Do not expand autoptr macros when running introspection
[glib.git] / tests / gen-casemap-txt.pl
blob3ae419ca9426db36e8a8ebf35ccdc37504bf7330
1 #! /usr/bin/perl -w
3 # Copyright (C) 1998, 1999 Tom Tromey
4 # Copyright (C) 2001 Red Hat Software
6 # This program is free software; you can redistribute it and/or modify
7 # it under the terms of the GNU General Public License as published by
8 # the Free Software Foundation; either version 2, or (at your option)
9 # any later version.
11 # This program is distributed in the hope that it will be useful,
12 # but WITHOUT ANY WARRANTY; without even the implied warranty of
13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 # GNU General Public License for more details.
16 # You should have received a copy of the GNU General Public License
17 # along with this program; if not, see <http://www.gnu.org/licenses/>.
19 # gen-casemap-test.pl - Generate test cases for case mapping from Unicode data.
20 # See http://www.unicode.org/Public/UNIDATA/UnicodeCharacterDatabase.html
21 # I consider the output of this program to be unrestricted. Use it as
22 # you will.
24 require 5.006;
25 use utf8;
27 if (@ARGV != 3) {
28 $0 =~ s@.*/@@;
29 die "Usage: $0 UNICODE-VERSION UnicodeData.txt SpecialCasing.txt\n";
32 use vars qw($CODE $NAME $CATEGORY $COMBINING_CLASSES $BIDI_CATEGORY $DECOMPOSITION $DECIMAL_VALUE $DIGIT_VALUE $NUMERIC_VALUE $MIRRORED $OLD_NAME $COMMENT $UPPER $LOWER $TITLE $BREAK_CODE $BREAK_CATEGORY $BREAK_NAME $CASE_CODE $CASE_LOWER $CASE_TITLE $CASE_UPPER $CASE_CONDITION);
34 # Names of fields in Unicode data table.
35 $CODE = 0;
36 $NAME = 1;
37 $CATEGORY = 2;
38 $COMBINING_CLASSES = 3;
39 $BIDI_CATEGORY = 4;
40 $DECOMPOSITION = 5;
41 $DECIMAL_VALUE = 6;
42 $DIGIT_VALUE = 7;
43 $NUMERIC_VALUE = 8;
44 $MIRRORED = 9;
45 $OLD_NAME = 10;
46 $COMMENT = 11;
47 $UPPER = 12;
48 $LOWER = 13;
49 $TITLE = 14;
51 # Names of fields in the SpecialCasing table
52 $CASE_CODE = 0;
53 $CASE_LOWER = 1;
54 $CASE_TITLE = 2;
55 $CASE_UPPER = 3;
56 $CASE_CONDITION = 4;
58 my @upper;
59 my @title;
60 my @lower;
62 binmode STDOUT, ":utf8";
63 open (INPUT, "< $ARGV[1]") || exit 1;
65 $last_code = -1;
66 while (<INPUT>)
68 chop;
69 @fields = split (';', $_, 30);
70 if ($#fields != 14)
72 printf STDERR ("Entry for $fields[$CODE] has wrong number of fields (%d)\n", $#fields);
75 $code = hex ($fields[$CODE]);
77 if ($code > $last_code + 1)
79 # Found a gap.
80 if ($fields[$NAME] =~ /Last>/)
82 # Fill the gap with the last character read,
83 # since this was a range specified in the char database
84 @gfields = @fields;
86 else
88 # The gap represents undefined characters. Only the type
89 # matters.
90 @gfields = ('', '', 'Cn', '0', '', '', '', '', '', '', '',
91 '', '', '', '');
93 for (++$last_code; $last_code < $code; ++$last_code)
95 $gfields{$CODE} = sprintf ("%04x", $last_code);
96 &process_one ($last_code, @gfields);
99 &process_one ($code, @fields);
100 $last_code = $code;
103 close INPUT;
105 open (INPUT, "< $ARGV[2]") || exit 1;
107 while (<INPUT>)
109 my $code;
111 chop;
113 next if /^#/;
114 next if /^\s*$/;
116 s/\s*#.*//;
118 @fields = split ('\s*;\s*', $_, 30);
120 $raw_code = $fields[$CASE_CODE];
121 $code = hex ($raw_code);
123 if ($#fields != 4 && $#fields != 5)
125 printf STDERR ("Entry for $raw_code has wrong number of fields (%d)\n", $#fields);
126 next;
129 if (defined $fields[5]) {
130 # Ignore conditional special cases - we'll handle them manually
131 next;
134 $upper[$code] = &make_hex ($fields[$CASE_UPPER]);
135 $lower[$code] = &make_hex ($fields[$CASE_LOWER]);
136 $title[$code] = &make_hex ($fields[$CASE_TITLE]);
139 close INPUT;
141 print <<EOT;
142 # Test cases generated from Unicode $ARGV[0] data
143 # by gen-case-tests.pl. Do not edit.
145 # Some special hand crafted tests
147 tr_TR\ti\ti\t\x{0130}\t\x{0130}\t# i => LATIN CAPITAL LETTER I WITH DOT ABOVE
148 tr_TR\tI\t\x{0131}\tI\tI\t# I => LATIN SMALL LETTER DOTLESS I
149 tr_TR\tI\x{0307}\ti\tI\x{0307}\tI\x{0307}\t# I => LATIN SMALL LETTER DOTLESS I
150 tr_TR.UTF-8\ti\ti\t\x{0130}\t\x{0130}\t# i => LATIN CAPITAL LETTER I WITH DOT ABOVE
151 tr_TR.UTF-8\tI\t\x{0131}\tI\tI\t# I => LATIN SMALL LETTER DOTLESS I
152 tr_TR.UTF-8\tI\x{0307}\ti\tI\x{0307}\tI\x{0307}\t# I => LATIN SMALL LETTER DOTLESS I
153 # Test reordering of YPOGEGRAMMENI across other accents
154 \t\x{03b1}\x{0345}\x{0314}\t\x{03b1}\x{0345}\x{314}\t\x{0391}\x{0345}\x{0314}\t\x{0391}\x{0314}\x{0399}\t
155 \t\x{03b1}\x{0314}\x{0345}\t\x{03b1}\x{314}\x{0345}\t\x{0391}\x{0314}\x{0345}\t\x{0391}\x{0314}\x{0399}\t
156 # Handling of final and nonfinal sigma
157 ΜΆΙΟΣ μάιος Μάιος ΜΆΙΟΣ
158 ΜΆΙΟΣ μάιος Μάιος ΜΆΙΟΣ
159 ΣΙΓΜΑ σιγμα Σιγμα ΣΙΓΜΑ
160 # Lithuanian rule of i followed by letter with dot. Not at all sure
161 # about the titlecase part here
162 lt_LT\ti\x{117}\ti\x{117}\tIe\tIE\t
163 lt_LT\tie\x{307}\tie\x{307}\tIe\tIE\t
164 lt_LT\t\x{00cc}\ti\x{0307}\x{0300}\t\x{00cc}\t\x{00cc}\t # LATIN CAPITAL LETTER I WITH GRAVE
165 lt_LT\t\x{00CD}\ti\x{0307}\x{0301}\t\x{00CD}\t\x{00CD}\t # LATIN CAPITAL LETTER I WITH ACUTE
166 lt_LT\t\x{0128}\ti\x{0307}\x{0303}\t\x{0128}\t\x{0128}\t # LATIN CAPITAL LETTER I WITH TILDE
167 lt_LT\tI\x{0301}\ti\x{0307}\x{0301}\tI\x{0301}\tI\x{0301}\t # LATIN CAPITAL LETTER I (with acute accent)
168 lt_LT\tI\x{0300}\ti\x{0307}\x{0300}\tI\x{0300}\tI\x{0300}\t # LATIN CAPITAL LETTER I (with grave accent)
169 lt_LT\tI\x{0303}\ti\x{0307}\x{0303}\tI\x{0303}\tI\x{0303}\t # LATIN CAPITAL LETTER I (with tilde above)
170 lt_LT\tI\x{0328}\x{0301}\ti\x{0307}\x{0328}\x{0301}\tI\x{0328}\x{0301}\tI\x{0328}\x{0301}\t # LATIN CAPITAL LETTER I (with ogonek and acute accent)
171 lt_LT\tJ\x{0301}\tj\x{0307}\x{0301}\tJ\x{0301}\tJ\x{0301}\t # LATIN CAPITAL LETTER J (with acute accent)
172 lt_LT\t\x{012e}\x{0301}\t\x{012f}\x{0307}\x{0301}\t\x{012e}\x{0301}\t\x{012e}\x{0301}\t # LATIN CAPITAL LETTER I WITH OGONEK (with acute accent)
173 lt_LT.UTF-8\ti\x{117}\ti\x{117}\tIe\tIE\t
174 lt_LT.UTF-8\tie\x{307}\tie\x{307}\tIe\tIE\t
175 lt_LT.UTF-8\t\x{00cc}\ti\x{0307}\x{0300}\t\x{00cc}\t\x{00cc}\t # LATIN CAPITAL LETTER I WITH GRAVE
176 lt_LT.UTF-8\t\x{00CD}\ti\x{0307}\x{0301}\t\x{00CD}\t\x{00CD}\t # LATIN CAPITAL LETTER I WITH ACUTE
177 lt_LT.UTF-8\t\x{0128}\ti\x{0307}\x{0303}\t\x{0128}\t\x{0128}\t # LATIN CAPITAL LETTER I WITH TILDE
178 lt_LT.UTF-8\tI\x{0301}\ti\x{0307}\x{0301}\tI\x{0301}\tI\x{0301}\t # LATIN CAPITAL LETTER I (with acute accent)
179 lt_LT.UTF-8\tI\x{0300}\ti\x{0307}\x{0300}\tI\x{0300}\tI\x{0300}\t # LATIN CAPITAL LETTER I (with grave accent)
180 lt_LT.UTF-8\tI\x{0303}\ti\x{0307}\x{0303}\tI\x{0303}\tI\x{0303}\t # LATIN CAPITAL LETTER I (with tilde above)
181 lt_LT.UTF-8\tI\x{0328}\x{0301}\ti\x{0307}\x{0328}\x{0301}\tI\x{0328}\x{0301}\tI\x{0328}\x{0301}\t # LATIN CAPITAL LETTER I (with ogonek and acute accent)
182 lt_LT.UTF-8\tJ\x{0301}\tj\x{0307}\x{0301}\tJ\x{0301}\tJ\x{0301}\t # LATIN CAPITAL LETTER J (with acute accent)
183 lt_LT.UTF-8\t\x{012e}\x{0301}\t\x{012f}\x{0307}\x{0301}\t\x{012e}\x{0301}\t\x{012e}\x{0301}\t # LATIN CAPITAL LETTER I WITH OGONEK (with acute accent)
184 # Special case not at initial position
185 \ta\x{fb04}\ta\x{fb04}\tAffl\tAFFL\t# FB04
187 # Now the automatic tests
190 &print_tests;
192 exit 0;
194 # Process a single character.
195 sub process_one
197 my ($code, @fields) = @_;
199 my $type = $fields[$CATEGORY];
200 if ($type eq 'Ll')
202 $upper[$code] = make_hex ($fields[$UPPER]);
203 $lower[$code] = pack ("U", $code);
204 $title[$code] = make_hex ($fields[$TITLE]);
206 elsif ($type eq 'Lu')
208 $lower[$code] = make_hex ($fields[$LOWER]);
209 $upper[$code] = pack ("U", $code);
210 $title[$code] = make_hex ($fields[$TITLE]);
213 if ($type eq 'Lt')
215 $upper[$code] = make_hex ($fields[$UPPER]);
216 $lower[$code] = pack ("U", hex ($fields[$LOWER]));
217 $title[$code] = make_hex ($fields[$LOWER]);
221 sub print_tests
223 for ($i = 0; $i < 0x10ffff; $i++) {
224 if ($i == 0x3A3) {
225 # Greek sigma needs special tests
226 next;
229 my $lower = $lower[$i];
230 my $title = $title[$i];
231 my $upper = $upper[$i];
233 if (defined $upper || defined $lower || defined $title) {
234 printf "\t%s\t%s\t%s\t%s\t# %4X\n",
235 pack ("U", $i),
236 (defined $lower ? $lower : ""),
237 (defined $title ? $title : ""),
238 (defined $upper ? $upper : ""),
244 sub make_hex
246 my $codes = shift;
248 $codes =~ s/^\s+//;
249 $codes =~ s/\s+$//;
251 if ($codes eq "0" || $codes eq "") {
252 return "";
253 } else {
254 return pack ("U*", map { hex ($_) } split /\s+/, $codes);