|
0
|
0
|
|
|
|
0
|
oops unless $@; |
1758
|
0
|
0
|
0
|
|
|
0
|
croak $@ unless $opthash->{optional} && $@ =~ /does not match/is; |
1759
|
|
|
|
|
|
|
# Always throw on other errors, e.g. regex matches more than one title |
1760
|
|
|
|
|
|
|
}; |
1761
|
|
|
|
|
|
|
|
1762
|
|
|
|
|
|
|
# Log each pair individually |
1763
|
0
|
|
|
|
|
0
|
$self->_logmethifv($opthash, \", $ident => ",\__fmt_colspec_cx($spec,$cx)); |
1764
|
|
|
|
|
|
|
|
1765
|
0
|
|
|
|
|
0
|
$colx->{$ident} = $cx; |
1766
|
0
|
|
|
|
|
0
|
$colx_desc->{$ident} = "alias for ".__fmt_cx($cx)." (".quotekey($spec).")"; |
1767
|
0
|
|
|
|
|
0
|
$useraliases->{$ident} = "${file}:$lno"; |
1768
|
0
|
|
|
|
|
0
|
push @cxlist, $cx; |
1769
|
|
|
|
|
|
|
} |
1770
|
0
|
|
|
|
|
0
|
$self->_rebuild_colx(); |
1771
|
|
|
|
|
|
|
|
1772
|
0
|
|
|
|
|
0
|
__first_ifnot_wantarray( @cxlist ) |
1773
|
|
|
|
|
|
|
}#alias |
1774
|
|
|
|
|
|
|
|
1775
|
|
|
|
|
|
|
sub unalias(@) { |
1776
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
1777
|
0
|
0
|
|
|
|
0
|
croak __methname, " does not accept an {OPTIONS} hash\n" |
1778
|
|
|
|
|
|
|
if ref($_[0]) eq 'HASH'; |
1779
|
|
|
|
|
|
|
|
1780
|
|
|
|
|
|
|
my ($colx, $colx_desc, $useraliases) |
1781
|
0
|
|
|
|
|
0
|
= @$$self{qw/colx colx_desc useraliases/}; |
1782
|
|
|
|
|
|
|
|
1783
|
0
|
|
|
|
|
0
|
foreach (@_) { |
1784
|
0
|
|
0
|
|
|
0
|
delete $useraliases->{$_} // croak "unalias: '$_' is not a column alias\n"; |
1785
|
0
|
|
|
|
|
0
|
$self->_logmethifv(\" Removing alias $_ => ", \$colx_desc->{$_}); |
1786
|
0
|
|
0
|
|
|
0
|
delete $colx->{$_} // oops; |
1787
|
0
|
|
0
|
|
|
0
|
delete $colx_desc->{$_} // oops; |
1788
|
|
|
|
|
|
|
} |
1789
|
0
|
|
|
|
|
0
|
$self->_rebuild_colx(); |
1790
|
0
|
|
|
|
|
0
|
$self |
1791
|
|
|
|
|
|
|
} |
1792
|
|
|
|
|
|
|
|
1793
|
|
|
|
|
|
|
# title_rx: Get/set the title row index |
1794
|
|
|
|
|
|
|
# |
1795
|
|
|
|
|
|
|
# $rx = title_rx ; # Retrieve |
1796
|
|
|
|
|
|
|
# |
1797
|
|
|
|
|
|
|
# title_rx undef # Set to no titles |
1798
|
|
|
|
|
|
|
# |
1799
|
|
|
|
|
|
|
# title_rx ROWINDEX # Set to specified rx |
1800
|
|
|
|
|
|
|
# |
1801
|
|
|
|
|
|
|
# title_rx 'auto' # Auto-detect the title row; an exception is thrown |
1802
|
|
|
|
|
|
|
# # if a plausible row is not found. |
1803
|
|
|
|
|
|
|
# |
1804
|
|
|
|
|
|
|
# {OPTARGS} may contain |
1805
|
|
|
|
|
|
|
# verbose, silent, debug (temporarily override the object's settings) |
1806
|
|
|
|
|
|
|
# |
1807
|
|
|
|
|
|
|
# Auto-detect options: |
1808
|
|
|
|
|
|
|
# required => [COLSPEC, ...] # required titles |
1809
|
|
|
|
|
|
|
# min_rx, max_rx => NUM # range of rows which may contain the row |
1810
|
|
|
|
|
|
|
# first_cx => NUM # first column ix which must contain a valid title |
1811
|
|
|
|
|
|
|
# last_cx => NUM # last column ix which must contain a valid title |
1812
|
|
|
|
|
|
|
# |
1813
|
|
|
|
|
|
|
# Note: This is called internally by read_spreadsheet(), passing 'auto' |
1814
|
|
|
|
|
|
|
# by default. Therefore users need not call this method explicitly |
1815
|
|
|
|
|
|
|
# except to change title row or if read_spreadsheet was not used at all. |
1816
|
|
|
|
|
|
|
# |
1817
|
|
|
|
|
|
|
sub title_rx(;$@) { |
1818
|
8
|
|
|
8
|
1
|
47
|
my ($self, $opthash_arg) = &__selfmust_opthash; |
1819
|
8
|
|
|
|
|
43
|
my $opthash = { %$opthash_arg }; # make copy so we can modify it |
1820
|
8
|
|
|
|
|
21
|
my @orig_args = @_; |
1821
|
|
|
|
|
|
|
|
1822
|
8
|
|
|
|
|
33
|
my $saved_stdopts = $self->_set_stdopts($opthash); |
1823
|
8
|
|
|
8
|
|
95
|
scope_guard{ $self->_restore_stdopts($saved_stdopts) }; |
|
8
|
|
|
|
|
37
|
|
1824
|
8
|
|
|
|
|
56
|
__validate_opthash( $opthash, |
1825
|
|
|
|
|
|
|
[qw(required min_rx max_rx first_cx last_cx)], |
1826
|
|
|
|
|
|
|
desc => "autodetect option", |
1827
|
|
|
|
|
|
|
undef_ok => [] ); |
1828
|
8
|
|
|
|
|
22
|
my $rx = -999; |
1829
|
8
|
100
|
|
|
|
35
|
if (@_ == 0) { |
1830
|
|
|
|
|
|
|
# A return value was requested |
1831
|
2
|
50
|
|
|
|
17
|
croak '{OPTARGS} passed to title_rx with no operator (get request?)' |
1832
|
|
|
|
|
|
|
if %$opthash; |
1833
|
2
|
|
|
|
|
17
|
$rx = $$self->{title_rx}; |
1834
|
2
|
|
|
|
|
29
|
$self->_logmethretifv([], [$rx]); |
1835
|
|
|
|
|
|
|
} else { |
1836
|
|
|
|
|
|
|
# N.B. undef arg means there are no titles |
1837
|
6
|
|
|
|
|
10
|
$rx = shift; |
1838
|
6
|
100
|
|
|
|
12
|
my $notie; $notie = shift() if u($_[0]) eq "_notie"; # during auto-detect probes |
|
6
|
|
|
|
|
39
|
|
1839
|
6
|
50
|
|
|
|
44
|
croak "Extraneous argument(s) to title_rx: ".avis(@_) if @_; |
1840
|
|
|
|
|
|
|
|
1841
|
6
|
50
|
|
|
|
25
|
if (defined $rx) { |
1842
|
6
|
100
|
|
|
|
48
|
if ($rx eq 'auto') { |
|
|
50
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
1843
|
1
|
|
|
|
|
5
|
$rx = $self->_autodetect_title_rx($opthash); |
1844
|
|
|
|
|
|
|
} |
1845
|
|
|
|
|
|
|
elsif ($rx !~ /^\d+$/) { |
1846
|
0
|
|
|
|
|
0
|
croak "Invalid title_rx argument: ", visq($rx); |
1847
|
|
|
|
|
|
|
} |
1848
|
5
|
|
|
|
|
24
|
elsif ($rx > $#{ $$self->{rows} }) { |
1849
|
0
|
|
|
|
|
0
|
croak "Rx $rx is beyond the end of the data", visq($rx); |
1850
|
|
|
|
|
|
|
} |
1851
|
|
|
|
|
|
|
} |
1852
|
6
|
|
|
|
|
18
|
$$self->{title_rx} = $rx; |
1853
|
6
|
|
|
|
|
27
|
$self->_logmethretifv([$opthash_arg, @orig_args], [$rx]); |
1854
|
6
|
|
|
|
|
36
|
$self->_rebuild_colx($notie); |
1855
|
|
|
|
|
|
|
} |
1856
|
8
|
|
|
|
|
43
|
$rx; |
1857
|
|
|
|
|
|
|
}#title_rx |
1858
|
|
|
|
|
|
|
|
1859
|
|
|
|
|
|
|
sub _autodetect_title_rx { |
1860
|
1
|
|
|
1
|
|
2
|
my ($self, $opthash) = @_; |
1861
|
|
|
|
|
|
|
|
1862
|
|
|
|
|
|
|
my ($title_rx, $rows, $colx, $num_cols, $verbose, $debug) = |
1863
|
1
|
|
|
|
|
5
|
@$$self{qw(title_rx rows colx num_cols verbose debug)}; |
1864
|
|
|
|
|
|
|
|
1865
|
1
|
50
|
|
|
|
4
|
if ($#$rows == -1) { |
1866
|
0
|
|
|
|
|
0
|
return undef; # completely empty |
1867
|
|
|
|
|
|
|
} |
1868
|
|
|
|
|
|
|
|
1869
|
|
|
|
|
|
|
# Filter out titles which can not be used as a COLSPEC |
1870
|
|
|
|
|
|
|
my @required_specs = $opthash->{required} |
1871
|
1
|
50
|
|
|
|
12
|
? to_array($opthash->{required}) : (); |
1872
|
1
|
50
|
|
|
|
5
|
croak "undef value in {required}" if grep{! defined} @required_specs; |
|
0
|
|
|
|
|
0
|
|
1873
|
1
|
|
|
|
|
2
|
@required_specs = grep{ !__unindexed_title($_, $num_cols) } @required_specs; |
|
0
|
|
|
|
|
0
|
|
1874
|
|
|
|
|
|
|
|
1875
|
1
|
|
50
|
|
|
8
|
my $min_rx = __validate_nonnegi($opthash->{min_rx}//0, "min_rx"); |
1876
|
1
|
|
33
|
|
|
11
|
my $max_rx = __validate_nonnegi($opthash->{max_rx}//$min_rx+3, "max_rx"); |
1877
|
1
|
50
|
|
|
|
4
|
$max_rx = $#$rows if $max_rx > $#$rows; |
1878
|
|
|
|
|
|
|
|
1879
|
1
|
|
50
|
|
|
14
|
my $first_cx = __validate_nonnegi($opthash->{first_cx}//0, "first_cx"); |
1880
|
1
|
|
50
|
|
|
5
|
my $last_cx = __validate_nonnegi($opthash->{last_cx}//INT_MAX, "last_cx"); |
1881
|
1
|
50
|
|
|
|
17
|
$last_cx = $num_cols-1 if $last_cx >= $num_cols; |
1882
|
|
|
|
|
|
|
|
1883
|
1
|
|
|
|
|
3
|
my @nd_reasons; |
1884
|
1
|
50
|
|
|
|
4
|
if ($min_rx > $#$rows) { |
|
|
50
|
|
|
|
|
|
1885
|
0
|
|
|
|
|
0
|
push @nd_reasons, "min_rx ($min_rx) is out of range"; |
1886
|
|
|
|
|
|
|
} |
1887
|
|
|
|
|
|
|
elsif ($min_rx > $max_rx) { |
1888
|
0
|
|
|
|
|
0
|
push @nd_reasons, |
1889
|
|
|
|
|
|
|
"min_rx ($min_rx) is greater than max_rx ($max_rx)" |
1890
|
|
|
|
|
|
|
} |
1891
|
1
|
50
|
|
|
|
4
|
if ($first_cx >= $num_cols) { |
|
|
50
|
|
|
|
|
|
1892
|
0
|
|
|
|
|
0
|
push @nd_reasons, "first_cx ($first_cx) is out of range" |
1893
|
|
|
|
|
|
|
} |
1894
|
|
|
|
|
|
|
elsif ($first_cx > $last_cx) { |
1895
|
0
|
|
|
|
|
0
|
push @nd_reasons, |
1896
|
|
|
|
|
|
|
"first_cx ($first_cx) is less than last_cx ($last_cx)" |
1897
|
|
|
|
|
|
|
} |
1898
|
|
|
|
|
|
|
|
1899
|
1
|
|
|
|
|
2
|
my $detected; |
1900
|
1
|
50
|
|
|
|
4
|
unless (@nd_reasons) { |
1901
|
|
|
|
|
|
|
# no logging during trial and error (except with debug) |
1902
|
1
|
|
|
|
|
4
|
local $$self->{verbose} = $debug; |
1903
|
1
|
|
|
|
|
4
|
local $$self->{silent} = !$debug; |
1904
|
1
|
|
|
|
|
3
|
RX: for my $rx ($min_rx .. $max_rx) { |
1905
|
1
|
50
|
33
|
|
|
4
|
warn "# ",$nd_reasons[-1],"\n" if $debug && @nd_reasons; |
1906
|
1
|
50
|
|
|
|
4
|
warn ivis '#autodetect: Trying RX $rx ...\n' if $debug; |
1907
|
|
|
|
|
|
|
|
1908
|
|
|
|
|
|
|
# Make $rx the title_rx so __specs2cxdesclist() can be used |
1909
|
|
|
|
|
|
|
# e.g. to handle regex COLSPECS. Pass special option to not tie |
1910
|
|
|
|
|
|
|
# user variables yet. |
1911
|
1
|
|
|
|
|
6
|
$self->title_rx($rx, "_notie"); |
1912
|
1
|
50
|
|
|
|
7
|
oops unless $rx == $$self->{title_rx}; |
1913
|
|
|
|
|
|
|
|
1914
|
1
|
|
|
|
|
5
|
my $row = $rows->[$rx]; |
1915
|
1
|
|
|
|
|
5
|
for my $cx ($first_cx .. $last_cx) { |
1916
|
2
|
50
|
|
|
|
5
|
if ($row->[$cx] eq "") { |
1917
|
0
|
|
|
|
|
0
|
push @nd_reasons, "rx $rx: col ".__fmt_cx($cx)." is empty"; |
1918
|
0
|
|
|
|
|
0
|
next RX; |
1919
|
|
|
|
|
|
|
} |
1920
|
|
|
|
|
|
|
} |
1921
|
1
|
|
|
|
|
3
|
foreach my $spec (@required_specs) { |
1922
|
0
|
|
|
|
|
0
|
my @list; # A regex might match multiple titles |
1923
|
0
|
|
|
|
|
0
|
eval { @list = $self->_specs2cxdesclist($spec) }; |
|
0
|
|
|
|
|
0
|
|
1924
|
0
|
0
|
|
|
|
0
|
warn ivis ' found $spec in @list\n' if $debug; |
1925
|
0
|
0
|
|
|
|
0
|
if (@list == 0) { |
1926
|
0
|
|
|
|
|
0
|
push @nd_reasons, ivis 'rx $rx: Required column \'$spec\' not found'; |
1927
|
|
|
|
|
|
|
next RX |
1928
|
0
|
|
|
|
|
0
|
} |
1929
|
0
|
0
|
|
|
|
0
|
my @shortlist = grep{ $_->[0] >= $first_cx && $_->[0] <= $last_cx } |
|
0
|
|
|
|
|
0
|
|
1930
|
|
|
|
|
|
|
@list; |
1931
|
0
|
0
|
|
|
|
0
|
if (@shortlist == 0) { |
1932
|
0
|
|
|
|
|
0
|
push @nd_reasons, ivis 'rx $rx: Matched \'$spec\' but in unacceptable cx '.avisl(map{$_->[0]} @list); |
|
0
|
|
|
|
|
0
|
|
1933
|
|
|
|
|
|
|
next RX |
1934
|
0
|
|
|
|
|
0
|
} |
1935
|
0
|
0
|
|
|
|
0
|
if (! grep{ $_->[1] =~ /title/i } @shortlist) { |
|
0
|
|
|
|
|
0
|
|
1936
|
|
|
|
|
|
|
### ??? Can this actually happen ??? |
1937
|
0
|
|
|
|
|
0
|
push @nd_reasons, ivis 'rx $rx: \'$spec\' resolved to something other than a title: '.__fmt_pairs(@shortlist); |
1938
|
0
|
|
|
|
|
0
|
next RX; |
1939
|
|
|
|
|
|
|
} |
1940
|
0
|
0
|
|
|
|
0
|
warn ivis ' <<cx is within $first_cx .. $last_cx>>\n' if $debug; |
1941
|
|
|
|
|
|
|
} |
1942
|
1
|
|
|
|
|
2
|
$detected = $rx; |
1943
|
|
|
|
|
|
|
last |
1944
|
1
|
|
|
|
|
2
|
} |
1945
|
1
|
|
|
|
|
5
|
$$self->{title_rx} = undef; # will re-do below |
1946
|
|
|
|
|
|
|
} |
1947
|
1
|
50
|
|
|
|
4
|
if (defined $detected) { |
1948
|
1
|
50
|
|
|
|
3
|
if ($verbose) { # should be $debug ?? |
1949
|
0
|
|
|
|
|
0
|
my ($fn, $lno, $methname) = __fn_ln_methname(); |
1950
|
0
|
|
|
|
|
0
|
print STDERR "[Auto-detected title_rx = $detected at ${fn}:$lno]\n"; |
1951
|
|
|
|
|
|
|
} |
1952
|
1
|
|
|
|
|
4
|
local $$self->{cmd_nesting} = $$self->{cmd_nesting} + 1; |
1953
|
1
|
|
|
|
|
3
|
local $$self->{verbose} = 0; # suppress normal logging |
1954
|
1
|
|
|
|
|
13
|
$self->title_rx($detected); # shows collision warnings unless {silent} |
1955
|
1
|
|
|
|
|
7
|
return $detected; |
1956
|
|
|
|
|
|
|
} else { |
1957
|
0
|
0
|
|
|
|
0
|
if (@nd_reasons == 0) { |
1958
|
0
|
|
|
|
|
0
|
push @nd_reasons, ivis '(BUG?) No rows checked! num_cols=$num_cols rows=$$self->{rows}'.dvis '\n##($min_rx $max_rx $first_cx $last_cx)' ; |
1959
|
|
|
|
|
|
|
} |
1960
|
0
|
|
|
|
|
0
|
croak("In ",qsh(fmt_sheet($self))," ...\n", |
1961
|
|
|
|
|
|
|
" Auto-detect of title_rx with options ",vis($opthash), |
1962
|
|
|
|
|
|
|
dvis ' @required_specs\n', |
1963
|
|
|
|
|
|
|
" failed because:\n ", join("\n ",@nd_reasons), |
1964
|
|
|
|
|
|
|
"\n" |
1965
|
|
|
|
|
|
|
); |
1966
|
|
|
|
|
|
|
} |
1967
|
|
|
|
|
|
|
} |
1968
|
|
|
|
|
|
|
|
1969
|
|
|
|
|
|
|
sub first_data_rx(;$) { |
1970
|
0
|
|
|
0
|
1
|
0
|
my $self = &__self; |
1971
|
0
|
|
|
|
|
0
|
my $first_data_rx = $$self->{first_data_rx}; |
1972
|
0
|
0
|
|
|
|
0
|
if (@_ == 0) { # 'get' request |
1973
|
0
|
|
|
|
|
0
|
$self->_logmethretifv([], [$first_data_rx]); |
1974
|
0
|
|
|
|
|
0
|
return $first_data_rx; |
1975
|
|
|
|
|
|
|
} |
1976
|
0
|
|
|
|
|
0
|
my $rx = __validate_nonnegi_or_undef( shift() ); |
1977
|
0
|
|
|
|
|
0
|
$self->_logmethretifv([$rx]); |
1978
|
|
|
|
|
|
|
# Okay if this points to one past the end |
1979
|
0
|
0
|
|
|
|
0
|
$self->_check_rx($rx, 1) if defined $rx; # one_past_end_ok=1 |
1980
|
0
|
|
|
|
|
0
|
$$self->{first_data_rx} = $rx; |
1981
|
0
|
|
|
|
|
0
|
$self |
1982
|
|
|
|
|
|
|
} |
1983
|
|
|
|
|
|
|
sub last_data_rx(;$) { |
1984
|
0
|
|
|
0
|
1
|
0
|
my $self = &__self; |
1985
|
0
|
|
|
|
|
0
|
my $last_data_rx = $$self->{last_data_rx}; |
1986
|
0
|
0
|
|
|
|
0
|
if (@_ == 0) { # 'get' request |
1987
|
0
|
|
|
|
|
0
|
$self->_logmethretifv([], [$last_data_rx]); |
1988
|
0
|
|
|
|
|
0
|
return $last_data_rx; |
1989
|
|
|
|
|
|
|
} |
1990
|
0
|
|
|
|
|
0
|
my $rx = __validate_nonnegi_or_undef( shift() ); |
1991
|
0
|
|
|
|
|
0
|
$self->_logmethretifv([$rx]); |
1992
|
0
|
0
|
|
|
|
0
|
if (defined $rx) { |
1993
|
0
|
|
|
|
|
0
|
$self->_check_rx($rx, 1); # one_past_end_ok=1 |
1994
|
|
|
|
|
|
|
confess "last_data_rx must be >= first_data_rx" |
1995
|
0
|
0
|
0
|
|
|
0
|
unless $rx >= ($$self->{first_data_rx}//0); |
1996
|
|
|
|
|
|
|
} |
1997
|
0
|
|
|
|
|
0
|
$$self->{last_data_rx} = $rx; |
1998
|
0
|
|
|
|
|
0
|
$self |
1999
|
|
|
|
|
|
|
} |
2000
|
|
|
|
|
|
|
|
2001
|
|
|
|
|
|
|
# move_cols ">COLSPEC",source cols... |
2002
|
|
|
|
|
|
|
# move_cols "absolute-position",source cols... |
2003
|
|
|
|
|
|
|
sub move_cols($@) { |
2004
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2005
|
0
|
|
|
|
|
0
|
my ($posn, @sources) = @_; |
2006
|
|
|
|
|
|
|
|
2007
|
0
|
|
|
|
|
0
|
my ($num_cols, $rows) = @$$self{qw/num_cols rows/}; |
2008
|
|
|
|
|
|
|
|
2009
|
0
|
|
|
|
|
0
|
my $to_cx = $self->_relspec2cx($posn); |
2010
|
|
|
|
|
|
|
|
2011
|
0
|
|
|
|
|
0
|
my @source_cxs = map { scalar $self->_spec2cx($_) } @sources; |
|
0
|
|
|
|
|
0
|
|
2012
|
0
|
|
|
|
|
0
|
my @source_cxs_before = grep { $_ < $to_cx } @source_cxs; |
|
0
|
|
|
|
|
0
|
|
2013
|
0
|
|
|
|
|
0
|
my $insert_offset = $to_cx - scalar(@source_cxs_before); |
2014
|
0
|
|
|
|
|
0
|
my @rsorted_source_cxs = sort { $b <=> $a } @source_cxs; |
|
0
|
|
|
|
|
0
|
|
2015
|
|
|
|
|
|
|
|
2016
|
|
|
|
|
|
|
$self->_logmethifv(\__fmt_colspec_cx($posn,$to_cx), \" <-- ", |
2017
|
0
|
|
|
|
|
0
|
\join(" ",map{"$source_cxs[$_]\[$_\]"} 0..$#source_cxs)); |
|
0
|
|
|
|
|
0
|
|
2018
|
|
|
|
|
|
|
|
2019
|
0
|
0
|
|
|
|
0
|
croak "move destination is too far to the right\n" |
2020
|
|
|
|
|
|
|
if $to_cx + @sources - @source_cxs_before > $num_cols; |
2021
|
|
|
|
|
|
|
|
2022
|
0
|
|
|
|
|
0
|
my @old_cxs = (0..$num_cols-1); |
2023
|
|
|
|
|
|
|
|
2024
|
0
|
|
|
|
|
0
|
foreach my $row (@$rows, \@old_cxs) { |
2025
|
0
|
|
|
|
|
0
|
my @moving_cells = @$row[@source_cxs]; # save |
2026
|
0
|
|
|
|
|
0
|
splice @$row, $_, 1 foreach (@rsorted_source_cxs); # delete |
2027
|
0
|
|
|
|
|
0
|
splice @$row, $insert_offset, 0, @moving_cells; # put back |
2028
|
|
|
|
|
|
|
}; |
2029
|
|
|
|
|
|
|
|
2030
|
0
|
|
|
|
|
0
|
$self->_adjust_colx(\@old_cxs); |
2031
|
0
|
|
|
|
|
0
|
$self |
2032
|
|
|
|
|
|
|
} |
2033
|
0
|
|
|
0
|
1
|
0
|
sub move_col($$) { goto &move_cols; } |
2034
|
|
|
|
|
|
|
|
2035
|
|
|
|
|
|
|
# insert_cols ">COLSPEC",new titles (or ""s or undefs if no title row) |
2036
|
|
|
|
|
|
|
# insert_cols "absolute-position",... |
2037
|
|
|
|
|
|
|
# RETURNS: The new colum indicies, or in scalar context the first cx |
2038
|
|
|
|
|
|
|
sub insert_cols($@) { |
2039
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2040
|
0
|
|
|
|
|
0
|
my ($posn, @new_titles) = @_; |
2041
|
0
|
|
|
|
|
0
|
my ($num_cols, $rows, $title_rx) = @$$self{qw/num_cols rows title_rx/}; |
2042
|
|
|
|
|
|
|
|
2043
|
0
|
|
|
|
|
0
|
my $to_cx = $self->_relspec2cx($posn); |
2044
|
|
|
|
|
|
|
|
2045
|
0
|
|
|
|
|
0
|
$self->_logmethifv(\__fmt_colspec_cx($posn,$to_cx), \" <-- ", \avis(@new_titles)); |
2046
|
|
|
|
|
|
|
|
2047
|
0
|
|
0
|
|
|
0
|
@new_titles = map { $_ // "" } @new_titles; # change undef to "" |
|
0
|
|
|
|
|
0
|
|
2048
|
0
|
|
|
0
|
|
0
|
my $have_new_titles = first { $_ ne "" } @new_titles; |
|
0
|
|
|
|
|
0
|
|
2049
|
0
|
0
|
0
|
|
|
0
|
if (!defined($title_rx) && $have_new_titles) { |
2050
|
0
|
|
|
|
|
0
|
croak "insert_cols: Can not specify non-undef titles if title_rx is not defined\n" |
2051
|
|
|
|
|
|
|
} |
2052
|
0
|
|
|
|
|
0
|
my $num_insert_cols = @new_titles; |
2053
|
|
|
|
|
|
|
|
2054
|
0
|
|
|
|
|
0
|
foreach my $row (@$rows) { |
2055
|
0
|
0
|
0
|
|
|
0
|
if (defined $title_rx && $row == $rows->[$title_rx]) { |
2056
|
0
|
|
|
|
|
0
|
splice @$row, $to_cx, 0, @new_titles; |
2057
|
|
|
|
|
|
|
} else { |
2058
|
0
|
|
|
|
|
0
|
splice @$row, $to_cx, 0, (("") x $num_insert_cols); |
2059
|
|
|
|
|
|
|
} |
2060
|
|
|
|
|
|
|
} |
2061
|
0
|
|
|
|
|
0
|
$$self->{num_cols} += $num_insert_cols; |
2062
|
|
|
|
|
|
|
|
2063
|
0
|
|
|
|
|
0
|
$self->_adjust_colx( |
2064
|
|
|
|
|
|
|
[ 0..$to_cx-1, ((undef) x $num_insert_cols), $to_cx..$num_cols-1 ] |
2065
|
|
|
|
|
|
|
); |
2066
|
|
|
|
|
|
|
|
2067
|
0
|
|
|
|
|
0
|
__first_ifnot_wantarray( $to_cx .. $to_cx+$num_insert_cols-1 ) |
2068
|
|
|
|
|
|
|
} |
2069
|
0
|
|
|
0
|
1
|
0
|
sub insert_col($$) { goto &insert_cols } |
2070
|
|
|
|
|
|
|
|
2071
|
|
|
|
|
|
|
# sort_rows {compare function} |
2072
|
|
|
|
|
|
|
# sort_rows {compare function} $first_rx, $last_rx |
2073
|
|
|
|
|
|
|
sub sort_rows(&) { |
2074
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2075
|
0
|
0
|
|
|
|
0
|
croak "bad args" unless @_ == 1; |
2076
|
0
|
|
|
|
|
0
|
my ($cmpfunc, $first_rx, $last_rx) = @_; |
2077
|
|
|
|
|
|
|
|
2078
|
|
|
|
|
|
|
my ($rows, $linenums, $title_rx, $first_data_rx, $last_data_rx) |
2079
|
0
|
|
|
|
|
0
|
= @$$self{qw/rows linenums title_rx first_data_rx last_data_rx/}; |
2080
|
|
|
|
|
|
|
|
2081
|
0
|
0
|
0
|
|
|
0
|
$first_rx //= $first_data_rx |
|
|
|
0
|
|
|
|
|
2082
|
|
|
|
|
|
|
// (defined($title_rx) ? $title_rx+1 : 0); |
2083
|
0
|
|
0
|
|
|
0
|
$last_rx //= $last_data_rx // $#$rows; |
|
|
|
0
|
|
|
|
|
2084
|
|
|
|
|
|
|
|
2085
|
0
|
|
|
|
|
0
|
$self->_logmethifv(\"(sorting rx ${first_rx}..${last_rx})"); |
2086
|
|
|
|
|
|
|
|
2087
|
0
|
0
|
|
|
|
0
|
oops unless defined($first_rx); |
2088
|
0
|
0
|
|
|
|
0
|
oops unless defined($last_rx); |
2089
|
0
|
|
|
|
|
0
|
my $pkg = caller; |
2090
|
|
|
|
|
|
|
my @indicies = sort { |
2091
|
0
|
|
|
|
|
0
|
my @row_indicies = ($a, $b); |
|
0
|
|
|
|
|
0
|
|
2092
|
3
|
|
|
3
|
|
31
|
no strict 'refs'; |
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
23186
|
|
2093
|
0
|
|
|
|
|
0
|
local ${ "$pkg\::a" } = $rows->[$a]; # actual row objects |
|
0
|
|
|
|
|
0
|
|
2094
|
0
|
|
|
|
|
0
|
local ${ "$pkg\::b" } = $rows->[$b]; |
|
0
|
|
|
|
|
0
|
|
2095
|
0
|
|
|
|
|
0
|
$cmpfunc->(@row_indicies) |
2096
|
|
|
|
|
|
|
} ($first_rx..$last_rx); |
2097
|
|
|
|
|
|
|
|
2098
|
0
|
|
|
|
|
0
|
@$rows[$first_rx..$#$rows] = @$rows[@indicies]; |
2099
|
0
|
|
|
|
|
0
|
@$linenums[$first_rx..$#$rows] = @$linenums[@indicies]; |
2100
|
|
|
|
|
|
|
|
2101
|
0
|
|
|
|
|
0
|
__validate_not_scalar_context(0..$first_rx-1, @indicies, $last_rx+1..$#$rows) |
2102
|
|
|
|
|
|
|
} |
2103
|
|
|
|
|
|
|
|
2104
|
|
|
|
|
|
|
sub delete_cols(@) { |
2105
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2106
|
0
|
|
|
|
|
0
|
my (@cols) = @_; |
2107
|
0
|
|
|
|
|
0
|
my ($num_cols, $rows) = @$$self{qw/num_cols rows/}; |
2108
|
|
|
|
|
|
|
|
2109
|
0
|
|
|
|
|
0
|
my @cxlist = $self->_colspecs_to_cxs_ckunique(\@cols); |
2110
|
|
|
|
|
|
|
|
2111
|
0
|
|
|
|
|
0
|
my @reverse_cxs = sort { $b <=> $a } @cxlist; |
|
0
|
|
|
|
|
0
|
|
2112
|
|
|
|
|
|
|
|
2113
|
0
|
|
|
|
|
0
|
$self->_logmethifv(reverse @reverse_cxs); |
2114
|
0
|
|
|
|
|
0
|
my @old_cxs = (0..$num_cols-1); |
2115
|
0
|
|
|
|
|
0
|
for my $row (@$rows, \@old_cxs) { |
2116
|
0
|
|
|
|
|
0
|
foreach my $cx (@reverse_cxs) { |
2117
|
0
|
0
|
|
|
|
0
|
oops if $cx > $#$row; |
2118
|
0
|
|
|
|
|
0
|
splice @$row, $cx, 1, (); |
2119
|
|
|
|
|
|
|
} |
2120
|
|
|
|
|
|
|
} |
2121
|
0
|
|
|
|
|
0
|
$$self->{num_cols} -= @reverse_cxs; |
2122
|
0
|
|
|
|
|
0
|
$self->_adjust_colx(\@old_cxs); |
2123
|
0
|
|
|
|
|
0
|
$self |
2124
|
|
|
|
|
|
|
} |
2125
|
0
|
|
|
0
|
1
|
0
|
sub delete_col($) { goto &delete_cols; } |
2126
|
|
|
|
|
|
|
|
2127
|
|
|
|
|
|
|
# Logic which forces verbose on when debug is on, etc. |
2128
|
|
|
|
|
|
|
# Used by new() and options() |
2129
|
|
|
|
|
|
|
sub _set_verbose_debug_silent(@) { |
2130
|
6
|
|
|
6
|
|
11
|
my $self = shift; |
2131
|
6
|
|
|
|
|
42
|
foreach (pairs @_) { |
2132
|
12
|
|
|
|
|
29
|
my ($key, $val) = @$_; |
2133
|
12
|
|
|
|
|
25
|
my $oldval = $$self->{$key}; |
2134
|
|
|
|
|
|
|
next |
2135
|
12
|
50
|
|
|
|
37
|
unless !!$oldval != !!$val; |
2136
|
0
|
0
|
|
|
|
0
|
if ($key eq "silent") { |
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
2137
|
0
|
|
|
|
|
0
|
$$self->{$key} = $val; |
2138
|
|
|
|
|
|
|
} |
2139
|
|
|
|
|
|
|
elsif ($key eq "verbose") { |
2140
|
0
|
0
|
|
|
|
0
|
if ($val) { |
2141
|
0
|
|
|
|
|
0
|
$$self->{saved_silent} = $$self->{silent}; |
2142
|
0
|
|
|
|
|
0
|
$$self->{silent} = 0; #?? might still want to suppress warnings |
2143
|
|
|
|
|
|
|
} else { |
2144
|
0
|
|
|
|
|
0
|
$$self->{silent} = delete $$self->{saved_silent}; |
2145
|
|
|
|
|
|
|
} |
2146
|
|
|
|
|
|
|
} |
2147
|
|
|
|
|
|
|
elsif ($key eq "debug") { |
2148
|
0
|
0
|
|
|
|
0
|
if ($val) { |
2149
|
0
|
|
|
|
|
0
|
$$self->{saved_verbose} = $$self->{verbose}; |
2150
|
0
|
|
|
|
|
0
|
$$self->{saved_silent} = $$self->{silent}; |
2151
|
0
|
|
|
|
|
0
|
$$self->{silent} = 0; |
2152
|
0
|
|
|
|
|
0
|
$$self->{verbose} = "forced by {debug}"; |
2153
|
|
|
|
|
|
|
} else { |
2154
|
0
|
|
|
|
|
0
|
$$self->{verbose} = delete $$self->{saved_verbose}; |
2155
|
0
|
|
|
|
|
0
|
$$self->{silent} = delete $$self->{saved_silent}; |
2156
|
|
|
|
|
|
|
} |
2157
|
|
|
|
|
|
|
} |
2158
|
0
|
|
|
|
|
0
|
else { confess "options: Unknown option key '$key'\n"; } |
2159
|
0
|
|
|
|
|
0
|
$$self->{$key} = $val; |
2160
|
|
|
|
|
|
|
} |
2161
|
|
|
|
|
|
|
} |
2162
|
|
|
|
|
|
|
|
2163
|
|
|
|
|
|
|
# Get or set option(s). |
2164
|
|
|
|
|
|
|
# New settings may be in an {OPTIONS} hash and/or linear args. |
2165
|
|
|
|
|
|
|
# Always returns the old options (key => value pairs). |
2166
|
|
|
|
|
|
|
sub options(@) { |
2167
|
0
|
|
|
0
|
1
|
0
|
my $self = &__self; # auto-create sheet if necessary |
2168
|
0
|
0
|
|
|
|
0
|
my @old = map{ exists($$self->{$_}) ? ($_ => $$self->{$_}) : () } |
|
0
|
|
|
|
|
0
|
|
2169
|
|
|
|
|
|
|
qw/verbose debug silent/; |
2170
|
|
|
|
|
|
|
|
2171
|
0
|
|
|
|
|
0
|
my %eff_args; |
2172
|
0
|
0
|
|
|
|
0
|
if (@_ == 0) { |
2173
|
0
|
0
|
|
|
|
0
|
croak "(list) returned but called in scalar or void context" |
2174
|
|
|
|
|
|
|
unless wantarray; |
2175
|
|
|
|
|
|
|
} else { |
2176
|
0
|
|
|
|
|
0
|
my $opthash = &__opthash; # shift off 1st arg iff it is a hashref |
2177
|
0
|
|
|
|
|
0
|
%eff_args = (%$opthash, &__validate_pairs); |
2178
|
0
|
|
|
|
|
0
|
$self->_set_verbose_debug_silent(%eff_args); |
2179
|
|
|
|
|
|
|
} |
2180
|
0
|
|
|
|
|
0
|
$self->_logmethretifv([\__fmt_pairlist(%eff_args)], [\hvis(@old)]); |
2181
|
0
|
|
|
|
|
0
|
@old; |
2182
|
|
|
|
|
|
|
} |
2183
|
|
|
|
|
|
|
|
2184
|
|
|
|
|
|
|
sub _colspecs_to_cxs_ckunique { |
2185
|
0
|
0
|
|
0
|
|
0
|
my ($self, $colspecs) = @_; oops unless @_==2; |
|
0
|
|
|
|
|
0
|
|
2186
|
0
|
|
|
|
|
0
|
my @cxlist; |
2187
|
|
|
|
|
|
|
my %seen; |
2188
|
0
|
|
|
|
|
0
|
foreach (@$colspecs) { |
2189
|
0
|
|
|
|
|
0
|
my $cx = $self->_spec2cx($_); # auto-detects title_rx if needed |
2190
|
0
|
0
|
|
|
|
0
|
if ($seen{$cx}) { |
2191
|
0
|
|
|
|
|
0
|
croak "cx $cx is specified by multiple COLSPECs: ", vis($_)," and ",vis($seen{$cx}),"\n"; |
2192
|
|
|
|
|
|
|
} |
2193
|
0
|
|
|
|
|
0
|
$seen{ $cx } = $_; |
2194
|
0
|
|
|
|
|
0
|
push @cxlist, $cx; |
2195
|
|
|
|
|
|
|
} |
2196
|
|
|
|
|
|
|
@cxlist |
2197
|
0
|
|
|
|
|
0
|
} |
2198
|
|
|
|
|
|
|
|
2199
|
|
|
|
|
|
|
sub only_cols(@) { |
2200
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2201
|
0
|
|
|
|
|
0
|
my @cols = @_; |
2202
|
0
|
|
|
|
|
0
|
my $rows = $self->rows; |
2203
|
|
|
|
|
|
|
|
2204
|
|
|
|
|
|
|
# Replace each row with just the surviving columns, in the order specified |
2205
|
0
|
|
|
|
|
0
|
my @cxlist = $self->_colspecs_to_cxs_ckunique(\@cols); |
2206
|
0
|
|
|
|
|
0
|
for my $row (@$rows) { |
2207
|
0
|
|
|
|
|
0
|
@$row = map{ $row->[$_] } @cxlist; |
|
0
|
|
|
|
|
0
|
|
2208
|
|
|
|
|
|
|
} |
2209
|
0
|
|
|
|
|
0
|
$$self->{num_cols} = scalar(@cxlist); |
2210
|
0
|
|
|
|
|
0
|
$self->_adjust_colx(\@cxlist); |
2211
|
0
|
|
|
|
|
0
|
$self |
2212
|
|
|
|
|
|
|
} |
2213
|
|
|
|
|
|
|
|
2214
|
|
|
|
|
|
|
# obj->join_cols separator_or_coderef, colspecs... |
2215
|
|
|
|
|
|
|
# If coderef: |
2216
|
|
|
|
|
|
|
# $_ is bound to the first-named column, and is the destination |
2217
|
|
|
|
|
|
|
# @_ is bound to all named columns, in the order named. |
2218
|
|
|
|
|
|
|
sub join_cols(&@) { |
2219
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2220
|
0
|
|
|
|
|
0
|
my ($separator, @sources) = @_; |
2221
|
0
|
|
|
|
|
0
|
my $hash = $$self; |
2222
|
|
|
|
|
|
|
|
2223
|
0
|
|
|
|
|
0
|
my ($num_cols, $rows) = @$hash{qw/num_cols rows/}; |
2224
|
|
|
|
|
|
|
|
2225
|
0
|
|
|
|
|
0
|
my @source_cxs = map { scalar $self->_spec2cx($_) } @sources; |
|
0
|
|
|
|
|
0
|
|
2226
|
|
|
|
|
|
|
$self->_logmethifv(\"'$separator' ", |
2227
|
0
|
|
|
|
|
0
|
\join(" ",map{"$source_cxs[$_]\[$_\]"} 0..$#source_cxs)); |
|
0
|
|
|
|
|
0
|
|
2228
|
|
|
|
|
|
|
|
2229
|
0
|
|
|
|
|
0
|
my $saved_v = $hash->{verbose}; $hash->{verbose} = 0; |
|
0
|
|
|
|
|
0
|
|
2230
|
|
|
|
|
|
|
|
2231
|
|
|
|
|
|
|
# Merge the content into the first column. N.B. EXCLUDES title row. |
2232
|
|
|
|
|
|
|
my $code = ref($separator) eq 'CODE' |
2233
|
|
|
|
|
|
|
? $separator |
2234
|
0
|
0
|
|
0
|
|
0
|
: sub{ $_ = join $separator, @_ } ; |
|
0
|
|
|
|
|
0
|
|
2235
|
|
|
|
|
|
|
|
2236
|
|
|
|
|
|
|
# Note first/last_data_rx are ignored |
2237
|
0
|
|
0
|
|
|
0
|
{ my $first_rx = ($hash->{title_rx} // -1)+1; |
|
0
|
|
|
|
|
0
|
|
2238
|
0
|
|
|
|
|
0
|
_apply_to_rows($self, $code, \@source_cxs, undef, $first_rx, undef); |
2239
|
|
|
|
|
|
|
} |
2240
|
|
|
|
|
|
|
|
2241
|
|
|
|
|
|
|
# Delete the other columns |
2242
|
0
|
|
|
|
|
0
|
$self->delete_cols(@source_cxs[1..$#source_cxs]); |
2243
|
|
|
|
|
|
|
|
2244
|
0
|
|
|
|
|
0
|
$$self->{verbose} = $saved_v; |
2245
|
0
|
|
|
|
|
0
|
$self |
2246
|
|
|
|
|
|
|
} |
2247
|
0
|
|
|
0
|
1
|
0
|
sub join_cols_sep($@) { goto &join_cols } # to match the functional API |
2248
|
|
|
|
|
|
|
|
2249
|
|
|
|
|
|
|
sub rename_cols(@) { |
2250
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2251
|
0
|
0
|
|
|
|
0
|
croak "rename_cols expects an even number of arguments\n" |
2252
|
|
|
|
|
|
|
unless scalar(@_ % 2)==0; |
2253
|
|
|
|
|
|
|
|
2254
|
0
|
|
|
|
|
0
|
my ($num_cols, $rows, $title_rx) = @$$self{qw/num_cols rows title_rx/}; |
2255
|
|
|
|
|
|
|
|
2256
|
0
|
0
|
|
|
|
0
|
croak "rename_cols: No title_rx is defined!\n" |
2257
|
|
|
|
|
|
|
unless defined $title_rx; |
2258
|
|
|
|
|
|
|
|
2259
|
0
|
|
|
|
|
0
|
my $title_row = $rows->[$title_rx]; |
2260
|
|
|
|
|
|
|
|
2261
|
0
|
|
|
|
|
0
|
while (@_) { |
2262
|
0
|
|
|
|
|
0
|
my $old_title = shift @_; |
2263
|
0
|
|
|
|
|
0
|
my $new_title = shift @_; |
2264
|
0
|
|
|
|
|
0
|
my $cx = $self->_spec2cx($old_title); |
2265
|
0
|
|
|
|
|
0
|
$self->_logmethifv($old_title, \" -> ", $new_title, \" [cx $cx]"); |
2266
|
0
|
0
|
|
|
|
0
|
croak "rename_cols: Column $old_title is too large\n" |
2267
|
|
|
|
|
|
|
if $cx > $#$title_row; # it must have been an absolute form |
2268
|
0
|
|
|
|
|
0
|
$title_row->[$cx] = $new_title; |
2269
|
|
|
|
|
|
|
|
2270
|
|
|
|
|
|
|
# N.B. aliases remain pointing to the same columns regardless of names |
2271
|
|
|
|
|
|
|
} |
2272
|
0
|
|
|
|
|
0
|
$self->_rebuild_colx(); |
2273
|
0
|
|
|
|
|
0
|
$self |
2274
|
|
|
|
|
|
|
} |
2275
|
|
|
|
|
|
|
|
2276
|
|
|
|
|
|
|
# apply {code}, colspec* |
2277
|
|
|
|
|
|
|
# @_ are bound to the columns in the order specified (if any) |
2278
|
|
|
|
|
|
|
# $_ is bound to the first such column |
2279
|
|
|
|
|
|
|
# Only visit rows bounded by first_data_rx and/or last_data_rx, |
2280
|
|
|
|
|
|
|
# starting with title_rx+1 if a title row is defined. |
2281
|
|
|
|
|
|
|
sub apply(&;@) { |
2282
|
1
|
|
|
1
|
1
|
6
|
my $self = &__selfmust; |
2283
|
1
|
|
|
|
|
8
|
my ($code, @cols) = @_; |
2284
|
1
|
|
|
|
|
5
|
my $hash = $$self; |
2285
|
1
|
|
|
|
|
8
|
my @cxs = map { scalar $self->_spec2cx($_) } @cols; |
|
0
|
|
|
|
|
0
|
|
2286
|
|
|
|
|
|
|
|
2287
|
1
|
|
50
|
|
|
30
|
my $first_rx = max(($hash->{title_rx} // -1)+1, $hash->{first_data_rx}//0); |
|
|
|
50
|
|
|
|
|
2288
|
|
|
|
|
|
|
|
2289
|
1
|
|
|
|
|
7
|
@_ = ($self, $code, \@cxs, undef, $first_rx, $hash->{last_data_rx}); |
2290
|
1
|
|
|
|
|
12
|
goto &_apply_to_rows |
2291
|
|
|
|
|
|
|
} |
2292
|
|
|
|
|
|
|
|
2293
|
|
|
|
|
|
|
# apply_all {code}, colspec* |
2294
|
|
|
|
|
|
|
# Like apply, but ALL rows are visited, inluding the title row if any |
2295
|
|
|
|
|
|
|
sub apply_all(&;@) { |
2296
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2297
|
0
|
|
|
|
|
0
|
my ($code, @cols) = @_; |
2298
|
0
|
|
|
|
|
0
|
my $hash = $$self; |
2299
|
0
|
|
|
|
|
0
|
my @cxs = map { scalar $self->_spec2cx($_) } @cols; |
|
0
|
|
|
|
|
0
|
|
2300
|
0
|
|
|
|
|
0
|
log_methcall $self, [\"rx 0..",$#{$hash->{rows}}, |
2301
|
|
|
|
|
|
|
@cxs > 0 ? \(" cxs=".avis(@cxs)) : ()] |
2302
|
0
|
0
|
|
|
|
0
|
if $$self->{verbose}; |
|
|
0
|
|
|
|
|
|
2303
|
0
|
|
|
|
|
0
|
@_ = ($self, $code, \@cxs); |
2304
|
0
|
|
|
|
|
0
|
goto &_apply_to_rows |
2305
|
|
|
|
|
|
|
} |
2306
|
|
|
|
|
|
|
|
2307
|
|
|
|
|
|
|
sub __arrify_checknotempty($) { |
2308
|
5
|
|
|
5
|
|
23
|
local $_ = shift; |
2309
|
5
|
100
|
|
|
|
48
|
my $result = ref($_) eq 'ARRAY' ? $_ : [ $_ ]; |
2310
|
|
|
|
|
|
|
croak "Invalid argument ",vis($_)," (expecting [array ref] or single value)\n" |
2311
|
5
|
50
|
33
|
|
|
66
|
unless @$result > 0 && !grep{ref($_) || $_ eq ""} @$result; |
|
5
|
50
|
|
|
|
80
|
|
2312
|
5
|
|
|
|
|
20
|
$result |
2313
|
|
|
|
|
|
|
} |
2314
|
|
|
|
|
|
|
|
2315
|
|
|
|
|
|
|
# apply_torx {code} rx, colspec* |
2316
|
|
|
|
|
|
|
# apply_torx {code} [rx list], colspec* |
2317
|
|
|
|
|
|
|
# Only the specified row(s) are visited |
2318
|
|
|
|
|
|
|
# first/last_data_rx are ignored. |
2319
|
|
|
|
|
|
|
sub apply_torx(&$;@) { |
2320
|
5
|
|
|
5
|
1
|
565
|
my $self = &__selfmust; |
2321
|
5
|
|
|
|
|
31
|
my ($code, $rxlist_arg, @cols) = @_; |
2322
|
5
|
50
|
|
|
|
47
|
croak "Missing rx (or [list of rx]) argument\n" unless defined $rxlist_arg; |
2323
|
5
|
|
|
|
|
49
|
my $rxlist = __arrify_checknotempty($rxlist_arg); |
2324
|
5
|
|
|
|
|
21
|
my @cxs = map { scalar $self->_spec2cx($_) } @cols; |
|
0
|
|
|
|
|
0
|
|
2325
|
|
|
|
|
|
|
log_methcall $self, [\vis($rxlist_arg), |
2326
|
|
|
|
|
|
|
@cxs > 0 ? \(" cxs=".avis(@cxs)) : ()] |
2327
|
5
|
0
|
|
|
|
45
|
if $$self->{verbose}; |
|
|
50
|
|
|
|
|
|
2328
|
5
|
|
|
|
|
34
|
@_ = ($self, $code, \@cxs, $rxlist); |
2329
|
5
|
|
|
|
|
62
|
goto &_apply_to_rows |
2330
|
|
|
|
|
|
|
} |
2331
|
|
|
|
|
|
|
|
2332
|
|
|
|
|
|
|
# apply_exceptrx {code} [rx list], colspec* |
2333
|
|
|
|
|
|
|
# All rows EXCEPT the specified rows are visited |
2334
|
|
|
|
|
|
|
sub apply_exceptrx(&$;@) { |
2335
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2336
|
0
|
|
|
|
|
0
|
my ($code, $exrxlist_arg, @cols) = @_; |
2337
|
0
|
0
|
|
|
|
0
|
croak "Missing rx (or [list of rx]) argument\n" unless defined $exrxlist_arg; |
2338
|
0
|
|
|
|
|
0
|
my $exrxlist = __arrify_checknotempty($exrxlist_arg); |
2339
|
0
|
|
|
|
|
0
|
my @cxs = map { scalar $self->_spec2cx($_) } @cols; |
|
0
|
|
|
|
|
0
|
|
2340
|
|
|
|
|
|
|
log_methcall $self, [\vis($exrxlist_arg), |
2341
|
|
|
|
|
|
|
@cxs > 0 ? \(" cxs=".avis(@cxs)) : ()] |
2342
|
0
|
0
|
|
|
|
0
|
if $$self->{verbose}; |
|
|
0
|
|
|
|
|
|
2343
|
0
|
|
|
|
|
0
|
my $hash = $$self; |
2344
|
0
|
|
|
|
|
0
|
my $max_rx = $#{ $hash->{rows} }; |
|
0
|
|
|
|
|
0
|
|
2345
|
0
|
|
|
|
|
0
|
foreach (@$exrxlist) { |
2346
|
0
|
0
|
0
|
|
|
0
|
croak "rx $_ is out of range\n" if $_ < 0 || $_ > $max_rx; |
2347
|
|
|
|
|
|
|
} |
2348
|
0
|
|
|
|
|
0
|
my %exrxlist = map{ $_ => 1 } @$exrxlist; |
|
0
|
|
|
|
|
0
|
|
2349
|
0
|
|
|
|
|
0
|
my $rxlist = [ grep{ ! exists $exrxlist{$_} } 0..$max_rx ]; |
|
0
|
|
|
|
|
0
|
|
2350
|
0
|
|
|
|
|
0
|
@_ = ($self, $code, \@cxs, $rxlist); |
2351
|
0
|
|
|
|
|
0
|
goto &_apply_to_rows |
2352
|
|
|
|
|
|
|
} |
2353
|
|
|
|
|
|
|
|
2354
|
|
|
|
|
|
|
# split_col {code} oldcol, newcol_start_position, new titles... |
2355
|
|
|
|
|
|
|
# {code} is called for each row with $_ bound to <oldcol> |
2356
|
|
|
|
|
|
|
# and @_ bound to the new column(s). |
2357
|
|
|
|
|
|
|
# The old column is left as-is (not deleted). |
2358
|
|
|
|
|
|
|
sub split_col(&$$$@) { |
2359
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2360
|
0
|
|
|
|
|
0
|
my ($code, $oldcol_posn, $newcols_posn, @new_titles) = @_; |
2361
|
|
|
|
|
|
|
|
2362
|
0
|
|
|
|
|
0
|
my $num_insert_cols = @new_titles; |
2363
|
0
|
|
|
|
|
0
|
my $old_cx = $self->_spec2cx($oldcol_posn); |
2364
|
0
|
|
|
|
|
0
|
my $newcols_first_cx = $self->_relspec2cx($newcols_posn); |
2365
|
|
|
|
|
|
|
|
2366
|
|
|
|
|
|
|
log_methcall $self, [\"... $oldcol_posn\[$old_cx] -> [$newcols_first_cx]", |
2367
|
|
|
|
|
|
|
avis(@new_titles)] |
2368
|
0
|
0
|
|
|
|
0
|
if $$self->{verbose}; |
2369
|
|
|
|
|
|
|
|
2370
|
0
|
|
|
|
|
0
|
local $$self->{verbose} = 0; |
2371
|
|
|
|
|
|
|
|
2372
|
0
|
|
|
|
|
0
|
$self->insert_cols($newcols_first_cx, @new_titles); |
2373
|
|
|
|
|
|
|
|
2374
|
0
|
0
|
|
|
|
0
|
$old_cx += $num_insert_cols if $old_cx >= $newcols_first_cx; |
2375
|
|
|
|
|
|
|
|
2376
|
0
|
|
|
|
|
0
|
$self->apply($code, |
2377
|
|
|
|
|
|
|
$old_cx, $newcols_first_cx..$newcols_first_cx+$num_insert_cols-1); |
2378
|
|
|
|
|
|
|
|
2379
|
0
|
|
|
|
|
0
|
$self |
2380
|
|
|
|
|
|
|
} |
2381
|
|
|
|
|
|
|
|
2382
|
|
|
|
|
|
|
sub reverse_cols() { |
2383
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2384
|
0
|
|
|
|
|
0
|
my ($rows, $num_cols) = @$$self{qw/rows num_cols/}; |
2385
|
0
|
0
|
|
|
|
0
|
log_methcall $self, [] if $$self->{verbose}; |
2386
|
0
|
|
|
|
|
0
|
for my $row (@$rows) { |
2387
|
0
|
|
|
|
|
0
|
@$row = reverse @$row; |
2388
|
|
|
|
|
|
|
} |
2389
|
0
|
|
|
|
|
0
|
$self->_adjust_colx([reverse 0..$num_cols-1]); |
2390
|
0
|
|
|
|
|
0
|
$self |
2391
|
|
|
|
|
|
|
} |
2392
|
|
|
|
|
|
|
|
2393
|
|
|
|
|
|
|
sub transpose() { |
2394
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2395
|
0
|
0
|
|
|
|
0
|
log_methcall $self, [] if $$self->{verbose}; |
2396
|
|
|
|
|
|
|
|
2397
|
0
|
|
|
|
|
0
|
my ($rows, $old_num_cols, $linenums) = @$$self{qw/rows num_cols linenums/}; |
2398
|
|
|
|
|
|
|
|
2399
|
0
|
|
|
|
|
0
|
$$self->{useraliases} = {}; |
2400
|
0
|
|
|
|
|
0
|
$$self->{title_rx} = undef; |
2401
|
0
|
|
|
|
|
0
|
$$self->{first_data_rx} = undef; |
2402
|
0
|
|
|
|
|
0
|
$$self->{last_data_rx} = undef; |
2403
|
|
|
|
|
|
|
|
2404
|
|
|
|
|
|
|
# Save a copy of the data |
2405
|
0
|
|
|
|
|
0
|
my @old_rows = ( map{ [ @$_ ] } @$rows ); |
|
0
|
|
|
|
|
0
|
|
2406
|
|
|
|
|
|
|
|
2407
|
|
|
|
|
|
|
# Rebuild the spreadsheet |
2408
|
0
|
|
|
|
|
0
|
@$rows = (); |
2409
|
0
|
|
|
|
|
0
|
$$self->{num_cols} = scalar @old_rows; |
2410
|
|
|
|
|
|
|
|
2411
|
0
|
|
|
|
|
0
|
for (my $ocx=0; $ocx < $old_num_cols; ++$ocx) { |
2412
|
0
|
|
|
|
|
0
|
my @nrow; |
2413
|
0
|
|
|
|
|
0
|
for my $row (@old_rows) { |
2414
|
0
|
|
0
|
|
|
0
|
push @nrow, $row->[$ocx] // ""; |
2415
|
|
|
|
|
|
|
} |
2416
|
0
|
|
|
|
|
0
|
push @$rows, \@nrow; |
2417
|
|
|
|
|
|
|
} |
2418
|
0
|
0
|
|
|
|
0
|
if ($$self->{saved_linenums}) { |
2419
|
0
|
|
|
|
|
0
|
@$linenums = @{ $$self->{saved_linenums} }; |
|
0
|
|
|
|
|
0
|
|
2420
|
0
|
|
|
|
|
0
|
delete $$self->{saved_linenums}; |
2421
|
|
|
|
|
|
|
} else { |
2422
|
0
|
|
|
|
|
0
|
$$self->{saved_linenums} = [ @$linenums ]; |
2423
|
0
|
|
|
|
|
0
|
@$linenums = ("?") x scalar @$rows; |
2424
|
|
|
|
|
|
|
} |
2425
|
0
|
|
0
|
|
|
0
|
$$self->{data_source} //= ""; |
2426
|
0
|
|
|
|
|
0
|
$$self->{data_source} .= " transposed"; |
2427
|
|
|
|
|
|
|
|
2428
|
0
|
|
|
|
|
0
|
$self->_rows_replaced; |
2429
|
0
|
|
|
|
|
0
|
$self |
2430
|
|
|
|
|
|
|
}#transpose |
2431
|
|
|
|
|
|
|
|
2432
|
|
|
|
|
|
|
# delete_rows rx ... |
2433
|
|
|
|
|
|
|
# delete_rows 'LAST' ... |
2434
|
|
|
|
|
|
|
# delete_rows '$' ... |
2435
|
|
|
|
|
|
|
sub delete_rows(@) { |
2436
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2437
|
0
|
|
|
|
|
0
|
my (@rowspecs) = @_; |
2438
|
|
|
|
|
|
|
|
2439
|
|
|
|
|
|
|
my ($rows, $linenums, $title_rx, $first_data_rx, $last_data_rx, $current_rx, $verbose) |
2440
|
0
|
|
|
|
|
0
|
= @$$self{qw/rows linenums title_rx first_data_rx last_data_rx current_rx verbose/}; |
2441
|
|
|
|
|
|
|
|
2442
|
0
|
|
|
|
|
0
|
foreach (@rowspecs) { |
2443
|
0
|
0
|
|
|
|
0
|
$_ = $#$rows if /^(?:LAST|\$)$/; |
2444
|
0
|
|
|
|
|
0
|
__validate_nonnegi($_, "rx to delete"); |
2445
|
0
|
0
|
|
|
|
0
|
croak "Invalid row index '$_'\n" unless $_ <= $#$rows; |
2446
|
|
|
|
|
|
|
} |
2447
|
0
|
|
|
|
|
0
|
my @rev_sorted_rxs = sort {$b <=> $a} @rowspecs; |
|
0
|
|
|
|
|
0
|
|
2448
|
0
|
0
|
|
|
|
0
|
log_methcall $self, [reverse @rev_sorted_rxs] if $$self->{verbose}; |
2449
|
|
|
|
|
|
|
|
2450
|
|
|
|
|
|
|
# Adjust if needed... |
2451
|
0
|
0
|
|
|
|
0
|
if (defined $title_rx) { |
2452
|
0
|
|
|
|
|
0
|
foreach (@rev_sorted_rxs) { |
2453
|
0
|
0
|
|
|
|
0
|
if ($_ < $title_rx) { --$title_rx } |
|
0
|
0
|
|
|
|
0
|
|
2454
|
|
|
|
|
|
|
elsif ($_ == $title_rx) { |
2455
|
|
|
|
|
|
|
$self->_log("Invalidating titles because rx $title_rx is being deleted\n") |
2456
|
0
|
0
|
|
|
|
0
|
if $$self->{verbose}; |
2457
|
0
|
|
|
|
|
0
|
$title_rx = undef; |
2458
|
0
|
|
|
|
|
0
|
last; |
2459
|
|
|
|
|
|
|
} |
2460
|
|
|
|
|
|
|
} |
2461
|
0
|
|
|
|
|
0
|
$$self->{title_rx} = $title_rx; |
2462
|
|
|
|
|
|
|
} |
2463
|
0
|
0
|
|
|
|
0
|
if (defined $first_data_rx) { |
2464
|
0
|
|
|
|
|
0
|
foreach (@rev_sorted_rxs) { |
2465
|
0
|
0
|
|
|
|
0
|
if ($_ <= $first_data_rx) { --$first_data_rx } |
|
0
|
|
|
|
|
0
|
|
2466
|
|
|
|
|
|
|
} |
2467
|
0
|
|
|
|
|
0
|
$$self->{first_data_rx} = $first_data_rx; |
2468
|
|
|
|
|
|
|
} |
2469
|
0
|
0
|
|
|
|
0
|
if (defined $last_data_rx) { |
2470
|
0
|
|
|
|
|
0
|
foreach (@rev_sorted_rxs) { |
2471
|
0
|
0
|
|
|
|
0
|
if ($_ <= $last_data_rx) { --$last_data_rx } |
|
0
|
|
|
|
|
0
|
|
2472
|
|
|
|
|
|
|
} |
2473
|
0
|
|
|
|
|
0
|
$$self->{last_data_rx} = $last_data_rx; |
2474
|
|
|
|
|
|
|
} |
2475
|
|
|
|
|
|
|
|
2476
|
|
|
|
|
|
|
# Back up $current_rx to account for deleted rows. |
2477
|
|
|
|
|
|
|
# $current_rx is left set to one less than the index of the "next" row if |
2478
|
|
|
|
|
|
|
# we are in an apply(). That is, current_rx will be left still pointing to |
2479
|
|
|
|
|
|
|
# the same row as before, or if that row has been deleted then the row |
2480
|
|
|
|
|
|
|
# before that (or -1 if row zero was deleted). |
2481
|
0
|
0
|
|
|
|
0
|
if (defined $current_rx) { |
2482
|
0
|
|
|
|
|
0
|
foreach (@rev_sorted_rxs) { |
2483
|
0
|
0
|
|
|
|
0
|
--$current_rx if ($_ <= $current_rx); |
2484
|
|
|
|
|
|
|
} |
2485
|
0
|
|
|
|
|
0
|
$$self->{current_rx} = $current_rx; |
2486
|
|
|
|
|
|
|
} |
2487
|
|
|
|
|
|
|
|
2488
|
|
|
|
|
|
|
#warn "### BEFORE delete_rows rx (@rev_sorted_rxs):\n", |
2489
|
|
|
|
|
|
|
# map( { " [$_]=(".join(",",@{$rows->[$_]}).")\n" } 0..$#$rows); |
2490
|
|
|
|
|
|
|
|
2491
|
0
|
|
|
|
|
0
|
for my $rx (@rev_sorted_rxs) { |
2492
|
0
|
|
|
|
|
0
|
splice @$rows, $rx, 1, (); |
2493
|
0
|
|
|
|
|
0
|
splice @$linenums, $rx, 1, (); |
2494
|
|
|
|
|
|
|
} |
2495
|
|
|
|
|
|
|
|
2496
|
|
|
|
|
|
|
#warn "### AFTER delete_rows:\n", |
2497
|
|
|
|
|
|
|
# map( { " [$_]=(".join(",",@{$rows->[$_]}).")\n" } 0..$#$rows); |
2498
|
|
|
|
|
|
|
$self |
2499
|
0
|
|
|
|
|
0
|
}#delete_rows |
2500
|
0
|
|
|
0
|
0
|
0
|
sub delete_row($) { goto &delete_rows; } |
2501
|
|
|
|
|
|
|
|
2502
|
|
|
|
|
|
|
# $firstrx = insert_rows [rx [,count]] |
2503
|
|
|
|
|
|
|
# $firstrx = insert_rows ['$'[,count]] |
2504
|
|
|
|
|
|
|
sub insert_rows(;$$) { |
2505
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2506
|
0
|
|
|
|
|
0
|
my ($rx, $count) = @_; |
2507
|
0
|
|
0
|
|
|
0
|
$rx //= 'END'; |
2508
|
0
|
|
0
|
|
|
0
|
$count //= 1; |
2509
|
|
|
|
|
|
|
|
2510
|
|
|
|
|
|
|
my ($rows, $linenums, $num_cols, $title_rx, $first_data_rx, $last_data_rx) |
2511
|
0
|
|
|
|
|
0
|
= @$$self{qw/rows linenums num_cols title_rx first_data_rx last_data_rx/}; |
2512
|
|
|
|
|
|
|
|
2513
|
0
|
0
|
|
|
|
0
|
$rx = @$rows if $rx =~ /^(?:END|\$)$/; |
2514
|
|
|
|
|
|
|
|
2515
|
|
|
|
|
|
|
log_methcall $self, [\"at rx $rx (count $count)"], [$rx] |
2516
|
0
|
0
|
|
|
|
0
|
if $$self->{verbose}; |
2517
|
|
|
|
|
|
|
|
2518
|
0
|
|
|
|
|
0
|
__validate_nonnegi($rx, "new rx"); |
2519
|
|
|
|
|
|
|
|
2520
|
0
|
0
|
0
|
|
|
0
|
if (defined($title_rx) && $rx <= $title_rx) { |
2521
|
0
|
|
|
|
|
0
|
$$self->{title_rx} = ($title_rx += $count); |
2522
|
|
|
|
|
|
|
} |
2523
|
0
|
0
|
0
|
|
|
0
|
if (defined($first_data_rx) && $rx <= $first_data_rx) { |
2524
|
0
|
|
|
|
|
0
|
$$self->{first_data_rx} = ($first_data_rx += $count); |
2525
|
|
|
|
|
|
|
} |
2526
|
0
|
0
|
0
|
|
|
0
|
if (defined($last_data_rx) && $rx <= $last_data_rx) { |
2527
|
0
|
|
|
|
|
0
|
$$self->{last_data_rx} = ($last_data_rx += $count); |
2528
|
|
|
|
|
|
|
} |
2529
|
|
|
|
|
|
|
|
2530
|
0
|
|
|
|
|
0
|
for (1..$count) { |
2531
|
0
|
|
|
|
|
0
|
splice @$rows, $rx, 0, [("") x $num_cols]; |
2532
|
0
|
|
|
|
|
0
|
splice @$linenums, $rx, 0, "??"; |
2533
|
|
|
|
|
|
|
} |
2534
|
|
|
|
|
|
|
|
2535
|
0
|
|
|
|
|
0
|
$rx; |
2536
|
|
|
|
|
|
|
} |
2537
|
0
|
|
|
0
|
1
|
0
|
sub insert_row(;$) { goto &insert_rows; } |
2538
|
|
|
|
|
|
|
|
2539
|
|
|
|
|
|
|
# read_spreadsheet $inpath [Spreadsheet::Edit::IO::OpenAsCSV options...] |
2540
|
|
|
|
|
|
|
# read_spreadsheet $inpath [,iolayers =>... or encoding =>...] |
2541
|
|
|
|
|
|
|
# read_spreadsheet $inpath [,{iolayers =>... or encoding =>... }] #OLD API |
2542
|
|
|
|
|
|
|
# read_spreadsheet [{iolayers =>... or encoding =>... }, ] $inpath #NEW API |
2543
|
|
|
|
|
|
|
# |
2544
|
|
|
|
|
|
|
# Titles are auto-detected by default, but this may be controlled |
2545
|
|
|
|
|
|
|
# via {OPTIONS}: |
2546
|
|
|
|
|
|
|
# title_rx => rx # Don't autodetect; set the specified title_rx |
2547
|
|
|
|
|
|
|
# title_rx => undef # Don't autodetect; no titles |
2548
|
|
|
|
|
|
|
# OR: autodetect options, passed thru to title_rx() |
2549
|
|
|
|
|
|
|
# |
2550
|
|
|
|
|
|
|
sub read_spreadsheet($;@) { |
2551
|
2
|
|
|
2
|
1
|
5
|
my ($self, $opthash, $inpath) = &__self_opthash_1arg; |
2552
|
2
|
|
|
|
|
5
|
my $orig_opthash = { %$opthash }; |
2553
|
|
|
|
|
|
|
|
2554
|
2
|
|
|
|
|
7
|
my $saved_stdopts = $self->_set_stdopts($opthash); |
2555
|
2
|
|
|
2
|
|
17
|
scope_guard{ $self->_restore_stdopts($saved_stdopts) }; |
|
2
|
|
|
|
|
6
|
|
2556
|
|
|
|
|
|
|
|
2557
|
2
|
|
|
|
|
14
|
my %csvopts = @sane_CSV_read_options; |
2558
|
|
|
|
|
|
|
# Separate out Text::CSV options from %$opthash |
2559
|
2
|
|
|
|
|
22
|
foreach my $key (Text::CSV::known_attributes()) { |
2560
|
|
|
|
|
|
|
#$csvopts{$key} = delete $opthash{$key} if exists $opthash{$key}; |
2561
|
62
|
50
|
|
|
|
185
|
$csvopts{$key} = $opthash->{$key} if defined $opthash->{$key}; |
2562
|
62
|
|
|
|
|
78
|
delete $opthash->{$key}; |
2563
|
|
|
|
|
|
|
} |
2564
|
2
|
|
|
|
|
9
|
$csvopts{escape_char} = $csvopts{quote_char}; # " : """ |
2565
|
|
|
|
|
|
|
|
2566
|
|
|
|
|
|
|
croak "Obsolete {sheet} key in options (use 'sheetname')" |
2567
|
2
|
50
|
|
|
|
7
|
if exists $opthash->{sheet}; |
2568
|
|
|
|
|
|
|
|
2569
|
2
|
|
|
|
|
23
|
__validate_opthash( $opthash, |
2570
|
|
|
|
|
|
|
[ |
2571
|
|
|
|
|
|
|
qw/title_rx/, |
2572
|
|
|
|
|
|
|
qw/iolayers encoding verbose silent debug/, |
2573
|
|
|
|
|
|
|
qw/tempdir use_gnumeric raw_values sheetname/, # for OpenAsCsv |
2574
|
|
|
|
|
|
|
qw/required min_rx max_rx first_cx last_cx/, # for title_rx |
2575
|
|
|
|
|
|
|
], |
2576
|
|
|
|
|
|
|
desc => "read_spreadsheet option", |
2577
|
|
|
|
|
|
|
undef_ok => [qw/title_rx verbose silent debug use_gnumeric/] ); |
2578
|
|
|
|
|
|
|
|
2579
|
|
|
|
|
|
|
# convert {encoding} to {iolayers} |
2580
|
2
|
50
|
|
|
|
10
|
if (my $enc = delete $opthash->{encoding}) { |
2581
|
|
|
|
|
|
|
#warn "Found OBSOLETE read_spreadsheet 'encoding' opt (use iolayers instead)\n"; |
2582
|
0
|
|
0
|
|
|
0
|
$opthash->{iolayers} = ($opthash->{iolayers}//"") . ":encoding($enc)"; |
2583
|
|
|
|
|
|
|
} |
2584
|
|
|
|
|
|
|
# Same as last-used, if any |
2585
|
|
|
|
|
|
|
# N.B. If user says nothing, OpenAsCsv() defaults to UTF-8 |
2586
|
2
|
|
50
|
|
|
31
|
$opthash->{iolayers} //= $$self->{iolayers} // ""; |
|
|
|
33
|
|
|
|
|
2587
|
|
|
|
|
|
|
|
2588
|
|
|
|
|
|
|
my ($rows, $linenums, $meta_info, $verbose, $debug) |
2589
|
2
|
|
|
|
|
7
|
= @$$self{qw/rows linenums meta_info verbose debug/}; |
2590
|
|
|
|
|
|
|
|
2591
|
|
|
|
|
|
|
##$self->_check_currsheet; |
2592
|
|
|
|
|
|
|
|
2593
|
2
|
|
|
|
|
4
|
my $hash; |
2594
|
2
|
|
|
|
|
6
|
{ local $$self->{verbose} = 0; |
|
2
|
|
|
|
|
4
|
|
2595
|
|
|
|
|
|
|
$hash = OpenAsCsv( |
2596
|
|
|
|
|
|
|
inpath => $inpath, |
2597
|
|
|
|
|
|
|
debug => $$self->{debug}, |
2598
|
|
|
|
|
|
|
silent => $$self->{silent}, |
2599
|
2
|
|
33
|
|
|
24
|
verbose => ($$self->{verbose} || $$self->{debug}), |
2600
|
|
|
|
|
|
|
%$opthash, # all our opts are valid here |
2601
|
|
|
|
|
|
|
); |
2602
|
|
|
|
|
|
|
} |
2603
|
|
|
|
|
|
|
|
2604
|
|
|
|
|
|
|
### TODO: Split off the following into a separate read_csvdata() method |
2605
|
|
|
|
|
|
|
### which takes a file handle? This might be useful so users |
2606
|
|
|
|
|
|
|
### can open arbitrary sources (even a pipe) and parse the data |
2607
|
|
|
|
|
|
|
### (e.g. /etc/passwd with : as the separator). |
2608
|
|
|
|
|
|
|
### ...but unclear how to handle encoding |
2609
|
|
|
|
|
|
|
|
2610
|
|
|
|
|
|
|
# Save possibly-defaulted iolayers for use in subsequent write_csv |
2611
|
2
|
|
33
|
|
|
23
|
$$self->{iolayers} //= $hash->{iolayers}; |
2612
|
|
|
|
|
|
|
|
2613
|
2
|
|
|
|
|
3
|
my $fh = $hash->{fh}; |
2614
|
|
|
|
|
|
|
|
2615
|
2
|
|
|
|
|
5
|
$csvopts{keep_meta_info} = 1; |
2616
|
2
|
50
|
|
|
|
10
|
my $csv = Text::CSV->new (\%csvopts) |
2617
|
|
|
|
|
|
|
or croak "read_spreadsheet: ".Text::CSV->error_diag () |
2618
|
|
|
|
|
|
|
.dvis('\n## %csvopts\n'); |
2619
|
|
|
|
|
|
|
|
2620
|
2
|
|
|
|
|
429
|
undef $$self->{num_cols}; |
2621
|
2
|
|
|
|
|
20
|
@$rows = (); |
2622
|
2
|
|
|
|
|
4
|
@$linenums = (); |
2623
|
2
|
|
|
|
|
4
|
my $lnum = 1; |
2624
|
2
|
|
|
|
|
80
|
while (my $F = $csv->getline( $fh )) { |
2625
|
6
|
|
|
|
|
368
|
push(@$linenums, $lnum); |
2626
|
6
|
|
|
|
|
24
|
my @minfo = $csv->meta_info(); |
2627
|
|
|
|
|
|
|
# Force quoting of fields which look like negative numbers with an ascii |
2628
|
|
|
|
|
|
|
# minus (\x{2D}) rather than Unicode math minus (\N{U+2212}). |
2629
|
|
|
|
|
|
|
# This prevents conversion to the Unicode math minus when LibreOffice |
2630
|
|
|
|
|
|
|
# reads the CSV. The assumption is that if the input, when converted |
2631
|
|
|
|
|
|
|
# TO a csv, has an ascii minus then the original spreadsheet cell format |
2632
|
|
|
|
|
|
|
# was "text" not numeric. |
2633
|
6
|
|
|
|
|
59
|
for my $cx (0..$#$F) { |
2634
|
|
|
|
|
|
|
#...TODO $minfo[$cx] |= 0x0001 if $F->[$cx] =~ /^-[\d.]+$/a; |
2635
|
|
|
|
|
|
|
} |
2636
|
6
|
|
|
|
|
11
|
push(@$meta_info, \@minfo); |
2637
|
6
|
|
|
|
|
16
|
$lnum = $.+1; |
2638
|
6
|
|
|
|
|
20
|
push(@$rows, $F); |
2639
|
|
|
|
|
|
|
} |
2640
|
2
|
50
|
|
|
|
214
|
close $fh || croak "Error reading $hash->{csvpath}: $!\n"; |
2641
|
|
|
|
|
|
|
|
2642
|
|
|
|
|
|
|
$$self->{data_source} = |
2643
|
2
|
|
|
|
|
17
|
form_spec_with_sheetname($hash->{inpath}, $hash->{sheetname}); |
2644
|
2
|
|
|
|
|
7
|
$$self->{sheetname} = $hash->{sheetname}; # possibly undef |
2645
|
|
|
|
|
|
|
|
2646
|
2
|
|
|
|
|
7
|
$self->_rows_replaced; |
2647
|
|
|
|
|
|
|
|
2648
|
|
|
|
|
|
|
# Set title_rx, either to a value explicitly given in OPTIONS (possibly |
2649
|
|
|
|
|
|
|
# undef, meaning no titles) or else auto-detect. |
2650
|
2
|
|
|
|
|
4
|
my %autodetect_opts; |
2651
|
2
|
|
|
|
|
5
|
foreach (qw/required min_rx max_rx first_cx last_cx/) { |
2652
|
10
|
50
|
|
|
|
27
|
$autodetect_opts{$_} = $opthash->{$_} if exists($opthash->{$_}); |
2653
|
|
|
|
|
|
|
} |
2654
|
|
|
|
|
|
|
|
2655
|
2
|
100
|
|
|
|
8
|
my $arg = exists($opthash->{title_rx}) ? $opthash->{title_rx} : 'auto'; |
2656
|
2
|
|
|
|
|
3
|
{ local $$self->{cmd_nesting} = $$self->{cmd_nesting} + 1; |
|
2
|
|
|
|
|
9
|
|
2657
|
2
|
|
|
|
|
7
|
$autodetect_opts{verbose} = 0; # suppress logging |
2658
|
2
|
|
|
|
|
10
|
$self->title_rx(\%autodetect_opts, $arg); |
2659
|
|
|
|
|
|
|
} |
2660
|
|
|
|
|
|
|
|
2661
|
|
|
|
|
|
|
log_methcall $self, [$orig_opthash, $inpath, |
2662
|
|
|
|
|
|
|
\" [title_rx set to ",vis($$self->{title_rx}),\"]"] |
2663
|
2
|
50
|
|
|
|
14
|
if $$self->{verbose}; |
2664
|
|
|
|
|
|
|
|
2665
|
2
|
|
|
|
|
25
|
$self |
2666
|
|
|
|
|
|
|
}#read_spreadsheet |
2667
|
|
|
|
|
|
|
|
2668
|
|
|
|
|
|
|
# write_csv {OPTHASH} "/path/to/output.csv" |
2669
|
|
|
|
|
|
|
# Cells will be quoted if the input was quoted, i.e. if indicated by meta_info. |
2670
|
|
|
|
|
|
|
sub write_csv(*;@) { |
2671
|
0
|
|
|
0
|
1
|
0
|
my $self = &__selfmust; |
2672
|
0
|
0
|
|
|
|
0
|
my $opts = ref($_[0]) eq 'HASH' ? shift() : {}; |
2673
|
0
|
|
|
|
|
0
|
my $dest = shift; |
2674
|
|
|
|
|
|
|
|
2675
|
0
|
|
|
|
|
0
|
my %csvopts = ( @sane_CSV_write_options, |
2676
|
|
|
|
|
|
|
quote_space => 0, # dont quote embedded spaces |
2677
|
|
|
|
|
|
|
); |
2678
|
|
|
|
|
|
|
# Separate out Text::CSV options from {OPTIONS} |
2679
|
0
|
|
|
|
|
0
|
foreach my $key (Text::CSV::known_attributes()) { |
2680
|
0
|
0
|
|
|
|
0
|
$csvopts{$key} = $opts->{$key} if defined $opts->{$key}; |
2681
|
0
|
|
|
|
|
0
|
delete $opts->{$key}; |
2682
|
|
|
|
|
|
|
} |
2683
|
|
|
|
|
|
|
|
2684
|
0
|
|
|
|
|
0
|
{ my %notok = %$opts; |
|
0
|
|
|
|
|
0
|
|
2685
|
0
|
|
|
|
|
0
|
delete $notok{$_} foreach ( |
2686
|
|
|
|
|
|
|
#removed above... Text::CSV::known_attributes(), |
2687
|
|
|
|
|
|
|
qw/verbose silent debug/, |
2688
|
|
|
|
|
|
|
); |
2689
|
0
|
0
|
|
|
|
0
|
croak "Unrecognized OPTION(s): ",avislq(keys %notok) if %notok; |
2690
|
|
|
|
|
|
|
} |
2691
|
|
|
|
|
|
|
|
2692
|
0
|
|
0
|
|
|
0
|
$opts->{iolayers} //= $$self->{iolayers} // ""; |
|
|
|
0
|
|
|
|
|
2693
|
|
|
|
|
|
|
# New API: opts->{iolayers} may have all 'binmode' arguments. |
2694
|
|
|
|
|
|
|
# If it does not include encoding(...) then insert default |
2695
|
0
|
0
|
|
|
|
0
|
if ($opts->{iolayers} !~ /encoding\(|:utf8/) { |
2696
|
0
|
|
0
|
|
|
0
|
$opts->{iolayers} .= ":encoding(". |
2697
|
|
|
|
|
|
|
($self->input_encoding() || DEFAULT_WRITE_ENCODING) |
2698
|
|
|
|
|
|
|
.")"; |
2699
|
|
|
|
|
|
|
} |
2700
|
0
|
0
|
|
|
|
0
|
if ($opts->{iolayers} !~ /:(?:crlf|raw)\b/) { |
2701
|
|
|
|
|
|
|
# Use platform default |
2702
|
|
|
|
|
|
|
#$opts->{iolayers} .= ":crlf"; |
2703
|
|
|
|
|
|
|
} |
2704
|
|
|
|
|
|
|
|
2705
|
|
|
|
|
|
|
my ($rows, $meta_info, $num_cols, $verbose, $debug) |
2706
|
0
|
|
|
|
|
0
|
= @$$self{qw/rows meta_info num_cols verbose debug/}; |
2707
|
|
|
|
|
|
|
|
2708
|
0
|
|
|
|
|
0
|
my $fh; |
2709
|
0
|
0
|
|
|
|
0
|
if (openhandle($dest)) { # an already-open file handle? |
2710
|
|
|
|
|
|
|
log_methcall $self, [$opts, \("<file handle specified> $opts->{iolayers} " |
2711
|
|
|
|
|
|
|
.scalar(@$rows)." rows, $num_cols columns)")] |
2712
|
0
|
0
|
|
|
|
0
|
if $$self->{verbose}; |
2713
|
0
|
|
|
|
|
0
|
$fh = $dest; |
2714
|
|
|
|
|
|
|
} else { |
2715
|
|
|
|
|
|
|
log_methcall $self, [$opts, \($dest." $opts->{iolayers} (" |
2716
|
|
|
|
|
|
|
.scalar(@$rows)." rows, $num_cols columns)")] |
2717
|
0
|
0
|
|
|
|
0
|
if $$self->{verbose}; |
2718
|
0
|
0
|
0
|
|
|
0
|
croak "Output path suffix must be *.csv, not\n ",qsh($dest),"\n" |
2719
|
|
|
|
|
|
|
if $dest =~ /\.([a-z]*)$/ && lc($1) ne "csv"; |
2720
|
0
|
0
|
|
|
|
0
|
open $fh,">$dest" or croak "$dest: $!\n"; |
2721
|
|
|
|
|
|
|
} |
2722
|
|
|
|
|
|
|
|
2723
|
0
|
0
|
|
|
|
0
|
binmode $fh, $opts->{iolayers} or die "binmode:$!"; |
2724
|
|
|
|
|
|
|
|
2725
|
|
|
|
|
|
|
# Arrgh. Although Text::CSV is huge and complex and implements a complicated |
2726
|
|
|
|
|
|
|
# meta_info mechanism to capture quoting details on input, there is no way to |
2727
|
|
|
|
|
|
|
# use the captured info to specify quoting of output fields! |
2728
|
|
|
|
|
|
|
# So we implement writing CSVs by hand here. |
2729
|
|
|
|
|
|
|
#my $csv = Text::CSV->new (\%csvopts) |
2730
|
|
|
|
|
|
|
# or die "write_csv: ".Text::CSV->error_diag (); |
2731
|
|
|
|
|
|
|
#foreach my $row (@$rows) { |
2732
|
|
|
|
|
|
|
# oops "UNDEF row" unless defined $row; # did user modify @rows? |
2733
|
|
|
|
|
|
|
# $csv->print ($fh, $row); |
2734
|
|
|
|
|
|
|
#}; |
2735
|
|
|
|
|
|
|
|
2736
|
|
|
|
|
|
|
# 5/2/22 FIXME: Maybe meta_info could be used when writing, albiet in |
2737
|
|
|
|
|
|
|
# a grotesque way: |
2738
|
|
|
|
|
|
|
# If keep_meta_info is set > 9, then the output quotation style is |
2739
|
|
|
|
|
|
|
# "like it was used in the input of the the last parsed record"; so |
2740
|
|
|
|
|
|
|
# we could "parse" a dummy record to set the quote style before writing |
2741
|
|
|
|
|
|
|
# each record, like this (see perldoc Text::CSV_XS "keep_meta_info"): |
2742
|
|
|
|
|
|
|
# my $csv = Text::CSV_XS->new({ binary=>1, keep_meta_info=>11, |
2743
|
|
|
|
|
|
|
# quote_space => 0 }); |
2744
|
|
|
|
|
|
|
# apply_all { |
2745
|
|
|
|
|
|
|
# my $minfo = $meta_info[$rx]; |
2746
|
|
|
|
|
|
|
# my @dummy = map{ '', 'x', '""' or '"x'' } @$minfo; # HOW? |
2747
|
|
|
|
|
|
|
# $csv->parse(join ",", @dummy); # set saved meta_info |
2748
|
|
|
|
|
|
|
# $csv->print(*OUTHANDLE, $row); |
2749
|
|
|
|
|
|
|
# } |
2750
|
|
|
|
|
|
|
# |
2751
|
|
|
|
|
|
|
|
2752
|
|
|
|
|
|
|
# Much of the option handling code was copied from Text::CSV_PP.pm |
2753
|
|
|
|
|
|
|
# which depends on default values of options we don't specify explicitly. |
2754
|
|
|
|
|
|
|
# So create a Text::CSV object just to get the effective option values... |
2755
|
0
|
|
|
|
|
0
|
{ my $o = Text::CSV->new( \%csvopts ); |
|
0
|
|
|
|
|
0
|
|
2756
|
0
|
|
|
|
|
0
|
foreach my $key (Text::CSV::known_attributes()) { |
2757
|
0
|
|
|
|
|
0
|
$csvopts{$key} = $o->{$key}; |
2758
|
|
|
|
|
|
|
} |
2759
|
|
|
|
|
|
|
} |
2760
|
|
|
|
|
|
|
|
2761
|
|
|
|
|
|
|
my $re_esc = ($csvopts{escape_char} ne '' and $csvopts{escape_char} ne "\0") |
2762
|
0
|
0
|
0
|
|
|
0
|
? ($csvopts{quote_char} ne '') ? qr/(\Q$csvopts{quote_char}\E|\Q$csvopts{escape_char}\E)/ : qr/(\Q$csvopts{escape_char}\E)/ |
|
|
0
|
|
|
|
|
|
2763
|
|
|
|
|
|
|
: qr/(*FAIL)/; |
2764
|
0
|
|
|
|
|
0
|
for my $rx (0..$#$rows) { |
2765
|
0
|
|
|
|
|
0
|
my $row = $rows->[$rx]; |
2766
|
0
|
|
|
|
|
0
|
my $minfo = $meta_info->[$rx]; |
2767
|
0
|
|
|
|
|
0
|
my @results; |
2768
|
0
|
|
|
|
|
0
|
for my $cx (0..$num_cols-1) { |
2769
|
0
|
|
|
|
|
0
|
my $value = $row->[$cx]; |
2770
|
0
|
0
|
|
|
|
0
|
confess "ERROR: rx $rx, cx $cx : undef cell value" unless defined($value); |
2771
|
0
|
|
|
|
|
0
|
my $mi = $minfo->[$cx]; # undef if input was missing columns in this row |
2772
|
|
|
|
|
|
|
my $must_be_quoted = $csvopts{always_quote} || |
2773
|
0
|
|
0
|
|
|
0
|
(($mi//0) & 0x0001); # was quoted on input |
2774
|
0
|
0
|
|
|
|
0
|
unless ($must_be_quoted) { |
2775
|
0
|
0
|
|
|
|
0
|
if ($value eq '') { |
2776
|
0
|
0
|
|
|
|
0
|
$must_be_quoted = 42 if $csvopts{quote_empty}; |
2777
|
|
|
|
|
|
|
} else { |
2778
|
0
|
0
|
|
|
|
0
|
if ($csvopts{quote_char} ne '') { |
2779
|
3
|
|
|
3
|
|
40
|
use bytes; |
|
3
|
|
|
|
|
8
|
|
|
3
|
|
|
|
|
30
|
|
2780
|
|
|
|
|
|
|
$must_be_quoted=43 if |
2781
|
|
|
|
|
|
|
($value =~ /\Q$csvopts{quote_char}\E/) || |
2782
|
|
|
|
|
|
|
($csvopts{sep_char} ne '' and $csvopts{sep_char} ne "\0" and $value =~ /\Q$csvopts{sep_char}\E/) || |
2783
|
|
|
|
|
|
|
($csvopts{escape_char} ne '' and $csvopts{escape_char} ne "\0" and $value =~ /\Q$csvopts{escape_char}\E/) || |
2784
|
|
|
|
|
|
|
($csvopts{quote_binary} && $value =~ /[\x00-\x1f\x7f-\xa0]/) || |
2785
|
0
|
0
|
0
|
|
|
0
|
($csvopts{quote_space} && $value =~ /[\x09\x20]/); |
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
2786
|
|
|
|
|
|
|
} |
2787
|
|
|
|
|
|
|
} |
2788
|
|
|
|
|
|
|
} |
2789
|
0
|
|
|
|
|
0
|
$value =~ s/($re_esc)/$csvopts{escape_char}$1/g; |
2790
|
0
|
0
|
|
|
|
0
|
if ($csvopts{escape_null}) { |
2791
|
0
|
|
|
|
|
0
|
$value =~ s/\0/$csvopts{escape_char}0/g; |
2792
|
|
|
|
|
|
|
} |
2793
|
0
|
0
|
|
|
|
0
|
if ($must_be_quoted) { |
2794
|
0
|
|
|
|
|
0
|
$value = $csvopts{quote_char} . $value . $csvopts{quote_char}; |
2795
|
|
|
|
|
|
|
} |
2796
|
0
|
0
|
|
|
|
0
|
$fh->print($csvopts{sep_char}) unless $cx==0; |
2797
|
0
|
|
|
|
|
0
|
$fh->print($value); |
2798
|
|
|
|
|
|
|
} |
2799
|
0
|
|
|
|
|
0
|
$fh->print($csvopts{eol}); |
2800
|
|
|
|
|
|
|
} |
2801
|
|
|
|
|
|
|
|
2802
|
0
|
0
|
|
|
|
0
|
if (! openhandle $dest) { |
2803
|
0
|
0
|
|
|
|
0
|
close $fh || croak "Error writing $dest : $!\n"; |
2804
|
|
|
|
|
|
|
} |
2805
|
|
|
|
|
|
|
$self |
2806
|
0
|
|
|
|
|
0
|
}#write_csv |
2807
|
|
|
|
|
|
|
|
2808
|
|
|
|
|
|
|
# Write spreadsheet with specified column formats |
2809
|
|
|
|
|
|
|
# {col_formats} is required |
2810
|
|
|
|
|
|
|
# Unless {sheetname} is specified, the sheet name is the outpath basename |
2811
|
|
|
|
|
|
|
# sans any suffix |
2812
|
|
|
|
|
|
|
sub write_spreadsheet(*;@) { |
2813
|
0
|
|
|
0
|
1
|
0
|
my ($self, $opts, $outpath) = &__self_opthash_1arg; |
2814
|
0
|
|
|
|
|
0
|
my $colx = $$self->{colx}; |
2815
|
|
|
|
|
|
|
|
2816
|
0
|
0
|
|
|
|
0
|
log_methcall $self, [$opts, $outpath] if $$self->{verbose}; |
2817
|
|
|
|
|
|
|
|
2818
|
|
|
|
|
|
|
# {col_formats} may be [list of formats in column order] |
2819
|
|
|
|
|
|
|
# or { COLSPEC => fmt, ..., __DEFAULT__ => fmt } |
2820
|
|
|
|
|
|
|
# Transform the latter to the former... |
2821
|
0
|
|
0
|
|
|
0
|
my $cf = $opts->{col_formats} // croak "{col_formats} is required"; |
2822
|
0
|
0
|
|
|
|
0
|
if (ref($cf) eq "HASH") { |
2823
|
0
|
|
|
|
|
0
|
my ($default, @ary); |
2824
|
0
|
|
|
|
|
0
|
while (my ($key, $fmt) = each %$cf) { |
2825
|
0
|
0
|
|
|
|
0
|
($default = $fmt),next if $key eq "__DEFAULT__"; |
2826
|
0
|
|
0
|
|
|
0
|
my $cx = $colx->{$key} // croak("Invalid COLSPEC '$key' in col_formats"); |
2827
|
0
|
|
|
|
|
0
|
$ary[$cx] = $fmt; |
2828
|
|
|
|
|
|
|
} |
2829
|
0
|
0
|
|
|
|
0
|
foreach (@ary) { $_ = $default if ! defined; } |
|
0
|
|
|
|
|
0
|
|
2830
|
0
|
|
|
|
|
0
|
$cf = \@ary; |
2831
|
|
|
|
|
|
|
} |
2832
|
0
|
|
|
|
|
0
|
local $opts->{col_formats} = $cf; |
2833
|
|
|
|
|
|
|
|
2834
|
0
|
|
|
|
|
0
|
my ($csvfh, $csvpath) = tempfile(SUFFIX => ".csv"); |
2835
|
0
|
|
|
|
|
0
|
{ local $$self->{verbose} = 0; |
|
0
|
|
|
|
|
0
|
|
2836
|
0
|
|
|
|
|
0
|
$self->write_csv($csvfh, silent => 1, iolayers => ':encoding(UTF-8)', |
2837
|
|
|
|
|
|
|
@sane_CSV_write_options); |
2838
|
|
|
|
|
|
|
} |
2839
|
0
|
0
|
|
|
|
0
|
close $csvfh or die "Error writing $csvpath : $!"; |
2840
|
|
|
|
|
|
|
|
2841
|
|
|
|
|
|
|
# Default sheet name to output file basename sans suffix |
2842
|
0
|
|
0
|
|
|
0
|
$opts->{sheetname} //= fileparse($outpath, qr/\.\w+/); |
2843
|
|
|
|
|
|
|
|
2844
|
0
|
|
|
|
|
0
|
convert_spreadsheet($csvpath, |
2845
|
|
|
|
|
|
|
%$opts, |
2846
|
|
|
|
|
|
|
iolayers => ':encoding(UTF-8)', |
2847
|
|
|
|
|
|
|
cvt_from => "csv", |
2848
|
|
|
|
|
|
|
outpath => $outpath, |
2849
|
|
|
|
|
|
|
); |
2850
|
0
|
|
|
|
|
0
|
$self |
2851
|
|
|
|
|
|
|
} |
2852
|
|
|
|
|
|
|
|
2853
|
|
|
|
|
|
|
#==================================================================== |
2854
|
|
|
|
|
|
|
# These helpers are used by predefined magic sheet variables. |
2855
|
|
|
|
|
|
|
# See code in Spreadsheet::Edit::import() |
2856
|
|
|
|
|
|
|
|
2857
|
|
|
|
|
|
|
# Return $self if during an apply, or if being examined by Data::Dumper ; |
2858
|
|
|
|
|
|
|
# otherwise croak |
2859
|
|
|
|
|
|
|
sub _onlyinapply { |
2860
|
7
|
|
|
7
|
|
20
|
my ($self, $accessor) = @_; |
2861
|
7
|
50
|
|
|
|
33
|
unless (defined $$self->{current_rx}) { |
2862
|
0
|
|
|
|
|
0
|
for (my $lvl=2; ;$lvl++) { |
2863
|
0
|
|
0
|
|
|
0
|
my $pkg = (caller($lvl))[0] || last; |
2864
|
0
|
0
|
0
|
|
|
0
|
return $self |
2865
|
|
|
|
|
|
|
if defined($pkg) && $pkg->isa("Data::Dumper") # perldoc UNIVERSAL |
2866
|
|
|
|
|
|
|
} |
2867
|
0
|
|
|
|
|
0
|
croak "Can't use $accessor now: Not during apply*\n" |
2868
|
|
|
|
|
|
|
} |
2869
|
|
|
|
|
|
|
$self |
2870
|
7
|
|
|
|
|
21
|
} |
2871
|
|
|
|
|
|
|
sub __getsheet($$$$) { |
2872
|
9
|
|
|
9
|
|
51
|
my ($mutating, $pkg, $uvar, $onlyinapply) = @_; |
2873
|
9
|
|
|
|
|
33
|
my $sheet = $pkg2currsheet{$pkg}; |
2874
|
9
|
50
|
|
|
|
46
|
croak("Modifying variable $uvar is not allowed\n") |
2875
|
|
|
|
|
|
|
if $mutating; |
2876
|
9
|
50
|
|
|
|
44
|
croak("Can not use $uvar: No sheet is currently valid for package $pkg\n") |
2877
|
|
|
|
|
|
|
unless defined $sheet; |
2878
|
9
|
100
|
|
|
|
52
|
$onlyinapply ? _onlyinapply($sheet, $uvar) : $sheet |
2879
|
|
|
|
|
|
|
} |
2880
|
|
|
|
|
|
|
sub _scal_tiehelper { # access a scalar sheet variable |
2881
|
5
|
|
|
5
|
|
618
|
my($mutating, $pkg, $uvar, $ident, $onlyinapply) = @_; |
2882
|
5
|
|
|
|
|
20
|
my $sheet = __getsheet($mutating, $pkg, $uvar, $onlyinapply); |
2883
|
5
|
50
|
|
|
|
20
|
confess avisq(@_) unless exists $$sheet->{$ident}; |
2884
|
5
|
|
|
|
|
42
|
return \$$sheet->{$ident}; # return ref to the scalar |
2885
|
|
|
|
|
|
|
} |
2886
|
|
|
|
|
|
|
sub _aryelem_tiehelper { # access an element of an array sheet variable |
2887
|
0
|
|
|
0
|
|
0
|
my($mutating, $pkg, $uvar, $index_ident, $array_ident, $onlyinapply) = @_; |
2888
|
|
|
|
|
|
|
# E.g. for $title_row : index_ident="title_rx" and array_ident="rows" |
2889
|
0
|
|
|
|
|
0
|
my $sheet = __getsheet($mutating, $pkg, $uvar, $onlyinapply); |
2890
|
0
|
|
0
|
|
|
0
|
my $aref = $$sheet->{$array_ident} // oops dvisq '$array_ident @_'; # e.g. {rows} |
2891
|
0
|
|
0
|
|
|
0
|
my $index = $$sheet->{$index_ident} // do{ |
2892
|
0
|
0
|
0
|
|
|
0
|
if ($index_ident eq "current_rx" or $index_ident eq "title_rx") { |
2893
|
|
|
|
|
|
|
return \undef # During Data::Dumper inspection of current_row? |
2894
|
0
|
|
|
|
|
0
|
} |
2895
|
0
|
|
|
|
|
0
|
oops dvis '$array_ident $index_ident'; # otherwise it's a bug |
2896
|
|
|
|
|
|
|
}; |
2897
|
0
|
0
|
|
|
|
0
|
oops(dvisq '@_ $index') if $index > $#$aref; |
2898
|
0
|
|
|
|
|
0
|
return \$aref->[$index]; # return ref to scalar (the element in the array) |
2899
|
|
|
|
|
|
|
} |
2900
|
|
|
|
|
|
|
sub _refval_tiehelper { # access a sheet variable which is a ref of some kind |
2901
|
4
|
|
|
4
|
|
1671240
|
my($mutating, $pkg, $uvar, $field_ident, $onlyinapply, $mutable) = @_; |
2902
|
4
|
50
|
|
|
|
51
|
$mutating = 0 if $mutable; |
2903
|
4
|
|
|
|
|
35
|
my $sheet = __getsheet($mutating, $pkg, $uvar, $onlyinapply); |
2904
|
4
|
|
|
|
|
173
|
return $$sheet->{$field_ident}; # return the value, which is itself a ref |
2905
|
|
|
|
|
|
|
} |
2906
|
|
|
|
|
|
|
|
2907
|
|
|
|
|
|
|
# Retrieve the sheet currently accessed by the functional API & tied globals |
2908
|
|
|
|
|
|
|
# in the caller's package (or the specified package). |
2909
|
|
|
|
|
|
|
# If an argument is passed, change the sheet to the specified sheet. |
2910
|
|
|
|
|
|
|
# |
2911
|
|
|
|
|
|
|
# Always returns the previous sheet (or undef) |
2912
|
|
|
|
|
|
|
sub sheet(;$$) { |
2913
|
14
|
|
|
14
|
1
|
536
|
my $opthash = &__opthash; |
2914
|
14
|
|
66
|
|
|
77
|
my $pkg = $opthash->{package} // caller(); |
2915
|
14
|
50
|
|
|
|
48
|
oops if index($pkg,__PACKAGE__) >= 0; # not us or sub-pkg |
2916
|
14
|
100
|
|
|
|
57
|
my $pkgmsg = $opthash->{package} ? " [for pkg $pkg]" : ""; |
2917
|
14
|
|
|
|
|
26
|
my $curr = $pkg2currsheet{$pkg}; |
2918
|
14
|
|
33
|
|
|
77
|
my $verbose = $opthash->{verbose} || ($curr && $$curr->{verbose}); |
2919
|
14
|
100
|
|
|
|
34
|
if (@_) { |
2920
|
12
|
|
|
|
|
101
|
my $new = __validate_sheet_arg(shift @_); |
2921
|
12
|
50
|
|
|
|
33
|
croak "Extraneous argument(s) in call to sheet()" if @_; |
2922
|
12
|
100
|
|
|
|
37
|
if (defined $new) { |
2923
|
9
|
50
|
|
|
|
34
|
oops if $$new->{cmd_nesting}; |
2924
|
9
|
|
33
|
|
|
63
|
$verbose ||= $$new->{verbose}; |
2925
|
|
|
|
|
|
|
} |
2926
|
|
|
|
|
|
|
|
2927
|
12
|
0
|
|
|
|
26
|
log_call [$opthash, \(" ".fmt_sheet($new)), |
|
|
50
|
|
|
|
|
|
2928
|
|
|
|
|
|
|
\(u($curr) eq u($new) |
2929
|
|
|
|
|
|
|
? " [no change]" |
2930
|
|
|
|
|
|
|
: " [previous: ".fmt_sheet($curr)."]"), |
2931
|
|
|
|
|
|
|
\$pkgmsg] |
2932
|
|
|
|
|
|
|
if $verbose; |
2933
|
|
|
|
|
|
|
|
2934
|
12
|
|
|
|
|
30
|
$pkg2currsheet{$pkg} = $new; |
2935
|
|
|
|
|
|
|
} else { |
2936
|
2
|
50
|
|
|
|
15
|
log_call [$opthash], [\fmt_sheet($curr), \$pkgmsg] |
2937
|
|
|
|
|
|
|
if $verbose; |
2938
|
|
|
|
|
|
|
} |
2939
|
14
|
|
|
|
|
118
|
$curr |
2940
|
|
|
|
|
|
|
} |
2941
|
|
|
|
|
|
|
|
2942
|
|
|
|
|
|
|
#==================================================================== |
2943
|
|
|
|
|
|
|
package |
2944
|
|
|
|
|
|
|
Spreadsheet::Edit::RowsTie; # implements @rows and @$sheet |
2945
|
3
|
|
|
3
|
|
6760
|
use parent 'Tie::Array'; |
|
3
|
|
|
|
|
8
|
|
|
3
|
|
|
|
|
19
|
|
2946
|
|
|
|
|
|
|
|
2947
|
3
|
|
|
3
|
|
2167
|
use Carp; |
|
3
|
|
|
|
|
9
|
|
|
3
|
|
|
|
|
303
|
|
2948
|
|
|
|
|
|
|
#our @CARP_NOT = qw(Tie::Indirect Tie::Indirect::Array |
2949
|
|
|
|
|
|
|
# Tie::Indirect::Hash Tie::Indirect::Scalar); |
2950
|
3
|
|
|
|
|
26
|
use Data::Dumper::Interp 6.004 qw/visnew |
2951
|
|
|
|
|
|
|
vis viso avis alvis ivis dvis hvis hlvis |
2952
|
|
|
|
|
|
|
visq visoq avisq alvisq ivisq dvisq hvisq hlvisq |
2953
|
3
|
|
|
3
|
|
24
|
addrvis rvis rvisq u quotekey qsh qshlist qshpath/; |
|
3
|
|
|
|
|
55
|
|
2954
|
|
|
|
|
|
|
|
2955
|
3
|
|
|
3
|
|
4701
|
use Scalar::Util qw(looks_like_number weaken); |
|
3
|
|
|
|
|
16
|
|
|
3
|
|
|
|
|
2265
|
|
2956
|
0
|
|
|
0
|
|
0
|
sub oops(@) { goto &Spreadsheet::Edit::oops } |
2957
|
|
|
|
|
|
|
|
2958
|
|
|
|
|
|
|
sub TIEARRAY { |
2959
|
4
|
|
|
4
|
|
13
|
my ($classname, $sheet) = @_; |
2960
|
4
|
|
|
|
|
12
|
my $o = bless [ [], $sheet], $classname; |
2961
|
4
|
|
|
|
|
31
|
weaken $o->[1]; |
2962
|
4
|
|
|
|
|
11
|
$o |
2963
|
|
|
|
|
|
|
} |
2964
|
|
|
|
|
|
|
sub FETCH { |
2965
|
47
|
|
|
47
|
|
87
|
my ($this, $index) = @_; |
2966
|
47
|
|
|
|
|
88
|
my $aref = $this->[0]; |
2967
|
47
|
50
|
33
|
|
|
166
|
croak "Index ",u($index)," is invalid or out of range" |
2968
|
|
|
|
|
|
|
unless $index >= 0 && $index <= $#$aref; |
2969
|
47
|
|
|
|
|
198
|
$aref->[$index]; |
2970
|
|
|
|
|
|
|
} |
2971
|
|
|
|
|
|
|
sub STORE { |
2972
|
12
|
|
|
12
|
|
60
|
my ($this, $index, $val) = @_; |
2973
|
12
|
|
|
|
|
33
|
my ($aref, $sheet) = @$this; |
2974
|
12
|
50
|
33
|
|
|
67
|
croak "Index ",u($index)," is invalid or out of range" |
2975
|
|
|
|
|
|
|
unless $index >= 0 && $index <= $#$aref+1; |
2976
|
12
|
50
|
|
|
|
29
|
croak "Value must be a ref to array of cell values (not $val)" |
2977
|
|
|
|
|
|
|
if ! Spreadsheet::Edit::__looks_like_aref($val); |
2978
|
|
|
|
|
|
|
croak "Cell values may not be undef" |
2979
|
12
|
50
|
|
|
|
29
|
if grep{! defined} @$val; |
|
30
|
|
|
|
|
74
|
|
2980
|
|
|
|
|
|
|
croak "Cell values must be strings or numbers" |
2981
|
12
|
50
|
|
|
|
29
|
if grep{ ref($_) && !looks_like_number($_) } @$val; |
|
30
|
50
|
|
|
|
79
|
|
2982
|
12
|
50
|
|
|
|
98
|
if (my $num_cols = $$sheet->{num_cols}) { |
2983
|
0
|
0
|
|
|
|
0
|
croak "New row must contain $num_cols cells (not ", $#$val+1, ")" |
2984
|
|
|
|
|
|
|
if @$val != $num_cols; |
2985
|
|
|
|
|
|
|
} |
2986
|
|
|
|
|
|
|
# else (0 or undef) someone promises to set it later |
2987
|
|
|
|
|
|
|
|
2988
|
|
|
|
|
|
|
# Store a *copy* of the data to dispose of a Magicrow wrapper, if present |
2989
|
12
|
|
|
|
|
48
|
my $cells = [ @$val ]; |
2990
|
12
|
|
|
|
|
56
|
$aref->[$index] = Spreadsheet::Edit::Magicrow->new($sheet, $cells); |
2991
|
|
|
|
|
|
|
} |
2992
|
245
|
|
|
245
|
|
437
|
sub FETCHSIZE { scalar @{ $_[0]->[0] } } |
|
245
|
|
|
|
|
974
|
|
2993
|
|
|
|
|
|
|
sub STORESIZE { |
2994
|
2
|
|
|
2
|
|
9
|
my ($this, $newlen) = @_; |
2995
|
2
|
|
|
|
|
4
|
$#{ $this->[0] } = $newlen-1; |
|
2
|
|
|
|
|
10
|
|
2996
|
|
|
|
|
|
|
} |
2997
|
|
|
|
|
|
|
# End packageSpreadsheet::Edit::RowsTie |
2998
|
|
|
|
|
|
|
|
2999
|
|
|
|
|
|
|
#==================================================================== |
3000
|
|
|
|
|
|
|
package |
3001
|
|
|
|
|
|
|
Spreadsheet::Edit::Magicrow; |
3002
|
|
|
|
|
|
|
|
3003
|
3
|
|
|
3
|
|
29
|
use Carp; |
|
3
|
|
|
|
|
16
|
|
|
3
|
|
|
|
|
421
|
|
3004
|
|
|
|
|
|
|
our @CARP_NOT = qw(Spreadsheet::Edit); |
3005
|
3
|
|
|
3
|
|
24
|
use Scalar::Util qw(weaken blessed looks_like_number); |
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
363
|
|
3006
|
0
|
|
|
0
|
|
0
|
sub oops(@) { goto &Spreadsheet::Edit::oops } |
3007
|
3
|
|
|
3
|
|
26
|
use Data::Dumper::Interp; |
|
3
|
|
|
|
|
6
|
|
|
3
|
|
|
|
|
21
|
|
3008
|
|
|
|
|
|
|
|
3009
|
|
|
|
|
|
|
sub new { |
3010
|
12
|
|
|
12
|
|
28
|
my ($classname, $sheet, $cells) = @_; |
3011
|
12
|
|
|
|
|
21
|
my %hashview; tie %hashview, __PACKAGE__, $cells, $sheet; |
|
12
|
|
|
|
|
40
|
|
3012
|
12
|
|
|
|
|
66
|
bless \ [$cells, \%hashview], $classname; |
3013
|
|
|
|
|
|
|
} |
3014
|
110
|
|
|
110
|
|
167
|
use overload '@{}' => sub { ${ shift() }->[0] }, |
|
110
|
|
|
|
|
335
|
|
3015
|
4
|
|
|
4
|
|
21
|
'%{}' => sub { ${ shift() }->[1] }, |
|
4
|
|
|
|
|
196
|
|
3016
|
|
|
|
|
|
|
#'""' => sub { shift }, # defeats vis overload eval! |
3017
|
|
|
|
|
|
|
#'0+' => sub { shift }, |
3018
|
|
|
|
|
|
|
#'==' => sub { my ($self, $other, $swap) = @_; $self == $other }, |
3019
|
|
|
|
|
|
|
#'eq' => sub { my ($self, $other, $swap) = @_; "$self" eq "$other" }, |
3020
|
3
|
|
|
|
|
26
|
fallback => 1, # for "" etc. FIXME: is this really ok? |
3021
|
3
|
|
|
3
|
|
1850
|
; |
|
3
|
|
|
|
|
9
|
|
3022
|
|
|
|
|
|
|
|
3023
|
|
|
|
|
|
|
sub TIEHASH { |
3024
|
12
|
|
|
12
|
|
25
|
my ($pkg, $cells, $sheet) = @_; |
3025
|
12
|
|
|
|
|
45
|
my $o = bless \ [$cells, $sheet], $pkg; |
3026
|
12
|
|
|
|
|
107
|
weaken $$o->[1]; |
3027
|
12
|
|
|
|
|
29
|
$o |
3028
|
|
|
|
|
|
|
} |
3029
|
|
|
|
|
|
|
sub _cellref { |
3030
|
4
|
|
|
4
|
|
8
|
my ($cells, $sheet) = @{ ${ shift() } }; # First arg is 'self' |
|
4
|
|
|
|
|
14
|
|
|
4
|
|
|
|
|
26
|
|
3031
|
4
|
|
|
|
|
10
|
my $key = shift; # Second arg is key |
3032
|
4
|
|
|
|
|
9
|
my $mutating = @_; # Third arg exists only for STORE |
3033
|
4
|
|
|
|
|
11
|
my $colx = $$sheet->{colx}; |
3034
|
4
|
|
|
|
|
9
|
my $cx = $colx->{$key}; |
3035
|
|
|
|
|
|
|
|
3036
|
4
|
50
|
|
|
|
21
|
if (! defined $cx) { |
3037
|
0
|
0
|
|
|
|
0
|
exists($colx->{$key}) |
3038
|
|
|
|
|
|
|
or croak "'$key' is an unknown COLSPEC. The valid keys are:\n", |
3039
|
|
|
|
|
|
|
$sheet->_fmt_colx(); |
3040
|
|
|
|
|
|
|
# Undef colx results from alias({optional => TRUE},...) which failed, |
3041
|
|
|
|
|
|
|
# or from an alias which became invalid because the column was deleted. |
3042
|
0
|
0
|
|
|
|
0
|
croak "Attempt to write to 'optional' alias '$key' which is currently NOT DEFINED" |
3043
|
|
|
|
|
|
|
if $mutating; |
3044
|
|
|
|
|
|
|
return \undef # Reading such a column returns undef |
3045
|
0
|
|
|
|
|
0
|
} |
3046
|
4
|
|
33
|
|
|
8
|
$cx <= $#{$cells} |
|
4
|
|
|
|
|
28
|
|
3047
|
|
|
|
|
|
|
// croak "BUG?? key '$key' maps to cx $cx which is out of range!"; |
3048
|
4
|
|
|
|
|
48
|
\$cells->[$cx] |
3049
|
|
|
|
|
|
|
} |
3050
|
|
|
|
|
|
|
sub FETCH { |
3051
|
4
|
|
|
4
|
|
22
|
${ &_cellref } |
|
4
|
|
|
|
|
9
|
|
3052
|
|
|
|
|
|
|
} |
3053
|
|
|
|
|
|
|
sub STORE { |
3054
|
0
|
|
|
0
|
|
|
my $r = &_cellref; |
3055
|
0
|
|
|
|
|
|
$$r = shift; |
3056
|
|
|
|
|
|
|
} |
3057
|
|
|
|
|
|
|
sub NEXTKEY { |
3058
|
0
|
|
|
0
|
|
|
my (undef, $sheet) = @{ ${ shift() } }; |
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
3059
|
0
|
|
|
|
|
|
each %{ $$sheet->{colx} } |
|
0
|
|
|
|
|
|
|
3060
|
|
|
|
|
|
|
} |
3061
|
|
|
|
|
|
|
sub FIRSTKEY { |
3062
|
0
|
|
|
0
|
|
|
my (undef, $sheet) = @{ ${ shift() } }; |
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
3063
|
0
|
|
|
|
|
|
my $colx = $$sheet->{colx}; |
3064
|
0
|
|
|
|
|
|
my $a = scalar keys %$colx; # reset iterator |
3065
|
0
|
|
|
|
|
|
each %$colx; |
3066
|
|
|
|
|
|
|
} |
3067
|
|
|
|
|
|
|
sub EXISTS { |
3068
|
0
|
|
|
0
|
|
|
my (undef, $sheet) = @{ ${ shift() } }; |
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
3069
|
0
|
|
|
|
|
|
my $key = shift; |
3070
|
0
|
|
|
|
|
|
exists $$sheet->{colx}->{$key} |
3071
|
|
|
|
|
|
|
} |
3072
|
|
|
|
|
|
|
sub SCALAR { |
3073
|
0
|
|
|
0
|
|
|
my (undef, $sheet) = @{ ${ shift() } }; |
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
3074
|
0
|
|
|
|
|
|
scalar %{ $$sheet->{colx} } |
|
0
|
|
|
|
|
|
|
3075
|
|
|
|
|
|
|
} |
3076
|
0
|
|
|
0
|
|
|
sub DELETE { confess "DELETE not allowed for ".__PACKAGE__ } |
3077
|
0
|
|
|
0
|
|
|
sub CLEAR { confess "CLEAR not allowed for ".__PACKAGE__ } |
3078
|
|
|
|
|
|
|
|
3079
|
|
|
|
|
|
|
# End package Spreadsheet::Edit::Magicrow; |
3080
|
|
|
|
|
|
|
#==================================================================== |
3081
|
|
|
|
|
|
|
|
3082
|
|
|
|
|
|
|
1; |
3083
|
|
|
|
|
|
|
__END__ |
3084
|
|
|
|
|
|
|
|
3085
|
|
|
|
|
|
|
=pod |
3086
|
|
|
|
|
|
|
|
3087
|
|
|
|
|
|
|
=encoding UTF-8 |
3088
|
|
|
|
|
|
|
|
3089
|
|
|
|
|
|
|
=head1 NAME |
3090
|
|
|
|
|
|
|
|
3091
|
|
|
|
|
|
|
Spreadsheet::Edit - Slice and dice spreadsheets, optionally using tied variables. |
3092
|
|
|
|
|
|
|
|
3093
|
|
|
|
|
|
|
=head1 NON-OO SYNOPSIS |
3094
|
|
|
|
|
|
|
|
3095
|
|
|
|
|
|
|
use Spreadsheet::Edit qw(:all); |
3096
|
|
|
|
|
|
|
|
3097
|
|
|
|
|
|
|
# Examples assume a spreadsheet with these titles in the first row: |
3098
|
|
|
|
|
|
|
# "Account Number" "Customer's Name" "Email" "Home-phone" "Income" |
3099
|
|
|
|
|
|
|
|
3100
|
|
|
|
|
|
|
read_spreadsheet "mailing_list.xls!Sheet1"; |
3101
|
|
|
|
|
|
|
|
3102
|
|
|
|
|
|
|
# alias an identifier to a long or complicated title |
3103
|
|
|
|
|
|
|
alias Name => qr/customer/i; # matches "Customer's Name" |
3104
|
|
|
|
|
|
|
|
3105
|
|
|
|
|
|
|
# ------------ without tied column variables ----------- |
3106
|
|
|
|
|
|
|
|
3107
|
|
|
|
|
|
|
# Print the data |
3108
|
|
|
|
|
|
|
printf "%20s %8s %8s %-13s %s\n", "Name","A/N","Income","Phone","Email"; |
3109
|
|
|
|
|
|
|
apply { |
3110
|
|
|
|
|
|
|
printf "%20s %8d %8.2f %-13s %s\n", |
3111
|
|
|
|
|
|
|
$crow{Name}, # this key is an explicit alias |
3112
|
|
|
|
|
|
|
$crow{"Account Number"}, # ...actual title |
3113
|
|
|
|
|
|
|
$crow{Income}, # ...actual title |
3114
|
|
|
|
|
|
|
$crow{Home_phone}, # ...auto-generated alias |
3115
|
|
|
|
|
|
|
$crow{Email} ; # ...actual title |
3116
|
|
|
|
|
|
|
}; |
3117
|
|
|
|
|
|
|
|
3118
|
|
|
|
|
|
|
# Randomly access rows. |
3119
|
|
|
|
|
|
|
print "Row 42: Column 'C' is ", $rows[41]{C}, "\n"; |
3120
|
|
|
|
|
|
|
print "Row 42: Customer's Name is ", $rows[41]{Name}, "\n"; |
3121
|
|
|
|
|
|
|
print "Row 42: 3rd column is ", $rows[41][2], "\n"; |
3122
|
|
|
|
|
|
|
|
3123
|
|
|
|
|
|
|
# Split the "Customer's Name" into separate FName and LName columns |
3124
|
|
|
|
|
|
|
insert_cols '>Name', "FName", "LName"; |
3125
|
|
|
|
|
|
|
apply { |
3126
|
|
|
|
|
|
|
($crow{FName}, $crow{LName}) = ($crow{Name} =~ /(.*) (.*)/) |
3127
|
|
|
|
|
|
|
or die logmsg "Could not parse Name"; # logmsg adds current row number |
3128
|
|
|
|
|
|
|
}; |
3129
|
|
|
|
|
|
|
delete_cols "Name"; |
3130
|
|
|
|
|
|
|
|
3131
|
|
|
|
|
|
|
# Sort by last name |
3132
|
|
|
|
|
|
|
sort_rows { $a->{LName} cmp $b->{LName} }; |
3133
|
|
|
|
|
|
|
|
3134
|
|
|
|
|
|
|
# ------------ using tied column variables ----------- |
3135
|
|
|
|
|
|
|
|
3136
|
|
|
|
|
|
|
our $Name; # 'Name' is the explicit alias created above |
3137
|
|
|
|
|
|
|
our $Account_Number; # Auto-generated alias for "Account Number" |
3138
|
|
|
|
|
|
|
our $Home_phone; # ditto |
3139
|
|
|
|
|
|
|
our $Income; # 'Income' is an actual title |
3140
|
|
|
|
|
|
|
our $Email; # ditto |
3141
|
|
|
|
|
|
|
our ($FName, $LName); # These columns do not yet exist |
3142
|
|
|
|
|
|
|
|
3143
|
|
|
|
|
|
|
tie_column_vars "Name", "Account_Number", qr/phone/, qr/^inc/i, "FName", "LName"; |
3144
|
|
|
|
|
|
|
|
3145
|
|
|
|
|
|
|
# Print the data |
3146
|
|
|
|
|
|
|
printf "%20s %8s %8s %-13s %s\n", "Name","A/N","Income","Phone","Email"; |
3147
|
|
|
|
|
|
|
apply { |
3148
|
|
|
|
|
|
|
printf "%20s %8d %8.2f %-13s%s\n", |
3149
|
|
|
|
|
|
|
$Name, $Account_Number, $Income, $Home_phone, $Email; |
3150
|
|
|
|
|
|
|
}; |
3151
|
|
|
|
|
|
|
|
3152
|
|
|
|
|
|
|
# Split the "Customer's Name" into separate FName and LName columns |
3153
|
|
|
|
|
|
|
insert_cols '>Name', "FName", "LName"; |
3154
|
|
|
|
|
|
|
apply { |
3155
|
|
|
|
|
|
|
($FName, $LName) = ($Name =~ /^(\S+) (\S+)$/) |
3156
|
|
|
|
|
|
|
or die logmsg "Could not parse Name"; |
3157
|
|
|
|
|
|
|
}; |
3158
|
|
|
|
|
|
|
delete_cols "Name"; |
3159
|
|
|
|
|
|
|
|
3160
|
|
|
|
|
|
|
# Simple mail-merge |
3161
|
|
|
|
|
|
|
use POSIX qw(strftime); |
3162
|
|
|
|
|
|
|
apply { |
3163
|
|
|
|
|
|
|
return |
3164
|
|
|
|
|
|
|
if $Income < 100000; # not in our audience |
3165
|
|
|
|
|
|
|
open SENDMAIL, "|sendmail -t -oi" || die "pipe:$!"; |
3166
|
|
|
|
|
|
|
print SENDMAIL "To: $FName $LName <$Email>\n"; |
3167
|
|
|
|
|
|
|
print SENDMAIL strftime("Date: %a, %d %b %y %T %z\n", localtime(time)); |
3168
|
|
|
|
|
|
|
print SENDMAIL <<EOF ; |
3169
|
|
|
|
|
|
|
From: sales\@example.com |
3170
|
|
|
|
|
|
|
Subject: Help for the 1% |
3171
|
|
|
|
|
|
|
|
3172
|
|
|
|
|
|
|
Dear $FName, |
3173
|
|
|
|
|
|
|
If you have disposable income, we can help with that. |
3174
|
|
|
|
|
|
|
Sincerely, |
3175
|
|
|
|
|
|
|
Your investment advisor. |
3176
|
|
|
|
|
|
|
EOF |
3177
|
|
|
|
|
|
|
close SENDMAIL || die "sendmail failed ($?)\n"; |
3178
|
|
|
|
|
|
|
}; |
3179
|
|
|
|
|
|
|
|
3180
|
|
|
|
|
|
|
# ------------ multiple sheets -------------- |
3181
|
|
|
|
|
|
|
|
3182
|
|
|
|
|
|
|
our ($Foo, $Bar, $Income); |
3183
|
|
|
|
|
|
|
|
3184
|
|
|
|
|
|
|
read_spreadsheet "file1.csv"; |
3185
|
|
|
|
|
|
|
tie_column_vars; # tie all vars that ever become valid |
3186
|
|
|
|
|
|
|
|
3187
|
|
|
|
|
|
|
my $s1 = sheet undef ; # Save ref to current sheet & forget it |
3188
|
|
|
|
|
|
|
|
3189
|
|
|
|
|
|
|
read_spreadsheet "file2.csv"; # Auto-creates sheet bc current is undef |
3190
|
|
|
|
|
|
|
tie_column_vars; |
3191
|
|
|
|
|
|
|
|
3192
|
|
|
|
|
|
|
my $s2 = sheet ; # Save ref to second sheet |
3193
|
|
|
|
|
|
|
|
3194
|
|
|
|
|
|
|
print "$Foo $Bar $Income\n"; # these refer to $s2, the current sheet |
3195
|
|
|
|
|
|
|
|
3196
|
|
|
|
|
|
|
sheet $s1 ; |
3197
|
|
|
|
|
|
|
print "$FName $LName $Income\n"; # these now refer to the original sheet |
3198
|
|
|
|
|
|
|
|
3199
|
|
|
|
|
|
|
# ------------ create sheet from memory -------------- |
3200
|
|
|
|
|
|
|
|
3201
|
|
|
|
|
|
|
my $s3 = new_sheet |
3202
|
|
|
|
|
|
|
data_source => "my own data", |
3203
|
|
|
|
|
|
|
rows => [ |
3204
|
|
|
|
|
|
|
["This is a row before the title row" ], |
3205
|
|
|
|
|
|
|
["Full Name", "Address", "City", "State", "Zip" ], |
3206
|
|
|
|
|
|
|
["Joe Smith", "123 Main St", "Boston", "CA", "12345"], |
3207
|
|
|
|
|
|
|
["Mary Jones", "999 Olive Drive", "Fenton", "OH", "67890"], |
3208
|
|
|
|
|
|
|
], |
3209
|
|
|
|
|
|
|
; |
3210
|
|
|
|
|
|
|
$s3=>title_rx(1); |
3211
|
|
|
|
|
|
|
... |
3212
|
|
|
|
|
|
|
|
3213
|
|
|
|
|
|
|
=head1 OO SYNOPSIS |
3214
|
|
|
|
|
|
|
|
3215
|
|
|
|
|
|
|
use Spreadsheet::Edit (); |
3216
|
|
|
|
|
|
|
|
3217
|
|
|
|
|
|
|
my $sheet = Spreadsheet::Edit->new(); |
3218
|
|
|
|
|
|
|
$sheet->read_spreadsheet("mailing_list.xls!sheet name"); |
3219
|
|
|
|
|
|
|
$sheet->alias( Name => qr/customer/i ); # matches "Customer's Name" |
3220
|
|
|
|
|
|
|
|
3221
|
|
|
|
|
|
|
# Randomly access rows. |
3222
|
|
|
|
|
|
|
# Sheet objects, when used as an ARRAYref, act like \@rows |
3223
|
|
|
|
|
|
|
print "Row 42: Name is ", $sheet->[41]{Name}, "\n"; |
3224
|
|
|
|
|
|
|
print "Row 42, Column 3 is ", $sheet->[41][2], "\n"; |
3225
|
|
|
|
|
|
|
|
3226
|
|
|
|
|
|
|
# Print the data. |
3227
|
|
|
|
|
|
|
# Sheet objects, when used as an HASHref, act like \%crow |
3228
|
|
|
|
|
|
|
printf "%20s %8s %8s %-13s %s\n", "Name","A/N","Income","Phone","Email"; |
3229
|
|
|
|
|
|
|
$sheet->apply( sub{ |
3230
|
|
|
|
|
|
|
printf "%20s %8d %8.2f %-13s%s\n", |
3231
|
|
|
|
|
|
|
$sheet->{Name}, |
3232
|
|
|
|
|
|
|
$sheet->{"Account Number"}, |
3233
|
|
|
|
|
|
|
$sheet->{Income}, |
3234
|
|
|
|
|
|
|
$sheet->{Home_phone}, |
3235
|
|
|
|
|
|
|
$sheet->{Email} ; |
3236
|
|
|
|
|
|
|
}); |
3237
|
|
|
|
|
|
|
|
3238
|
|
|
|
|
|
|
# Another way: |
3239
|
|
|
|
|
|
|
$sheet->apply( sub{ |
3240
|
|
|
|
|
|
|
my $r = $sheet->crow(); |
3241
|
|
|
|
|
|
|
printf "%20s %8d %8.2f %-13s%s\n", |
3242
|
|
|
|
|
|
|
$r->{Name}, $r->{"Account Number"}, $r->{Income}, |
3243
|
|
|
|
|
|
|
$r->{Home_phone}, $r->{Email} ; |
3244
|
|
|
|
|
|
|
}); |
3245
|
|
|
|
|
|
|
|
3246
|
|
|
|
|
|
|
# Another way: |
3247
|
|
|
|
|
|
|
$sheet->apply( sub{ |
3248
|
|
|
|
|
|
|
my $r = $sheet->crow(); |
3249
|
|
|
|
|
|
|
printf "%20s %8d %8.2f %-13s%s\n", |
3250
|
|
|
|
|
|
|
$r->[0], $r->[1], $r->[4], $r->[3], $r->[2] ; |
3251
|
|
|
|
|
|
|
}); |
3252
|
|
|
|
|
|
|
|
3253
|
|
|
|
|
|
|
# Split the "Customer's Name" into separate FName and LName columns |
3254
|
|
|
|
|
|
|
$sheet->insert_cols('>Name', "FName", "LName"); |
3255
|
|
|
|
|
|
|
$sheet->apply( sub { |
3256
|
|
|
|
|
|
|
my $r = $sheet->crow(); |
3257
|
|
|
|
|
|
|
($r->{FName}, $r->{LName}) = ($r->{Name} =~ /(.*) (.*)/) |
3258
|
|
|
|
|
|
|
or die Spreadsheet::Edit::logmsg("Could not parse Name"); |
3259
|
|
|
|
|
|
|
}); |
3260
|
|
|
|
|
|
|
$sheet->delete_cols( "Name" ); |
3261
|
|
|
|
|
|
|
|
3262
|
|
|
|
|
|
|
|
3263
|
|
|
|
|
|
|
=head1 INTRODUCTION |
3264
|
|
|
|
|
|
|
|
3265
|
|
|
|
|
|
|
=over |
3266
|
|
|
|
|
|
|
|
3267
|
|
|
|
|
|
|
|
3268
|
|
|
|
|
|
|
Skip ahead to "FUNCTIONS" for a list of operations (OO methods are |
3269
|
|
|
|
|
|
|
named the same). |
3270
|
|
|
|
|
|
|
|
3271
|
|
|
|
|
|
|
=back |
3272
|
|
|
|
|
|
|
|
3273
|
|
|
|
|
|
|
Columns may be referenced by title without knowing their absolute positions. |
3274
|
|
|
|
|
|
|
Optionally global (package) variables may be tied to columns. |
3275
|
|
|
|
|
|
|
Tabular data can come |
3276
|
|
|
|
|
|
|
from Spreadsheets, CSV files, or your code, and be written similarly. |
3277
|
|
|
|
|
|
|
Both Functional and Object-Oriented (OO) APIs are provided. |
3278
|
|
|
|
|
|
|
|
3279
|
|
|
|
|
|
|
A table in memory is stored in a C<sheet> object, which contains |
3280
|
|
|
|
|
|
|
an array of rows, each of which is an array of cell values. |
3281
|
|
|
|
|
|
|
|
3282
|
|
|
|
|
|
|
Cells within a row may be accessed by name (e.g. column titles) or |
3283
|
|
|
|
|
|
|
by absolute position (letter code like "A", "B" etc. or zero-based index). |
3284
|
|
|
|
|
|
|
|
3285
|
|
|
|
|
|
|
The usual paradigm is to iterate over rows applying a function |
3286
|
|
|
|
|
|
|
to each, vaguely inspired by 'sed' and 'awk' (see C<apply> below). |
3287
|
|
|
|
|
|
|
Random access is also supported. |
3288
|
|
|
|
|
|
|
|
3289
|
|
|
|
|
|
|
Tied variables can be used with the functional API. These variables |
3290
|
|
|
|
|
|
|
refer to columns in the current row during a C<apply> operation. |
3291
|
|
|
|
|
|
|
|
3292
|
|
|
|
|
|
|
Note: Only cell I<values> are handled; there is no provision |
3293
|
|
|
|
|
|
|
for processing formatting information from spreadsheets. |
3294
|
|
|
|
|
|
|
The author has a notion to add support for formats, |
3295
|
|
|
|
|
|
|
perhaps integrating with Spreadsheet::Read and Spreadsheet::Write |
3296
|
|
|
|
|
|
|
or the packages they use. Please contact the author if you want to help. |
3297
|
|
|
|
|
|
|
|
3298
|
|
|
|
|
|
|
=head3 HOW TO IMPORT |
3299
|
|
|
|
|
|
|
|
3300
|
|
|
|
|
|
|
By default only functions are imported, but most people will |
3301
|
|
|
|
|
|
|
|
3302
|
|
|
|
|
|
|
use Spreadsheet::Edit ':all'; |
3303
|
|
|
|
|
|
|
|
3304
|
|
|
|
|
|
|
to import functions and helper variables (see STANDARD SHEET VARIABLES |
3305
|
|
|
|
|
|
|
and VARIABLES USED DURING APPLY). |
3306
|
|
|
|
|
|
|
|
3307
|
|
|
|
|
|
|
You can rename imported items using the '-as' notation shown in |
3308
|
|
|
|
|
|
|
L<Exporter::Tiny::Manual::QuickStart>. |
3309
|
|
|
|
|
|
|
|
3310
|
|
|
|
|
|
|
Purely-OO applications can L<use Spreadsheet::Edit ();>. |
3311
|
|
|
|
|
|
|
|
3312
|
|
|
|
|
|
|
=head1 THE 'CURRENT SHEET' |
3313
|
|
|
|
|
|
|
|
3314
|
|
|
|
|
|
|
The I<Functions> and helper variables implicitly operate on a |
3315
|
|
|
|
|
|
|
package-global "current sheet" object, which can be switched at will. |
3316
|
|
|
|
|
|
|
OO I<Methods> operate on the C<sheet> object they are called on. |
3317
|
|
|
|
|
|
|
|
3318
|
|
|
|
|
|
|
Functions which operates on the "current sheet" have |
3319
|
|
|
|
|
|
|
corresponding OO methods with the same names and arguments |
3320
|
|
|
|
|
|
|
(note that method args must be enclosed by parenthesis). |
3321
|
|
|
|
|
|
|
|
3322
|
|
|
|
|
|
|
=head1 TIED COLUMN VARIABLES |
3323
|
|
|
|
|
|
|
|
3324
|
|
|
|
|
|
|
Package variables can refer directly to columns in the 'current sheet' |
3325
|
|
|
|
|
|
|
during C<apply>. For example C<$Email> and C<$FName> in |
3326
|
|
|
|
|
|
|
the SYNOPSIS above. |
3327
|
|
|
|
|
|
|
|
3328
|
|
|
|
|
|
|
See C<tie_column_vars> for details. |
3329
|
|
|
|
|
|
|
|
3330
|
|
|
|
|
|
|
=head1 THE FUNCTIONS |
3331
|
|
|
|
|
|
|
|
3332
|
|
|
|
|
|
|
In the following, {OPTIONS} refers to an optional first argument |
3333
|
|
|
|
|
|
|
which, if present, is a hashref giving additional parameters. |
3334
|
|
|
|
|
|
|
For example in |
3335
|
|
|
|
|
|
|
|
3336
|
|
|
|
|
|
|
read_spreadsheet {sheetname => 'Sheet1'}, '/path/to/file.xlsx'; |
3337
|
|
|
|
|
|
|
|
3338
|
|
|
|
|
|
|
the {...} hash is optional and specifies the sheet name. |
3339
|
|
|
|
|
|
|
|
3340
|
|
|
|
|
|
|
=head2 $curr_sheet = sheet ; |
3341
|
|
|
|
|
|
|
|
3342
|
|
|
|
|
|
|
=head2 $prev_sheet = sheet $another_sheet ; |
3343
|
|
|
|
|
|
|
|
3344
|
|
|
|
|
|
|
=head2 $prev_sheet = sheet undef ; |
3345
|
|
|
|
|
|
|
|
3346
|
|
|
|
|
|
|
[Functional API only] |
3347
|
|
|
|
|
|
|
Retrieve, change, or forget the 'current sheet' object used |
3348
|
|
|
|
|
|
|
by the functional API. |
3349
|
|
|
|
|
|
|
|
3350
|
|
|
|
|
|
|
Changing the current sheet immediately changes what is referenced by |
3351
|
|
|
|
|
|
|
tied column variables and STANDARD SHEET VARIABLES (described later). |
3352
|
|
|
|
|
|
|
|
3353
|
|
|
|
|
|
|
{OPTIONS} may specify C<< package => 'pkgname' >> to operate on the specified |
3354
|
|
|
|
|
|
|
package instead of the caller's package (useful in library code). |
3355
|
|
|
|
|
|
|
|
3356
|
|
|
|
|
|
|
=head2 read_spreadsheet CSVFILEPATH |
3357
|
|
|
|
|
|
|
|
3358
|
|
|
|
|
|
|
=head2 read_spreadsheet SPREADSHEETPATH |
3359
|
|
|
|
|
|
|
|
3360
|
|
|
|
|
|
|
=head2 read_spreadsheet "SPREADSHEETPATH!SHEETNAME" |
3361
|
|
|
|
|
|
|
|
3362
|
|
|
|
|
|
|
Replace any existing data with content from the given file. |
3363
|
|
|
|
|
|
|
|
3364
|
|
|
|
|
|
|
The Functional API will create a new sheet object if |
3365
|
|
|
|
|
|
|
there is no "current sheet". |
3366
|
|
|
|
|
|
|
|
3367
|
|
|
|
|
|
|
The file may be a .csv or any format supported by Libre Office or gnumeric. |
3368
|
|
|
|
|
|
|
|
3369
|
|
|
|
|
|
|
By default column titles are auto-detected and |
3370
|
|
|
|
|
|
|
an exception is thrown if a plausible title row can not be found. |
3371
|
|
|
|
|
|
|
|
3372
|
|
|
|
|
|
|
{OPTIONS} may include: |
3373
|
|
|
|
|
|
|
|
3374
|
|
|
|
|
|
|
Auto-detection options: |
3375
|
|
|
|
|
|
|
|
3376
|
|
|
|
|
|
|
=over 2 |
3377
|
|
|
|
|
|
|
|
3378
|
|
|
|
|
|
|
required => COLSPEC or [COLSPEC,...] # any required title(s) |
3379
|
|
|
|
|
|
|
min_rx => NUM, # first rx which may contain the title row. |
3380
|
|
|
|
|
|
|
max_rx => NUM, # maximum rx which may contain the title row. |
3381
|
|
|
|
|
|
|
first_cx => NUM, # first column ix which must contain required titles |
3382
|
|
|
|
|
|
|
last_cx => NUM, # last column ix which must contain required titles |
3383
|
|
|
|
|
|
|
|
3384
|
|
|
|
|
|
|
=back |
3385
|
|
|
|
|
|
|
|
3386
|
|
|
|
|
|
|
The first row is used which includes the C<required> title(s), if any, |
3387
|
|
|
|
|
|
|
and has non-empty titles in all columns, or columns |
3388
|
|
|
|
|
|
|
C<first_cx> through C<last_cx>. |
3389
|
|
|
|
|
|
|
|
3390
|
|
|
|
|
|
|
=over 2 |
3391
|
|
|
|
|
|
|
|
3392
|
|
|
|
|
|
|
title_rx => rx # specify title row (first row is 0) |
3393
|
|
|
|
|
|
|
title_rx => undef # no title row |
3394
|
|
|
|
|
|
|
|
3395
|
|
|
|
|
|
|
=back |
3396
|
|
|
|
|
|
|
|
3397
|
|
|
|
|
|
|
These disable auto-detection and explicitly specify the title row |
3398
|
|
|
|
|
|
|
index, or with C<undef>, that there are no titles. |
3399
|
|
|
|
|
|
|
See also the C<title_rx> function/method. |
3400
|
|
|
|
|
|
|
|
3401
|
|
|
|
|
|
|
Other options: |
3402
|
|
|
|
|
|
|
|
3403
|
|
|
|
|
|
|
=over 6 |
3404
|
|
|
|
|
|
|
|
3405
|
|
|
|
|
|
|
=item sheetname => SHEETNAME |
3406
|
|
|
|
|
|
|
|
3407
|
|
|
|
|
|
|
Specify which sheet in a multi-sheet workbook (i.e. spreadsheet file) to read. |
3408
|
|
|
|
|
|
|
Alternatively, the sheet name may be appended to the input |
3409
|
|
|
|
|
|
|
path after '!' as shown in the example. |
3410
|
|
|
|
|
|
|
|
3411
|
|
|
|
|
|
|
If no SHEETNAME is given then the sheet which was "active" when the |
3412
|
|
|
|
|
|
|
workbook was saved will be retrieved. |
3413
|
|
|
|
|
|
|
|
3414
|
|
|
|
|
|
|
=item silent => bool |
3415
|
|
|
|
|
|
|
|
3416
|
|
|
|
|
|
|
=item verbose => bool |
3417
|
|
|
|
|
|
|
|
3418
|
|
|
|
|
|
|
=item debug => bool |
3419
|
|
|
|
|
|
|
|
3420
|
|
|
|
|
|
|
Probably what you expect. |
3421
|
|
|
|
|
|
|
|
3422
|
|
|
|
|
|
|
=item Other C<< key => value >> pairs override details of CSV parsing. |
3423
|
|
|
|
|
|
|
|
3424
|
|
|
|
|
|
|
See L<Text::CSV>. UTF-8 encoding is assumed by default. |
3425
|
|
|
|
|
|
|
|
3426
|
|
|
|
|
|
|
=back |
3427
|
|
|
|
|
|
|
|
3428
|
|
|
|
|
|
|
Due to bugs in Libre/Open Office, spreadsheet files can not |
3429
|
|
|
|
|
|
|
be read if LO/OO is currently running, even |
3430
|
|
|
|
|
|
|
for unrelated purposes (see "BUGS"). |
3431
|
|
|
|
|
|
|
This problem does not occur with .csv files |
3432
|
|
|
|
|
|
|
|
3433
|
|
|
|
|
|
|
=head2 alias IDENT => COLSPEC, ... ; |
3434
|
|
|
|
|
|
|
|
3435
|
|
|
|
|
|
|
=head2 alias IDENT => qr/regexp/, ... ; |
3436
|
|
|
|
|
|
|
|
3437
|
|
|
|
|
|
|
Create alternate identifiers for specified columns. |
3438
|
|
|
|
|
|
|
|
3439
|
|
|
|
|
|
|
Each IDENT, which must be a valid Perl identifier, will henceforth |
3440
|
|
|
|
|
|
|
refer to the specified column even if the identifier is the same |
3441
|
|
|
|
|
|
|
as the title or letter code of a different column. |
3442
|
|
|
|
|
|
|
|
3443
|
|
|
|
|
|
|
C<$row{IDENT}> and a tied variable C<$IDENT> will refer to the specified column. |
3444
|
|
|
|
|
|
|
|
3445
|
|
|
|
|
|
|
Aliases automatically track the column if it's position changes. |
3446
|
|
|
|
|
|
|
|
3447
|
|
|
|
|
|
|
Regular Expressions are matched against titles only, and an exception is |
3448
|
|
|
|
|
|
|
thrown if more than one title matches. |
3449
|
|
|
|
|
|
|
|
3450
|
|
|
|
|
|
|
Otherwise, COLSPECs may be titles, existing alias names, column letters, etc. |
3451
|
|
|
|
|
|
|
(see "COLUMN SPECIFIERS" for details). |
3452
|
|
|
|
|
|
|
|
3453
|
|
|
|
|
|
|
The COLSPEC is evaluated before the alias is created, so |
3454
|
|
|
|
|
|
|
|
3455
|
|
|
|
|
|
|
alias B => "B"; |
3456
|
|
|
|
|
|
|
|
3457
|
|
|
|
|
|
|
would make "B" henceforth refer to the column with title "B", if one exists, |
3458
|
|
|
|
|
|
|
or the second column (treating COLSPEC as a letter-code), even if the column |
3459
|
|
|
|
|
|
|
is later moved. |
3460
|
|
|
|
|
|
|
|
3461
|
|
|
|
|
|
|
RETURNS: The 0-based column indices of the aliased column(s). |
3462
|
|
|
|
|
|
|
|
3463
|
|
|
|
|
|
|
=head2 unalias IDENT, ... ; |
3464
|
|
|
|
|
|
|
|
3465
|
|
|
|
|
|
|
Forget alias(es). Any masked COLSPECs become usable again. |
3466
|
|
|
|
|
|
|
|
3467
|
|
|
|
|
|
|
=head2 tie_column_vars VARNAME, ... |
3468
|
|
|
|
|
|
|
|
3469
|
|
|
|
|
|
|
Create tied package variables (scalars) for use during C<apply>. |
3470
|
|
|
|
|
|
|
|
3471
|
|
|
|
|
|
|
Each variable is a scalar corresponding to a column, and reading or writing |
3472
|
|
|
|
|
|
|
it accesses the corresponding cell in the row being visited during C<apply>. |
3473
|
|
|
|
|
|
|
|
3474
|
|
|
|
|
|
|
The '$' may be omitted in the VARNAME arguments to C<tie_column_vars>; |
3475
|
|
|
|
|
|
|
|
3476
|
|
|
|
|
|
|
You must separately declare these variables with C<our $NAME>, |
3477
|
|
|
|
|
|
|
except in the special case described |
3478
|
|
|
|
|
|
|
in "Use in BEGIN() or module import methods" later. |
3479
|
|
|
|
|
|
|
|
3480
|
|
|
|
|
|
|
The variable name itself implies the column it refers to. |
3481
|
|
|
|
|
|
|
|
3482
|
|
|
|
|
|
|
Variable names may be: |
3483
|
|
|
|
|
|
|
|
3484
|
|
|
|
|
|
|
=over |
3485
|
|
|
|
|
|
|
|
3486
|
|
|
|
|
|
|
=item * User-defined alias names (see "alias") |
3487
|
|
|
|
|
|
|
|
3488
|
|
|
|
|
|
|
=item * Titles which happen to be valid Perl identifiers |
3489
|
|
|
|
|
|
|
|
3490
|
|
|
|
|
|
|
=item * Identifiers derived from titles by replacing offending characters |
3491
|
|
|
|
|
|
|
with underscrores (see "AUTOMATIC ALIASES"), |
3492
|
|
|
|
|
|
|
|
3493
|
|
|
|
|
|
|
=item * Spreadsheet column letters like "A", "B" etc. |
3494
|
|
|
|
|
|
|
|
3495
|
|
|
|
|
|
|
=back |
3496
|
|
|
|
|
|
|
|
3497
|
|
|
|
|
|
|
See "CONFLICT RESOLUTION" for how ambiguity is resolved. |
3498
|
|
|
|
|
|
|
|
3499
|
|
|
|
|
|
|
Multiple calls accumulate, including with different sheets. |
3500
|
|
|
|
|
|
|
|
3501
|
|
|
|
|
|
|
Variable bindings are dynamically evaluated during each access by using the |
3502
|
|
|
|
|
|
|
variable's identifier as a COLSPEC with the 'current sheet' in your package. |
3503
|
|
|
|
|
|
|
This means that it does not matter which sheet |
3504
|
|
|
|
|
|
|
was 'current' when C<tie_column_vars> was called with a particular name; |
3505
|
|
|
|
|
|
|
it only matters that the name of a tied variable is a valid COLSPEC in |
3506
|
|
|
|
|
|
|
the 'current sheet' when that variable is referenced |
3507
|
|
|
|
|
|
|
(otherwise a read returns I<undef> and a write throws an exception). |
3508
|
|
|
|
|
|
|
[*Need clarification*] |
3509
|
|
|
|
|
|
|
|
3510
|
|
|
|
|
|
|
B<{OPTIONS}> may specify: |
3511
|
|
|
|
|
|
|
|
3512
|
|
|
|
|
|
|
=over |
3513
|
|
|
|
|
|
|
|
3514
|
|
|
|
|
|
|
=item package => "pkgname" |
3515
|
|
|
|
|
|
|
|
3516
|
|
|
|
|
|
|
Tie variables in the specified package instead of the caller's package. |
3517
|
|
|
|
|
|
|
|
3518
|
|
|
|
|
|
|
=item verbose => bool |
3519
|
|
|
|
|
|
|
|
3520
|
|
|
|
|
|
|
=item debug => bool |
3521
|
|
|
|
|
|
|
|
3522
|
|
|
|
|
|
|
Print trace messages. |
3523
|
|
|
|
|
|
|
|
3524
|
|
|
|
|
|
|
=back |
3525
|
|
|
|
|
|
|
|
3526
|
|
|
|
|
|
|
=head2 tie_column_vars ':all' |
3527
|
|
|
|
|
|
|
|
3528
|
|
|
|
|
|
|
With the B<:all> token I<all possible variables> are tied, corresponding |
3529
|
|
|
|
|
|
|
to the aliases, titles, non-conflicting column letters etc. which exist |
3530
|
|
|
|
|
|
|
for the current sheet. |
3531
|
|
|
|
|
|
|
|
3532
|
|
|
|
|
|
|
In addition, variables will be tied in the future I<whenever new identifiers |
3533
|
|
|
|
|
|
|
become valid> (for example when a new C<alias> is created, column added, |
3534
|
|
|
|
|
|
|
or another file is read into the same sheet). |
3535
|
|
|
|
|
|
|
|
3536
|
|
|
|
|
|
|
Although convenient this is B<insecure> because malicious |
3537
|
|
|
|
|
|
|
titles could clobber unintended globals. |
3538
|
|
|
|
|
|
|
|
3539
|
|
|
|
|
|
|
If VARNAMES are also specified, those variables will be tied |
3540
|
|
|
|
|
|
|
immediately even if not yet usable; an exception occurs if a tied variable |
3541
|
|
|
|
|
|
|
is referenced before the corresponding alias or title exists. |
3542
|
|
|
|
|
|
|
[*Need clarification* -- exception even for reads??] |
3543
|
|
|
|
|
|
|
|
3544
|
|
|
|
|
|
|
=head2 Use in BEGIN{} or module import methods |
3545
|
|
|
|
|
|
|
|
3546
|
|
|
|
|
|
|
C<tie_column_vars> B<imports> the tied variables into your module, |
3547
|
|
|
|
|
|
|
or the module specified with package => "pkgname" in {OPTIONS}. |
3548
|
|
|
|
|
|
|
|
3549
|
|
|
|
|
|
|
It is unnecessary to declare tied variables if the import |
3550
|
|
|
|
|
|
|
occurs before code is compiled which references the variables. This can |
3551
|
|
|
|
|
|
|
be the case if C<tie_column_vars> is called in a BEGIN{} block or in the |
3552
|
|
|
|
|
|
|
C<import> method of a module loaded with C<use>. |
3553
|
|
|
|
|
|
|
|
3554
|
|
|
|
|
|
|
L<Spreadsheet::Edit::Preload> makes use of this. |
3555
|
|
|
|
|
|
|
|
3556
|
|
|
|
|
|
|
=head2 $rowindex = title_rx ; |
3557
|
|
|
|
|
|
|
|
3558
|
|
|
|
|
|
|
Retrieve the current title row rx, or undef if there are no titles |
3559
|
|
|
|
|
|
|
|
3560
|
|
|
|
|
|
|
=head2 title_rx ROWINDEX ; |
3561
|
|
|
|
|
|
|
|
3562
|
|
|
|
|
|
|
Make the specified row be the title row. Titles in that row are |
3563
|
|
|
|
|
|
|
immediately (re-)examined and the corresponding COLSPECs become valid, |
3564
|
|
|
|
|
|
|
e.g. you can reference a column by it's title or a derived identifier. |
3565
|
|
|
|
|
|
|
|
3566
|
|
|
|
|
|
|
Note: Setting C<title_rx> this way is rarely needed because |
3567
|
|
|
|
|
|
|
by default C<read_spreadsheet> sets the title row. |
3568
|
|
|
|
|
|
|
|
3569
|
|
|
|
|
|
|
=head2 title_rx undef ; |
3570
|
|
|
|
|
|
|
|
3571
|
|
|
|
|
|
|
Titles are disabled and any existing COLSPECs derived from |
3572
|
|
|
|
|
|
|
titles are invalidated. |
3573
|
|
|
|
|
|
|
|
3574
|
|
|
|
|
|
|
=head2 title_rx {AUTODETECT_OPTIONS} 'auto'; |
3575
|
|
|
|
|
|
|
|
3576
|
|
|
|
|
|
|
Immediately perform auto-detection of the title row using |
3577
|
|
|
|
|
|
|
C<< {AUTODETECT_OPTIONS} >> to modify any existing auto-detect options |
3578
|
|
|
|
|
|
|
(see C<read_spreadsheet> for a description of the options). |
3579
|
|
|
|
|
|
|
|
3580
|
|
|
|
|
|
|
|
3581
|
|
|
|
|
|
|
=head2 apply {code} [COLSPEC*] ; |
3582
|
|
|
|
|
|
|
|
3583
|
|
|
|
|
|
|
=head2 apply_all {code} [COLSPEC*] ; |
3584
|
|
|
|
|
|
|
|
3585
|
|
|
|
|
|
|
=head2 apply_torx {code} RX-OR-RXLIST [,COLSPEC*] ; |
3586
|
|
|
|
|
|
|
|
3587
|
|
|
|
|
|
|
=head2 apply_exceptrx {code} RX-OR-RXLIST [,COLSPEC*] ; |
3588
|
|
|
|
|
|
|
|
3589
|
|
|
|
|
|
|
Execute the specified code block (or referenced sub) once for each row. |
3590
|
|
|
|
|
|
|
|
3591
|
|
|
|
|
|
|
Note that there is no comma after a bare {code} block. |
3592
|
|
|
|
|
|
|
|
3593
|
|
|
|
|
|
|
While executing your code, tied column variables and |
3594
|
|
|
|
|
|
|
the sheet variables C<@crow>, C<%crow>, C<$rx> and C<$linenum> |
3595
|
|
|
|
|
|
|
and corresponding OO methods will refer to the row being visited. |
3596
|
|
|
|
|
|
|
|
3597
|
|
|
|
|
|
|
If a list of COLSPECs is specified, then |
3598
|
|
|
|
|
|
|
|
3599
|
|
|
|
|
|
|
=over 2 |
3600
|
|
|
|
|
|
|
|
3601
|
|
|
|
|
|
|
@_ is bound to the columns in the order specified |
3602
|
|
|
|
|
|
|
$_ is bound to the first such column |
3603
|
|
|
|
|
|
|
|
3604
|
|
|
|
|
|
|
=back |
3605
|
|
|
|
|
|
|
|
3606
|
|
|
|
|
|
|
C<apply> normally visits all rows which follow the title row, or all rows |
3607
|
|
|
|
|
|
|
if there is no title row. |
3608
|
|
|
|
|
|
|
C<first_data_rx> and C<last_data_rx>, if defined, further limit the |
3609
|
|
|
|
|
|
|
range visited. |
3610
|
|
|
|
|
|
|
|
3611
|
|
|
|
|
|
|
C<apply_all> unconditionally visits every row, including any title row. |
3612
|
|
|
|
|
|
|
|
3613
|
|
|
|
|
|
|
C<apply_torx> or C<apply_exceptrx> visit exactly the indicated rows. |
3614
|
|
|
|
|
|
|
RX-OR-RXLIST may be either a single row index or a [list of rx]; |
3615
|
|
|
|
|
|
|
|
3616
|
|
|
|
|
|
|
Rows may be safely inserted or deleted during 'apply'; |
3617
|
|
|
|
|
|
|
rows inserted after the currently-being-visited row will be visited |
3618
|
|
|
|
|
|
|
at the proper time. |
3619
|
|
|
|
|
|
|
|
3620
|
|
|
|
|
|
|
Nested and recursive C<apply>s are allowed. |
3621
|
|
|
|
|
|
|
When an 'apply' changes the 'current sheet', |
3622
|
|
|
|
|
|
|
tied variables then refer to the other sheet and |
3623
|
|
|
|
|
|
|
any C<apply> active for that sheet. |
3624
|
|
|
|
|
|
|
With nested 'apply's, take care restore the original sheet before returning |
3625
|
|
|
|
|
|
|
(perhaps using C<Guard::scope_guard>). |
3626
|
|
|
|
|
|
|
|
3627
|
|
|
|
|
|
|
B<MAGIC VARIABLES USED DURING APPLY> |
3628
|
|
|
|
|
|
|
|
3629
|
|
|
|
|
|
|
These variables refer to the row currently being visited: |
3630
|
|
|
|
|
|
|
|
3631
|
|
|
|
|
|
|
=over 2 |
3632
|
|
|
|
|
|
|
|
3633
|
|
|
|
|
|
|
B<@crow> is an array aliased to the current row's cells. |
3634
|
|
|
|
|
|
|
|
3635
|
|
|
|
|
|
|
B<%crow> is a hash aliased to the same cells, |
3636
|
|
|
|
|
|
|
indexed by alias, title, letter code, etc. (any COLSPEC). |
3637
|
|
|
|
|
|
|
|
3638
|
|
|
|
|
|
|
B<$rx> is the 0-based index of the current row. |
3639
|
|
|
|
|
|
|
|
3640
|
|
|
|
|
|
|
B<$linenum> is the starting line number of the current row if the |
3641
|
|
|
|
|
|
|
data came from a .csv file. |
3642
|
|
|
|
|
|
|
|
3643
|
|
|
|
|
|
|
For example, the "Account Number" column in the SYNOPSIS may be accessed |
3644
|
|
|
|
|
|
|
many ways: |
3645
|
|
|
|
|
|
|
|
3646
|
|
|
|
|
|
|
alias AcNum => "Account Number"; |
3647
|
|
|
|
|
|
|
apply { |
3648
|
|
|
|
|
|
|
|
3649
|
|
|
|
|
|
|
$crow{"Account Number"} # %crow indexed by title |
3650
|
|
|
|
|
|
|
$crow{AcNum} # using an explicit alias |
3651
|
|
|
|
|
|
|
$crow{Account_Number} # using the AUTOMATIC ALIAS |
3652
|
|
|
|
|
|
|
|
3653
|
|
|
|
|
|
|
$crow[ $colx{"Account Number"} ]; # @crow indexed by a 0-based index |
3654
|
|
|
|
|
|
|
$crow[ $colx{"AcNum"} ]; # ...obtained from %colx |
3655
|
|
|
|
|
|
|
$crow[ $colx{"Account_Number"} ]; # |
3656
|
|
|
|
|
|
|
|
3657
|
|
|
|
|
|
|
$rows[$rx]->[ $colx{Account_Number} ] # Directly accessing @rows |
3658
|
|
|
|
|
|
|
|
3659
|
|
|
|
|
|
|
# See "TIED COLUMN VARIABLES" for a sweeter alternative |
3660
|
|
|
|
|
|
|
}; |
3661
|
|
|
|
|
|
|
|
3662
|
|
|
|
|
|
|
=back |
3663
|
|
|
|
|
|
|
|
3664
|
|
|
|
|
|
|
=head2 delete_col COLSPEC ; |
3665
|
|
|
|
|
|
|
|
3666
|
|
|
|
|
|
|
=head2 delete_cols COLSPEC+ ; |
3667
|
|
|
|
|
|
|
|
3668
|
|
|
|
|
|
|
The indicated columns are removed. Remaining title bindings |
3669
|
|
|
|
|
|
|
are adjusted to track shifted columns. |
3670
|
|
|
|
|
|
|
|
3671
|
|
|
|
|
|
|
=head2 only_cols COLSPEC+ ; |
3672
|
|
|
|
|
|
|
|
3673
|
|
|
|
|
|
|
All columns I<except> the specified columns are deleted. |
3674
|
|
|
|
|
|
|
|
3675
|
|
|
|
|
|
|
=head2 move_col POSITION, SOURCE ; |
3676
|
|
|
|
|
|
|
|
3677
|
|
|
|
|
|
|
=head2 move_cols POSITION, SOURCES... ; |
3678
|
|
|
|
|
|
|
|
3679
|
|
|
|
|
|
|
Relocate the indicated column(s) (C<SOURCES>) so they are adjacent, in |
3680
|
|
|
|
|
|
|
the order specified, starting at the position C<POSITION>. |
3681
|
|
|
|
|
|
|
|
3682
|
|
|
|
|
|
|
POSITION may be ">COLSPEC" to place moved column(s) |
3683
|
|
|
|
|
|
|
immediately after the indicated column (">$" to place at the end), |
3684
|
|
|
|
|
|
|
or POSITION may directly specify the destination column |
3685
|
|
|
|
|
|
|
using an unadorned COLSPEC. |
3686
|
|
|
|
|
|
|
|
3687
|
|
|
|
|
|
|
A non-absolute COLSPEC indicates the I<initial> position of the referenced column. |
3688
|
|
|
|
|
|
|
|
3689
|
|
|
|
|
|
|
=head2 insert_col POSITION, newtitle ; |
3690
|
|
|
|
|
|
|
|
3691
|
|
|
|
|
|
|
=head2 insert_cols POSITION, newtitles... ; |
3692
|
|
|
|
|
|
|
|
3693
|
|
|
|
|
|
|
One or more columns are created starting at a position |
3694
|
|
|
|
|
|
|
specified the same way as in C<move_cols> (later columns |
3695
|
|
|
|
|
|
|
are moved rightward). |
3696
|
|
|
|
|
|
|
|
3697
|
|
|
|
|
|
|
POSITION may be ">$" to place new column(s) at the far right. |
3698
|
|
|
|
|
|
|
|
3699
|
|
|
|
|
|
|
A new title must be specified for each new column. |
3700
|
|
|
|
|
|
|
If there is no title row, specify C<undef> for each position. |
3701
|
|
|
|
|
|
|
|
3702
|
|
|
|
|
|
|
Returns the new column index or indices. |
3703
|
|
|
|
|
|
|
|
3704
|
|
|
|
|
|
|
=head2 split_col {code} COLSPEC, POSITION, newtitles... ; |
3705
|
|
|
|
|
|
|
|
3706
|
|
|
|
|
|
|
New columns are created starting at POSITION as with C<insert_cols>, |
3707
|
|
|
|
|
|
|
and populated with data from column COLSPEC. |
3708
|
|
|
|
|
|
|
|
3709
|
|
|
|
|
|
|
C<{code}> is called for each row with $_ bound to the cell at COLSPEC |
3710
|
|
|
|
|
|
|
and @_ bound to cell(s) in the new column(s). It is up to your code to |
3711
|
|
|
|
|
|
|
read the old column ($_) and write into the new columns (@_). |
3712
|
|
|
|
|
|
|
|
3713
|
|
|
|
|
|
|
The old column is left as-is (not deleted). |
3714
|
|
|
|
|
|
|
|
3715
|
|
|
|
|
|
|
If there is no title row, specify C<undef> for each new title. |
3716
|
|
|
|
|
|
|
|
3717
|
|
|
|
|
|
|
=head2 sort_rows {rx cmp function} |
3718
|
|
|
|
|
|
|
|
3719
|
|
|
|
|
|
|
=head2 sort_rows {rx cmp function} $first_rx, $last_rx |
3720
|
|
|
|
|
|
|
|
3721
|
|
|
|
|
|
|
If no range is specified, then the range is the |
3722
|
|
|
|
|
|
|
same as for C<apply> (namely: All rows after the title row unless |
3723
|
|
|
|
|
|
|
limited by B<first_data_rx> .. B<last_data_rx>). |
3724
|
|
|
|
|
|
|
|
3725
|
|
|
|
|
|
|
In the comparison function, globals $a and $b will contain row objects, which |
3726
|
|
|
|
|
|
|
are dual-typed to act as either an array or hash ref to the cells |
3727
|
|
|
|
|
|
|
in their row. The corresponding original row indicies are also passed |
3728
|
|
|
|
|
|
|
as parameters in C<@_>. |
3729
|
|
|
|
|
|
|
|
3730
|
|
|
|
|
|
|
Rows are not actually moved until after all comparisons have finished. |
3731
|
|
|
|
|
|
|
|
3732
|
|
|
|
|
|
|
RETURNS: A list of the previous row indicies of all rows in the sheet. |
3733
|
|
|
|
|
|
|
|
3734
|
|
|
|
|
|
|
# Sort on the "LName" column using row indicies |
3735
|
|
|
|
|
|
|
# (contrast with the example in SYNOPSIS which uses $a and $b) |
3736
|
|
|
|
|
|
|
sort_rows { my ($rxa, $rxb) = @_; |
3737
|
|
|
|
|
|
|
$rows[$rxa]{LName} cmp $rows[$rxb]{LName} |
3738
|
|
|
|
|
|
|
}; |
3739
|
|
|
|
|
|
|
|
3740
|
|
|
|
|
|
|
=head2 rename_cols COLSPEC, "new title", ... ; |
3741
|
|
|
|
|
|
|
|
3742
|
|
|
|
|
|
|
Multiple pairs may be given. Title cell(s) are updated as indicated. |
3743
|
|
|
|
|
|
|
|
3744
|
|
|
|
|
|
|
Existing user-defined aliases are I<not> affected, i.e., |
3745
|
|
|
|
|
|
|
they continue to refer to the same columns as before even if the titles changed. |
3746
|
|
|
|
|
|
|
|
3747
|
|
|
|
|
|
|
=head2 join_cols_sep STRING COLSPEC+ ; |
3748
|
|
|
|
|
|
|
|
3749
|
|
|
|
|
|
|
=head2 join_cols {code} COLSPEC+ ; |
3750
|
|
|
|
|
|
|
|
3751
|
|
|
|
|
|
|
The specified columns are combined into the first-specified column and the other |
3752
|
|
|
|
|
|
|
columns are deleted. |
3753
|
|
|
|
|
|
|
|
3754
|
|
|
|
|
|
|
The first argument of C<join_cols_sep> should be a fixed separator. |
3755
|
|
|
|
|
|
|
The first argument of C<join_cols> may be a {code} block or subref; |
3756
|
|
|
|
|
|
|
|
3757
|
|
|
|
|
|
|
If a separator string is specified it is used to join column content |
3758
|
|
|
|
|
|
|
together. |
3759
|
|
|
|
|
|
|
|
3760
|
|
|
|
|
|
|
If a {code} block or sub ref is specified, |
3761
|
|
|
|
|
|
|
it is executed once for each row following the title row, |
3762
|
|
|
|
|
|
|
with $_ bound to the first-named column, i.e. the surviving column, |
3763
|
|
|
|
|
|
|
and @_ bound to all named columns in the order given. |
3764
|
|
|
|
|
|
|
|
3765
|
|
|
|
|
|
|
It is up to your code to combine the data by reading |
3766
|
|
|
|
|
|
|
@_ and writing $_ (or, equivalently, by writing $_[0]). |
3767
|
|
|
|
|
|
|
|
3768
|
|
|
|
|
|
|
C<first_data_rx> and C<last_data_rx> are ignored, and the title |
3769
|
|
|
|
|
|
|
is I<not> modified. |
3770
|
|
|
|
|
|
|
|
3771
|
|
|
|
|
|
|
=head2 reverse_cols |
3772
|
|
|
|
|
|
|
|
3773
|
|
|
|
|
|
|
The order of the columns is reversed. |
3774
|
|
|
|
|
|
|
|
3775
|
|
|
|
|
|
|
=head2 insert_row |
3776
|
|
|
|
|
|
|
|
3777
|
|
|
|
|
|
|
=head2 insert_row 'END' [,$count] |
3778
|
|
|
|
|
|
|
|
3779
|
|
|
|
|
|
|
=head2 insert_rows $rowx [,$count] |
3780
|
|
|
|
|
|
|
|
3781
|
|
|
|
|
|
|
Insert one or more empty rows at the indicated position |
3782
|
|
|
|
|
|
|
(default: at end). C<$rowx>, if specified, is either a 0-based offset |
3783
|
|
|
|
|
|
|
for the new row or 'END' to add the new row(s) at the end. |
3784
|
|
|
|
|
|
|
Returns the index of the first new row. |
3785
|
|
|
|
|
|
|
|
3786
|
|
|
|
|
|
|
=head2 delete_rows $rowx,... ; |
3787
|
|
|
|
|
|
|
|
3788
|
|
|
|
|
|
|
=head2 delete_rows 'LAST',... ; |
3789
|
|
|
|
|
|
|
|
3790
|
|
|
|
|
|
|
The indicated data rows are deleted. C<$rowx> is a zero-based row index |
3791
|
|
|
|
|
|
|
or the special token "LAST" to indicate the last row (same as C<$#rows>). |
3792
|
|
|
|
|
|
|
Any number of rows may be deleted in a single command, listed individually. |
3793
|
|
|
|
|
|
|
|
3794
|
|
|
|
|
|
|
=for Pod::Coverage delete_row |
3795
|
|
|
|
|
|
|
|
3796
|
|
|
|
|
|
|
=head2 transpose |
3797
|
|
|
|
|
|
|
|
3798
|
|
|
|
|
|
|
Invert the relation, i.e. rotate and flip the table. |
3799
|
|
|
|
|
|
|
Cells A1,B1,C1 etc. become A1,A2,A3 etc. |
3800
|
|
|
|
|
|
|
Any title_rx is forgotten. |
3801
|
|
|
|
|
|
|
|
3802
|
|
|
|
|
|
|
=head2 $href = read_workbook SPREADSHEETPATH |
3803
|
|
|
|
|
|
|
|
3804
|
|
|
|
|
|
|
**NOT YET IMPLEMENTED** |
3805
|
|
|
|
|
|
|
|
3806
|
|
|
|
|
|
|
[Function only, not callable as a method] |
3807
|
|
|
|
|
|
|
All sheets in the specified document are read into memory |
3808
|
|
|
|
|
|
|
without changing the 'current sheet'. A hashref is returned: |
3809
|
|
|
|
|
|
|
|
3810
|
|
|
|
|
|
|
{ |
3811
|
|
|
|
|
|
|
"sheet name" => (Spreadsheet::Edit object), |
3812
|
|
|
|
|
|
|
...for each sheet in the workbook... |
3813
|
|
|
|
|
|
|
} |
3814
|
|
|
|
|
|
|
|
3815
|
|
|
|
|
|
|
To access one of the workbook sheets, execute |
3816
|
|
|
|
|
|
|
|
3817
|
|
|
|
|
|
|
sheet $href->{"sheet name"}; # or call OO methods on it |
3818
|
|
|
|
|
|
|
|
3819
|
|
|
|
|
|
|
If SPREADSHEETPATH was a .csv file then the resulting hash will have only |
3820
|
|
|
|
|
|
|
one member with an indeterminate key. |
3821
|
|
|
|
|
|
|
|
3822
|
|
|
|
|
|
|
=head2 new_sheet |
3823
|
|
|
|
|
|
|
|
3824
|
|
|
|
|
|
|
[functional API only] |
3825
|
|
|
|
|
|
|
Create a new empty sheet and make it the 'current sheet', returning the |
3826
|
|
|
|
|
|
|
sheet object. |
3827
|
|
|
|
|
|
|
|
3828
|
|
|
|
|
|
|
Rarely used because a new sheet is automatically created by |
3829
|
|
|
|
|
|
|
C<read_spreadsheet> if your package has no current sheet. |
3830
|
|
|
|
|
|
|
|
3831
|
|
|
|
|
|
|
{OPTIONS} may include: |
3832
|
|
|
|
|
|
|
|
3833
|
|
|
|
|
|
|
=over 6 |
3834
|
|
|
|
|
|
|
|
3835
|
|
|
|
|
|
|
=item data_source => "text..." |
3836
|
|
|
|
|
|
|
|
3837
|
|
|
|
|
|
|
This string will be returned by the C<data_source> method, |
3838
|
|
|
|
|
|
|
overriding any default. |
3839
|
|
|
|
|
|
|
|
3840
|
|
|
|
|
|
|
=item rows => [[A1_value,B1_value,...], [A2_value,B2_value,...], ...], |
3841
|
|
|
|
|
|
|
|
3842
|
|
|
|
|
|
|
=item linenums => [...] #optional |
3843
|
|
|
|
|
|
|
|
3844
|
|
|
|
|
|
|
This makes the C<sheet> object hold data already in memory. |
3845
|
|
|
|
|
|
|
The data should not be modified directly while the sheet C<object> exists. |
3846
|
|
|
|
|
|
|
|
3847
|
|
|
|
|
|
|
=item clone => $existing_sheet |
3848
|
|
|
|
|
|
|
|
3849
|
|
|
|
|
|
|
A deep copy of an existing sheet is made. |
3850
|
|
|
|
|
|
|
|
3851
|
|
|
|
|
|
|
=item num_cols => $number # with no initial content |
3852
|
|
|
|
|
|
|
|
3853
|
|
|
|
|
|
|
An empty sheet is created but with a fixed number of columns. |
3854
|
|
|
|
|
|
|
When rows are later created they will be immediately padded with empty cells |
3855
|
|
|
|
|
|
|
if necessary to this width. |
3856
|
|
|
|
|
|
|
|
3857
|
|
|
|
|
|
|
=back |
3858
|
|
|
|
|
|
|
|
3859
|
|
|
|
|
|
|
|
3860
|
|
|
|
|
|
|
=head2 write_csv *FILEHANDLE |
3861
|
|
|
|
|
|
|
|
3862
|
|
|
|
|
|
|
=head2 write_csv $path |
3863
|
|
|
|
|
|
|
|
3864
|
|
|
|
|
|
|
Write the current data to the indicated path or open file handle as |
3865
|
|
|
|
|
|
|
a CSV text file. |
3866
|
|
|
|
|
|
|
The default encoding is UTF-8 or, if C<read_spreadsheet> was most-recently |
3867
|
|
|
|
|
|
|
used to read a csv file, the encoding used then. |
3868
|
|
|
|
|
|
|
|
3869
|
|
|
|
|
|
|
{OPTIONS} may include |
3870
|
|
|
|
|
|
|
|
3871
|
|
|
|
|
|
|
=over 6 |
3872
|
|
|
|
|
|
|
|
3873
|
|
|
|
|
|
|
=item options for Text::CSV |
3874
|
|
|
|
|
|
|
|
3875
|
|
|
|
|
|
|
Usually none need be specified because we supply sane defaults. |
3876
|
|
|
|
|
|
|
|
3877
|
|
|
|
|
|
|
=item silent => bool |
3878
|
|
|
|
|
|
|
|
3879
|
|
|
|
|
|
|
=item verbose => bool |
3880
|
|
|
|
|
|
|
|
3881
|
|
|
|
|
|
|
=item debug => bool |
3882
|
|
|
|
|
|
|
|
3883
|
|
|
|
|
|
|
=back |
3884
|
|
|
|
|
|
|
|
3885
|
|
|
|
|
|
|
=head2 write_spreadsheet OUTPUTPATH |
3886
|
|
|
|
|
|
|
|
3887
|
|
|
|
|
|
|
Write the current data to a spreadsheet (.ods, .xlsx, etc.) by |
3888
|
|
|
|
|
|
|
first writing to a temporary CSV file and then importing that file into |
3889
|
|
|
|
|
|
|
a new spreadsheet. |
3890
|
|
|
|
|
|
|
|
3891
|
|
|
|
|
|
|
{OPTIONS} may include |
3892
|
|
|
|
|
|
|
|
3893
|
|
|
|
|
|
|
=over 6 |
3894
|
|
|
|
|
|
|
|
3895
|
|
|
|
|
|
|
=item col_formats => [ LIST ] |
3896
|
|
|
|
|
|
|
|
3897
|
|
|
|
|
|
|
EXPERIMENTAL, likely to change when Spreadsheet::Read is integrated! |
3898
|
|
|
|
|
|
|
|
3899
|
|
|
|
|
|
|
Elements of LIST may be "" (Standard), "Text", "MM/DD/YY", "DD/MM/YY", or |
3900
|
|
|
|
|
|
|
"YY/MM/DD" to indicate the format of the corresponding column. The meaning |
3901
|
|
|
|
|
|
|
of "Standard" is not well documented but appears to mean "General Number" |
3902
|
|
|
|
|
|
|
in most cases. For details, see "Format Codes" in L<this old Open Office |
3903
|
|
|
|
|
|
|
documentation|https://wiki.openoffice.org/wiki/Documentation/DevGuide/Spreadsheets/Filter_Options#Filter_Options_for_the_CSV_Filter>. |
3904
|
|
|
|
|
|
|
|
3905
|
|
|
|
|
|
|
=item silent => bool |
3906
|
|
|
|
|
|
|
|
3907
|
|
|
|
|
|
|
=item verbose => bool |
3908
|
|
|
|
|
|
|
|
3909
|
|
|
|
|
|
|
=item debug => bool |
3910
|
|
|
|
|
|
|
|
3911
|
|
|
|
|
|
|
=back |
3912
|
|
|
|
|
|
|
|
3913
|
|
|
|
|
|
|
|
3914
|
|
|
|
|
|
|
=head2 options NAME => EXPR, ... ; |
3915
|
|
|
|
|
|
|
|
3916
|
|
|
|
|
|
|
=head2 options NAME ; |
3917
|
|
|
|
|
|
|
|
3918
|
|
|
|
|
|
|
Set or retrieve miscellaneous sheet-specific options. |
3919
|
|
|
|
|
|
|
When setting, the previous value of |
3920
|
|
|
|
|
|
|
the last option specified is returned. The only options currently defined |
3921
|
|
|
|
|
|
|
are I<silent>, I<verbose> and I<debug>. |
3922
|
|
|
|
|
|
|
|
3923
|
|
|
|
|
|
|
=head2 $hash = attributes ; |
3924
|
|
|
|
|
|
|
|
3925
|
|
|
|
|
|
|
Returns a reference to a hash in which you may store arbitrary data |
3926
|
|
|
|
|
|
|
associated with the sheet object. |
3927
|
|
|
|
|
|
|
|
3928
|
|
|
|
|
|
|
=head2 spectocx COLSPEC or qr/regexp/, ... ; |
3929
|
|
|
|
|
|
|
|
3930
|
|
|
|
|
|
|
Returns the 0-based indicies of the specified colomn(s). |
3931
|
|
|
|
|
|
|
Throws an exception if there is no such column. |
3932
|
|
|
|
|
|
|
A regexp may match multiple columns. |
3933
|
|
|
|
|
|
|
See also C<%colx>. |
3934
|
|
|
|
|
|
|
|
3935
|
|
|
|
|
|
|
=head2 logmsg [FOCUSARG,] string, string, ... |
3936
|
|
|
|
|
|
|
|
3937
|
|
|
|
|
|
|
(must be explicitly imported) |
3938
|
|
|
|
|
|
|
|
3939
|
|
|
|
|
|
|
Concatenate strings, prefixed by a description |
3940
|
|
|
|
|
|
|
of the 'current sheet' and row during C<apply>, if any (or with the |
3941
|
|
|
|
|
|
|
sheet and/or row given by FOCUSARG). |
3942
|
|
|
|
|
|
|
|
3943
|
|
|
|
|
|
|
The resulting string is returned, with "\n" appended if it was not |
3944
|
|
|
|
|
|
|
already terminated by a newline. |
3945
|
|
|
|
|
|
|
|
3946
|
|
|
|
|
|
|
The first argument is used as FOCUSARG if it is |
3947
|
|
|
|
|
|
|
a sheet object, [sheet_object], or [sheet_object, rowindex], and specifies |
3948
|
|
|
|
|
|
|
the sheet and/or row to describe in the message prefix. |
3949
|
|
|
|
|
|
|
Otherwise the first argument is not special and is simply |
3950
|
|
|
|
|
|
|
the first message string. |
3951
|
|
|
|
|
|
|
|
3952
|
|
|
|
|
|
|
The details of formatting the sheet may be customized with a call-back |
3953
|
|
|
|
|
|
|
given by a C<{logmsg_pfx_gen}> attribute. See comments |
3954
|
|
|
|
|
|
|
in the source for how this works. |
3955
|
|
|
|
|
|
|
|
3956
|
|
|
|
|
|
|
=head1 STANDARD SHEET VARIABLES |
3957
|
|
|
|
|
|
|
|
3958
|
|
|
|
|
|
|
These variables magically access the 'current sheet' in your package. |
3959
|
|
|
|
|
|
|
|
3960
|
|
|
|
|
|
|
=over |
3961
|
|
|
|
|
|
|
|
3962
|
|
|
|
|
|
|
=item @rows |
3963
|
|
|
|
|
|
|
|
3964
|
|
|
|
|
|
|
The spreadsheet data as an array of row objects. |
3965
|
|
|
|
|
|
|
|
3966
|
|
|
|
|
|
|
Each row object is "dual-typed" (overloaded) to act as either an ARRAY or HASH |
3967
|
|
|
|
|
|
|
reference to the cells in that row. |
3968
|
|
|
|
|
|
|
|
3969
|
|
|
|
|
|
|
When used as a HASH ref, the key may be a |
3970
|
|
|
|
|
|
|
alias, column title, letter-code etc. (any COLSPEC). |
3971
|
|
|
|
|
|
|
When used as an ARRAY ref, the 0-based index specifies the column. |
3972
|
|
|
|
|
|
|
|
3973
|
|
|
|
|
|
|
=item @linenums |
3974
|
|
|
|
|
|
|
|
3975
|
|
|
|
|
|
|
The first line numbers of corresponding rows (a row can contain |
3976
|
|
|
|
|
|
|
multiple lines if cells contain embedded newlines). Valid only if |
3977
|
|
|
|
|
|
|
the data came from a CSV file. |
3978
|
|
|
|
|
|
|
|
3979
|
|
|
|
|
|
|
=item $num_cols |
3980
|
|
|
|
|
|
|
|
3981
|
|
|
|
|
|
|
The number of columns in the widest input row. Shorter rows are |
3982
|
|
|
|
|
|
|
padded with empty cells when read so that all rows have the same number |
3983
|
|
|
|
|
|
|
of columns in memory. |
3984
|
|
|
|
|
|
|
|
3985
|
|
|
|
|
|
|
=item $title_rx and $title_row |
3986
|
|
|
|
|
|
|
|
3987
|
|
|
|
|
|
|
C<$title_rx> contains the 0-based row index of the title row |
3988
|
|
|
|
|
|
|
and C<$title_row> is an alias for C<$rows[ $title_rx ]>. |
3989
|
|
|
|
|
|
|
|
3990
|
|
|
|
|
|
|
The title row is auto-detected by default. |
3991
|
|
|
|
|
|
|
See C<read_spreadsheet> and C<title_rx> for how to control this. |
3992
|
|
|
|
|
|
|
|
3993
|
|
|
|
|
|
|
If a column title is modified, set C<$title_rx = undef;> to force re-detection. |
3994
|
|
|
|
|
|
|
|
3995
|
|
|
|
|
|
|
=item $first_data_rx and $last_data_rx |
3996
|
|
|
|
|
|
|
|
3997
|
|
|
|
|
|
|
Optional limits on the range of rows visited by C<apply()> |
3998
|
|
|
|
|
|
|
or sorted by C<sort_rows()>. By default $first_data_rx |
3999
|
|
|
|
|
|
|
is the first row following the title row (or 0 if no title row). |
4000
|
|
|
|
|
|
|
|
4001
|
|
|
|
|
|
|
=item %colx (column key => column index) |
4002
|
|
|
|
|
|
|
|
4003
|
|
|
|
|
|
|
C<< %colx >> maps aliases, titles, etc. (all currently-valid COLSPECs) |
4004
|
|
|
|
|
|
|
to the corresponding zero-based column indicies. See "COLSPECS" . |
4005
|
|
|
|
|
|
|
|
4006
|
|
|
|
|
|
|
=item %colx_desc (column key => "debugging info") |
4007
|
|
|
|
|
|
|
|
4008
|
|
|
|
|
|
|
=back |
4009
|
|
|
|
|
|
|
|
4010
|
|
|
|
|
|
|
=head1 COLSPECs (COLUMN SPECIFIERS) |
4011
|
|
|
|
|
|
|
|
4012
|
|
|
|
|
|
|
Arguments which specify columns may be: |
4013
|
|
|
|
|
|
|
|
4014
|
|
|
|
|
|
|
=over |
4015
|
|
|
|
|
|
|
|
4016
|
|
|
|
|
|
|
=item (1) a user-defined alias identifier |
4017
|
|
|
|
|
|
|
|
4018
|
|
|
|
|
|
|
=item (2) an actual "column title" ** |
4019
|
|
|
|
|
|
|
|
4020
|
|
|
|
|
|
|
=item (3) an actual title with any leading & trailing spaces removed * |
4021
|
|
|
|
|
|
|
|
4022
|
|
|
|
|
|
|
=item (4) an AUTOMATIC ALIAS identifier * |
4023
|
|
|
|
|
|
|
|
4024
|
|
|
|
|
|
|
|
4025
|
|
|
|
|
|
|
=item (6) a Regexp (qr/.../) which matches an actual title |
4026
|
|
|
|
|
|
|
|
4027
|
|
|
|
|
|
|
=item (7) a numeric column index (0-based) |
4028
|
|
|
|
|
|
|
|
4029
|
|
|
|
|
|
|
=item (8) '^' or '$' (means first or last column, respectively) |
4030
|
|
|
|
|
|
|
|
4031
|
|
|
|
|
|
|
|
4032
|
|
|
|
|
|
|
=back |
4033
|
|
|
|
|
|
|
|
4034
|
|
|
|
|
|
|
*These may only be used if they do not conflict with an |
4035
|
|
|
|
|
|
|
item listed higher up. |
4036
|
|
|
|
|
|
|
|
4037
|
|
|
|
|
|
|
**Titles may be used directly if they can not be confused with |
4038
|
|
|
|
|
|
|
a user-defined alias, the special names '^' or '$' or a numeric |
4039
|
|
|
|
|
|
|
column index. See "CONFLICT RESOLUTION". |
4040
|
|
|
|
|
|
|
|
4041
|
|
|
|
|
|
|
B<AUTOMATIC ALIASES> are Perl I<identifiers> derived from column titles by |
4042
|
|
|
|
|
|
|
first removing leading or trailing spaces, and then |
4043
|
|
|
|
|
|
|
replacing non-word characters with underscores and prepending |
4044
|
|
|
|
|
|
|
an underscore if necessary. |
4045
|
|
|
|
|
|
|
For example: |
4046
|
|
|
|
|
|
|
|
4047
|
|
|
|
|
|
|
Title Automatic Alias |
4048
|
|
|
|
|
|
|
|
4049
|
|
|
|
|
|
|
"Address" Address (no change needed) |
4050
|
|
|
|
|
|
|
" First Name " First_Name |
4051
|
|
|
|
|
|
|
"First & Last" First___Last |
4052
|
|
|
|
|
|
|
"+sizes" _sizes |
4053
|
|
|
|
|
|
|
"1000s" _1000s (underscore avoids leading digit) |
4054
|
|
|
|
|
|
|
|
4055
|
|
|
|
|
|
|
Aliases (both automatic and user-defined) are valid identifiers, |
4056
|
|
|
|
|
|
|
so can be used as the names of tied variables, |
4057
|
|
|
|
|
|
|
bareword keys to C<%colx> and C<%crow>, and related OO interfaces, |
4058
|
|
|
|
|
|
|
|
4059
|
|
|
|
|
|
|
CONFLICT RESOLUTION |
4060
|
|
|
|
|
|
|
|
4061
|
|
|
|
|
|
|
A conflict occurs when a column key potentially refers to multiple |
4062
|
|
|
|
|
|
|
columns. For example, "A", "B" etc. are standard column |
4063
|
|
|
|
|
|
|
names, but they might also be the actual titles of other columns. |
4064
|
|
|
|
|
|
|
Warnings are printed about conflicts unless the B<silent> option |
4065
|
|
|
|
|
|
|
is true (see C<options>). |
4066
|
|
|
|
|
|
|
|
4067
|
|
|
|
|
|
|
=over |
4068
|
|
|
|
|
|
|
|
4069
|
|
|
|
|
|
|
B<User alias identifiers> (defined using C<alias>) are always valid. |
4070
|
|
|
|
|
|
|
|
4071
|
|
|
|
|
|
|
'^' and '$' always refer to the first and last column. |
4072
|
|
|
|
|
|
|
|
4073
|
|
|
|
|
|
|
Numeric "names" 0, 1, etc. always give a 0-based column index |
4074
|
|
|
|
|
|
|
if the value is between 0 and num_cols (i.e. one past the end). |
4075
|
|
|
|
|
|
|
|
4076
|
|
|
|
|
|
|
B<Actual Titles> refer to to their columns, except if they: |
4077
|
|
|
|
|
|
|
|
4078
|
|
|
|
|
|
|
=over |
4079
|
|
|
|
|
|
|
|
4080
|
|
|
|
|
|
|
are the same as a user-defined alias |
4081
|
|
|
|
|
|
|
|
4082
|
|
|
|
|
|
|
are '^' or '$' |
4083
|
|
|
|
|
|
|
|
4084
|
|
|
|
|
|
|
consist only of digits (without leading 0s) corresponding |
4085
|
|
|
|
|
|
|
to a valid column index. |
4086
|
|
|
|
|
|
|
|
4087
|
|
|
|
|
|
|
=back |
4088
|
|
|
|
|
|
|
|
4089
|
|
|
|
|
|
|
B<Automatic Aliases> and B<Standard column names> ("A", "B", etc.) |
4090
|
|
|
|
|
|
|
are available as column keys |
4091
|
|
|
|
|
|
|
unless they conflict with a user-defined alias or an actual title. |
4092
|
|
|
|
|
|
|
|
4093
|
|
|
|
|
|
|
=back |
4094
|
|
|
|
|
|
|
|
4095
|
|
|
|
|
|
|
Note: To unconditionally refer to numeric titles or titles which |
4096
|
|
|
|
|
|
|
look like '^' or '$', use a Regexp B<qr/.../>. |
4097
|
|
|
|
|
|
|
Automatic Aliases can also refer to such titles if there are no conflicts. |
4098
|
|
|
|
|
|
|
|
4099
|
|
|
|
|
|
|
Column positions always refer to the data before a command is |
4100
|
|
|
|
|
|
|
executed. This is relevant for commands which re-number or delete columns. |
4101
|
|
|
|
|
|
|
|
4102
|
|
|
|
|
|
|
=head1 OO DESCRIPTION (OBJECT-ORIENTED INTERFACE) |
4103
|
|
|
|
|
|
|
|
4104
|
|
|
|
|
|
|
All the Functions listed above (except for C<new_sheet>) have |
4105
|
|
|
|
|
|
|
corresponding methods with the same arguments. |
4106
|
|
|
|
|
|
|
|
4107
|
|
|
|
|
|
|
However method arguments must be enclosed in parenthesis |
4108
|
|
|
|
|
|
|
and bare {code} blocks may not be used; a sub{...} ref |
4109
|
|
|
|
|
|
|
should be passed to C<apply> etc. |
4110
|
|
|
|
|
|
|
|
4111
|
|
|
|
|
|
|
=head1 OO-SPECIFIC METHODS |
4112
|
|
|
|
|
|
|
|
4113
|
|
|
|
|
|
|
=head2 Spreadsheet::Edit->new(OPTIONS...) |
4114
|
|
|
|
|
|
|
|
4115
|
|
|
|
|
|
|
Creates a new "sheet" object. |
4116
|
|
|
|
|
|
|
|
4117
|
|
|
|
|
|
|
OPTIONS are the same as described for the C<new_sheet> Function above, |
4118
|
|
|
|
|
|
|
except that they may be specified as key => value pairs of arguments |
4119
|
|
|
|
|
|
|
instead of (or in addition to) an {OPTIONS} hashref. |
4120
|
|
|
|
|
|
|
|
4121
|
|
|
|
|
|
|
=head2 $sheet->rows() ; # Analogous to to \@rows |
4122
|
|
|
|
|
|
|
|
4123
|
|
|
|
|
|
|
=head2 $sheet->linenums() ; # Analogous to \@linenums |
4124
|
|
|
|
|
|
|
|
4125
|
|
|
|
|
|
|
=head2 $sheet->num_cols() ; # Analogous to $num_cols |
4126
|
|
|
|
|
|
|
|
4127
|
|
|
|
|
|
|
=head2 $sheet->colx() ; # Analogous to \%colx |
4128
|
|
|
|
|
|
|
|
4129
|
|
|
|
|
|
|
=head2 $sheet->colx_desc() ; # Analogous to \%colx_desc |
4130
|
|
|
|
|
|
|
|
4131
|
|
|
|
|
|
|
=head2 $sheet->first_data_rx() ; # Analogous to $first_data_rx |
4132
|
|
|
|
|
|
|
|
4133
|
|
|
|
|
|
|
=head2 $sheet->last_data_rx() ; # Analogous to $last_data_rx |
4134
|
|
|
|
|
|
|
|
4135
|
|
|
|
|
|
|
=head2 $sheet->title_rx() ; # Analogous to to $title_rx |
4136
|
|
|
|
|
|
|
|
4137
|
|
|
|
|
|
|
=head2 $sheet->title_row() ; # Analogous to $title_row |
4138
|
|
|
|
|
|
|
|
4139
|
|
|
|
|
|
|
=head2 $sheet->rx() ; # Current rx in apply, analogous to to $rx |
4140
|
|
|
|
|
|
|
|
4141
|
|
|
|
|
|
|
=head2 $sheet->title_rx(rxvalue) ; # Analogous to assigning to $title_rx (changes title row) |
4142
|
|
|
|
|
|
|
|
4143
|
|
|
|
|
|
|
=head2 $sheet->crow(); # Current row in apply (a dual-typed row object) |
4144
|
|
|
|
|
|
|
|
4145
|
|
|
|
|
|
|
=head2 $sheet->linenum() ; # Analogous to to $linenum |
4146
|
|
|
|
|
|
|
|
4147
|
|
|
|
|
|
|
=head2 $sheet->get(rx,ident) ; # Analogous to to $rows[rx]{ident} |
4148
|
|
|
|
|
|
|
|
4149
|
|
|
|
|
|
|
=head2 $sheet->set(rx,ident,value); # Analogous to to $rows[rx]{ident} = value |
4150
|
|
|
|
|
|
|
|
4151
|
|
|
|
|
|
|
=head2 $sheet->data_source(); # Returns "description of sheet" (e.g. path read) |
4152
|
|
|
|
|
|
|
|
4153
|
|
|
|
|
|
|
=head2 $sheet->sheetname(); # valid if input was a spreadsheet, else undef |
4154
|
|
|
|
|
|
|
|
4155
|
|
|
|
|
|
|
=head2 |
4156
|
|
|
|
|
|
|
|
4157
|
|
|
|
|
|
|
|
4158
|
|
|
|
|
|
|
=head1 SEE ALSO |
4159
|
|
|
|
|
|
|
|
4160
|
|
|
|
|
|
|
L<Spreadsheet::Edit::Preload> |
4161
|
|
|
|
|
|
|
|
4162
|
|
|
|
|
|
|
=head1 BUGS |
4163
|
|
|
|
|
|
|
|
4164
|
|
|
|
|
|
|
Some vestigial support for formats remains from an earlier implementation, |
4165
|
|
|
|
|
|
|
but this support is likely to be entirely replaced at some point. |
4166
|
|
|
|
|
|
|
|
4167
|
|
|
|
|
|
|
Reading a spreadsheet (but not csv) may fail if I<Libre Office> |
4168
|
|
|
|
|
|
|
or I<Open Office> are currently running for any purpose; this seems to be |
4169
|
|
|
|
|
|
|
a bug or limitation where batch-mode operations share the same profile as |
4170
|
|
|
|
|
|
|
interactive sessions. In any case, I<ssconvert> (gnumeric) will be used |
4171
|
|
|
|
|
|
|
if it is installed, and does not have this limitation. |
4172
|
|
|
|
|
|
|
In the future Spreadsheet::Read might be used instead of external programs, |
4173
|
|
|
|
|
|
|
although it uses Twig and is quite a bit slower. |
4174
|
|
|
|
|
|
|
|
4175
|
|
|
|
|
|
|
=head1 THREAD SAFETY |
4176
|
|
|
|
|
|
|
|
4177
|
|
|
|
|
|
|
Unknown, and probably not worth the trouble to find out. |
4178
|
|
|
|
|
|
|
|
4179
|
|
|
|
|
|
|
=head1 FUTURE IDEAS |
4180
|
|
|
|
|
|
|
|
4181
|
|
|
|
|
|
|
=over 4 |
4182
|
|
|
|
|
|
|
|
4183
|
|
|
|
|
|
|
=item Add format-processing support. |
4184
|
|
|
|
|
|
|
|
4185
|
|
|
|
|
|
|
=item Add "column-major" views. |
4186
|
|
|
|
|
|
|
|
4187
|
|
|
|
|
|
|
This would allow accessing a whole column as an array. |
4188
|
|
|
|
|
|
|
Perhaps C<@cols> and C<%cols> would be sets of column arrays |
4189
|
|
|
|
|
|
|
(@cols indexed by column index, %cols indexed by any COLSPEC). |
4190
|
|
|
|
|
|
|
And C<tie_column_vars '@NAME'> would tie user array variables to columns. |
4191
|
|
|
|
|
|
|
|
4192
|
|
|
|
|
|
|
=back |
4193
|
|
|
|
|
|
|
|
4194
|
|
|
|
|
|
|
=head1 AUTHOR |
4195
|
|
|
|
|
|
|
|
4196
|
|
|
|
|
|
|
Jim Avera (jim.avera at gmail) |
4197
|
|
|
|
|
|
|
|
4198
|
|
|
|
|
|
|
=head1 LICENSE |
4199
|
|
|
|
|
|
|
|
4200
|
|
|
|
|
|
|
Public Domain or CC0. |
4201
|
|
|
|
|
|
|
|
4202
|
|
|
|
|
|
|
=for Pod::Coverage meta_info |
4203
|
|
|
|
|
|
|
|
4204
|
|
|
|
|
|
|
=for Pod::Coverage iolayers input_encoding |
4205
|
|
|
|
|
|
|
|
4206
|
|
|
|
|
|
|
=for Pod::Coverage oops btw fmt_sheet fmt_list |
4207
|
|
|
|
|
|
|
|
4208
|
|
|
|
|
|
|
=for Pod::Coverage to_aref to_array to_wanted to_hash |
4209
|
|
|
|
|
|
|
|
4210
|
|
|
|
|
|
|
=for Pod::Coverage tied_varnames title2ident let2cx cx2let |
4211
|
|
|
|
|
|
|
|
4212
|
|
|
|
|
|
|
=cut |
4213
|
|
|
|
|
|
|
|