line |
stmt |
bran |
cond |
sub |
pod |
time |
code |
1
|
|
|
|
|
|
|
# |
2
|
|
|
|
|
|
|
# FILE %gg/perl/HP200LX/DB.pm |
3
|
|
|
|
|
|
|
# |
4
|
|
|
|
|
|
|
# access HP 200LX database files |
5
|
|
|
|
|
|
|
# See POD Section for a few more details |
6
|
|
|
|
|
|
|
# |
7
|
|
|
|
|
|
|
# work area: |
8
|
|
|
|
|
|
|
# decode_type14 |
9
|
|
|
|
|
|
|
# dump_type .. export everything in ASCII format |
10
|
|
|
|
|
|
|
# loader .. import everything from ASCII format |
11
|
|
|
|
|
|
|
# |
12
|
|
|
|
|
|
|
# written: 1997-12-28 (c) g.gonter@ieee.org |
13
|
|
|
|
|
|
|
# latest update: 2001-02-09 17:22:39 |
14
|
|
|
|
|
|
|
# $Id: DB.pm,v 1.13 2001/03/05 01:52:39 gonter Exp $ |
15
|
|
|
|
|
|
|
# |
16
|
|
|
|
|
|
|
|
17
|
|
|
|
|
|
|
package HP200LX::DB; |
18
|
|
|
|
|
|
|
|
19
|
1
|
|
|
1
|
|
1548
|
use strict; |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
190
|
|
20
|
1
|
|
|
1
|
|
5
|
use vars qw($VERSION @ISA @EXPORT_OK @REC_TYPE); |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
115
|
|
21
|
1
|
|
|
1
|
|
6
|
use Exporter; |
|
1
|
|
|
|
|
80
|
|
|
1
|
|
|
|
|
119
|
|
22
|
|
|
|
|
|
|
|
23
|
|
|
|
|
|
|
$VERSION= '0.09'; |
24
|
|
|
|
|
|
|
@ISA= qw(Exporter); |
25
|
|
|
|
|
|
|
@EXPORT_OK= qw(openDB saveDB |
26
|
|
|
|
|
|
|
fmt_date fmt_time pack_date hex_dump |
27
|
|
|
|
|
|
|
); |
28
|
|
|
|
|
|
|
|
29
|
1
|
|
|
1
|
|
676
|
use HP200LX::DB::vpt; # view point management, including vpt definition |
|
1
|
|
|
|
|
2
|
|
|
1
|
|
|
|
|
12428
|
|
30
|
|
|
|
|
|
|
|
31
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
32
|
|
|
|
|
|
|
my $no_note= 65535; # note number if there is no note |
33
|
|
|
|
|
|
|
my $no_val= 65535; # NIL, empty list, -1 etc. |
34
|
|
|
|
|
|
|
my $no_time= 32768; # empty time field |
35
|
|
|
|
|
|
|
my $no_year= 255; # empty year, mon, day elements |
36
|
|
|
|
|
|
|
my $no_mon= 255; |
37
|
|
|
|
|
|
|
my $no_day= 255; |
38
|
|
|
|
|
|
|
my $no_date= 255; # ... no_date values |
39
|
|
|
|
|
|
|
my $delim= '-'x 74; # optic delimiter |
40
|
|
|
|
|
|
|
|
41
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
42
|
|
|
|
|
|
|
my @REC_TYPE= # HP's internal record type definitions |
43
|
|
|
|
|
|
|
( |
44
|
|
|
|
|
|
|
'DBHEADER', # 0 |
45
|
|
|
|
|
|
|
'PASSWORD', # 1: only present when a password was set |
46
|
|
|
|
|
|
|
'', # 2 |
47
|
|
|
|
|
|
|
'', # 3 |
48
|
|
|
|
|
|
|
'CARDDEF', # 4 |
49
|
|
|
|
|
|
|
'CATEGORY', # 5 |
50
|
|
|
|
|
|
|
'FIELDDEF', # 6 |
51
|
|
|
|
|
|
|
'VIEWPTDEF', # 7 sort and subset |
52
|
|
|
|
|
|
|
'', # 8 |
53
|
|
|
|
|
|
|
'NOTE', # 9 |
54
|
|
|
|
|
|
|
'VIEWPTTABLE', # 10 table of viewpoint entries |
55
|
|
|
|
|
|
|
'DATA', # 11 |
56
|
|
|
|
|
|
|
'LINKDEF', # 12: usually smart clips |
57
|
|
|
|
|
|
|
'CARDPAGEDEF', # 13 |
58
|
|
|
|
|
|
|
'', # 14 APP: |
59
|
|
|
|
|
|
|
# + ADB: appt_info |
60
|
|
|
|
|
|
|
'SMART_CLIP', # 15 APP: smart clip def in appt.adb (GG) |
61
|
|
|
|
|
|
|
# + ADB: appt_list (adbio) |
62
|
|
|
|
|
|
|
'', # 16 APP |
63
|
|
|
|
|
|
|
'', # 17 APP |
64
|
|
|
|
|
|
|
'', # 18 APP |
65
|
|
|
|
|
|
|
'', # 19 APP |
66
|
|
|
|
|
|
|
'', # 20 APP |
67
|
|
|
|
|
|
|
'', # 21 APP |
68
|
|
|
|
|
|
|
'', # 22 APP |
69
|
|
|
|
|
|
|
'', # 23 APP |
70
|
|
|
|
|
|
|
'', # 24 APP |
71
|
|
|
|
|
|
|
'', # 25 APP |
72
|
|
|
|
|
|
|
'', # 26 APP |
73
|
|
|
|
|
|
|
'', # 27 APP |
74
|
|
|
|
|
|
|
'', # 28 APP |
75
|
|
|
|
|
|
|
'', # 29 APP |
76
|
|
|
|
|
|
|
'', # 30 APP |
77
|
|
|
|
|
|
|
'LOOKUPTABLE' # 31 |
78
|
|
|
|
|
|
|
# 14..30 application specific! |
79
|
|
|
|
|
|
|
); |
80
|
0
|
0
|
|
0
|
0
|
0
|
sub REC_TYPE { my $num= shift; $REC_TYPE[$num] || "USER_TYPE_$num"; } |
|
0
|
|
|
|
|
0
|
|
81
|
|
|
|
|
|
|
|
82
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
83
|
|
|
|
|
|
|
my @FIELD_TYPE= # HP's internal field type definitions |
84
|
|
|
|
|
|
|
( |
85
|
|
|
|
|
|
|
{ 'Desc' => 'BYTEBOOL', 'Size' => 1, }, # 0 |
86
|
|
|
|
|
|
|
{ 'Desc' => 'WORDBOOL', 'Size' => 2, }, # 1 .. e.g. check box |
87
|
|
|
|
|
|
|
{ 'Desc' => 'STRING', 'Size' => 2, }, # 2 |
88
|
|
|
|
|
|
|
{ 'Desc' => 'PHONE', 'Size' => 2, }, # 3 |
89
|
|
|
|
|
|
|
{ 'Desc' => 'NUMBER', 'Size' => 2, }, # 4 |
90
|
|
|
|
|
|
|
{ 'Desc' => 'CURRENCY', 'Size' => 2, }, # 5 |
91
|
|
|
|
|
|
|
{ 'Desc' => 'CATEGORY', 'Size' => 2, }, # 6 |
92
|
|
|
|
|
|
|
{ 'Desc' => 'TIME', 'Size' => 2, }, # 7 Test: store |
93
|
|
|
|
|
|
|
{ 'Desc' => 'DATE', 'Size' => 3, }, # 8 Test: store |
94
|
|
|
|
|
|
|
{ 'Desc' => 'RADIO_BUTTON', 'Size' => 2, }, # 9 Note: should be 1 byte but it uses 2 bytes! |
95
|
|
|
|
|
|
|
{ 'Desc' => 'NOTE', 'Size' => 2, }, # 10 Store: seems to work now |
96
|
|
|
|
|
|
|
{ 'Desc' => 'GROUP', 'Size' => 0, }, # 11 |
97
|
|
|
|
|
|
|
{ 'Desc' => 'STATIC', 'Size' => 0, }, # 12: Label |
98
|
|
|
|
|
|
|
{ 'Desc' => 'MULTILINE', 'Size' => 0, }, # 13 ?? |
99
|
|
|
|
|
|
|
{ 'Desc' => 'LIST', 'Size' => 0, }, # 14 |
100
|
|
|
|
|
|
|
{ 'Desc' => 'COMBO', 'Size' => 0, }, # 15 |
101
|
|
|
|
|
|
|
{ 'Desc' => 'U16', 'Size' => 0, }, # 16: WDB time zone difference |
102
|
|
|
|
|
|
|
{ 'Desc' => 'U17', 'Size' => 0, }, # 17 |
103
|
|
|
|
|
|
|
{ 'Desc' => 'U18', 'Size' => 1, }, # 18: ADB "Repeat Status" |
104
|
|
|
|
|
|
|
{ 'Desc' => 'U19', 'Size' => 3, }, # 19: ADB "Start Date" |
105
|
|
|
|
|
|
|
{ 'Desc' => 'U20', 'Size' => 2, }, # 20: ADB "Due Date" |
106
|
|
|
|
|
|
|
{ 'Desc' => 'U21', 'Size' => 0, }, # 21 |
107
|
|
|
|
|
|
|
{ 'Desc' => 'U22', 'Size' => 2, }, # 22: ADB "Priority" |
108
|
|
|
|
|
|
|
{ 'Desc' => 'U23', 'Size' => 2, }, # 23: ADB "#consecutive days" |
109
|
|
|
|
|
|
|
{ 'Desc' => 'U24', 'Size' => 2, }, # 24: ADB "Leadtime" |
110
|
|
|
|
|
|
|
{ 'Desc' => 'U25', 'Size' => 0, }, # 25 |
111
|
|
|
|
|
|
|
); |
112
|
|
|
|
|
|
|
|
113
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
114
|
|
|
|
|
|
|
# The HP-LX's password protection engine uses a two constant code blocks: |
115
|
|
|
|
|
|
|
# CODE_A is 127 byte long, CODE_B is 17 byte long |
116
|
|
|
|
|
|
|
my @CODE_A= |
117
|
|
|
|
|
|
|
( |
118
|
|
|
|
|
|
|
0xe8, 0xa3, 0xfe, 0x1b, 0x02, 0xce, 0x40, 0x35, |
119
|
|
|
|
|
|
|
0xa4, 0x7b, 0xf2, 0xa1, 0x70, 0xd5, 0x40, 0x65, |
120
|
|
|
|
|
|
|
0x09, 0x42, 0x23, 0xff, 0xaa, 0xed, 0xf0, 0x2a, |
121
|
|
|
|
|
|
|
0xa2, 0xa9, 0x38, 0xd7, 0xe5, 0x95, 0xea, 0x8c, |
122
|
|
|
|
|
|
|
0x46, 0xdd, 0x90, 0x94, 0x5e, 0x6b, 0x5d, 0xa4, |
123
|
|
|
|
|
|
|
0x7b, 0x8c, 0xea, 0x24, 0xa1, 0x7c, 0xaf, 0x30, |
124
|
|
|
|
|
|
|
0x62, 0x2a, 0xa5, 0x8e, 0xad, 0x67, 0xde, 0x3f, |
125
|
|
|
|
|
|
|
0xb3, 0xe3, 0x53, 0xde, 0x19, 0x42, 0xf8, 0x40, |
126
|
|
|
|
|
|
|
0x96, 0xe8, 0x15, 0x75, 0x43, 0x08, 0x2f, 0xe9, |
127
|
|
|
|
|
|
|
0xb1, 0x4f, 0x1d, 0xd5, 0xa9, 0x16, 0x2c, 0xfb, |
128
|
|
|
|
|
|
|
0x9f, 0x0f, 0xb2, 0xcc, 0xe4, 0x27, 0xbc, 0x1b, |
129
|
|
|
|
|
|
|
0x49, 0xa6, 0x90, 0x79, 0x03, 0x9a, 0xa6, 0x1a, |
130
|
|
|
|
|
|
|
0x70, 0x89, 0x9d, 0x35, 0x81, 0xad, 0x80, 0xb0, |
131
|
|
|
|
|
|
|
0x79, 0x45, 0x21, 0x5f, 0x94, 0x1c, 0xd1, 0x3f, |
132
|
|
|
|
|
|
|
0xdf, 0xa8, 0xa3, 0x40, 0x31, 0x34, 0x66, 0x84, |
133
|
|
|
|
|
|
|
0x85, 0x28, 0xf1, 0x8d, 0x82, 0x04, 0xa4 |
134
|
|
|
|
|
|
|
); |
135
|
|
|
|
|
|
|
|
136
|
|
|
|
|
|
|
my @CODE_B= |
137
|
|
|
|
|
|
|
( |
138
|
|
|
|
|
|
|
0x09, 0x0b, 0x09, 0x0f, 0x09, 0x0b, 0x09, 0x77, |
139
|
|
|
|
|
|
|
0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, 0x08, |
140
|
|
|
|
|
|
|
0x78 |
141
|
|
|
|
|
|
|
); |
142
|
|
|
|
|
|
|
my @DIAG_K; # used for diagnosing the decryption functions |
143
|
|
|
|
|
|
|
|
144
|
|
|
|
|
|
|
# 17 byte code to decrypt the password |
145
|
|
|
|
|
|
|
my @PW_CODE= |
146
|
|
|
|
|
|
|
( |
147
|
|
|
|
|
|
|
0xE1, 0xA8, 0xF4, 0x17, 0x0B, 0xE7, 0x09, 0x75, # 0x00 |
148
|
|
|
|
|
|
|
0xD2, 0x6B, 0x9F, 0x84, 0x2D, 0x9A, 0x3F, 0x05, # 0x08 |
149
|
|
|
|
|
|
|
0x71 |
150
|
|
|
|
|
|
|
); |
151
|
|
|
|
|
|
|
|
152
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
153
|
|
|
|
|
|
|
my %XHDR= # debugging: headers that will not be printed |
154
|
|
|
|
|
|
|
( |
155
|
|
|
|
|
|
|
'sig' => 1, 'time' => 1, 'lookup_table_offset' => 1, |
156
|
|
|
|
|
|
|
'recheader' => 1, 'file_type' => 1, |
157
|
|
|
|
|
|
|
); |
158
|
|
|
|
|
|
|
|
159
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
160
|
|
|
|
|
|
|
# create a new (empty) database object |
161
|
|
|
|
|
|
|
sub new |
162
|
|
|
|
|
|
|
{ |
163
|
1
|
|
|
1
|
0
|
2
|
my $class= shift; |
164
|
1
|
|
|
|
|
1
|
my $fnm= shift; |
165
|
1
|
|
33
|
|
|
11
|
my $apt= shift || &derive_apt ($fnm); |
166
|
|
|
|
|
|
|
|
167
|
|
|
|
|
|
|
# print ">>> NEW: fnm='$fnm' apt='$apt'\n"; |
168
|
1
|
|
|
|
|
2
|
my $i; |
169
|
1
|
|
|
|
|
2
|
my $Types= []; |
170
|
1
|
|
|
|
|
214
|
my @t= localtime (time); |
171
|
|
|
|
|
|
|
|
172
|
1
|
|
|
|
|
6
|
for ($i= 0; $i < 32; $i++) { push (@$Types, []); } |
|
32
|
|
|
|
|
93
|
|
173
|
|
|
|
|
|
|
|
174
|
1
|
|
|
|
|
19
|
my $obj= |
175
|
|
|
|
|
|
|
{ |
176
|
|
|
|
|
|
|
'Filename' => $fnm, |
177
|
|
|
|
|
|
|
|
178
|
|
|
|
|
|
|
'APT' => $apt, # application type |
179
|
|
|
|
|
|
|
# GDB: generic database (default) |
180
|
|
|
|
|
|
|
# NDB: note taker (NDB == GDB) |
181
|
|
|
|
|
|
|
# ADB: appointment book |
182
|
|
|
|
|
|
|
# WDB: world time |
183
|
|
|
|
|
|
|
'APT_Data' => {}, # application specific extension data |
184
|
|
|
|
|
|
|
|
185
|
|
|
|
|
|
|
'Header' => # see loader, save |
186
|
|
|
|
|
|
|
{ |
187
|
|
|
|
|
|
|
'sig' => "hcD\000", |
188
|
|
|
|
|
|
|
'recheader' => |
189
|
|
|
|
|
|
|
{ |
190
|
|
|
|
|
|
|
'type' => 0, |
191
|
|
|
|
|
|
|
'status' => 0, |
192
|
|
|
|
|
|
|
'length' => 25, |
193
|
|
|
|
|
|
|
'idx' => 0, |
194
|
|
|
|
|
|
|
}, |
195
|
|
|
|
|
|
|
|
196
|
|
|
|
|
|
|
'time' => |
197
|
|
|
|
|
|
|
{ |
198
|
|
|
|
|
|
|
'year' => $t[5]+1900, |
199
|
|
|
|
|
|
|
'mon' => $t[4]+1, |
200
|
|
|
|
|
|
|
'day' => $t[3], |
201
|
|
|
|
|
|
|
'min' => $t[2]*60 + $t[1], |
202
|
|
|
|
|
|
|
}, |
203
|
|
|
|
|
|
|
|
204
|
|
|
|
|
|
|
# guessed data from other examples |
205
|
|
|
|
|
|
|
'release_version' => 0x0102, |
206
|
|
|
|
|
|
|
'file_type' => &get_apt ($apt), |
207
|
|
|
|
|
|
|
'file_status' => 0, |
208
|
|
|
|
|
|
|
'cur_viewpt' => 0, |
209
|
|
|
|
|
|
|
'num_recs' => 0, |
210
|
|
|
|
|
|
|
'lookup_table_offset' => 0, |
211
|
|
|
|
|
|
|
'viewpt_hash' => 0x8525, # "Magic Code" |
212
|
|
|
|
|
|
|
# 0x8437 for US american 100LX |
213
|
|
|
|
|
|
|
}, |
214
|
|
|
|
|
|
|
|
215
|
|
|
|
|
|
|
'Types' => $Types, # DB records of each type |
216
|
|
|
|
|
|
|
|
217
|
|
|
|
|
|
|
# pre-processed internal datatypes |
218
|
|
|
|
|
|
|
'fielddef' => [], # data descriptions of fields |
219
|
|
|
|
|
|
|
'carddef' => [], # window descriptions of fields |
220
|
|
|
|
|
|
|
'cardpagedef' => [], # description for the four cards |
221
|
|
|
|
|
|
|
'viewptdef' => [], # view point definitins; list/sort/filter |
222
|
|
|
|
|
|
|
'viewpttable' => [], # cached view point table |
223
|
|
|
|
|
|
|
|
224
|
|
|
|
|
|
|
'update' => 0, # number of items modified |
225
|
|
|
|
|
|
|
}; |
226
|
|
|
|
|
|
|
|
227
|
1
|
|
|
|
|
7
|
bless $obj, $class; |
228
|
|
|
|
|
|
|
} |
229
|
|
|
|
|
|
|
|
230
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
231
|
|
|
|
|
|
|
sub get_apt |
232
|
|
|
|
|
|
|
{ |
233
|
1
|
|
50
|
1
|
0
|
5
|
my $APT= shift || 'GDB'; |
234
|
1
|
|
|
|
|
2
|
my $code= 0x44; # generic database, GDB and PDB |
235
|
|
|
|
|
|
|
|
236
|
1
|
50
|
|
|
|
9
|
if ($APT eq 'ADB') { $code= 0x32; } |
|
0
|
50
|
|
|
|
0
|
|
|
|
50
|
|
|
|
|
|
237
|
0
|
|
|
|
|
0
|
elsif ($APT eq 'NDB') { $code= 0x4E; } |
238
|
0
|
|
|
|
|
0
|
elsif ($APT eq 'WDB') { $code= 0x57; } |
239
|
|
|
|
|
|
|
# else: gdb, pdb: GDB (generic data base) |
240
|
|
|
|
|
|
|
|
241
|
1
|
|
|
|
|
18
|
$code; |
242
|
|
|
|
|
|
|
} |
243
|
|
|
|
|
|
|
|
244
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
245
|
|
|
|
|
|
|
sub decode_apt |
246
|
|
|
|
|
|
|
{ |
247
|
1
|
|
|
1
|
0
|
3
|
my $code= shift; |
248
|
1
|
|
|
|
|
2
|
my $APT= 'GDB'; |
249
|
|
|
|
|
|
|
|
250
|
1
|
50
|
|
|
|
10
|
if ($code == 0x32) { $APT= 'ADB'; } |
|
0
|
50
|
|
|
|
0
|
|
|
|
50
|
|
|
|
|
|
251
|
0
|
|
|
|
|
0
|
elsif ($code eq 0x4E) { $APT= 'NDB'; } |
252
|
0
|
|
|
|
|
0
|
elsif ($code eq 0x57) { $APT= 'WDB'; } |
253
|
|
|
|
|
|
|
|
254
|
1
|
|
|
|
|
4
|
$APT; |
255
|
|
|
|
|
|
|
} |
256
|
|
|
|
|
|
|
|
257
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
258
|
|
|
|
|
|
|
sub derive_apt |
259
|
|
|
|
|
|
|
{ |
260
|
1
|
|
|
1
|
0
|
2
|
my $fnm= shift; |
261
|
1
|
|
|
|
|
2
|
my $APT= 'GDB'; # generic database |
262
|
|
|
|
|
|
|
|
263
|
1
|
50
|
|
|
|
17
|
if ($fnm =~ m/\.adb$/i) { $APT= 'ADB'; } # appointment book |
|
0
|
50
|
|
|
|
0
|
|
|
|
50
|
|
|
|
|
|
264
|
0
|
|
|
|
|
0
|
elsif ($fnm =~ m/\.ndb$/i) { $APT= 'NDB'; } # note taker |
265
|
0
|
|
|
|
|
0
|
elsif ($fnm =~ m/\.wdb$/i) { $APT= 'WDB'; } # world time application |
266
|
|
|
|
|
|
|
# else: gdb, pdb: GDB (generic data base) |
267
|
|
|
|
|
|
|
|
268
|
1
|
|
|
|
|
5
|
$APT; |
269
|
|
|
|
|
|
|
} |
270
|
|
|
|
|
|
|
|
271
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
272
|
|
|
|
|
|
|
# open a given file and read the database into memory |
273
|
|
|
|
|
|
|
sub openDB |
274
|
|
|
|
|
|
|
{ |
275
|
1
|
|
|
1
|
0
|
12
|
my $fnm= shift; |
276
|
1
|
|
|
|
|
2
|
my $APT= shift; |
277
|
1
|
|
|
|
|
2
|
my $dont_decrypt= shift; |
278
|
|
|
|
|
|
|
|
279
|
1
|
|
|
|
|
10
|
my $obj= new HP200LX::DB ($fnm, $APT); |
280
|
1
|
|
|
|
|
10
|
$APT= $obj->{APT}; # use application detection logic in new |
281
|
1
|
|
|
|
|
22
|
my $b; |
282
|
|
|
|
|
|
|
my $sig; |
283
|
1
|
|
|
|
|
3
|
local *FI; |
284
|
|
|
|
|
|
|
|
285
|
1
|
50
|
|
|
|
35
|
unless (open (FI, $fnm)) |
286
|
|
|
|
|
|
|
{ |
287
|
0
|
|
|
|
|
0
|
print "ERROR: could not open DB file '$fnm'!\n"; |
288
|
0
|
|
|
|
|
0
|
return undef; |
289
|
|
|
|
|
|
|
} |
290
|
1
|
|
|
|
|
5
|
binmode (FI); # MS-DOS systems need this, T2D: how about Mac? |
291
|
|
|
|
|
|
|
|
292
|
1
|
|
|
|
|
26
|
read (FI, $sig, 4); |
293
|
|
|
|
|
|
|
|
294
|
|
|
|
|
|
|
# BEGIN to read the db header; see save |
295
|
1
|
|
|
|
|
5
|
my $recheader= &get_recheader (*FI); |
296
|
1
|
|
|
|
|
3
|
my $lng= $recheader->{'length'}; |
297
|
1
|
50
|
|
|
|
6
|
print "WARNING lng=$lng, 25 expected!\n" unless ($lng == 25); |
298
|
|
|
|
|
|
|
|
299
|
1
|
|
|
|
|
4
|
read (FI, $b, 19); # lng minus length of record header: 19+6= 25 |
300
|
1
|
|
|
|
|
5
|
my ($release_version, $file_type, $file_status, |
301
|
|
|
|
|
|
|
$cur_viewpt, $num_recs, $lookup_table_offset, |
302
|
|
|
|
|
|
|
$year, $mon, $day, $min, $viewpt_hash)= unpack ('vCCvvVCCCvv', $b); |
303
|
|
|
|
|
|
|
# END to read the record header |
304
|
|
|
|
|
|
|
|
305
|
1
|
|
|
|
|
5
|
my $time= |
306
|
|
|
|
|
|
|
{ |
307
|
|
|
|
|
|
|
'year' => $year+1900, |
308
|
|
|
|
|
|
|
'mon' => $mon+1, |
309
|
|
|
|
|
|
|
'day' => $day+1, |
310
|
|
|
|
|
|
|
'min' => $min, |
311
|
|
|
|
|
|
|
}; |
312
|
|
|
|
|
|
|
|
313
|
1
|
|
|
|
|
13
|
my $hdr= |
314
|
|
|
|
|
|
|
{ |
315
|
|
|
|
|
|
|
'sig' => $sig, |
316
|
|
|
|
|
|
|
'time' => $time, |
317
|
|
|
|
|
|
|
'recheader' => $recheader, |
318
|
|
|
|
|
|
|
|
319
|
|
|
|
|
|
|
'release_version' => $release_version, |
320
|
|
|
|
|
|
|
'file_type' => $file_type, |
321
|
|
|
|
|
|
|
'file_status' => $file_status, |
322
|
|
|
|
|
|
|
'cur_viewpt' => $cur_viewpt, |
323
|
|
|
|
|
|
|
'num_recs' => $num_recs, |
324
|
|
|
|
|
|
|
'lookup_table_offset' => $lookup_table_offset, |
325
|
|
|
|
|
|
|
'viewpt_hash' => $viewpt_hash, |
326
|
|
|
|
|
|
|
}; |
327
|
|
|
|
|
|
|
|
328
|
1
|
|
|
|
|
5
|
$obj->{Header}= $hdr; |
329
|
1
|
|
|
|
|
6
|
$APT= $obj->{APT}= &decode_apt ($file_type); |
330
|
|
|
|
|
|
|
# &hex_dump ($b); |
331
|
|
|
|
|
|
|
# print "APT=$APT file_type=$file_type num_recs=$num_recs", |
332
|
|
|
|
|
|
|
# " cur_viewpt=$cur_viewpt\n"; |
333
|
|
|
|
|
|
|
# printf ("lookup_table_offset= 0x%08lX\n", $lookup_table_offset); |
334
|
|
|
|
|
|
|
|
335
|
|
|
|
|
|
|
# read lookup table |
336
|
1
|
|
|
|
|
2
|
my ($v, $i, $xrec); |
337
|
1
|
|
|
|
|
2
|
my @ltbl= (); # lookup table |
338
|
1
|
|
|
|
|
4
|
my @ftbl= (); # "type first" table |
339
|
|
|
|
|
|
|
|
340
|
1
|
50
|
|
|
|
3
|
if ($lookup_table_offset > 0) |
341
|
|
|
|
|
|
|
{ |
342
|
1
|
|
|
|
|
10
|
seek (FI, $lookup_table_offset, 0); |
343
|
1
|
|
|
|
|
12
|
$xrec= &get_recheader (*FI); |
344
|
|
|
|
|
|
|
# &print_recheader (*STDOUT, "lookup table (offset=$lookup_table_offset)", $xrec); |
345
|
1
|
|
|
|
|
3
|
$lng= $xrec->{'length'}-6; |
346
|
1
|
|
|
|
|
3
|
$i= read (FI, $b, $lng); |
347
|
|
|
|
|
|
|
|
348
|
1
|
50
|
|
|
|
3
|
print "WARNING: could not read complete lookup table; read=$i lng=$lng\n" |
349
|
|
|
|
|
|
|
unless ($i == $lng); |
350
|
|
|
|
|
|
|
|
351
|
1
|
|
|
|
|
2
|
$i= $num_recs * 8; # 8 byte per lookup table entry |
352
|
1
|
50
|
|
|
|
5
|
print "WARNING: lookup table size seems wrong;", |
353
|
|
|
|
|
|
|
" lng=$lng num_recs=$num_recs $num_recs*8=$i\n" |
354
|
|
|
|
|
|
|
unless ($i == $lng); |
355
|
|
|
|
|
|
|
|
356
|
1
|
|
|
|
|
5
|
for ($i= 0; $i < $num_recs; $i++) |
357
|
|
|
|
|
|
|
{ |
358
|
22
|
|
|
|
|
57
|
my ($size, $filters, $flags, $off_low, $off)= |
359
|
|
|
|
|
|
|
unpack ('vvCCv', substr ($b, $i*8, 8)); |
360
|
22
|
|
|
|
|
25
|
$off= $off*256+$off_low; |
361
|
|
|
|
|
|
|
|
362
|
|
|
|
|
|
|
# print "lut [$i] off=$off size=$size\n"; |
363
|
22
|
|
|
|
|
47
|
my $lut= |
364
|
|
|
|
|
|
|
{ |
365
|
|
|
|
|
|
|
'siz' => $size, |
366
|
|
|
|
|
|
|
'off' => $off, |
367
|
|
|
|
|
|
|
'filters' => $filters, |
368
|
|
|
|
|
|
|
'flags' => $flags, |
369
|
|
|
|
|
|
|
} ; |
370
|
|
|
|
|
|
|
|
371
|
22
|
|
|
|
|
56
|
push (@ltbl, $lut); |
372
|
|
|
|
|
|
|
} |
373
|
|
|
|
|
|
|
# $hdr->{lookup_table_header}= $xrec; |
374
|
|
|
|
|
|
|
# $hdr->{lookup_table}= \@ltbl; |
375
|
|
|
|
|
|
|
|
376
|
|
|
|
|
|
|
# typefirst table |
377
|
|
|
|
|
|
|
# |
378
|
|
|
|
|
|
|
# Purpose: |
379
|
|
|
|
|
|
|
# This table points into the lookup table at the position of the |
380
|
|
|
|
|
|
|
# first record of each record type |
381
|
|
|
|
|
|
|
# Example: |
382
|
|
|
|
|
|
|
# lookup data for record 3 of type 4 is at: ltbl [ftbl [4] + 3] |
383
|
|
|
|
|
|
|
# NOTE: |
384
|
|
|
|
|
|
|
# this is not used here! |
385
|
|
|
|
|
|
|
# |
386
|
|
|
|
|
|
|
# printf ("typefirst table: 0x%08lX\n", $lookup_table_offset + $lng + 6); |
387
|
1
|
|
|
|
|
2
|
$i= read (FI, $b, 64); |
388
|
1
|
50
|
|
|
|
3
|
print "WARNING: could not read complete typefirst table; read=$i lng=64\n" |
389
|
|
|
|
|
|
|
unless ($i == 64); |
390
|
1
|
|
|
|
|
3
|
for ($i= 0; $i < 32; $i++) |
391
|
|
|
|
|
|
|
{ |
392
|
32
|
|
|
|
|
35
|
$v= unpack ('v', substr ($b, $i*2, 2)); |
393
|
32
|
|
|
|
|
50
|
push (@ftbl, $v); |
394
|
|
|
|
|
|
|
# print "ftbl[$i]= $v\n"; |
395
|
|
|
|
|
|
|
} |
396
|
|
|
|
|
|
|
# $hdr->{typefirst_table}= \@ftbl; |
397
|
|
|
|
|
|
|
} # lookup table read |
398
|
|
|
|
|
|
|
# else { print "no lookup table present!\n"; } |
399
|
|
|
|
|
|
|
|
400
|
1
|
|
|
|
|
2
|
$obj->{Meta}= 'Plaintext'; |
401
|
1
|
|
|
|
|
12
|
$obj->{dont_decrypt}= $dont_decrypt; |
402
|
1
|
|
|
|
|
2
|
my ($CODE, $CODE_SIZE); # used to decrypt data records |
403
|
|
|
|
|
|
|
|
404
|
1
|
|
|
|
|
2
|
for ($i= 0;; $i++) |
405
|
|
|
|
|
|
|
{ |
406
|
23
|
|
|
|
|
25
|
my ($off, $siz, $type, $lut); |
407
|
|
|
|
|
|
|
|
408
|
23
|
50
|
|
|
|
45
|
if ($lookup_table_offset > 0) |
409
|
|
|
|
|
|
|
{ # use lookup table to seek each record otherwise read file seqentially |
410
|
23
|
100
|
|
|
|
43
|
last if ($i > $#ltbl); |
411
|
|
|
|
|
|
|
|
412
|
22
|
|
|
|
|
35
|
$lut= $ltbl [$i]; |
413
|
|
|
|
|
|
|
|
414
|
22
|
|
|
|
|
34
|
$off= $lut->{off}; |
415
|
22
|
|
|
|
|
27
|
$siz= $lut->{siz} - 6; |
416
|
|
|
|
|
|
|
|
417
|
22
|
50
|
33
|
|
|
110
|
if ($siz < 0 || $off < 0) |
418
|
|
|
|
|
|
|
{ # empty record |
419
|
|
|
|
|
|
|
# print "[$i] type=???? siz=$siz off=$off\n"; |
420
|
0
|
|
|
|
|
0
|
next; |
421
|
|
|
|
|
|
|
} |
422
|
|
|
|
|
|
|
|
423
|
22
|
|
|
|
|
195
|
seek (FI, $off, 0); |
424
|
|
|
|
|
|
|
} |
425
|
|
|
|
|
|
|
|
426
|
22
|
50
|
|
|
|
609
|
last unless (defined ($xrec= &get_recheader (*FI))); |
427
|
|
|
|
|
|
|
|
428
|
22
|
|
|
|
|
40
|
$siz= $xrec->{length}- 6; |
429
|
22
|
|
|
|
|
33
|
$type= $xrec->{type}; |
430
|
|
|
|
|
|
|
# the real record data! |
431
|
22
|
|
|
|
|
33
|
read (FI, $b, $siz); |
432
|
|
|
|
|
|
|
|
433
|
22
|
50
|
33
|
|
|
94
|
if ($type < 0 || $type >= 32) |
434
|
|
|
|
|
|
|
{ |
435
|
0
|
|
|
|
|
0
|
print "WARNING: unknown record type: $type; IGNORED\n"; |
436
|
0
|
|
|
|
|
0
|
&print_recheader (*STDOUT, |
437
|
|
|
|
|
|
|
"record [$i] type=$type siz=$siz off=$off", |
438
|
|
|
|
|
|
|
$xrec); |
439
|
0
|
|
|
|
|
0
|
&hex_dump ($b); |
440
|
0
|
|
|
|
|
0
|
next; |
441
|
|
|
|
|
|
|
} |
442
|
|
|
|
|
|
|
|
443
|
22
|
50
|
|
|
|
43
|
if (defined ($lut)) |
444
|
|
|
|
|
|
|
{ # additional record data from the LUT |
445
|
22
|
|
|
|
|
35
|
$xrec->{off}= $off; |
446
|
22
|
|
|
|
|
59
|
$xrec->{flags}= $lut->{flags}; |
447
|
22
|
|
|
|
|
38
|
$xrec->{filters}= $lut->{filters}; |
448
|
|
|
|
|
|
|
} |
449
|
|
|
|
|
|
|
|
450
|
22
|
|
|
|
|
43
|
&analyze_record ($obj, $xrec, $i, $b); |
451
|
|
|
|
|
|
|
} |
452
|
|
|
|
|
|
|
# print "LUT table size: i=$i\n"; |
453
|
|
|
|
|
|
|
|
454
|
1
|
|
|
|
|
19
|
close (FI); |
455
|
|
|
|
|
|
|
|
456
|
1
|
|
|
|
|
18
|
$obj; |
457
|
|
|
|
|
|
|
} |
458
|
|
|
|
|
|
|
|
459
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
460
|
|
|
|
|
|
|
sub analyze_record |
461
|
|
|
|
|
|
|
{ |
462
|
22
|
|
|
22
|
0
|
38
|
my ($obj, $xrec, $i, $b)= @_; |
463
|
|
|
|
|
|
|
|
464
|
22
|
|
|
|
|
25
|
my $type= $xrec->{type}; |
465
|
22
|
|
|
|
|
37
|
my $siz= $xrec->{length}-6; |
466
|
|
|
|
|
|
|
|
467
|
|
|
|
|
|
|
# $xrec only contains only fields from the LUT |
468
|
|
|
|
|
|
|
# filters:length:type:off:status:flags:idx |
469
|
|
|
|
|
|
|
# inserts only $xrec->{data} which contains the (decrypted) data |
470
|
|
|
|
|
|
|
|
471
|
22
|
50
|
66
|
|
|
108
|
if ($type > 1 && $obj->{Meta} eq 'Encrypted' && !$obj->{dont_decrypt}) |
|
|
|
33
|
|
|
|
|
472
|
|
|
|
|
|
|
{ |
473
|
|
|
|
|
|
|
# print "DATA encoded \n"; &hex_dump ($b); |
474
|
|
|
|
|
|
|
|
475
|
0
|
|
|
|
|
0
|
$b= &decrypt_data ($b, $siz, $obj->{Key}); |
476
|
|
|
|
|
|
|
|
477
|
|
|
|
|
|
|
# print "DATA decoded\n"; &hex_dump ($b); print "\n"; |
478
|
|
|
|
|
|
|
} |
479
|
|
|
|
|
|
|
|
480
|
22
|
|
|
|
|
70
|
$xrec->{data}= $b; |
481
|
|
|
|
|
|
|
|
482
|
|
|
|
|
|
|
# specially handled objects |
483
|
22
|
100
|
|
|
|
47
|
if ($type == 9) # NOTE |
484
|
|
|
|
|
|
|
{ # note records may be missing, but they are accessed according |
485
|
|
|
|
|
|
|
# to their index, thus leave the blank entries in the table. |
486
|
1
|
|
|
|
|
3
|
$obj->{Types}->[9]->[$xrec->{idx}]= $xrec; |
487
|
1
|
|
|
|
|
4
|
return; |
488
|
|
|
|
|
|
|
} |
489
|
|
|
|
|
|
|
|
490
|
21
|
|
|
|
|
23
|
push (@{$obj->{Types}->[$type]}, $xrec); |
|
21
|
|
|
|
|
59
|
|
491
|
|
|
|
|
|
|
|
492
|
21
|
50
|
66
|
|
|
104
|
if ($type > 1 |
|
|
|
33
|
|
|
|
|
493
|
|
|
|
|
|
|
&& $obj->{Meta} eq 'Encrypted' |
494
|
|
|
|
|
|
|
&& $obj->{dont_decrypt}) |
495
|
|
|
|
|
|
|
{ # no usuefull data to process if encrypted |
496
|
0
|
|
|
|
|
0
|
return; |
497
|
|
|
|
|
|
|
} |
498
|
|
|
|
|
|
|
|
499
|
|
|
|
|
|
|
# Main DB type decoder |
500
|
21
|
100
|
|
|
|
101
|
if ($type == 0) |
|
|
50
|
|
|
|
|
|
|
|
100
|
|
|
|
|
|
|
|
100
|
|
|
|
|
|
|
|
100
|
|
|
|
|
|
|
|
100
|
|
|
|
|
|
|
|
50
|
|
|
|
|
|
501
|
|
|
|
|
|
|
{ # record header; this is actually read twice and was already |
502
|
|
|
|
|
|
|
# decoded, see above |
503
|
|
|
|
|
|
|
# NOTE: The DB header seems to get modified as soon as an |
504
|
|
|
|
|
|
|
# application opens the database to indicate it is busy |
505
|
|
|
|
|
|
|
# by setting the viewpoint table offset to NULL |
506
|
|
|
|
|
|
|
} |
507
|
|
|
|
|
|
|
elsif ($type == 1) |
508
|
|
|
|
|
|
|
{ # password record; this code is very experimental! |
509
|
0
|
|
|
|
|
0
|
$obj->{Meta}= 'Encrypted'; |
510
|
|
|
|
|
|
|
|
511
|
0
|
0
|
|
|
|
0
|
if ($obj->{dont_decrypt}) |
512
|
|
|
|
|
|
|
{ # do not attempt to decrypt this password |
513
|
0
|
|
|
|
|
0
|
return; |
514
|
|
|
|
|
|
|
} |
515
|
|
|
|
|
|
|
|
516
|
|
|
|
|
|
|
# decode and print the password |
517
|
0
|
|
|
|
|
0
|
my ($pass, $key)= &decrypt_password ($b, $siz); |
518
|
0
|
|
|
|
|
0
|
$obj->{Password}= $pass; |
519
|
0
|
|
|
|
|
0
|
$obj->{Key}= $key; |
520
|
|
|
|
|
|
|
# print "session key:\n"; |
521
|
|
|
|
|
|
|
# &hex_dump ($key); |
522
|
|
|
|
|
|
|
} # END of type == 1 processing; password record |
523
|
|
|
|
|
|
|
|
524
|
|
|
|
|
|
|
elsif ($type == 4) # CARDDEF |
525
|
|
|
|
|
|
|
{ # only one record of this type allowed!! |
526
|
1
|
|
|
|
|
4
|
$obj->{carddef}= &get_carddef ($b); |
527
|
|
|
|
|
|
|
} |
528
|
|
|
|
|
|
|
elsif ($type == 6) # FIELDDEF |
529
|
|
|
|
|
|
|
{ |
530
|
13
|
|
|
|
|
24
|
my ($fdef, $rec_size)= &get_fielddef ($b); |
531
|
13
|
|
|
|
|
17
|
push (@{$obj->{fielddef}}, $fdef); |
|
13
|
|
|
|
|
29
|
|
532
|
13
|
100
|
|
|
|
42
|
$obj->{rec_size}= $rec_size if ($rec_size > $obj->{rec_size}); |
533
|
|
|
|
|
|
|
} |
534
|
|
|
|
|
|
|
elsif ($type == 7) # VIEWPTDEF |
535
|
|
|
|
|
|
|
{ |
536
|
|
|
|
|
|
|
# print ">>> view point defintion\n"; &hex_dump ($b); |
537
|
1
|
|
|
|
|
7
|
my $vptd= &get_viewptdef ($b); |
538
|
|
|
|
|
|
|
# $vptd->show_viewptdef (*STDOUT); |
539
|
1
|
|
|
|
|
2
|
push (@{$obj->{viewptdef}}, $vptd); |
|
1
|
|
|
|
|
3
|
|
540
|
1
|
|
|
|
|
2
|
$vptd->{index}= $#{$obj->{viewptdef}}; |
|
1
|
|
|
|
|
9
|
|
541
|
|
|
|
|
|
|
} |
542
|
|
|
|
|
|
|
elsif ($type == 10) # VIEWPTTABLE |
543
|
|
|
|
|
|
|
{ |
544
|
|
|
|
|
|
|
# print ">>> view point table\n"; &hex_dump ($b); |
545
|
1
|
|
|
|
|
2
|
push (@{$obj->{viewpttable}}, &get_viewpttable ($b)); |
|
1
|
|
|
|
|
6
|
|
546
|
|
|
|
|
|
|
} |
547
|
|
|
|
|
|
|
elsif ($type == 13) # CARDPAGEDEF |
548
|
|
|
|
|
|
|
{ # only none or one record of this type allowed!! |
549
|
0
|
|
|
|
|
0
|
$obj->{cardpagedef}= &get_cardpagedef ($b); |
550
|
|
|
|
|
|
|
} |
551
|
|
|
|
|
|
|
|
552
|
21
|
50
|
|
|
|
72
|
unless ($REC_TYPE[$type]) |
553
|
|
|
|
|
|
|
{ |
554
|
|
|
|
|
|
|
# application specific data |
555
|
0
|
|
|
|
|
0
|
my $APT= $obj->{APT}; |
556
|
|
|
|
|
|
|
|
557
|
0
|
0
|
0
|
|
|
0
|
if ($type == 14 && $APT eq 'ADB') |
558
|
|
|
|
|
|
|
{ |
559
|
0
|
|
|
|
|
0
|
$obj->decode_type14 (*STDOUT, $b); |
560
|
|
|
|
|
|
|
} |
561
|
|
|
|
|
|
|
else |
562
|
|
|
|
|
|
|
{ # dump info about other unknown field types |
563
|
0
|
|
0
|
|
|
0
|
my $off= $xrec->{off} || 'SEQ'; |
564
|
0
|
|
|
|
|
0
|
print "[$i] off=$off siz=$siz type=$type APT='$APT'\n"; |
565
|
0
|
|
|
|
|
0
|
&print_recheader (*STDOUT, "record [$i]:", $xrec); |
566
|
|
|
|
|
|
|
|
567
|
|
|
|
|
|
|
# print "b='$b'\n"; |
568
|
0
|
|
|
|
|
0
|
&hex_dump ($b); |
569
|
0
|
|
|
|
|
0
|
$obj->{has_unknown_records}++; |
570
|
|
|
|
|
|
|
} |
571
|
|
|
|
|
|
|
} |
572
|
|
|
|
|
|
|
} |
573
|
|
|
|
|
|
|
|
574
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
575
|
|
|
|
|
|
|
sub has_errors |
576
|
|
|
|
|
|
|
{ |
577
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
578
|
0
|
0
|
|
|
|
0
|
return 1 if ($self->{has_unknown_records}); |
579
|
0
|
|
|
|
|
0
|
0; |
580
|
|
|
|
|
|
|
} |
581
|
|
|
|
|
|
|
|
582
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
583
|
|
|
|
|
|
|
sub saveDB |
584
|
|
|
|
|
|
|
{ |
585
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
586
|
0
|
|
0
|
|
|
0
|
my $fnmo= shift || $self->{Filename}; |
587
|
|
|
|
|
|
|
|
588
|
0
|
|
|
|
|
0
|
my $hdr= $self->{Header}; |
589
|
0
|
|
|
|
|
0
|
my $Types= $self->{Types}; |
590
|
|
|
|
|
|
|
|
591
|
0
|
|
|
|
|
0
|
my ($type, $Data, $rec, $lng, $idx); |
592
|
|
|
|
|
|
|
|
593
|
|
|
|
|
|
|
# fixup header if necessary |
594
|
0
|
|
|
|
|
0
|
$Data= $Types->[0]; |
595
|
|
|
|
|
|
|
|
596
|
0
|
|
|
|
|
0
|
my ($off)= 4; |
597
|
0
|
|
|
|
|
0
|
my (@lut, @ftype, $ftype); # lookup table and first type table |
598
|
0
|
|
|
|
|
0
|
my $lut= 0; |
599
|
0
|
|
|
|
|
0
|
my $num_recs= 0; |
600
|
|
|
|
|
|
|
|
601
|
|
|
|
|
|
|
# calculate lookup table and firsttype table |
602
|
|
|
|
|
|
|
# . for each record type: calculate size of each entry |
603
|
|
|
|
|
|
|
# print "lut_size= $#lut $lut\n"; |
604
|
0
|
|
|
|
|
0
|
for ($type= 0; $type < 32; $type++) |
605
|
|
|
|
|
|
|
{ |
606
|
0
|
|
|
|
|
0
|
push (@ftype, $lut); |
607
|
0
|
|
|
|
|
0
|
$Data= $Types->[$type]; |
608
|
|
|
|
|
|
|
|
609
|
0
|
|
|
|
|
0
|
for ($idx= 0; $idx <= $#$Data; $idx++) |
610
|
|
|
|
|
|
|
{ |
611
|
0
|
|
|
|
|
0
|
$rec= $Data->[$idx]; |
612
|
|
|
|
|
|
|
|
613
|
|
|
|
|
|
|
# print ">>> save: type=$type idx=$idx\n"; |
614
|
|
|
|
|
|
|
|
615
|
|
|
|
|
|
|
# T2D, TEST: note records may be blank!! |
616
|
0
|
0
|
|
|
|
0
|
if (defined ($rec)) |
617
|
|
|
|
|
|
|
{ # populated record to be saved |
618
|
0
|
|
|
|
|
0
|
$lng= length ($rec->{data}); |
619
|
|
|
|
|
|
|
|
620
|
0
|
|
|
|
|
0
|
$rec->{off}= $off; |
621
|
0
|
|
|
|
|
0
|
$off += ($rec->{'length'}= $lng + 6); # 6 off ??? |
622
|
0
|
|
|
|
|
0
|
$rec->{idx}= $idx; |
623
|
|
|
|
|
|
|
|
624
|
0
|
0
|
|
|
|
0
|
unless (defined ($rec->{type})) |
625
|
|
|
|
|
|
|
{ # set type if not alrady done |
626
|
0
|
|
|
|
|
0
|
$rec->{type}= $type; |
627
|
|
|
|
|
|
|
} |
628
|
|
|
|
|
|
|
|
629
|
0
|
0
|
|
|
|
0
|
unless (defined ($rec->{status})) |
630
|
|
|
|
|
|
|
{ # set type if not alrady done |
631
|
0
|
|
|
|
|
0
|
$rec->{status}= 2; # T2D: status == 2 means what ??? |
632
|
|
|
|
|
|
|
} |
633
|
|
|
|
|
|
|
} |
634
|
|
|
|
|
|
|
else |
635
|
|
|
|
|
|
|
{ # empty record, set up an entry for the lookup table |
636
|
0
|
|
|
|
|
0
|
print ">>>>> save rec type=$type idx=$idx undefined!\n"; |
637
|
|
|
|
|
|
|
|
638
|
0
|
|
|
|
|
0
|
$rec= |
639
|
|
|
|
|
|
|
{ |
640
|
|
|
|
|
|
|
off => 0, |
641
|
|
|
|
|
|
|
'length'=> 0, |
642
|
|
|
|
|
|
|
flags => 0, |
643
|
|
|
|
|
|
|
filters => 0, |
644
|
|
|
|
|
|
|
}; |
645
|
|
|
|
|
|
|
} |
646
|
|
|
|
|
|
|
|
647
|
0
|
|
|
|
|
0
|
$lut [$lut++]= $rec; |
648
|
0
|
|
|
|
|
0
|
$num_recs++; |
649
|
|
|
|
|
|
|
} |
650
|
|
|
|
|
|
|
} |
651
|
|
|
|
|
|
|
|
652
|
|
|
|
|
|
|
# print "lut_size= $#lut $lut num_recs=$num_recs off=$off\n"; |
653
|
|
|
|
|
|
|
|
654
|
0
|
|
|
|
|
0
|
$hdr->{lookup_table_offset}= $off; |
655
|
0
|
|
|
|
|
0
|
$hdr->{num_recs}= $num_recs; |
656
|
|
|
|
|
|
|
|
657
|
0
|
|
|
|
|
0
|
local *FO; |
658
|
0
|
0
|
|
|
|
0
|
open (FO, ">$fnmo") || die; |
659
|
0
|
|
|
|
|
0
|
binmode (FI); # MS-DOS systems need this, T2D: how about Mac? |
660
|
|
|
|
|
|
|
|
661
|
|
|
|
|
|
|
# save db header; see also loader |
662
|
0
|
|
|
|
|
0
|
print FO $hdr->{sig}; |
663
|
0
|
|
|
|
|
0
|
&put_recheader (*FO, $hdr->{recheader}); |
664
|
0
|
|
|
|
|
0
|
my $time= $hdr->{'time'}; |
665
|
0
|
|
|
|
|
0
|
my $b= pack ('vCCvvVCCCvv', |
666
|
|
|
|
|
|
|
$hdr->{release_version}, |
667
|
|
|
|
|
|
|
$hdr->{file_type}, $hdr->{file_status}, |
668
|
|
|
|
|
|
|
$hdr->{cur_viewpt}, $hdr->{num_recs}, |
669
|
|
|
|
|
|
|
$off, # lookup_table_offset |
670
|
|
|
|
|
|
|
$time->{year}-1900, $time->{mon}-1, |
671
|
|
|
|
|
|
|
$time->{day}-1, $time->{min}, |
672
|
|
|
|
|
|
|
$hdr->{viewpt_hash}, |
673
|
|
|
|
|
|
|
); |
674
|
0
|
|
|
|
|
0
|
print FO $b; |
675
|
|
|
|
|
|
|
|
676
|
|
|
|
|
|
|
# save each record for each type |
677
|
0
|
|
|
|
|
0
|
for ($type= 1; $type < 32; $type++) |
678
|
|
|
|
|
|
|
{ |
679
|
0
|
|
|
|
|
0
|
$Data= $Types->[$type]; |
680
|
|
|
|
|
|
|
|
681
|
0
|
|
|
|
|
0
|
for ($idx= 0; $idx <= $#$Data; $idx++) |
682
|
|
|
|
|
|
|
{ |
683
|
0
|
|
|
|
|
0
|
$rec= $Data->[$idx]; |
684
|
|
|
|
|
|
|
|
685
|
0
|
0
|
|
|
|
0
|
next unless (defined ($rec->{data})); # empty records |
686
|
|
|
|
|
|
|
# print ">>> save data records type=$type idx=$idx\n"; |
687
|
0
|
|
|
|
|
0
|
&put_recheader (*FO, $rec); |
688
|
0
|
|
|
|
|
0
|
print FO $rec->{data}; |
689
|
|
|
|
|
|
|
} |
690
|
|
|
|
|
|
|
} |
691
|
|
|
|
|
|
|
|
692
|
|
|
|
|
|
|
# print "lut_size= $#lut $lut\n"; |
693
|
|
|
|
|
|
|
|
694
|
|
|
|
|
|
|
# save lookup table |
695
|
0
|
|
|
|
|
0
|
$rec= |
696
|
|
|
|
|
|
|
{ |
697
|
|
|
|
|
|
|
'type' => 31, |
698
|
|
|
|
|
|
|
'status' => 0, |
699
|
|
|
|
|
|
|
'length' => ($#lut+1)*8+6, |
700
|
|
|
|
|
|
|
'idx' => 0, |
701
|
|
|
|
|
|
|
}; |
702
|
|
|
|
|
|
|
|
703
|
0
|
|
|
|
|
0
|
&put_recheader (*FO, $rec); |
704
|
0
|
|
|
|
|
0
|
foreach $lut (@lut) |
705
|
|
|
|
|
|
|
{ |
706
|
0
|
|
|
|
|
0
|
my $off_low= $lut->{off}%256; |
707
|
0
|
|
|
|
|
0
|
my $off= $lut->{off}/256; |
708
|
|
|
|
|
|
|
|
709
|
0
|
|
|
|
|
0
|
my $b= pack ('vvCCv', |
710
|
|
|
|
|
|
|
$lut->{'length'}, |
711
|
|
|
|
|
|
|
$lut->{filters}, $lut->{flags}, |
712
|
|
|
|
|
|
|
$off_low, $off |
713
|
|
|
|
|
|
|
); |
714
|
|
|
|
|
|
|
|
715
|
0
|
|
|
|
|
0
|
print FO $b; |
716
|
|
|
|
|
|
|
} |
717
|
|
|
|
|
|
|
|
718
|
|
|
|
|
|
|
# save firsttype table |
719
|
0
|
|
|
|
|
0
|
foreach $ftype (@ftype) |
720
|
|
|
|
|
|
|
{ |
721
|
0
|
|
|
|
|
0
|
my $b= pack ('v', $ftype); |
722
|
0
|
|
|
|
|
0
|
print FO $b; |
723
|
|
|
|
|
|
|
} |
724
|
|
|
|
|
|
|
|
725
|
0
|
|
|
|
|
0
|
close (FO); |
726
|
|
|
|
|
|
|
} |
727
|
|
|
|
|
|
|
|
728
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
729
|
|
|
|
|
|
|
sub print_summary |
730
|
|
|
|
|
|
|
{ |
731
|
0
|
|
|
0
|
0
|
0
|
my $db= shift; |
732
|
0
|
|
|
|
|
0
|
my $prt_hdr= shift; |
733
|
|
|
|
|
|
|
|
734
|
0
|
|
|
|
|
0
|
my $hdr= $db->{Header}; |
735
|
0
|
|
|
|
|
0
|
my $t= $hdr->{time}; |
736
|
0
|
|
|
|
|
0
|
my $min= $t->{min}; |
737
|
0
|
|
|
|
|
0
|
my $h= int ($min/60); |
738
|
0
|
|
|
|
|
0
|
$min= $min%60; |
739
|
|
|
|
|
|
|
|
740
|
0
|
0
|
|
|
|
0
|
printf ("Type %-24s Recs View Hash %-16s Comment\n", |
741
|
|
|
|
|
|
|
'Filename', 'created') |
742
|
|
|
|
|
|
|
if ($prt_hdr); |
743
|
|
|
|
|
|
|
|
744
|
0
|
|
|
|
|
0
|
my $Comment; |
745
|
0
|
0
|
|
|
|
0
|
$Comment .= ' CORRUPTED!' if ($db->has_errors); |
746
|
0
|
0
|
|
|
|
0
|
$Comment .= ' Password' if ($db->{Meta} eq 'Encrypted'); |
747
|
|
|
|
|
|
|
|
748
|
0
|
|
|
|
|
0
|
printf ("%-4s %-24s %5d %4d 0x%04X %4d-%02d-%02d %2d:%02d%s\n", |
749
|
|
|
|
|
|
|
$db->{APT}, $db->{Filename}, |
750
|
|
|
|
|
|
|
$hdr->{num_recs}, |
751
|
|
|
|
|
|
|
$hdr->{cur_viewpt}, $hdr->{viewpt_hash}, |
752
|
|
|
|
|
|
|
$t->{year}, $t->{mon}, $t->{day}, $h, $min, |
753
|
|
|
|
|
|
|
$Comment, |
754
|
|
|
|
|
|
|
); |
755
|
|
|
|
|
|
|
} |
756
|
|
|
|
|
|
|
|
757
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
758
|
|
|
|
|
|
|
sub get_field_def |
759
|
|
|
|
|
|
|
{ |
760
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
761
|
0
|
|
|
|
|
0
|
my $num= shift; |
762
|
|
|
|
|
|
|
|
763
|
0
|
|
|
|
|
0
|
$self->{fielddef}->[$num]; |
764
|
|
|
|
|
|
|
} |
765
|
|
|
|
|
|
|
|
766
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
767
|
|
|
|
|
|
|
sub show_db_def |
768
|
|
|
|
|
|
|
{ |
769
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
770
|
0
|
|
|
|
|
0
|
local *FO= shift; |
771
|
|
|
|
|
|
|
|
772
|
0
|
|
|
|
|
0
|
my $Fdef= $self->{'fielddef'}; |
773
|
0
|
|
|
|
|
0
|
my $field; |
774
|
0
|
|
|
|
|
0
|
my $num= 0; |
775
|
0
|
|
|
|
|
0
|
my %off= (); # sorted by offset |
776
|
0
|
|
|
|
|
0
|
my $off; |
777
|
|
|
|
|
|
|
|
778
|
0
|
|
|
|
|
0
|
my $hdr= sprintf ("[##] ## %-12s Siz %-24s FID Off Res Flg\n", |
779
|
|
|
|
|
|
|
"Type", "Name"); |
780
|
0
|
|
|
|
|
0
|
print FO $delim, "\n"; |
781
|
0
|
|
|
|
|
0
|
print FO "DB def by field number\n", $hdr; |
782
|
|
|
|
|
|
|
|
783
|
0
|
|
|
|
|
0
|
foreach $field (@$Fdef) |
784
|
|
|
|
|
|
|
{ |
785
|
0
|
|
|
|
|
0
|
$off= &show_field_def (*FO, $field, $num++); |
786
|
0
|
|
|
|
|
0
|
push (@{$off{$off}}, $field); |
|
0
|
|
|
|
|
0
|
|
787
|
|
|
|
|
|
|
} |
788
|
|
|
|
|
|
|
|
789
|
0
|
|
|
|
|
0
|
$num= 0; |
790
|
|
|
|
|
|
|
|
791
|
0
|
|
|
|
|
0
|
print FO $delim, "\n", "DB def by offset position\n", $hdr; |
792
|
0
|
|
|
|
|
0
|
foreach $off (sort keys %off) |
793
|
|
|
|
|
|
|
{ |
794
|
0
|
|
|
|
|
0
|
foreach $field (@{$off{$off}}) |
|
0
|
|
|
|
|
0
|
|
795
|
|
|
|
|
|
|
{ |
796
|
0
|
|
|
|
|
0
|
&show_field_def (*FO, $field, $num); |
797
|
|
|
|
|
|
|
} |
798
|
0
|
|
|
|
|
0
|
$num++ |
799
|
|
|
|
|
|
|
} |
800
|
|
|
|
|
|
|
|
801
|
0
|
|
|
|
|
0
|
print FO $delim, "\n"; |
802
|
|
|
|
|
|
|
} |
803
|
|
|
|
|
|
|
|
804
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
805
|
|
|
|
|
|
|
sub show_card_def |
806
|
|
|
|
|
|
|
{ |
807
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
808
|
0
|
|
|
|
|
0
|
local *FO= shift; |
809
|
|
|
|
|
|
|
|
810
|
0
|
|
|
|
|
0
|
my $Cdef= $self->{'carddef'}; |
811
|
0
|
0
|
|
|
|
0
|
return if ($#$Cdef < 0); |
812
|
0
|
|
|
|
|
0
|
my ($field, $f); |
813
|
|
|
|
|
|
|
|
814
|
0
|
|
|
|
|
0
|
print FO "card definition:\n"; |
815
|
0
|
|
|
|
|
0
|
my $i= 0; |
816
|
0
|
|
|
|
|
0
|
foreach $field (@$Cdef) |
817
|
|
|
|
|
|
|
{ |
818
|
|
|
|
|
|
|
# &show_field_window ($field); |
819
|
0
|
|
|
|
|
0
|
printf FO ("field [%2d]:", $i++); |
820
|
0
|
|
|
|
|
0
|
foreach $f (sort keys %$field) |
821
|
|
|
|
|
|
|
{ |
822
|
0
|
0
|
0
|
|
|
0
|
if ($f eq 'Parent' || $f eq 'Style') |
823
|
|
|
|
|
|
|
{ |
824
|
0
|
|
|
|
|
0
|
printf (" %s=%8X,", $f, $field->{$f}); |
825
|
|
|
|
|
|
|
} else { |
826
|
0
|
|
|
|
|
0
|
printf (" %s=%3d,", $f, $field->{$f}); |
827
|
|
|
|
|
|
|
} |
828
|
|
|
|
|
|
|
} |
829
|
0
|
|
|
|
|
0
|
print "\n"; |
830
|
|
|
|
|
|
|
} |
831
|
|
|
|
|
|
|
} |
832
|
|
|
|
|
|
|
|
833
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
834
|
|
|
|
|
|
|
sub dump_data |
835
|
|
|
|
|
|
|
{ |
836
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
837
|
|
|
|
|
|
|
|
838
|
0
|
|
|
|
|
0
|
my $APT= $self->{APT}; |
839
|
0
|
|
0
|
|
|
0
|
my $T= $self->{Types} || die; |
840
|
0
|
|
|
|
|
0
|
my $D= $T->[11]; # array of data records |
841
|
0
|
|
|
|
|
0
|
my $N= $T->[9]; # array of note records |
842
|
|
|
|
|
|
|
|
843
|
0
|
|
0
|
|
|
0
|
my $rec_beg= shift || 0; |
844
|
0
|
|
0
|
|
|
0
|
my $rec_end= shift || $#$D; |
845
|
0
|
|
0
|
|
|
0
|
my $Fdef= shift || $self->{fielddef}; # array of field definitions |
846
|
|
|
|
|
|
|
|
847
|
0
|
|
|
|
|
0
|
my ($rec, $field); |
848
|
|
|
|
|
|
|
|
849
|
0
|
|
|
|
|
0
|
print "show_data\n"; |
850
|
0
|
|
|
|
|
0
|
foreach $rec ($rec_beg .. $rec_end) |
851
|
|
|
|
|
|
|
{ |
852
|
0
|
|
0
|
|
|
0
|
my $d= $D->[$rec] || next; |
853
|
0
|
|
0
|
|
|
0
|
my $b= $d->{data} || next; |
854
|
|
|
|
|
|
|
|
855
|
0
|
|
|
|
|
0
|
my ($ok, $o)= &fetch_data ($b, $Fdef, $N, $APT); |
856
|
0
|
|
|
|
|
0
|
&dump_data_record ($b, $ok, $o); |
857
|
|
|
|
|
|
|
} |
858
|
|
|
|
|
|
|
} |
859
|
|
|
|
|
|
|
|
860
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
861
|
|
|
|
|
|
|
sub dump_type |
862
|
|
|
|
|
|
|
{ |
863
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
864
|
0
|
|
|
|
|
0
|
local *FO= shift; |
865
|
0
|
|
|
|
|
0
|
my $Ty= shift; # if undef, dump all items |
866
|
0
|
|
0
|
|
|
0
|
my $Format= shift || 'auto'; |
867
|
|
|
|
|
|
|
|
868
|
|
|
|
|
|
|
# print '# ', join (' ', keys %$self), "\n"; |
869
|
0
|
|
|
|
|
0
|
my ($T, $Ty_from, $Ty_end); |
870
|
|
|
|
|
|
|
|
871
|
0
|
0
|
|
|
|
0
|
unless (defined ($T= $self->{Types})) |
872
|
|
|
|
|
|
|
{ |
873
|
0
|
|
|
|
|
0
|
print STDERR "can't access Type table in $self\n"; |
874
|
0
|
|
|
|
|
0
|
return; |
875
|
|
|
|
|
|
|
} |
876
|
|
|
|
|
|
|
|
877
|
0
|
0
|
|
|
|
0
|
if (defined ($Ty)) { $Ty_from= $Ty_end= $Ty; } |
|
0
|
|
|
|
|
0
|
|
878
|
0
|
|
|
|
|
0
|
else { $Ty_from= 0; $Ty_end= 255; } |
|
0
|
|
|
|
|
0
|
|
879
|
|
|
|
|
|
|
|
880
|
0
|
|
|
|
|
0
|
for ($Ty= $Ty_from; $Ty <= $Ty_end; $Ty++) |
881
|
|
|
|
|
|
|
{ |
882
|
0
|
|
|
|
|
0
|
my $D= $T->[$Ty]; |
883
|
0
|
|
|
|
|
0
|
my $c= $#$D; |
884
|
0
|
0
|
|
|
|
0
|
next if ($c == -1); |
885
|
|
|
|
|
|
|
|
886
|
0
|
|
|
|
|
0
|
my $format= $Format; |
887
|
0
|
0
|
|
|
|
0
|
if ($Format eq 'auto') |
888
|
|
|
|
|
|
|
{ # see @REC_TYPE |
889
|
0
|
0
|
0
|
|
|
0
|
if ($Ty == 5 || $Ty == 9 || $Ty == 11) { $format= 'QP'; } |
|
0
|
|
0
|
|
|
0
|
|
890
|
0
|
|
|
|
|
0
|
else { $format= 'HEX'; } |
891
|
|
|
|
|
|
|
} |
892
|
|
|
|
|
|
|
|
893
|
0
|
|
0
|
|
|
0
|
my $ty_str= $REC_TYPE[$Ty] || "USER$Ty"; |
894
|
|
|
|
|
|
|
|
895
|
0
|
|
|
|
|
0
|
my ($i, $Dk, $Dv, $cp, $ch, $cv, $lng, $llng); |
896
|
0
|
|
|
|
|
0
|
for ($i= 0; $i <= $c; $i++) |
897
|
|
|
|
|
|
|
{ |
898
|
0
|
|
|
|
|
0
|
print FO "$Ty $ty_str $i/$c\n"; |
899
|
|
|
|
|
|
|
|
900
|
0
|
|
|
|
|
0
|
$Dv= $D->[$i]; |
901
|
|
|
|
|
|
|
# NOTE: fields not written: off (completely redundant) |
902
|
|
|
|
|
|
|
# off, filters, and flags come from the LUT |
903
|
|
|
|
|
|
|
# print FO '# ZZ ', join (' ', keys %$Dv), "\n"; |
904
|
0
|
|
|
|
|
0
|
foreach $Dk (qw(type idx length status filters flags)) |
905
|
|
|
|
|
|
|
{ |
906
|
0
|
0
|
|
|
|
0
|
next unless (defined ($Dv->{$Dk})); |
907
|
0
|
|
|
|
|
0
|
print FO "<$Dk>$Dv->{$Dk}\n"; |
908
|
|
|
|
|
|
|
} |
909
|
|
|
|
|
|
|
|
910
|
0
|
|
|
|
|
0
|
print FO "\n"; |
911
|
0
|
0
|
|
|
|
0
|
if ($format eq 'HEX') |
912
|
|
|
|
|
|
|
{ |
913
|
0
|
|
|
|
|
0
|
&hex_dump ($Dv->{data}, *FO); |
914
|
|
|
|
|
|
|
} |
915
|
|
|
|
|
|
|
else # especially if ($format eq 'QP') |
916
|
|
|
|
|
|
|
{ |
917
|
0
|
|
|
|
|
0
|
my $data= $Dv->{data}; |
918
|
0
|
|
|
|
|
0
|
$lng= length ($data); |
919
|
0
|
|
|
|
|
0
|
for ($cp= 0; $cp < $lng; $cp++) |
920
|
|
|
|
|
|
|
{ |
921
|
0
|
|
|
|
|
0
|
$cv= unpack ('C', $ch= substr ($data, $cp, 1)); |
922
|
|
|
|
|
|
|
|
923
|
0
|
0
|
0
|
|
|
0
|
if (($cv >= 0x00 && $cv <= 0x1F) |
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
924
|
|
|
|
|
|
|
|| ($cv >= 0x3C && $cv <= 0x3E) |
925
|
|
|
|
|
|
|
|| ($cv >= 0x7F && $cv <= 0xFF) |
926
|
|
|
|
|
|
|
) |
927
|
|
|
|
|
|
|
{ |
928
|
0
|
|
|
|
|
0
|
$ch= sprintf ("=%02X", $cv); |
929
|
0
|
|
|
|
|
0
|
$llng += 3; |
930
|
|
|
|
|
|
|
} |
931
|
0
|
|
|
|
|
0
|
else { $llng++; } |
932
|
|
|
|
|
|
|
|
933
|
0
|
|
|
|
|
0
|
print FO $ch; |
934
|
0
|
0
|
|
|
|
0
|
if ($llng > 72) { print FO "=\n"; $llng= 0; } |
|
0
|
|
|
|
|
0
|
|
|
0
|
|
|
|
|
0
|
|
935
|
|
|
|
|
|
|
} |
936
|
0
|
0
|
|
|
|
0
|
if ($llng > 0) { print FO "\n"; $llng= 0; } |
|
0
|
|
|
|
|
0
|
|
|
0
|
|
|
|
|
0
|
|
937
|
|
|
|
|
|
|
} |
938
|
|
|
|
|
|
|
|
939
|
0
|
|
|
|
|
0
|
print FO "\n\n\n"; |
940
|
|
|
|
|
|
|
} |
941
|
|
|
|
|
|
|
} |
942
|
|
|
|
|
|
|
} |
943
|
|
|
|
|
|
|
|
944
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
945
|
|
|
|
|
|
|
# load ASCII file; name should be changed... |
946
|
|
|
|
|
|
|
sub loader |
947
|
|
|
|
|
|
|
{ |
948
|
0
|
|
|
0
|
0
|
0
|
my $self= shift; |
949
|
0
|
|
|
|
|
0
|
local *FI= shift; |
950
|
|
|
|
|
|
|
|
951
|
0
|
|
|
|
|
0
|
my $status= 'undef'; |
952
|
0
|
|
|
|
|
0
|
my ($rec, $counter, $b, $format); |
953
|
0
|
|
|
|
|
0
|
while () |
954
|
|
|
|
|
|
|
{ |
955
|
0
|
|
|
|
|
0
|
chomp; |
956
|
|
|
|
|
|
|
# print ">>> $_\n"; |
957
|
|
|
|
|
|
|
|
958
|
0
|
0
|
0
|
|
|
0
|
if (m##) |
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
959
|
|
|
|
|
|
|
{ |
960
|
0
|
|
|
|
|
0
|
$rec= {}; |
961
|
0
|
|
|
|
|
0
|
$status= 'record'; |
962
|
0
|
|
|
|
|
0
|
$counter++; |
963
|
0
|
|
|
|
|
0
|
$b= ''; |
964
|
|
|
|
|
|
|
} |
965
|
|
|
|
|
|
|
elsif (m##) |
966
|
|
|
|
|
|
|
{ |
967
|
0
|
0
|
0
|
|
|
0
|
if ($status ne 'record' && $status ne 'data') |
968
|
|
|
|
|
|
|
{ |
969
|
0
|
|
|
|
|
0
|
print "WARNING: unexpected status $status\n"; |
970
|
|
|
|
|
|
|
} |
971
|
|
|
|
|
|
|
|
972
|
|
|
|
|
|
|
# analyze header if necessary: |
973
|
|
|
|
|
|
|
# filters:length:type:off:status:flags:idx |
974
|
|
|
|
|
|
|
# print ">>> insert record: ", join (':', %$rec), "\n"; |
975
|
|
|
|
|
|
|
# &hex_dump ($b); |
976
|
0
|
|
|
|
|
0
|
&analyze_record ($self, $rec, $counter, $b); |
977
|
0
|
|
|
|
|
0
|
$status= 'undef'; |
978
|
|
|
|
|
|
|
} |
979
|
|
|
|
|
|
|
elsif (m##) |
980
|
|
|
|
|
|
|
{ |
981
|
0
|
|
|
|
|
0
|
$format= $1; |
982
|
0
|
|
|
|
|
0
|
$status= 'data'; |
983
|
|
|
|
|
|
|
} |
984
|
|
|
|
|
|
|
elsif (m##) |
985
|
|
|
|
|
|
|
{ |
986
|
0
|
|
|
|
|
0
|
$status= 'record'; |
987
|
|
|
|
|
|
|
} |
988
|
|
|
|
|
|
|
elsif (m#<(type|idx|length|status|filters|flags)>(.*)#) |
989
|
|
|
|
|
|
|
{ |
990
|
0
|
|
|
|
|
0
|
$rec->{$1}= $2; |
991
|
|
|
|
|
|
|
} |
992
|
|
|
|
|
|
|
elsif ($status eq 'data') |
993
|
|
|
|
|
|
|
{ |
994
|
0
|
0
|
|
|
|
0
|
if ($format eq 'QP') |
|
|
0
|
|
|
|
|
|
995
|
|
|
|
|
|
|
{ |
996
|
0
|
|
|
|
|
0
|
s/=$//; |
997
|
0
|
|
|
|
|
0
|
s/=([a-fA-F0-9][a-fA-F0-9])/pack("C", hex($1))/eg; |
|
0
|
|
|
|
|
0
|
|
998
|
0
|
|
|
|
|
0
|
$b .= $_; |
999
|
|
|
|
|
|
|
} |
1000
|
|
|
|
|
|
|
elsif ($format eq 'HEX') |
1001
|
|
|
|
|
|
|
{ |
1002
|
0
|
|
|
|
|
0
|
my @x= split (/\|/); |
1003
|
0
|
|
|
|
|
0
|
@x= split (' ', $x[0]); |
1004
|
0
|
|
|
|
|
0
|
shift (@x); |
1005
|
|
|
|
|
|
|
# print "[", join (':', @x), "]\n"; |
1006
|
0
|
|
|
|
|
0
|
$b .= pack ("C*", map { hex ($_); } @x); |
|
0
|
|
|
|
|
0
|
|
1007
|
|
|
|
|
|
|
} |
1008
|
|
|
|
|
|
|
else |
1009
|
|
|
|
|
|
|
{ |
1010
|
0
|
|
|
|
|
0
|
print "WARNING: unexpected data format: '$format'\n"; |
1011
|
|
|
|
|
|
|
} |
1012
|
|
|
|
|
|
|
} |
1013
|
|
|
|
|
|
|
elsif (/^#/ || /^[ \t]*$/) {} # comment |
1014
|
|
|
|
|
|
|
else |
1015
|
|
|
|
|
|
|
{ |
1016
|
0
|
|
|
|
|
0
|
print "WARNING: unexpected data: '$_'\n"; |
1017
|
|
|
|
|
|
|
} |
1018
|
|
|
|
|
|
|
} |
1019
|
|
|
|
|
|
|
} |
1020
|
|
|
|
|
|
|
|
1021
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1022
|
|
|
|
|
|
|
sub TIEARRAY |
1023
|
|
|
|
|
|
|
{ |
1024
|
1
|
|
|
1
|
|
15
|
return $_[1]; |
1025
|
|
|
|
|
|
|
} |
1026
|
|
|
|
|
|
|
|
1027
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1028
|
|
|
|
|
|
|
sub FETCH |
1029
|
|
|
|
|
|
|
{ |
1030
|
1
|
|
|
1
|
|
11
|
my $db= shift; |
1031
|
1
|
|
|
|
|
2
|
my $idx= shift; |
1032
|
|
|
|
|
|
|
|
1033
|
1
|
|
50
|
|
|
4
|
my $T= $db->{Types} || die 'not a database'; |
1034
|
1
|
|
|
|
|
2
|
my $D= $T->[11]; # array of data records |
1035
|
1
|
50
|
|
|
|
4
|
return undef if ($idx > $#$D); |
1036
|
|
|
|
|
|
|
|
1037
|
1
|
|
|
|
|
2
|
my $Dx= $D->[$idx]; # data record for the given index |
1038
|
1
|
|
|
|
|
2
|
my $rv; |
1039
|
|
|
|
|
|
|
|
1040
|
1
|
50
|
|
|
|
3
|
unless (defined ($rv= $Dx->{obj})) |
1041
|
|
|
|
|
|
|
{ # no record data was previously stored, fetch that |
1042
|
1
|
|
|
|
|
1
|
my $N= $T->[9]; # array of note records |
1043
|
1
|
|
|
|
|
3
|
my $F= $db->{fielddef}; |
1044
|
1
|
|
|
|
|
2
|
my $b= $Dx->{data}; |
1045
|
1
|
|
|
|
|
2
|
my $APT= $db->{APT}; |
1046
|
|
|
|
|
|
|
|
1047
|
|
|
|
|
|
|
# print "FETCH: T=$T D=$D N=$N F=$F b=$b\n"; |
1048
|
1
|
|
|
|
|
5
|
my ($ok, $o)= &fetch_data ($b, $F, $N, $APT); |
1049
|
|
|
|
|
|
|
# &dump_data_record ($b, $ok, $o); |
1050
|
|
|
|
|
|
|
|
1051
|
1
|
|
|
|
|
3
|
$Dx->{obj}= $rv= $o; |
1052
|
1
|
|
|
|
|
2
|
$Dx->{ok}= $ok; |
1053
|
|
|
|
|
|
|
} |
1054
|
|
|
|
|
|
|
|
1055
|
1
|
|
|
|
|
5
|
return $rv; |
1056
|
|
|
|
|
|
|
} |
1057
|
|
|
|
|
|
|
|
1058
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1059
|
|
|
|
|
|
|
sub FETCH_data_raw |
1060
|
|
|
|
|
|
|
{ |
1061
|
0
|
|
|
0
|
0
|
0
|
my $db= shift; |
1062
|
0
|
|
|
|
|
0
|
my $idx= shift; |
1063
|
|
|
|
|
|
|
|
1064
|
0
|
|
0
|
|
|
0
|
my $T= $db->{Types} || return undef; |
1065
|
0
|
|
|
|
|
0
|
my $D= $T->[11]; # array of data records |
1066
|
0
|
0
|
|
|
|
0
|
return undef if ($idx > $#$D); |
1067
|
|
|
|
|
|
|
|
1068
|
0
|
|
|
|
|
0
|
$D->[$idx]->{data}; # data record for the given index |
1069
|
|
|
|
|
|
|
} |
1070
|
|
|
|
|
|
|
|
1071
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1072
|
|
|
|
|
|
|
sub FETCH_note_raw |
1073
|
|
|
|
|
|
|
{ |
1074
|
0
|
|
|
0
|
0
|
0
|
my $db= shift; |
1075
|
0
|
|
|
|
|
0
|
my $idx= shift; |
1076
|
|
|
|
|
|
|
|
1077
|
0
|
|
0
|
|
|
0
|
my $T= $db->{Types} || return undef; |
1078
|
0
|
|
|
|
|
0
|
my $N= $T->[9]; # array of note records |
1079
|
0
|
0
|
|
|
|
0
|
return undef if ($idx > $#$N); |
1080
|
|
|
|
|
|
|
|
1081
|
0
|
|
|
|
|
0
|
$N->[$idx]->{data}; # data record for the given index |
1082
|
|
|
|
|
|
|
} |
1083
|
|
|
|
|
|
|
|
1084
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1085
|
|
|
|
|
|
|
sub STORE |
1086
|
|
|
|
|
|
|
{ |
1087
|
0
|
|
|
0
|
|
0
|
my $db= shift; |
1088
|
0
|
|
|
|
|
0
|
my $idx= shift; |
1089
|
0
|
|
|
|
|
0
|
my $val= shift; |
1090
|
|
|
|
|
|
|
# print "STORE: ", join (':', %$val), "\n"; |
1091
|
|
|
|
|
|
|
|
1092
|
0
|
|
0
|
|
|
0
|
my $T= $db->{Types} || die; |
1093
|
0
|
|
|
|
|
0
|
my $D= $T->[11]; # array of data records |
1094
|
0
|
|
|
|
|
0
|
my $N= $T->[9]; # array of note records |
1095
|
0
|
|
|
|
|
0
|
my $F= $db->{fielddef}; |
1096
|
0
|
|
|
|
|
0
|
my $APT= $db->{APT}; |
1097
|
|
|
|
|
|
|
|
1098
|
0
|
|
|
|
|
0
|
my $Dx; |
1099
|
|
|
|
|
|
|
|
1100
|
0
|
0
|
|
|
|
0
|
if ($idx > $#$D) |
1101
|
|
|
|
|
|
|
{ |
1102
|
|
|
|
|
|
|
# print "adding records: num=$#$D idx=$idx\n"; |
1103
|
0
|
|
|
|
|
0
|
$Dx= { 'data' => '' }; |
1104
|
|
|
|
|
|
|
} |
1105
|
|
|
|
|
|
|
else |
1106
|
|
|
|
|
|
|
{ |
1107
|
0
|
|
|
|
|
0
|
$Dx= $D->[$idx]; # data record for the given index |
1108
|
|
|
|
|
|
|
} |
1109
|
|
|
|
|
|
|
|
1110
|
0
|
|
|
|
|
0
|
my ($ok, $b)= &store_data ($val, $F, $N, $APT, $db->{rec_size}); |
1111
|
|
|
|
|
|
|
|
1112
|
0
|
|
|
|
|
0
|
$Dx->{data}= $b; |
1113
|
0
|
|
|
|
|
0
|
undef ($Dx->{obj}); |
1114
|
0
|
|
|
|
|
0
|
undef ($Dx->{ok}); |
1115
|
0
|
|
|
|
|
0
|
$D->[$idx]= $Dx; |
1116
|
|
|
|
|
|
|
|
1117
|
|
|
|
|
|
|
# T2D: unfinished |
1118
|
|
|
|
|
|
|
# missing items: refreshing and/or invalidating view points |
1119
|
0
|
|
|
|
|
0
|
$db->{update}++; |
1120
|
|
|
|
|
|
|
} |
1121
|
|
|
|
|
|
|
|
1122
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1123
|
|
|
|
|
|
|
sub STORE_data_raw |
1124
|
|
|
|
|
|
|
{ |
1125
|
0
|
|
|
0
|
0
|
0
|
my $db= shift; |
1126
|
0
|
|
|
|
|
0
|
my $idx= shift; |
1127
|
0
|
|
|
|
|
0
|
my $data= shift; |
1128
|
|
|
|
|
|
|
|
1129
|
0
|
|
0
|
|
|
0
|
my $T= $db->{Types} || die; |
1130
|
0
|
|
|
|
|
0
|
my $D= $T->[11]; # array of data records |
1131
|
0
|
|
|
|
|
0
|
$D->[$idx]->{data}= $data; |
1132
|
0
|
|
|
|
|
0
|
$db->{update}++; |
1133
|
|
|
|
|
|
|
} |
1134
|
|
|
|
|
|
|
|
1135
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1136
|
|
|
|
|
|
|
sub STORE_note_raw |
1137
|
|
|
|
|
|
|
{ |
1138
|
0
|
|
|
0
|
0
|
0
|
my $db= shift; |
1139
|
0
|
|
|
|
|
0
|
my $idx= shift; |
1140
|
0
|
|
|
|
|
0
|
my $data= shift; |
1141
|
|
|
|
|
|
|
|
1142
|
0
|
|
0
|
|
|
0
|
my $T= $db->{Types} || die; |
1143
|
0
|
|
|
|
|
0
|
my $N= $T->[9]; # array of note records |
1144
|
0
|
|
|
|
|
0
|
$N->[$idx]->{data}= $data; |
1145
|
0
|
|
|
|
|
0
|
$db->{update}++; |
1146
|
|
|
|
|
|
|
} |
1147
|
|
|
|
|
|
|
|
1148
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1149
|
|
|
|
|
|
|
sub FETCHSIZE |
1150
|
|
|
|
|
|
|
{ |
1151
|
0
|
|
|
0
|
|
0
|
my $db= shift; |
1152
|
0
|
|
|
|
|
0
|
return 1 + $db->get_last_index(); |
1153
|
|
|
|
|
|
|
} |
1154
|
|
|
|
|
|
|
|
1155
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1156
|
|
|
|
|
|
|
sub get_last_index |
1157
|
|
|
|
|
|
|
{ |
1158
|
0
|
|
|
0
|
0
|
0
|
my $db= shift; |
1159
|
|
|
|
|
|
|
|
1160
|
0
|
|
0
|
|
|
0
|
my $T= $db->{Types} || die; |
1161
|
0
|
|
|
|
|
0
|
my $D= $T->[11]; # array of data records |
1162
|
0
|
|
|
|
|
0
|
return $#$D; |
1163
|
|
|
|
|
|
|
} |
1164
|
|
|
|
|
|
|
|
1165
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1166
|
|
|
|
|
|
|
sub get_str |
1167
|
|
|
|
|
|
|
{ |
1168
|
3
|
|
|
3
|
0
|
4
|
my $b= shift; |
1169
|
3
|
|
|
|
|
3
|
my $off= shift; |
1170
|
|
|
|
|
|
|
|
1171
|
3
|
|
|
|
|
6
|
my $res= substr ($$b, $off); |
1172
|
3
|
|
|
|
|
7
|
my $idx= index ($res, "\000"); |
1173
|
3
|
50
|
|
|
|
8
|
$res= substr ($res, 0, $idx) if ($idx >= 0); |
1174
|
3
|
|
|
|
|
7
|
$res; |
1175
|
|
|
|
|
|
|
} |
1176
|
|
|
|
|
|
|
|
1177
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1178
|
|
|
|
|
|
|
sub fmt_date |
1179
|
|
|
|
|
|
|
{ |
1180
|
1
|
|
|
1
|
0
|
3
|
my $str= shift; |
1181
|
|
|
|
|
|
|
|
1182
|
1
|
|
|
|
|
3
|
my ($year, $mon, $day)= unpack ('CCC', $str); |
1183
|
1
|
50
|
33
|
|
|
13
|
($year == $no_year && $mon == $no_mon && $day == $no_day) |
1184
|
|
|
|
|
|
|
? '' # empty date field |
1185
|
|
|
|
|
|
|
: sprintf ("%d-%02d-%02d", 1900 + $year, $mon+1, $day+1); |
1186
|
|
|
|
|
|
|
} |
1187
|
|
|
|
|
|
|
|
1188
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1189
|
|
|
|
|
|
|
sub pack_date |
1190
|
|
|
|
|
|
|
{ |
1191
|
0
|
|
|
0
|
0
|
0
|
my $val= shift; |
1192
|
0
|
|
|
|
|
0
|
my ($year, $mon, $day); |
1193
|
|
|
|
|
|
|
|
1194
|
0
|
|
|
|
|
0
|
$year= $mon= $day= $no_date; |
1195
|
0
|
0
|
|
|
|
0
|
if ($val =~ /(\d+)-(\d+)-(\d+)/) |
1196
|
|
|
|
|
|
|
{ |
1197
|
0
|
|
|
|
|
0
|
($year, $mon, $day)= ($1, $2, $3); |
1198
|
|
|
|
|
|
|
# check for valid dates otherwise set no_date value |
1199
|
0
|
0
|
0
|
|
|
0
|
$year= $mon= $day= $no_date |
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
1200
|
|
|
|
|
|
|
if ($year < 1900 || $year > 2155 |
1201
|
|
|
|
|
|
|
|| $mon < 1 || $mon > 12 |
1202
|
|
|
|
|
|
|
|| $day < 1 || $day > 31); |
1203
|
|
|
|
|
|
|
|
1204
|
0
|
|
|
|
|
0
|
$year -= 1900; |
1205
|
0
|
|
|
|
|
0
|
$mon--; |
1206
|
0
|
|
|
|
|
0
|
$day--; |
1207
|
|
|
|
|
|
|
} |
1208
|
|
|
|
|
|
|
|
1209
|
0
|
|
|
|
|
0
|
pack ('CCC', $year, $mon, $day); |
1210
|
|
|
|
|
|
|
} |
1211
|
|
|
|
|
|
|
|
1212
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1213
|
|
|
|
|
|
|
sub fmt_time |
1214
|
|
|
|
|
|
|
{ |
1215
|
1
|
|
|
1
|
0
|
3
|
my $str= shift; |
1216
|
|
|
|
|
|
|
|
1217
|
1
|
|
|
|
|
2
|
my $val= unpack ('v', $str); |
1218
|
1
|
50
|
33
|
|
|
7
|
return '' if ($val == $no_time || $val == $no_val); |
1219
|
|
|
|
|
|
|
|
1220
|
1
|
|
|
|
|
3
|
my $min= $val % 60; |
1221
|
1
|
|
|
|
|
3
|
my $xval= int ($val / 60); |
1222
|
1
|
|
|
|
|
4
|
sprintf ("%d:%02d", $xval, $min); |
1223
|
|
|
|
|
|
|
} |
1224
|
|
|
|
|
|
|
|
1225
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1226
|
|
|
|
|
|
|
sub fetch_data |
1227
|
|
|
|
|
|
|
{ |
1228
|
1
|
|
|
1
|
0
|
2
|
my $b= shift; # raw binary data |
1229
|
1
|
|
|
|
|
1
|
my $Fdef= shift; # Field Definitions |
1230
|
1
|
|
|
|
|
1
|
my $N= shift; # Notes Data |
1231
|
1
|
|
|
|
|
2
|
my $APT= shift; # application type |
1232
|
|
|
|
|
|
|
|
1233
|
1
|
|
|
|
|
2
|
my $ok= 1; |
1234
|
1
|
|
|
|
|
1
|
my %o; |
1235
|
|
|
|
|
|
|
my %RB; # radio button at offset |
1236
|
0
|
|
|
|
|
0
|
my $field; |
1237
|
|
|
|
|
|
|
|
1238
|
1
|
|
|
|
|
4
|
my @Fdef= @$Fdef; # Field Definition List |
1239
|
1
|
|
|
|
|
2
|
my $APT2; |
1240
|
|
|
|
|
|
|
|
1241
|
1
|
50
|
|
|
|
4
|
if ($APT eq 'ADB') |
1242
|
|
|
|
|
|
|
{ # For appointment book entries we have to analyze if |
1243
|
|
|
|
|
|
|
# the record describes a to-do item or a date or event |
1244
|
|
|
|
|
|
|
|
1245
|
0
|
|
|
|
|
0
|
my $val= unpack ('C', substr ($b, 0x0E, 1)); |
1246
|
0
|
|
|
|
|
0
|
my @TLT= (); |
1247
|
|
|
|
|
|
|
|
1248
|
|
|
|
|
|
|
# if ($val & 0x02) { $APT2= 'Done'; } # checked to-do entry |
1249
|
0
|
0
|
|
|
|
0
|
if ($val & 0x10) { $APT2= 'To-Do'; @TLT= (0, 1, 8..12); } |
|
0
|
0
|
|
|
|
0
|
|
|
0
|
0
|
|
|
|
0
|
|
1250
|
0
|
|
|
|
|
0
|
elsif ($val & 0x20) { $APT2= 'Event'; @TLT= (0..7, 12, 14, 15); } |
|
0
|
|
|
|
|
0
|
|
1251
|
0
|
|
|
|
|
0
|
elsif ($val & 0x80) { $APT2= 'Date'; @TLT= (0..7, 12, 14, 15); } |
|
0
|
|
|
|
|
0
|
|
1252
|
|
|
|
|
|
|
|
1253
|
0
|
|
|
|
|
0
|
$o{'type'}= $APT2; |
1254
|
0
|
|
|
|
|
0
|
$o{'repeat'}= unpack ('C', substr ($b, 0x1A, 1)); |
1255
|
|
|
|
|
|
|
|
1256
|
0
|
|
|
|
|
0
|
@Fdef= map { $Fdef[$_] } @TLT; |
|
0
|
|
|
|
|
0
|
|
1257
|
|
|
|
|
|
|
} |
1258
|
|
|
|
|
|
|
|
1259
|
1
|
|
|
|
|
2
|
FIELD: foreach $field (@Fdef) |
1260
|
|
|
|
|
|
|
{ |
1261
|
13
|
|
|
|
|
18
|
my $type= $field->{ftype}; |
1262
|
13
|
|
|
|
|
19
|
my $off= $field->{off}; |
1263
|
13
|
|
|
|
|
18
|
my $name= $field->{name}; |
1264
|
13
|
|
|
|
|
13
|
my $res; |
1265
|
|
|
|
|
|
|
# printf ("APT= 0x%02X %2d '%s'\n", $off, $type, $name); |
1266
|
|
|
|
|
|
|
|
1267
|
13
|
50
|
66
|
|
|
191
|
if ($type == 0) # BYTE_BOOL |
|
|
100
|
33
|
|
|
|
|
|
|
50
|
66
|
|
|
|
|
|
|
100
|
100
|
|
|
|
|
|
|
100
|
100
|
|
|
|
|
|
|
100
|
33
|
|
|
|
|
|
|
100
|
66
|
|
|
|
|
|
|
100
|
33
|
|
|
|
|
|
|
50
|
66
|
|
|
|
|
|
|
0
|
100
|
|
|
|
|
|
|
0
|
66
|
|
|
|
|
|
|
0
|
33
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
33
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
1268
|
|
|
|
|
|
|
{ |
1269
|
0
|
|
|
|
|
0
|
my $val= unpack ('C', substr ($b, $off, 1)); |
1270
|
0
|
0
|
|
|
|
0
|
$res= ($val) ? 'X' : ''; |
1271
|
|
|
|
|
|
|
} |
1272
|
|
|
|
|
|
|
elsif ($type == 1) # WORD_BOOL |
1273
|
|
|
|
|
|
|
{ |
1274
|
2
|
|
|
|
|
5
|
my $val= unpack ('v', substr ($b, $off, 2)); |
1275
|
2
|
100
|
|
|
|
13
|
$res= ($val) ? 'X' : ''; |
1276
|
|
|
|
|
|
|
} |
1277
|
|
|
|
|
|
|
elsif ($type == 2 && $APT eq 'ADB' && $off eq 0x1B) |
1278
|
|
|
|
|
|
|
{ # Beschreibung bei ADB geht ohne Offset! |
1279
|
0
|
|
|
|
|
0
|
$res= &get_str (\$b, $off); |
1280
|
|
|
|
|
|
|
} |
1281
|
|
|
|
|
|
|
elsif ($type == 2 # STRING |
1282
|
|
|
|
|
|
|
|| $type == 3 # PHONE |
1283
|
|
|
|
|
|
|
|| $type == 4 # NUMBER |
1284
|
|
|
|
|
|
|
|| $type == 6 # CATEGORY |
1285
|
|
|
|
|
|
|
) |
1286
|
|
|
|
|
|
|
{ |
1287
|
3
|
|
|
|
|
7
|
my $offs= unpack ('v', substr ($b, $off, 2)); |
1288
|
3
|
|
|
|
|
10
|
$res= &get_str (\$b, $offs); |
1289
|
|
|
|
|
|
|
} |
1290
|
|
|
|
|
|
|
elsif ($type == 7 # TIME |
1291
|
|
|
|
|
|
|
|| ($type == 24 && $APT eq 'ADB') # Vorlauf |
1292
|
|
|
|
|
|
|
) |
1293
|
|
|
|
|
|
|
{ |
1294
|
|
|
|
|
|
|
#??? next if ($APT eq 'APT' && $APT2 eq 'To-Do'); # overlapping fields |
1295
|
1
|
|
|
|
|
5
|
$res= &fmt_time (substr ($b, $off, 2)); |
1296
|
|
|
|
|
|
|
} |
1297
|
|
|
|
|
|
|
elsif ($type == 8 # DATE |
1298
|
|
|
|
|
|
|
|| ($type == 19 && $APT eq 'ADB') # Beginndatum |
1299
|
|
|
|
|
|
|
) |
1300
|
|
|
|
|
|
|
{ |
1301
|
1
|
|
|
|
|
5
|
$res= &fmt_date (substr ($b, $off, 3)); |
1302
|
|
|
|
|
|
|
} |
1303
|
|
|
|
|
|
|
elsif ($type == 9) # RADIO_BUTTON |
1304
|
|
|
|
|
|
|
{ |
1305
|
2
|
|
|
|
|
5
|
my $val= unpack ('C', substr ($b, $off, 1)); # 2 or 1 byte?? |
1306
|
2
|
|
|
|
|
5
|
my $cnt= ++$RB{$off}; |
1307
|
2
|
100
|
|
|
|
6
|
$res= ($cnt == $val) ? 'X' : ''; |
1308
|
|
|
|
|
|
|
} |
1309
|
|
|
|
|
|
|
elsif ($type == 10) # NOTE |
1310
|
|
|
|
|
|
|
{ |
1311
|
1
|
|
|
|
|
3
|
my $note_number= unpack ('v', substr ($b, $off, 2)); |
1312
|
1
|
|
|
|
|
12
|
$o{"$name&nr"}= $note_number; |
1313
|
1
|
50
|
|
|
|
4
|
unless ($note_number eq $no_note) |
1314
|
|
|
|
|
|
|
{ |
1315
|
1
|
|
|
|
|
1
|
my $nr; |
1316
|
1
|
|
|
|
|
2
|
$nr= $N->[$note_number]; # $nr should be a valid reference! |
1317
|
1
|
50
|
|
|
|
10
|
$res= (defined ($nr)) ? $nr->{data} : ''; |
1318
|
|
|
|
|
|
|
} |
1319
|
|
|
|
|
|
|
} |
1320
|
|
|
|
|
|
|
elsif ($type == 11 # GROUP |
1321
|
|
|
|
|
|
|
|| $type == 12 # STATIC (e.g. Label) |
1322
|
|
|
|
|
|
|
|| $type == 14 # LIST |
1323
|
|
|
|
|
|
|
|| $type == 15 # COMBO |
1324
|
|
|
|
|
|
|
|| ($type == 18 && $APT eq 'ADB') # repeat factor |
1325
|
|
|
|
|
|
|
) # no action ?!?!? |
1326
|
|
|
|
|
|
|
{ |
1327
|
3
|
|
|
|
|
6
|
next FIELD; |
1328
|
|
|
|
|
|
|
} |
1329
|
|
|
|
|
|
|
elsif ($type == 16 && $APT == 'WDB') |
1330
|
|
|
|
|
|
|
{ |
1331
|
0
|
|
|
|
|
0
|
$res= unpack ('v', substr ($b, $off, 2)); |
1332
|
|
|
|
|
|
|
} |
1333
|
|
|
|
|
|
|
elsif ($APT eq 'ADB' |
1334
|
|
|
|
|
|
|
&& ($type == 23 # number of days |
1335
|
|
|
|
|
|
|
|| $type == 20 # date due Faelligkeitsdatum |
1336
|
|
|
|
|
|
|
) |
1337
|
|
|
|
|
|
|
) |
1338
|
|
|
|
|
|
|
{ |
1339
|
0
|
0
|
0
|
|
|
0
|
next if ($type == 23 && $APT2 eq 'To-Do'); |
1340
|
0
|
0
|
0
|
|
|
0
|
next if ($type == 20 && $APT2 ne 'To-Do'); |
1341
|
|
|
|
|
|
|
|
1342
|
0
|
|
|
|
|
0
|
$res= unpack ('v', substr ($b, $off, 2)); # 2 byte integer value |
1343
|
|
|
|
|
|
|
} |
1344
|
|
|
|
|
|
|
elsif ($APT eq 'ADB' && $type == 22) |
1345
|
|
|
|
|
|
|
{ |
1346
|
|
|
|
|
|
|
# print "\n", $delim, "\n>>> U22: APT2='$APT2'\n"; |
1347
|
0
|
0
|
|
|
|
0
|
next unless ($APT2 eq 'To-Do'); # priority code |
1348
|
0
|
|
|
|
|
0
|
$res= substr ($b, $off, 2); |
1349
|
0
|
|
|
|
|
0
|
$res=~ s/\x00//g; |
1350
|
|
|
|
|
|
|
} |
1351
|
|
|
|
|
|
|
else |
1352
|
|
|
|
|
|
|
{ |
1353
|
0
|
|
|
|
|
0
|
$res= "unknown type $type"; |
1354
|
0
|
|
|
|
|
0
|
&show_field_def (*STDOUT, $field, -1); |
1355
|
0
|
|
|
|
|
0
|
$ok= 0; |
1356
|
|
|
|
|
|
|
} |
1357
|
|
|
|
|
|
|
|
1358
|
|
|
|
|
|
|
# print "fetch: name=$name res=$res\n"; |
1359
|
10
|
|
|
|
|
31
|
$o{$name}= $res; |
1360
|
|
|
|
|
|
|
} |
1361
|
|
|
|
|
|
|
|
1362
|
1
|
|
|
|
|
5
|
return ($ok, \%o); |
1363
|
|
|
|
|
|
|
} |
1364
|
|
|
|
|
|
|
|
1365
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1366
|
|
|
|
|
|
|
sub store_data |
1367
|
|
|
|
|
|
|
{ |
1368
|
0
|
|
|
0
|
0
|
0
|
my $data= shift; # record data to be stored into the database |
1369
|
0
|
|
|
|
|
0
|
my $Fdef= shift; # Field Definitions |
1370
|
0
|
|
|
|
|
0
|
my $N= shift; # Notes Data; array of references |
1371
|
0
|
|
|
|
|
0
|
my $APT= shift; # application type |
1372
|
0
|
|
|
|
|
0
|
my $rec_size= shift; # standard record size and next string position |
1373
|
|
|
|
|
|
|
|
1374
|
0
|
|
|
|
|
0
|
my $b_off= 0; # offset into binary data |
1375
|
0
|
|
|
|
|
0
|
my @b= # binary data at each offset |
1376
|
|
|
|
|
|
|
my $b; # final binary data |
1377
|
0
|
|
|
|
|
0
|
my $nil_addr; # address of the NIL string record |
1378
|
|
|
|
|
|
|
# this is set up when there are actually strings |
1379
|
|
|
|
|
|
|
# see notes below |
1380
|
|
|
|
|
|
|
|
1381
|
0
|
|
|
|
|
0
|
my $ok= 1; |
1382
|
0
|
|
|
|
|
0
|
my %RB; |
1383
|
|
|
|
|
|
|
my $field; |
1384
|
|
|
|
|
|
|
|
1385
|
|
|
|
|
|
|
# print "rec_size= $rec_size\n"; |
1386
|
|
|
|
|
|
|
|
1387
|
|
|
|
|
|
|
# NOTE: ADB records should possibly not be handled here at all!!! |
1388
|
|
|
|
|
|
|
|
1389
|
0
|
|
|
|
|
0
|
FIELD: foreach $field (@$Fdef) |
1390
|
|
|
|
|
|
|
{ |
1391
|
0
|
|
|
|
|
0
|
my $type= $field->{ftype}; |
1392
|
0
|
|
|
|
|
0
|
my $off= $field->{off}; |
1393
|
0
|
|
|
|
|
0
|
my $name= $field->{name}; |
1394
|
0
|
0
|
|
|
|
0
|
my $ex= (exists ($data->{$name})) ? 1 : 0; # data value present? |
1395
|
0
|
|
|
|
|
0
|
my $val= $data->{$name}; # actual value |
1396
|
0
|
|
|
|
|
0
|
my $APT2; |
1397
|
|
|
|
|
|
|
|
1398
|
0
|
0
|
|
|
|
0
|
$APT2= $data->{type} if ($APT eq 'ADB'); |
1399
|
|
|
|
|
|
|
|
1400
|
|
|
|
|
|
|
# print "offset= $off type=$type name=$name val='$val'\n"; |
1401
|
|
|
|
|
|
|
|
1402
|
0
|
0
|
0
|
|
|
0
|
if ($type == 0) # BYTEBOOL |
|
|
0
|
0
|
|
|
|
|
|
|
0
|
0
|
|
|
|
|
|
|
0
|
0
|
|
|
|
|
|
|
0
|
0
|
|
|
|
|
|
|
0
|
0
|
|
|
|
|
|
|
0
|
|
|
|
|
|
|
|
0
|
|
|
|
|
|
1403
|
|
|
|
|
|
|
{ |
1404
|
0
|
0
|
|
|
|
0
|
$b [$off]= pack ('C', ($val) ? 1 : 0); |
1405
|
|
|
|
|
|
|
} |
1406
|
|
|
|
|
|
|
elsif ($type == 1) # WORDBOOL |
1407
|
|
|
|
|
|
|
{ |
1408
|
0
|
0
|
|
|
|
0
|
$b [$off]= pack ('v', ($val) ? 1 : 0); |
1409
|
|
|
|
|
|
|
} |
1410
|
|
|
|
|
|
|
elsif ($type == 2 # STRING |
1411
|
|
|
|
|
|
|
|| $type == 3 # PHONE |
1412
|
|
|
|
|
|
|
|| $type == 4 # NUMBER |
1413
|
|
|
|
|
|
|
|| $type == 6 # CATEGORY |
1414
|
|
|
|
|
|
|
) |
1415
|
|
|
|
|
|
|
{ |
1416
|
0
|
0
|
|
|
|
0
|
if ($nil_addr eq '') |
1417
|
|
|
|
|
|
|
{ # create empty string which is used for all other empty strings |
1418
|
|
|
|
|
|
|
# see note below |
1419
|
0
|
|
|
|
|
0
|
$nil_addr= $rec_size; |
1420
|
0
|
|
|
|
|
0
|
$b [$rec_size++]= "\000"; |
1421
|
|
|
|
|
|
|
# print "insert nil at $nil_addr, rec_size=$rec_size\n"; |
1422
|
|
|
|
|
|
|
} |
1423
|
|
|
|
|
|
|
|
1424
|
0
|
0
|
|
|
|
0
|
if ($val) |
1425
|
|
|
|
|
|
|
{ |
1426
|
0
|
|
|
|
|
0
|
$b [$off] = pack ('v', $rec_size); |
1427
|
0
|
|
|
|
|
0
|
$b [$rec_size]= $val . "\000"; |
1428
|
0
|
|
|
|
|
0
|
$rec_size += length ($val) + 1; |
1429
|
|
|
|
|
|
|
} |
1430
|
|
|
|
|
|
|
else |
1431
|
|
|
|
|
|
|
{ # store pointer to the empty string record |
1432
|
0
|
|
|
|
|
0
|
$b [$off] = pack ('v', $nil_addr); |
1433
|
|
|
|
|
|
|
} |
1434
|
|
|
|
|
|
|
# &hex_dump ($b[$off]); |
1435
|
|
|
|
|
|
|
} |
1436
|
|
|
|
|
|
|
elsif ($type == 7) # TIME |
1437
|
|
|
|
|
|
|
{ |
1438
|
0
|
0
|
0
|
|
|
0
|
next if ($APT eq 'ADB' && $APT2 eq 'To-Do'); |
1439
|
|
|
|
|
|
|
|
1440
|
0
|
|
|
|
|
0
|
my ($h, $m, $t); |
1441
|
0
|
|
|
|
|
0
|
$h= $val; |
1442
|
0
|
0
|
|
|
|
0
|
($h, $m)= ($1, $2) if ($val =~ /(\d+)[:\.](\d+)/); |
1443
|
0
|
|
|
|
|
0
|
$t= $h*60+$m; |
1444
|
0
|
0
|
0
|
|
|
0
|
$t= $no_time if (!$ex || $t < 0 || $t > $no_time); |
|
|
|
0
|
|
|
|
|
1445
|
0
|
|
|
|
|
0
|
$b [$off]= pack ('v', $t); |
1446
|
|
|
|
|
|
|
} |
1447
|
|
|
|
|
|
|
elsif ($type == 8) # DATE |
1448
|
|
|
|
|
|
|
{ |
1449
|
0
|
|
|
|
|
0
|
$b [$off]= &pack_date ($val); |
1450
|
|
|
|
|
|
|
} |
1451
|
|
|
|
|
|
|
elsif ($type == 9) # RADIO_BUTTON |
1452
|
|
|
|
|
|
|
{ # several radio buttons point to the same offset |
1453
|
|
|
|
|
|
|
# the value can be the number of the button pointing there |
1454
|
|
|
|
|
|
|
# or 0 when no button is checked |
1455
|
|
|
|
|
|
|
|
1456
|
0
|
|
|
|
|
0
|
my $v; # value to be stored |
1457
|
0
|
0
|
|
|
|
0
|
my $checked= ($val) ? 1 : 0; |
1458
|
0
|
0
|
|
|
|
0
|
$checked= 0 if ($v= $RB{$off}); # only the first button is valid |
1459
|
0
|
0
|
|
|
|
0
|
$RB{$off}= $v= $field->{res} if ($checked); |
1460
|
|
|
|
|
|
|
|
1461
|
0
|
|
|
|
|
0
|
$b [$off]= pack ('v', $v); # Note: should be 'c' ?!?! |
1462
|
|
|
|
|
|
|
} |
1463
|
|
|
|
|
|
|
elsif ($type == 10) # NOTE |
1464
|
|
|
|
|
|
|
{ # store note record |
1465
|
|
|
|
|
|
|
|
1466
|
|
|
|
|
|
|
# possible cases: |
1467
|
|
|
|
|
|
|
# stored | new | action |
1468
|
|
|
|
|
|
|
# no | no | no action, $no_note is already stored |
1469
|
|
|
|
|
|
|
# no | yes | store new note number |
1470
|
|
|
|
|
|
|
# yes | no | T2D: delete old note, but how?? |
1471
|
|
|
|
|
|
|
# yes | yes | store note number and replace the note |
1472
|
|
|
|
|
|
|
|
1473
|
0
|
|
|
|
|
0
|
my $note_nr= $no_note; |
1474
|
0
|
|
|
|
|
0
|
my $xn= "$name&nr"; |
1475
|
0
|
0
|
|
|
|
0
|
$note_nr= $data->{$xn} if (defined ($data->{$xn})); # stored note |
1476
|
|
|
|
|
|
|
|
1477
|
0
|
0
|
0
|
|
|
0
|
if ($note_nr == $no_note && $val ne '') |
|
|
0
|
0
|
|
|
|
|
|
|
0
|
0
|
|
|
|
|
1478
|
|
|
|
|
|
|
{ # no note before but a valid note: create new note record |
1479
|
0
|
|
|
|
|
0
|
push (@$N, { data => $val }); |
1480
|
0
|
|
|
|
|
0
|
$data->{$xn}= $note_nr= $#$N; |
1481
|
|
|
|
|
|
|
} |
1482
|
|
|
|
|
|
|
elsif ($note_nr != $no_note && $val eq '') |
1483
|
|
|
|
|
|
|
{ # T2D: delete note!! |
1484
|
|
|
|
|
|
|
# this leaves an empty note record in the database !!! |
1485
|
0
|
|
|
|
|
0
|
undef ($N->[$note_nr]->{data}); # T2D, Test |
1486
|
0
|
|
|
|
|
0
|
$data->{$xn}= $note_nr= $no_note; |
1487
|
|
|
|
|
|
|
} |
1488
|
|
|
|
|
|
|
elsif ($note_nr != $no_note && $val ne '') |
1489
|
|
|
|
|
|
|
{ # replace existing note |
1490
|
0
|
|
|
|
|
0
|
$N->[$note_nr]->{data}= $val; |
1491
|
|
|
|
|
|
|
} |
1492
|
|
|
|
|
|
|
|
1493
|
0
|
|
|
|
|
0
|
$b [$off]= pack ('v', $note_nr); |
1494
|
|
|
|
|
|
|
} |
1495
|
|
|
|
|
|
|
elsif ($type == 11 # GROUP |
1496
|
|
|
|
|
|
|
|| $type == 12 # STATIC |
1497
|
|
|
|
|
|
|
|| $type == 14 # LIST |
1498
|
|
|
|
|
|
|
|| $type == 15 # COMBO |
1499
|
|
|
|
|
|
|
) # no action ?!?!? |
1500
|
|
|
|
|
|
|
{ |
1501
|
0
|
|
|
|
|
0
|
next FIELD; |
1502
|
|
|
|
|
|
|
} |
1503
|
|
|
|
|
|
|
else |
1504
|
|
|
|
|
|
|
{ |
1505
|
0
|
|
|
|
|
0
|
print "store_data: ERROR! unknown type $type\n"; |
1506
|
0
|
|
|
|
|
0
|
&show_field_def (*STDOUT, $field, -1); |
1507
|
0
|
|
|
|
|
0
|
print "value: $val\n"; |
1508
|
0
|
|
|
|
|
0
|
$ok= 0; |
1509
|
|
|
|
|
|
|
} |
1510
|
|
|
|
|
|
|
} |
1511
|
|
|
|
|
|
|
|
1512
|
0
|
0
|
|
|
|
0
|
if ($ok) |
1513
|
|
|
|
|
|
|
{ |
1514
|
0
|
|
|
|
|
0
|
$b= join ('', @b); |
1515
|
|
|
|
|
|
|
|
1516
|
0
|
0
|
|
|
|
0
|
if (length ($b) != $rec_size) |
1517
|
|
|
|
|
|
|
{ |
1518
|
0
|
|
|
|
|
0
|
print "ERROR: resulting record size does not match!\n", |
1519
|
|
|
|
|
|
|
"length=", length ($b), " rec_size=$rec_size\n"; |
1520
|
0
|
|
|
|
|
0
|
&hex_dump ($b); |
1521
|
0
|
|
|
|
|
0
|
my ($x, $y); |
1522
|
0
|
|
|
|
|
0
|
for ($x= 0; $x <= $#b; $x++) |
1523
|
|
|
|
|
|
|
{ |
1524
|
0
|
0
|
|
|
|
0
|
next unless ($y= $b[$x]); |
1525
|
0
|
|
|
|
|
0
|
printf ("[%02d] %2d '%s'\n", $x, length ($y), $y); |
1526
|
|
|
|
|
|
|
} |
1527
|
|
|
|
|
|
|
} |
1528
|
|
|
|
|
|
|
} |
1529
|
|
|
|
|
|
|
|
1530
|
|
|
|
|
|
|
# T2D: unfinished |
1531
|
0
|
|
|
|
|
0
|
return ($ok, $b); |
1532
|
|
|
|
|
|
|
} |
1533
|
|
|
|
|
|
|
|
1534
|
|
|
|
|
|
|
# NOTES: |
1535
|
|
|
|
|
|
|
# Empty Strings are stored as null character at the beginning of the |
1536
|
|
|
|
|
|
|
# extended data record. All empty strings point to the same address. |
1537
|
|
|
|
|
|
|
# An empty string is stored even when all strings have a value. |
1538
|
|
|
|
|
|
|
|
1539
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1540
|
|
|
|
|
|
|
# read a 6 byte record header |
1541
|
|
|
|
|
|
|
sub get_recheader |
1542
|
|
|
|
|
|
|
{ |
1543
|
24
|
|
|
24
|
0
|
42
|
local *F= shift; |
1544
|
24
|
|
|
|
|
128
|
my $b; |
1545
|
|
|
|
|
|
|
|
1546
|
24
|
50
|
|
|
|
186
|
read (F, $b, 6) || return undef; |
1547
|
24
|
|
|
|
|
70
|
my ($type, $status, $length, $idx)= unpack ('CCvv', $b); |
1548
|
|
|
|
|
|
|
|
1549
|
24
|
|
|
|
|
101
|
my $rec= |
1550
|
|
|
|
|
|
|
{ |
1551
|
|
|
|
|
|
|
'type' => $type, |
1552
|
|
|
|
|
|
|
'status' => $status, |
1553
|
|
|
|
|
|
|
'length' => $length, |
1554
|
|
|
|
|
|
|
'idx' => $idx, |
1555
|
|
|
|
|
|
|
}; |
1556
|
|
|
|
|
|
|
|
1557
|
24
|
|
|
|
|
86
|
$rec; |
1558
|
|
|
|
|
|
|
} |
1559
|
|
|
|
|
|
|
|
1560
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1561
|
|
|
|
|
|
|
# write a 6 byte record header |
1562
|
|
|
|
|
|
|
sub put_recheader |
1563
|
|
|
|
|
|
|
{ |
1564
|
0
|
|
|
0
|
0
|
0
|
local *F= shift; |
1565
|
0
|
|
|
|
|
0
|
my $r= shift; |
1566
|
|
|
|
|
|
|
|
1567
|
0
|
|
|
|
|
0
|
my $b= pack ('CCvv', $r->{'type'}, $r->{'status'}, |
1568
|
|
|
|
|
|
|
$r->{'length'}, $r->{'idx'}); |
1569
|
0
|
|
|
|
|
0
|
print F $b; |
1570
|
|
|
|
|
|
|
} |
1571
|
|
|
|
|
|
|
|
1572
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1573
|
|
|
|
|
|
|
sub fmt_time_stamp |
1574
|
|
|
|
|
|
|
{ |
1575
|
0
|
|
|
0
|
0
|
0
|
my $time= shift; |
1576
|
0
|
|
|
|
|
0
|
my $Time= sprintf ("%d-%02d-%02d %2d:%02d", |
1577
|
|
|
|
|
|
|
$time->{'year'}, $time->{'mon'}+1, $time->{'day'}+1, |
1578
|
|
|
|
|
|
|
$time->{'min'} / 60, $time->{'min'} % 60); |
1579
|
|
|
|
|
|
|
|
1580
|
0
|
|
|
|
|
0
|
$Time; |
1581
|
|
|
|
|
|
|
} |
1582
|
|
|
|
|
|
|
|
1583
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1584
|
|
|
|
|
|
|
sub get_carddef |
1585
|
|
|
|
|
|
|
{ |
1586
|
1
|
|
|
1
|
0
|
3
|
my $def= shift; |
1587
|
1
|
|
|
|
|
2
|
my @wins; |
1588
|
1
|
|
|
|
|
2
|
my $num= 0; |
1589
|
|
|
|
|
|
|
|
1590
|
|
|
|
|
|
|
# print ">>> processing card definition\n"; |
1591
|
1
|
|
|
|
|
3
|
while ($def) |
1592
|
|
|
|
|
|
|
{ |
1593
|
13
|
|
|
|
|
18
|
my $pw= substr ($def, 0, 20); |
1594
|
13
|
|
|
|
|
17
|
$def= substr ($def, 20); |
1595
|
|
|
|
|
|
|
|
1596
|
13
|
|
|
|
|
31
|
my ($u, $x, $y, $w, $h, $Lsize, $style, $parent)= |
1597
|
|
|
|
|
|
|
unpack ('VvvvvvVv', $pw); |
1598
|
|
|
|
|
|
|
|
1599
|
|
|
|
|
|
|
# printf ("[%3d] x=%3d y=%3d w=%3d h=%3d L=%3d S=0x%08lX P=0x%04X\n", |
1600
|
|
|
|
|
|
|
# $num, $x, $y, $w, $h, $Lsize, $style, $parent); |
1601
|
|
|
|
|
|
|
|
1602
|
13
|
|
|
|
|
16
|
$num++; |
1603
|
|
|
|
|
|
|
|
1604
|
13
|
|
|
|
|
40
|
my $win= |
1605
|
|
|
|
|
|
|
{ |
1606
|
|
|
|
|
|
|
'x' => $x, |
1607
|
|
|
|
|
|
|
'y' => $y, |
1608
|
|
|
|
|
|
|
'w' => $w, |
1609
|
|
|
|
|
|
|
'h' => $h, |
1610
|
|
|
|
|
|
|
'Lsize' => $Lsize, |
1611
|
|
|
|
|
|
|
'Style' => $style, |
1612
|
|
|
|
|
|
|
'Parent' => $parent, |
1613
|
|
|
|
|
|
|
}; |
1614
|
|
|
|
|
|
|
|
1615
|
13
|
|
|
|
|
32
|
push (@wins, $win); |
1616
|
|
|
|
|
|
|
} |
1617
|
|
|
|
|
|
|
|
1618
|
1
|
|
|
|
|
4
|
\@wins; |
1619
|
|
|
|
|
|
|
} |
1620
|
|
|
|
|
|
|
|
1621
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1622
|
|
|
|
|
|
|
sub get_fielddef |
1623
|
|
|
|
|
|
|
{ |
1624
|
13
|
|
|
13
|
0
|
22
|
my $def= shift; |
1625
|
|
|
|
|
|
|
|
1626
|
13
|
|
|
|
|
42
|
my ($ftype, $fid, $off, $flg, $res)= unpack ('CCvCv', $def); |
1627
|
13
|
|
|
|
|
28
|
my $name= substr ($def, 7, length ($def)-8); |
1628
|
13
|
|
|
|
|
21
|
$name=~ s/\&//g; |
1629
|
|
|
|
|
|
|
|
1630
|
13
|
|
|
|
|
80
|
my $fd= |
1631
|
|
|
|
|
|
|
{ |
1632
|
|
|
|
|
|
|
'ftype' => $ftype, |
1633
|
|
|
|
|
|
|
'Ftype' => $FIELD_TYPE [$ftype]->{Desc}, |
1634
|
|
|
|
|
|
|
'fid' => $fid, |
1635
|
|
|
|
|
|
|
'off' => $off, |
1636
|
|
|
|
|
|
|
'flg' => $flg, |
1637
|
|
|
|
|
|
|
'res' => $res, |
1638
|
|
|
|
|
|
|
'name' => $name, |
1639
|
|
|
|
|
|
|
}; |
1640
|
|
|
|
|
|
|
|
1641
|
13
|
|
|
|
|
24
|
$off += $FIELD_TYPE [$ftype]->{Size}; |
1642
|
13
|
|
|
|
|
38
|
($fd, $off); |
1643
|
|
|
|
|
|
|
} |
1644
|
|
|
|
|
|
|
|
1645
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1646
|
|
|
|
|
|
|
sub get_cardpagedef |
1647
|
|
|
|
|
|
|
{ |
1648
|
0
|
|
|
0
|
0
|
|
my $def= shift; |
1649
|
|
|
|
|
|
|
|
1650
|
|
|
|
|
|
|
# print ">>> processing card page definition\n"; |
1651
|
0
|
|
|
|
|
|
my @pages; |
1652
|
0
|
|
|
|
|
|
my ($PW, $CP, $PC, @ps, @pc, $i); |
1653
|
|
|
|
|
|
|
|
1654
|
0
|
|
|
|
|
|
($PW, $CP, $PC, |
1655
|
|
|
|
|
|
|
$ps[1], $ps[2], $ps[3], $ps[4], |
1656
|
|
|
|
|
|
|
$pc[1], $pc[2], $pc[3], $pc[4])= unpack ('vvvvvvvvvv', $def); |
1657
|
|
|
|
|
|
|
|
1658
|
|
|
|
|
|
|
# print ">>>> CP=$CP PC=$PC\n"; |
1659
|
0
|
|
|
|
|
|
for ($i= 1; $i <= $PC; $i++) |
1660
|
|
|
|
|
|
|
{ |
1661
|
0
|
|
|
|
|
|
push (@pages, { 'num' => $i, 'start' => $ps[$i], 'size' => $pc[$i] }); |
1662
|
|
|
|
|
|
|
# print ">>>>> [$i] start=$ps[$i] size=$pc[$i]\n"; |
1663
|
|
|
|
|
|
|
} |
1664
|
|
|
|
|
|
|
|
1665
|
0
|
|
|
|
|
|
\@pages; |
1666
|
|
|
|
|
|
|
} |
1667
|
|
|
|
|
|
|
|
1668
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1669
|
|
|
|
|
|
|
sub show_field_def |
1670
|
|
|
|
|
|
|
{ |
1671
|
0
|
|
|
0
|
0
|
|
local *FO= shift; |
1672
|
0
|
|
|
|
|
|
my $fdef= shift; |
1673
|
0
|
|
|
|
|
|
my $num= shift; |
1674
|
|
|
|
|
|
|
|
1675
|
0
|
|
|
|
|
|
my $type= $fdef->{'ftype'}; |
1676
|
0
|
|
|
|
|
|
my $ftype= $FIELD_TYPE[$type]; |
1677
|
0
|
|
0
|
|
|
|
my $ttype= $ftype->{Desc} || "USER$type"; |
1678
|
0
|
|
|
|
|
|
my $x_siz= $ftype->{Size}; |
1679
|
0
|
|
|
|
|
|
my $x_off= sprintf ('0x%02X', $fdef->{off}); |
1680
|
0
|
|
|
|
|
|
my $x_flg= sprintf ('0x%02X', $fdef->{flg}); |
1681
|
0
|
|
|
|
|
|
my $x_name= $fdef->{name}; |
1682
|
0
|
|
|
|
|
|
$x_name=~ s/[\x80-\xFF]/?/g; |
1683
|
|
|
|
|
|
|
|
1684
|
0
|
|
|
|
|
|
printf FO "[%02d] %2d %-12s %3s %-24s %3d %s 0x%02X %s\n", |
1685
|
|
|
|
|
|
|
$num, $type, $ttype, $x_siz, "'$x_name'", |
1686
|
|
|
|
|
|
|
$fdef->{fid}, $x_off, $fdef->{res}, $x_flg; |
1687
|
|
|
|
|
|
|
|
1688
|
|
|
|
|
|
|
# print FO " | '$x_name'\n"; |
1689
|
|
|
|
|
|
|
# print FO "[$num] type= $ttype ($type) name='$fdef->{name}'" |
1690
|
|
|
|
|
|
|
# " id=$fdef->{fid} off=$x_off res=$fdef->{res} flg=$x_flg\n"; |
1691
|
|
|
|
|
|
|
|
1692
|
0
|
|
|
|
|
|
$x_off; |
1693
|
|
|
|
|
|
|
} |
1694
|
|
|
|
|
|
|
|
1695
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1696
|
|
|
|
|
|
|
sub decode_type14 # analyze application specific field type 14 |
1697
|
|
|
|
|
|
|
{ |
1698
|
0
|
|
|
0
|
0
|
|
my $obj= shift; |
1699
|
0
|
|
|
|
|
|
local *FO= shift; |
1700
|
0
|
|
|
|
|
|
my $b= shift; |
1701
|
|
|
|
|
|
|
|
1702
|
0
|
|
|
|
|
|
my $AD= $obj->{APT_Data}; |
1703
|
0
|
|
|
|
|
|
my $lng= length ($b); |
1704
|
|
|
|
|
|
|
|
1705
|
0
|
|
|
|
|
|
my ($off, $d, $v); |
1706
|
0
|
0
|
|
|
|
|
if (defined ($AD->{View_Table})) |
|
|
0
|
|
|
|
|
|
1707
|
|
|
|
|
|
|
{ |
1708
|
0
|
|
|
|
|
|
print <
|
1709
|
|
|
|
|
|
|
Warning: type 14 in data base appears more than twice. |
1710
|
|
|
|
|
|
|
Please send a sample of your database to the author |
1711
|
|
|
|
|
|
|
&hex_dump ($b); |
1712
|
|
|
|
|
|
|
EOX |
1713
|
|
|
|
|
|
|
} |
1714
|
|
|
|
|
|
|
elsif (defined ($AD->{Header})) |
1715
|
|
|
|
|
|
|
{ |
1716
|
0
|
|
|
|
|
|
my @View_Table; |
1717
|
0
|
|
|
|
|
|
for ($off= 0; $off+5 <= $lng; $off += 5) |
1718
|
|
|
|
|
|
|
{ |
1719
|
0
|
|
|
|
|
|
$d= &fmt_date (substr ($b, $off, 3)); |
1720
|
0
|
|
|
|
|
|
$v= unpack ('v', substr ($b, $off+3, 2)); |
1721
|
0
|
0
|
|
|
|
|
last if ($v eq $no_val); # end marker |
1722
|
0
|
|
|
|
|
|
push (@View_Table, { 'date' => $d, num => $v } ); |
1723
|
|
|
|
|
|
|
# print FO " date=$d num=$v\n"; |
1724
|
|
|
|
|
|
|
} |
1725
|
0
|
|
|
|
|
|
$AD->{View_Table}= \@View_Table; |
1726
|
|
|
|
|
|
|
# &hex_dump ($b); |
1727
|
|
|
|
|
|
|
} |
1728
|
|
|
|
|
|
|
else |
1729
|
|
|
|
|
|
|
{ |
1730
|
0
|
|
|
|
|
|
$d= &fmt_date (substr ($b, 0, 3)); |
1731
|
0
|
|
|
|
|
|
$AD->{Head_Date}= $d; |
1732
|
0
|
|
|
|
|
|
$AD->{Header}= $b; |
1733
|
|
|
|
|
|
|
} |
1734
|
|
|
|
|
|
|
} |
1735
|
|
|
|
|
|
|
|
1736
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1737
|
|
|
|
|
|
|
sub print_recheader |
1738
|
|
|
|
|
|
|
{ |
1739
|
0
|
|
|
0
|
0
|
|
local *FH= shift; |
1740
|
0
|
|
|
|
|
|
my $txt= shift; |
1741
|
0
|
|
|
|
|
|
my $r= shift; |
1742
|
|
|
|
|
|
|
|
1743
|
0
|
|
|
|
|
|
my @extra= @_; |
1744
|
0
|
|
|
|
|
|
my $fld; |
1745
|
0
|
|
|
|
|
|
my $type= $r->{'type'}; |
1746
|
0
|
|
0
|
|
|
|
my $ttype= $REC_TYPE[$type] || "USER$type"; |
1747
|
|
|
|
|
|
|
|
1748
|
0
|
|
|
|
|
|
print "$txt\n"; |
1749
|
|
|
|
|
|
|
|
1750
|
0
|
|
|
|
|
|
print " type= $ttype ($type)\n"; |
1751
|
0
|
|
|
|
|
|
foreach $fld ('status', 'length', 'idx', @extra) |
1752
|
|
|
|
|
|
|
{ |
1753
|
0
|
|
|
|
|
|
print " $fld= $r->{$fld}\n"; |
1754
|
|
|
|
|
|
|
} |
1755
|
|
|
|
|
|
|
} |
1756
|
|
|
|
|
|
|
|
1757
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1758
|
|
|
|
|
|
|
sub dump_def |
1759
|
|
|
|
|
|
|
{ |
1760
|
0
|
|
|
0
|
0
|
|
my $self= shift; |
1761
|
0
|
|
|
|
|
|
local *FO= shift; |
1762
|
0
|
|
|
|
|
|
my $level= shift; |
1763
|
|
|
|
|
|
|
|
1764
|
0
|
|
|
|
|
|
my $hdr= $self->{Header}; |
1765
|
0
|
|
|
|
|
|
my $Time= &fmt_time_stamp ($hdr->{'time'}); |
1766
|
|
|
|
|
|
|
|
1767
|
0
|
|
|
|
|
|
my $fld; |
1768
|
0
|
|
|
|
|
|
my $sig= substr ($hdr->{sig}, 0, 3); |
1769
|
0
|
|
|
|
|
|
my $x_ltable= sprintf ("0x%08lX", $hdr->{lookup_table_offset}); |
1770
|
0
|
|
|
|
|
|
my $APT= &decode_apt ($hdr->{file_type}); |
1771
|
|
|
|
|
|
|
|
1772
|
0
|
|
|
|
|
|
print FO <
|
1773
|
|
|
|
|
|
|
Filename: $self->{Filename} |
1774
|
|
|
|
|
|
|
Meta: $self->{Meta} |
1775
|
|
|
|
|
|
|
DB Header: |
1776
|
|
|
|
|
|
|
sig= $sig |
1777
|
|
|
|
|
|
|
time= $Time |
1778
|
|
|
|
|
|
|
lookup_table_offset= $x_ltable |
1779
|
|
|
|
|
|
|
file_type= $hdr->{file_type} $APT |
1780
|
|
|
|
|
|
|
EOX |
1781
|
|
|
|
|
|
|
|
1782
|
0
|
|
|
|
|
|
foreach $fld (sort keys %$hdr) |
1783
|
|
|
|
|
|
|
{ |
1784
|
0
|
0
|
|
|
|
|
print FO " $fld= $hdr->{$fld}\n" unless (defined ($XHDR{$fld})); |
1785
|
|
|
|
|
|
|
} |
1786
|
|
|
|
|
|
|
|
1787
|
|
|
|
|
|
|
# &print_recheader (*FO, 'record header:', $hdr->{recheader}); |
1788
|
|
|
|
|
|
|
# print FO 'self:: ', join (',', sort keys %$self), "\n"; |
1789
|
|
|
|
|
|
|
|
1790
|
|
|
|
|
|
|
# $level= 0 if ($self->{Meta} eq 'Encrypted' && $level < 10); |
1791
|
|
|
|
|
|
|
|
1792
|
0
|
0
|
|
|
|
|
if ($level > 0) |
1793
|
|
|
|
|
|
|
{ |
1794
|
0
|
|
|
|
|
|
$self->show_db_def (*FO); |
1795
|
|
|
|
|
|
|
# $self-> show_card_def (*FO); |
1796
|
|
|
|
|
|
|
} |
1797
|
|
|
|
|
|
|
|
1798
|
0
|
0
|
|
|
|
|
if ($level > 1) |
1799
|
|
|
|
|
|
|
{ |
1800
|
0
|
|
|
|
|
|
print FO $delim, "\n\n"; |
1801
|
0
|
|
|
|
|
|
for ($fld= 0; $fld < 32; $fld++) |
1802
|
|
|
|
|
|
|
{ |
1803
|
0
|
|
|
|
|
|
$self->dump_db (*FO, $fld); |
1804
|
|
|
|
|
|
|
} |
1805
|
|
|
|
|
|
|
} |
1806
|
|
|
|
|
|
|
} |
1807
|
|
|
|
|
|
|
|
1808
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1809
|
|
|
|
|
|
|
sub dump_db |
1810
|
|
|
|
|
|
|
{ |
1811
|
0
|
|
|
0
|
0
|
|
my $self= shift; |
1812
|
0
|
|
|
|
|
|
local *FO= shift; |
1813
|
0
|
|
|
|
|
|
my $type= shift; |
1814
|
0
|
|
|
|
|
|
my $idx= shift; |
1815
|
|
|
|
|
|
|
|
1816
|
0
|
|
|
|
|
|
my $Types= $self->{Types}; |
1817
|
0
|
|
|
|
|
|
my $Data= $Types->[$type]; |
1818
|
0
|
|
|
|
|
|
my ($el, $i); |
1819
|
|
|
|
|
|
|
|
1820
|
0
|
0
|
|
|
|
|
if (defined ($idx)) |
1821
|
|
|
|
|
|
|
{ |
1822
|
0
|
|
|
|
|
|
$el= $Data->[$idx]; |
1823
|
0
|
|
|
|
|
|
&dump_db_rec (*FO, $idx, $el); |
1824
|
0
|
|
|
|
|
|
return; |
1825
|
|
|
|
|
|
|
} |
1826
|
|
|
|
|
|
|
|
1827
|
0
|
|
|
|
|
|
$idx= 0; |
1828
|
0
|
|
|
|
|
|
foreach $el (@$Data) |
1829
|
|
|
|
|
|
|
{ |
1830
|
0
|
|
|
|
|
|
&dump_db_rec (*FO, $idx, $el); |
1831
|
0
|
|
|
|
|
|
$idx++; |
1832
|
|
|
|
|
|
|
} |
1833
|
|
|
|
|
|
|
} |
1834
|
|
|
|
|
|
|
|
1835
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1836
|
|
|
|
|
|
|
sub dump_db_rec |
1837
|
|
|
|
|
|
|
{ |
1838
|
0
|
|
|
0
|
0
|
|
local *FO= shift; |
1839
|
0
|
|
|
|
|
|
my $i= shift; |
1840
|
0
|
|
|
|
|
|
my $el= shift; |
1841
|
|
|
|
|
|
|
|
1842
|
0
|
0
|
|
|
|
|
unless (defined ($el)) |
1843
|
|
|
|
|
|
|
{ |
1844
|
0
|
|
|
|
|
|
print FO "data record [$i] not defined!\n"; |
1845
|
0
|
|
|
|
|
|
return; |
1846
|
|
|
|
|
|
|
} |
1847
|
|
|
|
|
|
|
|
1848
|
0
|
|
|
|
|
|
&print_recheader (*FO, "data record [$i]", $el, 'filters', 'flags'); |
1849
|
|
|
|
|
|
|
# print FO "el= ", join (':', keys %$el), "\n"; |
1850
|
0
|
|
|
|
|
|
print FO "data=\n"; |
1851
|
0
|
|
|
|
|
|
&hex_dump ($el->{data}, *FO); |
1852
|
0
|
|
|
|
|
|
print FO $delim, "\n\n"; |
1853
|
|
|
|
|
|
|
} |
1854
|
|
|
|
|
|
|
|
1855
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1856
|
|
|
|
|
|
|
sub dump_data_record |
1857
|
|
|
|
|
|
|
{ |
1858
|
0
|
|
|
0
|
0
|
|
my $b= shift; |
1859
|
0
|
|
|
|
|
|
my $ok= shift; |
1860
|
0
|
|
|
|
|
|
my $o= shift; |
1861
|
|
|
|
|
|
|
|
1862
|
0
|
|
|
|
|
|
print "dump_data_record:\n"; |
1863
|
0
|
|
|
|
|
|
print join (':', %$o), "\n"; |
1864
|
|
|
|
|
|
|
# print "note: $nd\n" if ($nd); |
1865
|
|
|
|
|
|
|
|
1866
|
0
|
0
|
0
|
|
|
|
unless ($ok && 0) |
1867
|
|
|
|
|
|
|
{ |
1868
|
0
|
|
|
|
|
|
&hex_dump ($b); |
1869
|
|
|
|
|
|
|
} |
1870
|
|
|
|
|
|
|
} |
1871
|
|
|
|
|
|
|
|
1872
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1873
|
|
|
|
|
|
|
sub hex_dump |
1874
|
|
|
|
|
|
|
{ |
1875
|
0
|
|
|
0
|
0
|
|
my $data= shift; |
1876
|
0
|
|
0
|
|
|
|
local *FX= shift || *STDOUT; |
1877
|
|
|
|
|
|
|
|
1878
|
0
|
|
|
|
|
|
my $off= 0; |
1879
|
0
|
|
|
|
|
|
my ($i, $c, $v); |
1880
|
|
|
|
|
|
|
|
1881
|
0
|
|
|
|
|
|
while ($data) |
1882
|
|
|
|
|
|
|
{ |
1883
|
0
|
|
|
|
|
|
my $char= ''; |
1884
|
0
|
|
|
|
|
|
my $hex= ''; |
1885
|
0
|
|
|
|
|
|
my $offx= sprintf ('%08X', $off); |
1886
|
0
|
|
|
|
|
|
$off += 0x10; |
1887
|
|
|
|
|
|
|
|
1888
|
0
|
|
|
|
|
|
for ($i= 0; $i < 16; $i++) |
1889
|
|
|
|
|
|
|
{ |
1890
|
0
|
|
|
|
|
|
$c= substr ($data, 0, 1); |
1891
|
|
|
|
|
|
|
|
1892
|
0
|
0
|
|
|
|
|
if ($c ne '') |
1893
|
|
|
|
|
|
|
{ |
1894
|
0
|
|
|
|
|
|
$data= substr ($data, 1); |
1895
|
0
|
|
|
|
|
|
$v= unpack ('C', $c); |
1896
|
0
|
0
|
0
|
|
|
|
$c= '.' if ($v < 0x20 || $v >= 0x7F); |
1897
|
|
|
|
|
|
|
|
1898
|
0
|
|
|
|
|
|
$char .= $c; |
1899
|
0
|
|
|
|
|
|
$hex .= sprintf (' %02X', $v); |
1900
|
|
|
|
|
|
|
} |
1901
|
|
|
|
|
|
|
else |
1902
|
|
|
|
|
|
|
{ |
1903
|
0
|
|
|
|
|
|
$char .= ' '; |
1904
|
0
|
|
|
|
|
|
$hex .= ' '; |
1905
|
|
|
|
|
|
|
} |
1906
|
|
|
|
|
|
|
} |
1907
|
|
|
|
|
|
|
|
1908
|
0
|
|
|
|
|
|
print FX "$offx $hex |$char|\n"; |
1909
|
|
|
|
|
|
|
} |
1910
|
|
|
|
|
|
|
} |
1911
|
|
|
|
|
|
|
|
1912
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1913
|
|
|
|
|
|
|
# Decrypt the password of a HP 200LX Database. |
1914
|
|
|
|
|
|
|
# This function implements the algorithm in Curtis Cameron's dbcheck program. |
1915
|
|
|
|
|
|
|
# Returns a session key and the original password. I'm not quite sure |
1916
|
|
|
|
|
|
|
# if the original password is correct in all cases, this needs more testing. |
1917
|
|
|
|
|
|
|
sub decrypt_password |
1918
|
|
|
|
|
|
|
{ |
1919
|
0
|
|
|
0
|
0
|
|
my ($b, $siz)= @_; |
1920
|
0
|
|
|
|
|
|
my ($pass, $key); |
1921
|
|
|
|
|
|
|
|
1922
|
0
|
0
|
|
|
|
|
if ($siz != 17) |
1923
|
|
|
|
|
|
|
{ |
1924
|
0
|
|
|
|
|
|
print "WARNING: decrypt_password (siz=$siz): ", |
1925
|
|
|
|
|
|
|
"password block size should be 17 byte!\n"; |
1926
|
|
|
|
|
|
|
} |
1927
|
|
|
|
|
|
|
|
1928
|
0
|
|
|
|
|
|
my ($i, $c, $k, $p); |
1929
|
0
|
|
|
|
|
|
for ($i= 0; $i < 17; $i++) |
1930
|
|
|
|
|
|
|
{ |
1931
|
0
|
|
|
|
|
|
$c= unpack ('C', substr ($b, $i, 1)); |
1932
|
0
|
|
|
|
|
|
$k= $c ^ $i ^ $CODE_A[$i]; |
1933
|
|
|
|
|
|
|
# my $diag= sprintf ("%02X ^ %02X ^ A[%3d]=%02X", $c, $i, $i, $CODE_A[$i]); |
1934
|
|
|
|
|
|
|
|
1935
|
|
|
|
|
|
|
# this CODE_B round cancels the effect of the same thing in decrypt_data |
1936
|
|
|
|
|
|
|
# $k ^= $CODE_B[$i]; |
1937
|
|
|
|
|
|
|
# $diag .= sprintf (" ^ B[%2d]=%02X", $i, $CODE_B[$i]); |
1938
|
|
|
|
|
|
|
|
1939
|
0
|
|
|
|
|
|
push (@$key, $k); |
1940
|
|
|
|
|
|
|
# push (@DIAG_K, $diag); |
1941
|
|
|
|
|
|
|
|
1942
|
0
|
|
|
|
|
|
$p= $PW_CODE [$i] ^ $c; |
1943
|
0
|
0
|
|
|
|
|
$pass .= pack ('C', $p) if ($p > 0x00); |
1944
|
|
|
|
|
|
|
} |
1945
|
|
|
|
|
|
|
|
1946
|
0
|
|
|
|
|
|
print "database is encrypted\npassword record, encrypted, siz=$siz\n"; |
1947
|
0
|
|
|
|
|
|
&hex_dump ($b); |
1948
|
0
|
|
|
|
|
|
print "password record, decryption attempted (1)\n"; |
1949
|
0
|
|
|
|
|
|
&hex_dump ($pass); |
1950
|
0
|
|
|
|
|
|
print "password= '$pass'\n"; |
1951
|
|
|
|
|
|
|
|
1952
|
0
|
|
|
|
|
|
($pass, $key); |
1953
|
|
|
|
|
|
|
} |
1954
|
|
|
|
|
|
|
|
1955
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
1956
|
|
|
|
|
|
|
# Decrypt the data portion of a HP 200LX Database record. |
1957
|
|
|
|
|
|
|
# This function implements the algorithm in Curtis Cameron's dbcheck program. |
1958
|
|
|
|
|
|
|
sub decrypt_data |
1959
|
|
|
|
|
|
|
{ |
1960
|
0
|
|
|
0
|
0
|
|
my ($b, $siz, $code_ref)= @_; |
1961
|
|
|
|
|
|
|
|
1962
|
0
|
|
|
|
|
|
my ($cc, $c0, $bb); |
1963
|
0
|
|
|
|
|
|
my ($c_a, $c_b, $c_k); |
1964
|
0
|
|
|
|
|
|
my ($ii, $i_127, $i_17); |
1965
|
0
|
|
|
|
|
|
for ($ii= 0; $ii < $siz; $ii++) |
1966
|
|
|
|
|
|
|
{ |
1967
|
0
|
|
|
|
|
|
$c0= unpack ('C', substr ($b, $ii, 1)); |
1968
|
|
|
|
|
|
|
|
1969
|
0
|
|
|
|
|
|
$c_a= $CODE_A [$i_127]; |
1970
|
0
|
|
|
|
|
|
$c_k= $code_ref->[$i_17]; |
1971
|
0
|
|
|
|
|
|
$cc= $c0 ^ $c_k ^ $c_a; |
1972
|
|
|
|
|
|
|
|
1973
|
|
|
|
|
|
|
# my $diag= sprintf ("[%4d] %02X ^ K[%2d]=(%s)=%02X ^ A[%3d]=%02X", |
1974
|
|
|
|
|
|
|
# $ii, $c0, |
1975
|
|
|
|
|
|
|
# $i_17, $DIAG_K[$i_17], $c_k, |
1976
|
|
|
|
|
|
|
# $i_127, $c_a); |
1977
|
|
|
|
|
|
|
|
1978
|
|
|
|
|
|
|
# this CODE_B round cancels the effect of the same thing in decrypt_password |
1979
|
|
|
|
|
|
|
# $c_b= $CODE_B [$i_17]; |
1980
|
|
|
|
|
|
|
# $cc ^= $c_b; |
1981
|
|
|
|
|
|
|
# $diag .= sprintf (" ^ B[%3d]=%02X", $i_17, $c_b); |
1982
|
|
|
|
|
|
|
|
1983
|
0
|
0
|
|
|
|
|
if ($ii > 126) |
1984
|
|
|
|
|
|
|
{ |
1985
|
0
|
|
|
|
|
|
my $ti; |
1986
|
0
|
|
|
|
|
|
for ($ti= $ii-127; $ti >= 0; $ti -= 127) |
1987
|
|
|
|
|
|
|
{ |
1988
|
0
|
|
|
|
|
|
$c_b= $CODE_B [$ti % 17]; |
1989
|
0
|
|
|
|
|
|
$cc ^= $c_b; |
1990
|
|
|
|
|
|
|
# $diag .= sprintf (" ^ B[%3d]=%02X", $ti%17, $c_b); |
1991
|
|
|
|
|
|
|
} |
1992
|
|
|
|
|
|
|
} |
1993
|
|
|
|
|
|
|
# $diag .= sprintf (" =: %02X %c", $cc, $cc); print $diag, "\n"; |
1994
|
|
|
|
|
|
|
|
1995
|
0
|
|
|
|
|
|
$bb .= pack ('C', $cc); |
1996
|
0
|
0
|
|
|
|
|
$i_17= 0 if (++$i_17 >= 17); |
1997
|
0
|
0
|
|
|
|
|
$i_127= 0 if (++$i_127 >= 127); |
1998
|
|
|
|
|
|
|
} |
1999
|
|
|
|
|
|
|
|
2000
|
0
|
|
|
|
|
|
$bb; |
2001
|
|
|
|
|
|
|
} |
2002
|
|
|
|
|
|
|
|
2003
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
2004
|
|
|
|
|
|
|
sub recover_password |
2005
|
|
|
|
|
|
|
{ |
2006
|
0
|
|
|
0
|
0
|
|
my $self= shift; |
2007
|
0
|
|
|
|
|
|
my $note_nr= shift; |
2008
|
0
|
|
|
|
|
|
my $ptx_fnm= shift; |
2009
|
0
|
|
|
|
|
|
my $key_fnm= shift; |
2010
|
|
|
|
|
|
|
|
2011
|
|
|
|
|
|
|
# fetch encrypted note |
2012
|
0
|
|
0
|
|
|
|
my $T= $self->{Types} || die; |
2013
|
0
|
|
|
|
|
|
my $D= $T->[11]; # array of data records |
2014
|
0
|
|
|
|
|
|
my $N= $T->[9]; # array of note records |
2015
|
0
|
|
|
|
|
|
my $enc_txt= $N->[$note_nr]->{data}; |
2016
|
|
|
|
|
|
|
# print "encrypted text:\n"; &hex_dump ($enc_txt); |
2017
|
|
|
|
|
|
|
|
2018
|
|
|
|
|
|
|
# fetch plain text |
2019
|
0
|
|
|
|
|
|
my $ptx_txt; |
2020
|
0
|
0
|
|
|
|
|
open (FI, $ptx_fnm) || die; |
2021
|
0
|
|
|
|
|
|
while () { $ptx_txt .= $_; } |
|
0
|
|
|
|
|
|
|
2022
|
0
|
|
|
|
|
|
close (FI); |
2023
|
|
|
|
|
|
|
# print "plain text:\n"; &hex_dump ($ptx_txt); |
2024
|
|
|
|
|
|
|
|
2025
|
|
|
|
|
|
|
# recover the key |
2026
|
0
|
|
|
|
|
|
my ($pp, $cc, $ee, $ii, $key); |
2027
|
0
|
|
|
|
|
|
my $ll_enc= length ($enc_txt); |
2028
|
0
|
|
|
|
|
|
my $ll_ptx= length ($ptx_txt); |
2029
|
0
|
|
|
|
|
|
print "text size enc=$ll_enc plain=$ll_ptx\n"; |
2030
|
|
|
|
|
|
|
|
2031
|
0
|
|
|
|
|
|
for ($ii= 0; $ii < $ll_ptx; $ii++) |
2032
|
|
|
|
|
|
|
{ |
2033
|
0
|
|
|
|
|
|
$pp= unpack ('C', substr ($ptx_txt, $ii, 1)); |
2034
|
0
|
|
|
|
|
|
$ee= unpack ('C', substr ($enc_txt, $ii, 1)); |
2035
|
0
|
|
|
|
|
|
$cc= $pp ^ $ee ^ $ii; |
2036
|
0
|
|
|
|
|
|
$key .= pack ('C', $cc); |
2037
|
|
|
|
|
|
|
} |
2038
|
|
|
|
|
|
|
|
2039
|
|
|
|
|
|
|
# print "the key is\n"; &hex_dump ($key); |
2040
|
|
|
|
|
|
|
|
2041
|
0
|
|
|
|
|
|
print "dumping key to $key_fnm\n"; |
2042
|
0
|
0
|
|
|
|
|
open (FO, ">$key_fnm") || die; |
2043
|
0
|
|
|
|
|
|
binmode (FI); # MS-DOS systems need this, T2D: how about Mac? |
2044
|
0
|
|
|
|
|
|
print FO $key; |
2045
|
0
|
|
|
|
|
|
close (FO); |
2046
|
|
|
|
|
|
|
} |
2047
|
|
|
|
|
|
|
|
2048
|
|
|
|
|
|
|
# ---------------------------------------------------------------------------- |
2049
|
|
|
|
|
|
|
sub get_field_type |
2050
|
|
|
|
|
|
|
{ |
2051
|
0
|
|
|
0
|
0
|
|
my $ty= shift; |
2052
|
0
|
|
|
|
|
|
$FIELD_TYPE[$ty]; |
2053
|
|
|
|
|
|
|
} |
2054
|
|
|
|
|
|
|
|
2055
|
|
|
|
|
|
|
# Autoload methods go after =cut, and are processed by the autosplit program. |
2056
|
|
|
|
|
|
|
|
2057
|
|
|
|
|
|
|
1; |
2058
|
|
|
|
|
|
|
__END__ |