line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
package Astro::STSDAS::Table::Binary; |
2
|
|
|
|
|
|
|
|
3
|
|
|
|
|
|
|
our $VERSION = '0.13'; |
4
|
|
|
|
|
|
|
|
5
|
1
|
|
|
1
|
|
4
|
use strict; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
30
|
|
6
|
1
|
|
|
1
|
|
5
|
use warnings; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
22
|
|
7
|
1
|
|
|
1
|
|
846
|
use FileHandle; |
|
1
|
|
|
|
|
11580
|
|
|
1
|
|
|
|
|
5
|
|
8
|
1
|
|
|
1
|
|
478
|
use Carp qw( carp croak ); |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
59
|
|
9
|
|
|
|
|
|
|
|
10
|
|
|
|
|
|
|
|
11
|
|
|
|
|
|
|
our @ISA = qw( Astro::STSDAS::Table::Base ); |
12
|
|
|
|
|
|
|
|
13
|
1
|
|
|
1
|
|
842
|
use Astro::STSDAS::Table::Base; |
|
1
|
|
|
|
|
179
|
|
|
1
|
|
|
|
|
57
|
|
14
|
1
|
|
|
1
|
|
8
|
use Astro::STSDAS::Table::Constants; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
1888
|
|
15
|
|
|
|
|
|
|
|
16
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
# things read in from the table |
18
|
|
|
|
|
|
|
our @hdr_fields = ( |
19
|
|
|
|
|
|
|
'nhp', # number of header parameters |
20
|
|
|
|
|
|
|
'nhp_a', # number of header parameters allocated |
21
|
|
|
|
|
|
|
'nrows', # number of rows written to table |
22
|
|
|
|
|
|
|
'nrows_a', # number of rows allocated |
23
|
|
|
|
|
|
|
'ncols', # number of column descriptors in table |
24
|
|
|
|
|
|
|
'ncols_a', # number of column descriptors allocated |
25
|
|
|
|
|
|
|
'row_used', # size in CHAR_SZ of space used in row |
26
|
|
|
|
|
|
|
'row_len', # size in CHAR_SZ of row length |
27
|
|
|
|
|
|
|
'ttype', # type of table (row or column ordered) |
28
|
|
|
|
|
|
|
'version', # STSDAS software version number |
29
|
|
|
|
|
|
|
); |
30
|
|
|
|
|
|
|
|
31
|
|
|
|
|
|
|
|
32
|
|
|
|
|
|
|
|
33
|
|
|
|
|
|
|
|
34
|
|
|
|
|
|
|
# row_len - the row length, in bytes, for row-ordered tables |
35
|
|
|
|
|
|
|
# row_used - the actual length of the row in the file, in bytes, for |
36
|
|
|
|
|
|
|
# row-ordered tables |
37
|
|
|
|
|
|
|
# row_els - the number of elements in a row (includes vector elements) |
38
|
|
|
|
|
|
|
# nrows_a - the number of rows allocated (in a column ordered table) |
39
|
|
|
|
|
|
|
# ttype - the type of table (either TT_ROW_ORDER or TT_COL_ORDER) |
40
|
|
|
|
|
|
|
# version - "table software version number" from STSDAS created tables |
41
|
|
|
|
|
|
|
# row - the next record (zero based) to be read in |
42
|
|
|
|
|
|
|
# last_col_idx - index of the last column read, for column ordered tables |
43
|
|
|
|
|
|
|
# last_col - the last column read, for column ordered tables |
44
|
|
|
|
|
|
|
# buf - the input buffer, row_len bytes wide. |
45
|
|
|
|
|
|
|
# have_vecs - the table has vectors |
46
|
|
|
|
|
|
|
|
47
|
|
|
|
|
|
|
|
48
|
|
|
|
|
|
|
sub new |
49
|
|
|
|
|
|
|
{ |
50
|
0
|
|
|
0
|
1
|
|
my $this = shift; |
51
|
0
|
|
0
|
|
|
|
my $class = ref($this) || $this; |
52
|
|
|
|
|
|
|
|
53
|
0
|
|
|
|
|
|
my $self = $class->SUPER::new(); |
54
|
|
|
|
|
|
|
|
55
|
0
|
|
|
|
|
|
$self->{last_col_idx} = -1; |
56
|
0
|
|
|
|
|
|
$self->{last_col} = undef; |
57
|
0
|
|
|
|
|
|
$self->{row} = 0; |
58
|
|
|
|
|
|
|
|
59
|
0
|
|
|
|
|
|
bless $self, $class; |
60
|
|
|
|
|
|
|
} |
61
|
|
|
|
|
|
|
|
62
|
|
|
|
|
|
|
|
63
|
|
|
|
|
|
|
# _read_hdr |
64
|
|
|
|
|
|
|
|
65
|
|
|
|
|
|
|
# _read_hdr is an internal routine which digests the binary table |
66
|
|
|
|
|
|
|
# header. besides stocking the table hash with the information, it |
67
|
|
|
|
|
|
|
# converts lengths into bytes and creates a pack() compatible format |
68
|
|
|
|
|
|
|
# for reading in rows. It also initializes various things. |
69
|
|
|
|
|
|
|
|
70
|
|
|
|
|
|
|
sub _read_hdr |
71
|
|
|
|
|
|
|
{ |
72
|
0
|
|
|
0
|
|
|
my $self = shift; |
73
|
|
|
|
|
|
|
|
74
|
0
|
|
|
|
|
|
my $buf; |
75
|
0
|
0
|
|
|
|
|
read( $self->{fh}, $buf, 12 * $TypeSize{TY_INT()} ) == |
76
|
|
|
|
|
|
|
12 * $TypeSize{TY_INT()} or |
77
|
|
|
|
|
|
|
croak( "no data or error reading header\n"); |
78
|
|
|
|
|
|
|
|
79
|
0
|
|
|
|
|
|
my %rawhdr; |
80
|
0
|
|
|
|
|
|
@rawhdr{@hdr_fields} = unpack( 'i10', $buf ); |
81
|
|
|
|
|
|
|
|
82
|
|
|
|
|
|
|
# save a few of the values |
83
|
0
|
|
|
|
|
|
$self->{row_len} = $rawhdr{row_len} * CHAR_SZ; |
84
|
0
|
|
|
|
|
|
$self->{row_used} = $rawhdr{row_used} * CHAR_SZ; |
85
|
0
|
|
|
|
|
|
$self->{nrows} = $rawhdr{nrows}; |
86
|
0
|
|
|
|
|
|
$self->{nrows_a} = $rawhdr{nrows_a}; |
87
|
0
|
|
|
|
|
|
$self->{ttype} = $rawhdr{ttype}; |
88
|
0
|
|
|
|
|
|
$self->{version} = $rawhdr{version}; |
89
|
|
|
|
|
|
|
|
90
|
|
|
|
|
|
|
|
91
|
0
|
0
|
|
|
|
|
if ( $rawhdr{nhp} ) |
92
|
|
|
|
|
|
|
{ |
93
|
0
|
|
|
|
|
|
my $pars = $self->{pars}; |
94
|
|
|
|
|
|
|
|
95
|
0
|
|
|
|
|
|
for my $i ( 1 .. $rawhdr{nhp} ) |
96
|
|
|
|
|
|
|
{ |
97
|
0
|
0
|
|
|
|
|
read( $self->{fh}, $buf, 80 ) == 80 or |
98
|
|
|
|
|
|
|
croak( "ran out of data reading header parameter $i\n" ); |
99
|
|
|
|
|
|
|
|
100
|
0
|
|
|
|
|
|
my $name = unpack('A*', substr($buf, 0, 8)); |
101
|
0
|
|
|
|
|
|
my $type = $HdrType{substr( $buf, 8, 1 )}; |
102
|
0
|
|
|
|
|
|
( my $value = substr( $buf, 9, 71 ) ) =~ s/\0.*//g; |
103
|
|
|
|
|
|
|
|
104
|
0
|
0
|
|
|
|
|
if ( $type eq TY_STRING ) |
105
|
|
|
|
|
|
|
{ |
106
|
0
|
|
|
|
|
|
$value =~ s/^'|'$//g; |
107
|
|
|
|
|
|
|
} |
108
|
|
|
|
|
|
|
|
109
|
0
|
|
|
|
|
|
$pars->add( $name, $value, undef, $type ); |
110
|
|
|
|
|
|
|
} |
111
|
|
|
|
|
|
|
} |
112
|
|
|
|
|
|
|
|
113
|
0
|
0
|
|
|
|
|
if ( $rawhdr{nhp_a} > $rawhdr{nhp} ) |
114
|
|
|
|
|
|
|
{ |
115
|
0
|
|
|
|
|
|
my $i = 1; |
116
|
0
|
|
|
|
|
|
while ( $i++ <= $rawhdr{nhp_a} - $rawhdr{nhp} ) |
117
|
|
|
|
|
|
|
{ |
118
|
0
|
0
|
|
|
|
|
read( $self->{fh}, $buf, 80 ) == 80 or |
119
|
|
|
|
|
|
|
croak( "ran out of data reading padding header record $i\n" ); |
120
|
|
|
|
|
|
|
} |
121
|
|
|
|
|
|
|
} |
122
|
|
|
|
|
|
|
|
123
|
0
|
|
|
|
|
|
$self->{row_els} = 0; |
124
|
0
|
|
|
|
|
|
for my $coln ( 1 .. $rawhdr{ncols} ) |
125
|
|
|
|
|
|
|
{ |
126
|
0
|
0
|
|
|
|
|
read( $self->{fh}, $buf, 16 * $TypeSize{TY_INT()} ) == |
127
|
|
|
|
|
|
|
16 * $TypeSize{TY_INT()} or |
128
|
|
|
|
|
|
|
croak( "ran out of data reading column definition $coln\n" ); |
129
|
|
|
|
|
|
|
|
130
|
0
|
|
|
|
|
|
my ( $idx, $offset, $space, $type ) = unpack( 'i4', $buf ); |
131
|
|
|
|
|
|
|
|
132
|
0
|
|
|
|
|
|
my $nelem; |
133
|
0
|
|
|
|
|
|
( my $name = substr( $buf, 4 * $TypeSize{TY_INT()}, 20 ) ) =~ s/\0.*//g; |
134
|
0
|
|
|
|
|
|
( my $units = substr( $buf, 9 * $TypeSize{TY_INT()}, 20 ) ) =~ s/\0.*//g; |
135
|
0
|
|
|
|
|
|
( my $format = substr( $buf, 14 * $TypeSize{TY_INT()}, 8 ) ) =~ s/\0.*//g; |
136
|
|
|
|
|
|
|
|
137
|
|
|
|
|
|
|
|
138
|
|
|
|
|
|
|
# if type is negative, it's a string; $type also gives length |
139
|
0
|
0
|
|
|
|
|
if ( $type < 0 ) |
140
|
|
|
|
|
|
|
{ |
141
|
0
|
|
|
|
|
|
$type = TY_STRING; |
142
|
0
|
|
|
|
|
|
$nelem = -$type; |
143
|
|
|
|
|
|
|
} |
144
|
|
|
|
|
|
|
|
145
|
|
|
|
|
|
|
# nope. get the length from the number of bytes in the |
146
|
|
|
|
|
|
|
# element |
147
|
|
|
|
|
|
|
else |
148
|
|
|
|
|
|
|
{ |
149
|
0
|
|
|
|
|
|
$nelem = $space * CHAR_SZ / $TypeSize{$type}; |
150
|
|
|
|
|
|
|
} |
151
|
|
|
|
|
|
|
|
152
|
0
|
|
|
|
|
|
my $col = $self->{cols}->add( $name, $units, $format, $idx, |
153
|
|
|
|
|
|
|
$offset * CHAR_SZ, |
154
|
|
|
|
|
|
|
$type, $nelem ); |
155
|
|
|
|
|
|
|
|
156
|
0
|
|
|
|
|
|
$self->{row_els} += $nelem; |
157
|
|
|
|
|
|
|
|
158
|
0
|
|
|
|
|
|
$self->{row_extract} .= $col->fmt; |
159
|
|
|
|
|
|
|
} |
160
|
|
|
|
|
|
|
|
161
|
0
|
0
|
|
|
|
|
if ( $rawhdr{ncols_a} > $rawhdr{ncols} ) |
162
|
|
|
|
|
|
|
{ |
163
|
0
|
|
|
|
|
|
my $nbytes = ($rawhdr{ncols_a} - $rawhdr{ncols}) * 16 * $TypeSize{TY_INT()} ; |
164
|
0
|
0
|
|
|
|
|
read( $self->{fh}, $buf, $nbytes ) == $nbytes |
165
|
|
|
|
|
|
|
or |
166
|
|
|
|
|
|
|
croak( "ran out of data reading padding column definitions\n" ); |
167
|
|
|
|
|
|
|
} |
168
|
|
|
|
|
|
|
|
169
|
|
|
|
|
|
|
# reuse buffers for speed (Perl will size them correctly the first time |
170
|
|
|
|
|
|
|
# they're used |
171
|
|
|
|
|
|
|
|
172
|
|
|
|
|
|
|
# input raw buffer |
173
|
0
|
|
|
|
|
|
$self->{buf} = ''; |
174
|
|
|
|
|
|
|
|
175
|
|
|
|
|
|
|
# input extracted data buffer (inlined vector elements) |
176
|
0
|
|
|
|
|
|
$self->{data} = []; |
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
# input data, vector elements split out |
179
|
0
|
|
|
|
|
|
$self->{row_arr} = []; |
180
|
0
|
|
|
|
|
|
$self->{row_hash} = {}; |
181
|
|
|
|
|
|
|
|
182
|
0
|
|
|
|
|
|
$self->{have_vecs} = grep { $_->is_vector } $self->{cols}->cols; |
|
0
|
|
|
|
|
|
|
183
|
|
|
|
|
|
|
} |
184
|
|
|
|
|
|
|
|
185
|
0
|
|
|
0
|
1
|
|
sub is_row_order { $_[0]->{ttype} == TT_ROW_ORDER } |
186
|
0
|
|
|
0
|
1
|
|
sub is_col_order { $_[0]->{ttype} == TT_COL_ORDER } |
187
|
|
|
|
|
|
|
|
188
|
|
|
|
|
|
|
sub read_rows_hash |
189
|
|
|
|
|
|
|
{ |
190
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
191
|
|
|
|
|
|
|
|
192
|
|
|
|
|
|
|
# pre extend |
193
|
0
|
|
|
|
|
|
my @rows; |
194
|
0
|
|
|
|
|
|
$#rows = $self->{nrows} - 1; |
195
|
0
|
|
|
|
|
|
@rows = (); |
196
|
|
|
|
|
|
|
|
197
|
0
|
0
|
|
|
|
|
if ( $self->is_row_order ) |
198
|
|
|
|
|
|
|
{ |
199
|
0
|
|
|
|
|
|
my $row; |
200
|
0
|
|
|
|
|
|
push @rows, $row while $row = $self->read_row_row_hash( {} ) ; |
201
|
|
|
|
|
|
|
} |
202
|
|
|
|
|
|
|
|
203
|
|
|
|
|
|
|
else |
204
|
|
|
|
|
|
|
{ |
205
|
0
|
|
|
|
|
|
1 while $self->read_col_row_hash( \@rows ); |
206
|
|
|
|
|
|
|
} |
207
|
|
|
|
|
|
|
|
208
|
0
|
|
|
|
|
|
\@rows; |
209
|
|
|
|
|
|
|
} |
210
|
|
|
|
|
|
|
|
211
|
|
|
|
|
|
|
sub read_rows_array |
212
|
|
|
|
|
|
|
{ |
213
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
214
|
0
|
|
|
|
|
|
my %attr = ( VecSplit => 1, |
215
|
0
|
0
|
0
|
|
|
|
( @_ && 'HASH' eq ref($_[-1]) ? %{pop @_} : () ) ); |
216
|
|
|
|
|
|
|
|
217
|
0
|
|
|
|
|
|
my @rows; |
218
|
0
|
|
|
|
|
|
$#rows = $self->{nrows} - 1; |
219
|
0
|
|
|
|
|
|
@rows = (); |
220
|
|
|
|
|
|
|
|
221
|
0
|
0
|
|
|
|
|
if ( $self->is_row_order ) |
222
|
|
|
|
|
|
|
{ |
223
|
0
|
|
|
|
|
|
my $idx = 0; |
224
|
0
|
|
|
|
|
|
my $row; |
225
|
0
|
|
|
|
|
|
push @rows, $row |
226
|
|
|
|
|
|
|
while ( $row = $self->read_row_row_array( [], \%attr ) ); |
227
|
|
|
|
|
|
|
} |
228
|
|
|
|
|
|
|
|
229
|
|
|
|
|
|
|
else |
230
|
|
|
|
|
|
|
{ |
231
|
|
|
|
|
|
|
# pre extend row arrays. |
232
|
0
|
|
|
|
|
|
@rows = map { |
233
|
0
|
|
|
|
|
|
my @row; |
234
|
0
|
|
|
|
|
|
$#row = $self->{ncols} - 1; |
235
|
0
|
|
|
|
|
|
@row = (); |
236
|
0
|
|
|
|
|
|
\@row; |
237
|
|
|
|
|
|
|
} ( 0..($self->{nrows}-1) ); |
238
|
|
|
|
|
|
|
|
239
|
0
|
|
|
|
|
|
1 while $self->read_col_row_array( \@rows, \%attr ); |
240
|
|
|
|
|
|
|
} |
241
|
|
|
|
|
|
|
|
242
|
0
|
|
|
|
|
|
\@rows; |
243
|
|
|
|
|
|
|
} |
244
|
|
|
|
|
|
|
|
245
|
|
|
|
|
|
|
sub read_cols_hash |
246
|
|
|
|
|
|
|
{ |
247
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
248
|
|
|
|
|
|
|
|
249
|
0
|
|
|
|
|
|
my $cols_arr = $self->read_cols_array; |
250
|
|
|
|
|
|
|
|
251
|
0
|
|
|
|
|
|
my %cols; |
252
|
0
|
|
|
|
|
|
@cols{ map { lc $_ } $self->{cols}->names } = @{$cols_arr}; |
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
253
|
|
|
|
|
|
|
|
254
|
0
|
|
|
|
|
|
\%cols; |
255
|
|
|
|
|
|
|
} |
256
|
|
|
|
|
|
|
|
257
|
|
|
|
|
|
|
sub read_cols_array |
258
|
|
|
|
|
|
|
{ |
259
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
260
|
|
|
|
|
|
|
|
261
|
0
|
|
|
|
|
|
my @cols; |
262
|
|
|
|
|
|
|
|
263
|
0
|
0
|
|
|
|
|
if ( $self->is_row_order ) |
264
|
|
|
|
|
|
|
{ |
265
|
0
|
|
|
|
|
|
@cols = map { my @a; $#a = $self->{nrows} - 1; \@a } |
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
266
|
|
|
|
|
|
|
1..($self->{cols}->ncols - 1); |
267
|
|
|
|
|
|
|
|
268
|
1
|
|
|
1
|
|
1025
|
use integer; |
|
1
|
|
|
|
|
10
|
|
|
1
|
|
|
|
|
6
|
|
269
|
0
|
|
|
|
|
|
my $idx; |
270
|
0
|
|
|
|
|
|
while( my $row = $self->read_row_row_array ) |
271
|
|
|
|
|
|
|
{ |
272
|
0
|
|
|
|
|
|
$cols[$_][$idx] = $row->[$_] for 0..($self->{cols}->ncols-1); |
273
|
0
|
|
|
|
|
|
$idx++; |
274
|
|
|
|
|
|
|
} |
275
|
|
|
|
|
|
|
} |
276
|
|
|
|
|
|
|
|
277
|
|
|
|
|
|
|
else |
278
|
|
|
|
|
|
|
{ |
279
|
0
|
|
|
|
|
|
my $data; |
280
|
0
|
|
|
|
|
|
push @cols, $data |
281
|
|
|
|
|
|
|
while ( $data = $self->read_col_col_array ); |
282
|
|
|
|
|
|
|
} |
283
|
|
|
|
|
|
|
|
284
|
0
|
|
|
|
|
|
\@cols; |
285
|
|
|
|
|
|
|
} |
286
|
|
|
|
|
|
|
|
287
|
|
|
|
|
|
|
# read a column from a column oriented table into a row hash |
288
|
|
|
|
|
|
|
sub read_col_row_hash |
289
|
|
|
|
|
|
|
{ |
290
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
291
|
0
|
|
|
|
|
|
my $row = shift; |
292
|
|
|
|
|
|
|
|
293
|
0
|
|
|
|
|
|
my $data = $self->read_col_col_array; |
294
|
0
|
|
|
|
|
|
my $name = $self->{last_col}->name; |
295
|
|
|
|
|
|
|
|
296
|
0
|
|
|
|
|
|
eval qq{ |
297
|
|
|
|
|
|
|
use integer; |
298
|
|
|
|
|
|
|
\$row->[\$_]{$name} = \$data->[\$_] |
299
|
|
|
|
|
|
|
foreach 0..($self->{nrows}-1) ; |
300
|
|
|
|
|
|
|
}; |
301
|
|
|
|
|
|
|
|
302
|
0
|
|
|
|
|
|
1; |
303
|
|
|
|
|
|
|
} |
304
|
|
|
|
|
|
|
|
305
|
|
|
|
|
|
|
# read a column from a column oriented table into a row array |
306
|
|
|
|
|
|
|
sub read_col_row_array |
307
|
|
|
|
|
|
|
{ |
308
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
309
|
|
|
|
|
|
|
|
310
|
0
|
|
|
|
|
|
my %attr = ( VecSplit => 1, |
311
|
0
|
0
|
0
|
|
|
|
( @_ && 'HASH' eq ref($_[-1]) ? %{pop @_} : () ) ); |
312
|
|
|
|
|
|
|
|
313
|
0
|
|
|
|
|
|
my $row = shift; |
314
|
0
|
|
|
|
|
|
my $col = $self->{last_col}; |
315
|
|
|
|
|
|
|
|
316
|
0
|
|
|
|
|
|
my $data = $self->read_col_col_array( \%attr ); |
317
|
|
|
|
|
|
|
|
318
|
0
|
0
|
0
|
|
|
|
if ( !$self->{have_vecs} || $attr{VecSplit} ) |
319
|
|
|
|
|
|
|
{ |
320
|
0
|
|
|
|
|
|
eval qq{ |
321
|
|
|
|
|
|
|
use integer; |
322
|
|
|
|
|
|
|
push \@{\$row->[\$_]}, \$data->[\$_] foreach 0..$self->{nrows}-1 ; |
323
|
|
|
|
|
|
|
}; |
324
|
|
|
|
|
|
|
} |
325
|
|
|
|
|
|
|
|
326
|
|
|
|
|
|
|
else |
327
|
|
|
|
|
|
|
{ |
328
|
|
|
|
|
|
|
# make special code; don't know yet if this is worth it |
329
|
0
|
|
|
|
|
|
my $dp = 0; |
330
|
0
|
|
|
|
|
|
my $dpd = $col->nelem; |
331
|
0
|
|
|
|
|
|
my $dpd1 = $col->nelem - 1; |
332
|
|
|
|
|
|
|
|
333
|
0
|
|
|
|
|
|
eval qq{ |
334
|
|
|
|
|
|
|
use integer; |
335
|
0
|
|
|
|
|
|
for my \$idx ( 0..@{\$self->{nrows}}-1 ) |
336
|
|
|
|
|
|
|
{ |
337
|
|
|
|
|
|
|
push \@{\$row->[\$idx]}, [ \@{\$data}[\$dp .. (\$dp + $dpd1) ] ] ; |
338
|
|
|
|
|
|
|
\$dp += $dpd |
339
|
|
|
|
|
|
|
} |
340
|
|
|
|
|
|
|
}; |
341
|
|
|
|
|
|
|
|
342
|
|
|
|
|
|
|
} |
343
|
|
|
|
|
|
|
|
344
|
0
|
|
|
|
|
|
1; |
345
|
|
|
|
|
|
|
} |
346
|
|
|
|
|
|
|
|
347
|
|
|
|
|
|
|
sub read_col_col_array |
348
|
|
|
|
|
|
|
{ |
349
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
350
|
0
|
|
|
|
|
|
my $uattr = shift; |
351
|
|
|
|
|
|
|
|
352
|
0
|
0
|
|
|
|
|
my %attr = { VecSplit => 1, defined $uattr ? %$uattr : () }; |
353
|
|
|
|
|
|
|
|
354
|
0
|
|
|
|
|
|
my $data = $self->_read_next_col; |
355
|
0
|
|
|
|
|
|
my $col = $self->{last_col}; |
356
|
|
|
|
|
|
|
|
357
|
|
|
|
|
|
|
# if there are no vector elements, just return the data as is |
358
|
0
|
0
|
0
|
|
|
|
return $data if 1 == $col->{nelem} || ! $attr{VecSplit}; |
359
|
|
|
|
|
|
|
|
360
|
|
|
|
|
|
|
# deal with the vector elements |
361
|
0
|
|
|
|
|
|
my @vec_data; |
362
|
0
|
|
|
|
|
|
$#vec_data = $self->{nrows} - 1; |
363
|
|
|
|
|
|
|
|
364
|
|
|
|
|
|
|
# make special code; don't know yet if this is worth it |
365
|
0
|
|
|
|
|
|
my $dp = 0; |
366
|
0
|
|
|
|
|
|
my $dpd = $col->nelem; |
367
|
0
|
|
|
|
|
|
my $dpd1 = $col->nelem - 1; |
368
|
0
|
|
|
|
|
|
eval qq{ |
369
|
|
|
|
|
|
|
use integer; |
370
|
0
|
|
|
|
|
|
for my \$idx ( 0..@{\$self->{nrows}}-1 ) |
371
|
|
|
|
|
|
|
{ |
372
|
|
|
|
|
|
|
\$vec_data[\$idx] = [ \@{\$data}[\$dp .. (\$dp + $dpd1) ] ] ; |
373
|
|
|
|
|
|
|
\$dp += $dpd; |
374
|
|
|
|
|
|
|
} |
375
|
|
|
|
|
|
|
}; |
376
|
0
|
|
|
|
|
|
return \@vec_data; |
377
|
|
|
|
|
|
|
} |
378
|
|
|
|
|
|
|
|
379
|
|
|
|
|
|
|
|
380
|
|
|
|
|
|
|
sub _read_next_col |
381
|
|
|
|
|
|
|
{ |
382
|
0
|
|
|
0
|
|
|
my $self = shift; |
383
|
|
|
|
|
|
|
|
384
|
|
|
|
|
|
|
# if we're all done, don't bother |
385
|
0
|
0
|
|
|
|
|
return () if $self->{last_col_idx} + 1 == $self->{cols}->ncols; |
386
|
|
|
|
|
|
|
|
387
|
0
|
|
|
|
|
|
my $buf; |
388
|
|
|
|
|
|
|
|
389
|
0
|
|
|
|
|
|
my $col = $self->{cols}->byidx($self->{last_col_idx} + 1); |
390
|
|
|
|
|
|
|
|
391
|
0
|
|
|
|
|
|
my $ndata = $self->{nrows_a} * $col->nelem; |
392
|
0
|
|
|
|
|
|
my $nbytes = $ndata * $col->size; |
393
|
|
|
|
|
|
|
|
394
|
0
|
|
|
|
|
|
my $nread = read( $self->{fh}, $buf, $nbytes ); |
395
|
|
|
|
|
|
|
|
396
|
0
|
0
|
|
|
|
|
unless( $nbytes == $nread ) |
397
|
|
|
|
|
|
|
{ |
398
|
|
|
|
|
|
|
# gotta be exactly $nbytes or we're loused |
399
|
0
|
|
|
|
|
|
croak( "incomplete read of column ", $self->{last_col_idx} + 2, "\n" ) |
400
|
|
|
|
|
|
|
} |
401
|
|
|
|
|
|
|
|
402
|
0
|
|
|
|
|
|
my @data = unpack( $col->ifmt . $ndata , $buf ); |
403
|
|
|
|
|
|
|
|
404
|
|
|
|
|
|
|
# clean the data; |
405
|
0
|
0
|
|
|
|
|
unless ( $col->is_string ) |
406
|
|
|
|
|
|
|
{ |
407
|
0
|
|
0
|
|
|
|
$col->is_indef($_) && ($_ = undef) foreach @data; |
408
|
|
|
|
|
|
|
} |
409
|
|
|
|
|
|
|
|
410
|
0
|
|
|
|
|
|
$self->{last_col_idx}++; |
411
|
0
|
|
|
|
|
|
$self->{last_col} = $col; |
412
|
|
|
|
|
|
|
|
413
|
0
|
|
|
|
|
|
\@data; |
414
|
|
|
|
|
|
|
} |
415
|
|
|
|
|
|
|
|
416
|
|
|
|
|
|
|
|
417
|
|
|
|
|
|
|
sub read_row_row_array |
418
|
|
|
|
|
|
|
{ |
419
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
420
|
0
|
|
|
|
|
|
$self->_read_next_row( @_ ); |
421
|
|
|
|
|
|
|
} |
422
|
|
|
|
|
|
|
|
423
|
|
|
|
|
|
|
|
424
|
|
|
|
|
|
|
|
425
|
|
|
|
|
|
|
sub read_row_row_hash |
426
|
|
|
|
|
|
|
{ |
427
|
0
|
|
|
0
|
1
|
|
my $self = shift; |
428
|
0
|
|
0
|
|
|
|
my $row = shift || $self->{row_hash}; |
429
|
|
|
|
|
|
|
|
430
|
0
|
|
|
|
|
|
my $row_arr = $self->_read_next_row( $row ); |
431
|
|
|
|
|
|
|
|
432
|
0
|
0
|
|
|
|
|
return undef unless $row; |
433
|
|
|
|
|
|
|
|
434
|
0
|
|
|
|
|
|
@{$row}{ map { lc $_ } $self->{cols}->names } = @$row_arr; |
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
435
|
0
|
|
|
|
|
|
return $row; |
436
|
|
|
|
|
|
|
} |
437
|
|
|
|
|
|
|
|
438
|
|
|
|
|
|
|
# _read_row |
439
|
|
|
|
|
|
|
|
440
|
|
|
|
|
|
|
# This reads the next row from a row-ordered table into an array, in the |
441
|
|
|
|
|
|
|
# same order as that of the columns in the table. Vector elements are |
442
|
|
|
|
|
|
|
# stored as described above. It returns the undefined value if there are |
443
|
|
|
|
|
|
|
# no more data. |
444
|
|
|
|
|
|
|
|
445
|
|
|
|
|
|
|
sub _read_next_row |
446
|
|
|
|
|
|
|
{ |
447
|
0
|
|
|
0
|
|
|
my $self = shift; |
448
|
|
|
|
|
|
|
|
449
|
0
|
|
|
|
|
|
my %attr = ( VecSplit => 1, |
450
|
0
|
0
|
0
|
|
|
|
( @_ && 'HASH' eq ref($_[-1]) ? %{pop @_} : () ) ); |
451
|
|
|
|
|
|
|
|
452
|
|
|
|
|
|
|
# store the row data in what the caller wants, or the object's buffer. |
453
|
0
|
|
0
|
|
|
|
my $row = shift || $self->{row_arr}; |
454
|
|
|
|
|
|
|
|
455
|
|
|
|
|
|
|
# guess what? there's (possibly only sometimes) an extra row filled |
456
|
|
|
|
|
|
|
# with indefs at the end of the file! so we can't actually use |
457
|
|
|
|
|
|
|
# the end of file condition to stop reading. ACKCKCKCCKKC! |
458
|
0
|
0
|
|
|
|
|
return undef if $self->{row} == $self->{nrows}; |
459
|
|
|
|
|
|
|
|
460
|
|
|
|
|
|
|
|
461
|
0
|
|
|
|
|
|
my $nread = read( $self->{fh}, $self->{buf}, $self->{row_len}); |
462
|
|
|
|
|
|
|
|
463
|
0
|
0
|
|
|
|
|
unless( $nread == $self->{row_len} ) |
464
|
|
|
|
|
|
|
{ |
465
|
|
|
|
|
|
|
# if it's not zero, then we've read too little, and that's a no-no |
466
|
0
|
0
|
|
|
|
|
croak( "incomplete last record (", $self->{row}+1, ")\n" ) |
467
|
|
|
|
|
|
|
if 0 != $nread; |
468
|
|
|
|
|
|
|
|
469
|
|
|
|
|
|
|
# EOF |
470
|
0
|
|
|
|
|
|
return undef; |
471
|
|
|
|
|
|
|
} |
472
|
|
|
|
|
|
|
|
473
|
|
|
|
|
|
|
# if we're not splitting vectors up, just read into the final destination |
474
|
0
|
0
|
|
|
|
|
my $data = $attr{VecSplit} ? $self->{data} : $row; |
475
|
|
|
|
|
|
|
|
476
|
|
|
|
|
|
|
# pre-extend. should only hurt once |
477
|
0
|
|
|
|
|
|
$#{$data} = $self->{row_els}; |
|
0
|
|
|
|
|
|
|
478
|
0
|
|
|
|
|
|
@{$data} = unpack( $self->{row_extract}, $self->{buf} ); |
|
0
|
|
|
|
|
|
|
479
|
|
|
|
|
|
|
|
480
|
0
|
0
|
|
|
|
|
if ( $attr{VecSplit} ) |
481
|
|
|
|
|
|
|
{ |
482
|
|
|
|
|
|
|
# this is slow, but it works. clean it up someday |
483
|
|
|
|
|
|
|
|
484
|
|
|
|
|
|
|
# prextend the row. should only hurt once. |
485
|
0
|
|
|
|
|
|
$#{$row} = $self->{cols}->ncols; |
|
0
|
|
|
|
|
|
|
486
|
0
|
|
|
|
|
|
@$row = (); |
487
|
0
|
|
|
|
|
|
for my $col ( $self->{cols}->cols ) |
488
|
|
|
|
|
|
|
{ |
489
|
0
|
0
|
|
|
|
|
if ( $col->nelem == 1 ) |
490
|
|
|
|
|
|
|
{ |
491
|
0
|
|
|
|
|
|
my $elem = shift @$data; |
492
|
0
|
0
|
|
|
|
|
push @$row, $col->is_indef($elem) ? undef : $elem; |
493
|
|
|
|
|
|
|
} |
494
|
|
|
|
|
|
|
else |
495
|
|
|
|
|
|
|
{ |
496
|
0
|
0
|
|
|
|
|
push @$row, |
497
|
0
|
|
|
|
|
|
[ map { $col->is_indef($_) ? undef : $_ } |
498
|
|
|
|
|
|
|
splice( @$data, 0, $col->nelem ) ]; |
499
|
|
|
|
|
|
|
} |
500
|
|
|
|
|
|
|
} |
501
|
|
|
|
|
|
|
} |
502
|
|
|
|
|
|
|
else |
503
|
|
|
|
|
|
|
{ |
504
|
0
|
|
|
|
|
|
my $idx = 0; |
505
|
0
|
|
|
|
|
|
for my $col ( $self->{cols}->cols ) |
506
|
|
|
|
|
|
|
{ |
507
|
0
|
|
|
|
|
|
for ( my $nelem = $col->nelem; $nelem ; $nelem--, $idx++ ) |
508
|
|
|
|
|
|
|
{ |
509
|
0
|
0
|
|
|
|
|
$data->[$idx] = undef if $col->is_indef($data->[$idx]); |
510
|
|
|
|
|
|
|
} |
511
|
|
|
|
|
|
|
} |
512
|
|
|
|
|
|
|
} |
513
|
|
|
|
|
|
|
|
514
|
0
|
|
|
|
|
|
$self->{row}++; |
515
|
0
|
|
|
|
|
|
return $row; |
516
|
|
|
|
|
|
|
} |
517
|
|
|
|
|
|
|
|
518
|
|
|
|
|
|
|
|
519
|
|
|
|
|
|
|
1; |
520
|
|
|
|
|
|
|
|
521
|
|
|
|
|
|
|
__END__ |