init_upcase(): Fix and cleanup
[wimlib] / src / encoding.c
1 /*
2  * encoding.c
3  */
4
5 /*
6  * Copyright (C) 2012, 2013 Eric Biggers
7  *
8  * This file is part of wimlib, a library for working with WIM files.
9  *
10  * wimlib is free software; you can redistribute it and/or modify it under the
11  * terms of the GNU General Public License as published by the Free
12  * Software Foundation; either version 3 of the License, or (at your option)
13  * any later version.
14  *
15  * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
16  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
17  * A PARTICULAR PURPOSE. See the GNU General Public License for more
18  * details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with wimlib; if not, see http://www.gnu.org/licenses/.
22  */
23
24 #ifdef HAVE_CONFIG_H
25 #  include "config.h"
26 #endif
27
28 #include "wimlib.h"
29 #include "wimlib/assert.h"
30 #include "wimlib/encoding.h"
31 #include "wimlib/endianness.h"
32 #include "wimlib/error.h"
33 #include "wimlib/list.h"
34 #include "wimlib/util.h"
35
36 #include <errno.h>
37 #include <iconv.h>
38 #include <pthread.h>
39 #include <stdlib.h>
40 #include <string.h>
41
42 bool wimlib_mbs_is_utf8 = !TCHAR_IS_UTF16LE;
43
44 /* List of iconv_t conversion descriptors for a specific character conversion.
45  * The idea is that it is not thread-safe to have just one conversion
46  * descriptor, but it also is inefficient to open a new conversion descriptor to
47  * convert every string.  Both these problems can be solved by maintaining a
48  * list of conversion descriptors; then, a thread can use an existing conversion
49  * descriptor if available. */
50 struct iconv_list_head {
51         const char *from_encoding;
52         const char *to_encoding;
53         struct list_head list;
54         pthread_mutex_t mutex;
55 };
56
57 struct iconv_node {
58         iconv_t cd;
59         struct list_head list;
60         struct iconv_list_head *head;
61 };
62
63 #define ICONV_LIST(name, from, to)                      \
64 struct iconv_list_head name = {                         \
65         .from_encoding = from,                          \
66         .to_encoding = to,                              \
67         .list = LIST_HEAD_INIT(name.list),              \
68         .mutex = PTHREAD_MUTEX_INITIALIZER,             \
69 }
70
71 static iconv_t *
72 get_iconv(struct iconv_list_head *head)
73 {
74         iconv_t cd;
75         iconv_t *cd_p;
76         struct iconv_node *i;
77
78         pthread_mutex_lock(&head->mutex);
79         if (list_empty(&head->list)) {
80                 cd = iconv_open(head->to_encoding, head->from_encoding);
81                 if (cd == (iconv_t)-1) {
82                         ERROR_WITH_ERRNO("Failed to open iconv from %s to %s",
83                                          head->from_encoding, head->to_encoding);
84                         cd_p = NULL;
85                 } else {
86                         i = MALLOC(sizeof(struct iconv_node));
87                         if (i) {
88                                 i->head = head;
89                                 i->cd = cd;
90                                 cd_p = &i->cd;
91                         } else {
92                                 iconv_close(cd);
93                                 cd_p = NULL;
94                         }
95                 }
96         } else {
97                 i = container_of(head->list.next, struct iconv_node, list);
98                 list_del(head->list.next);
99                 cd_p = &i->cd;
100         }
101         pthread_mutex_unlock(&head->mutex);
102         return cd_p;
103 }
104
105 static void
106 put_iconv(iconv_t *cd)
107 {
108         int errno_save = errno;
109         struct iconv_node *i = container_of(cd, struct iconv_node, cd);
110         struct iconv_list_head *head = i->head;
111
112         pthread_mutex_lock(&head->mutex);
113         list_add(&i->list, &head->list);
114         pthread_mutex_unlock(&head->mutex);
115         errno = errno_save;
116 }
117
118 /* Prevent printing an error message if a character conversion error occurs
119  * while printing an error message.  (This variable is not per-thread but it
120  * doesn't matter too much since it's just the error messages.) */
121 static bool error_message_being_printed = false;
122
123 #define DEFINE_CHAR_CONVERSION_FUNCTIONS(varname1, longname1, chartype1,\
124                                          varname2, longname2, chartype2,\
125                                          earlyreturn_on_utf8_locale,    \
126                                          earlyreturn_expr,              \
127                                          worst_case_len_expr,           \
128                                          err_return,                    \
129                                          err_msg,                       \
130                                          modifier)                      \
131 static ICONV_LIST(iconv_##varname1##_to_##varname2,                     \
132                   longname1, longname2);                                \
133                                                                         \
134 modifier int                                                            \
135 varname1##_to_##varname2##_nbytes(const chartype1 *in, size_t in_nbytes,\
136                                   size_t *out_nbytes_ret)               \
137 {                                                                       \
138         iconv_t *cd = get_iconv(&iconv_##varname1##_to_##varname2);     \
139         if (cd == NULL)                                                 \
140                 return WIMLIB_ERR_ICONV_NOT_AVAILABLE;                  \
141                                                                         \
142         /* Worst case length */                                         \
143         chartype2 buf[worst_case_len_expr];                             \
144         char *inbuf = (char*)in;                                        \
145         size_t inbytesleft = in_nbytes;                                 \
146         char *outbuf = (char*)buf;                                      \
147         size_t outbytesleft = sizeof(buf);                              \
148         size_t len;                                                     \
149         int ret;                                                        \
150                                                                         \
151         len = iconv(*cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft); \
152         if (len == (size_t)-1) {                                        \
153                 if (!error_message_being_printed) {                     \
154                         error_message_being_printed = true;             \
155                         err_msg;                                        \
156                         error_message_being_printed = false;            \
157                 }                                                       \
158                 ret = err_return;                                       \
159         } else {                                                        \
160                 *out_nbytes_ret = sizeof(buf) - outbytesleft;           \
161                 ret = 0;                                                \
162         }                                                               \
163         put_iconv(cd);                                                  \
164         return ret;                                                     \
165 }                                                                       \
166                                                                         \
167 modifier int                                                            \
168 varname1##_to_##varname2##_buf(const chartype1 *in, size_t in_nbytes,   \
169                                chartype2 *out)                          \
170 {                                                                       \
171         iconv_t *cd = get_iconv(&iconv_##varname1##_to_##varname2);     \
172         if (cd == NULL)                                                 \
173                 return WIMLIB_ERR_ICONV_NOT_AVAILABLE;                  \
174                                                                         \
175         char *inbuf = (char*)in;                                        \
176         size_t inbytesleft = in_nbytes;                                 \
177         char *outbuf = (char*)out;                                      \
178         const size_t LARGE_NUMBER = 1000000000;                         \
179         size_t outbytesleft = LARGE_NUMBER;                             \
180         size_t len;                                                     \
181         int ret;                                                        \
182                                                                         \
183         len = iconv(*cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft); \
184         if (len == (size_t)-1) {                                        \
185                 if (!error_message_being_printed) {                     \
186                         error_message_being_printed = true;             \
187                         err_msg;                                        \
188                         error_message_being_printed = false;            \
189                 }                                                       \
190                 ret = err_return;                                       \
191         } else {                                                        \
192                 out[(LARGE_NUMBER-outbytesleft)/sizeof(chartype2)] = 0; \
193                 ret = 0;                                                \
194         }                                                               \
195         put_iconv(cd);                                                  \
196         return ret;                                                     \
197 }                                                                       \
198                                                                         \
199 modifier int                                                            \
200 varname1##_to_##varname2(const chartype1 *in, size_t in_nbytes,         \
201                          chartype2 **out_ret,                           \
202                          size_t *out_nbytes_ret)                        \
203 {                                                                       \
204         int ret;                                                        \
205         chartype2 *out;                                                 \
206         size_t out_nbytes;                                              \
207                                                                         \
208         if (earlyreturn_on_utf8_locale && wimlib_mbs_is_utf8) {         \
209                 earlyreturn_expr;                                       \
210                 /* Out same as in */                                    \
211                 out = MALLOC(in_nbytes + sizeof(chartype2));            \
212                 if (!out)                                               \
213                         return WIMLIB_ERR_NOMEM;                        \
214                 memcpy(out, in, in_nbytes);                             \
215                 out[in_nbytes / sizeof(chartype2)] = 0;                 \
216                 *out_ret = out;                                         \
217                 *out_nbytes_ret = in_nbytes;                            \
218                 return 0;                                               \
219         }                                                               \
220                                                                         \
221         ret = varname1##_to_##varname2##_nbytes(in, in_nbytes,          \
222                                                 &out_nbytes);           \
223         if (ret)                                                        \
224                 return ret;                                             \
225                                                                         \
226         out = MALLOC(out_nbytes + sizeof(chartype2));                   \
227         if (!out)                                                       \
228                 return WIMLIB_ERR_NOMEM;                                \
229                                                                         \
230         ret = varname1##_to_##varname2##_buf(in, in_nbytes, out);       \
231         if (ret) {                                                      \
232                 FREE(out);                                              \
233         } else {                                                        \
234                 *out_ret = out;                                         \
235                 *out_nbytes_ret = out_nbytes;                           \
236         }                                                               \
237         return ret;                                                     \
238 }
239
240 #if !TCHAR_IS_UTF16LE
241
242 /* UNIX */
243
244 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", tchar,
245                                  utf16le, "UTF-16LE", utf16lechar,
246                                  false,
247                                  ,
248                                  in_nbytes * 2,
249                                  WIMLIB_ERR_INVALID_UTF8_STRING,
250                                  ERROR_WITH_ERRNO("Failed to convert UTF-8 string "
251                                                   "to UTF-16LE string!"),
252                                  static)
253
254 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf16le, "UTF-16LE", utf16lechar,
255                                  utf8, "UTF-8", tchar,
256                                  false,
257                                  ,
258                                  in_nbytes * 2,
259                                  WIMLIB_ERR_INVALID_UTF16_STRING,
260                                  ERROR_WITH_ERRNO("Failed to convert UTF-16LE string "
261                                                   "to UTF-8 string!"),
262                                  static)
263
264 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "", tchar,
265                                  utf16le, "UTF-16LE", utf16lechar,
266                                  true,
267                                  return utf8_to_utf16le(in, in_nbytes, out_ret, out_nbytes_ret),
268                                  in_nbytes * 2,
269                                  WIMLIB_ERR_INVALID_MULTIBYTE_STRING,
270                                  ERROR_WITH_ERRNO("Failed to convert multibyte "
271                                                   "string \"%"TS"\" to UTF-16LE string!", in);
272                                  ERROR("If the data you provided was UTF-8, please make sure "
273                                        "the character encoding\n"
274                                        "        of your current locale is UTF-8."),
275                                  )
276
277 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf16le, "UTF-16LE", utf16lechar,
278                                  tstr, "", tchar,
279                                  true,
280                                  return utf16le_to_utf8(in, in_nbytes, out_ret, out_nbytes_ret),
281                                  in_nbytes * 2,
282                                  WIMLIB_ERR_UNICODE_STRING_NOT_REPRESENTABLE,
283                                  ERROR("Failed to convert UTF-16LE string to "
284                                        "multibyte string!");
285                                  ERROR("This may be because the UTF-16LE string "
286                                        "could not be represented\n"
287                                        "        in your locale's character encoding."),
288                                  )
289 #endif
290
291 /* tchar to UTF-8 and back */
292 #if TCHAR_IS_UTF16LE
293
294 /* Windows */
295 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "UTF-16LE", tchar,
296                                  utf8, "UTF-8", char,
297                                  false,
298                                  ,
299                                  in_nbytes * 2,
300                                  WIMLIB_ERR_INVALID_UTF16_STRING,
301                                  ERROR_WITH_ERRNO("Failed to convert UTF-16LE "
302                                                   "string \"%"TS"\" to UTF-8 string!", in),
303                                  )
304
305 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", char,
306                                  tstr, "UTF-16LE", tchar,
307                                  false,
308                                  ,
309                                  in_nbytes * 2,
310                                  WIMLIB_ERR_INVALID_UTF8_STRING,
311                                  ERROR_WITH_ERRNO("Failed to convert UTF-8 string "
312                                                   "to UTF-16LE string!"),
313                                  )
314 #else
315
316 /* UNIX */
317
318 DEFINE_CHAR_CONVERSION_FUNCTIONS(tstr, "", tchar,
319                                  utf8, "UTF-8", char,
320                                  true,
321                                  ,
322                                  in_nbytes * 4,
323                                  WIMLIB_ERR_INVALID_MULTIBYTE_STRING,
324                                  ERROR_WITH_ERRNO("Failed to convert multibyte "
325                                                   "string \"%"TS"\" to UTF-8 string!", in);
326                                  ERROR("If the data you provided was UTF-8, please make sure "
327                                        "the character\n"
328                                        "        encoding of your current locale is UTF-8."),
329                                  )
330
331 DEFINE_CHAR_CONVERSION_FUNCTIONS(utf8, "UTF-8", char,
332                                  tstr, "", tchar,
333                                  true,
334                                  ,
335                                  in_nbytes * 4,
336                                  WIMLIB_ERR_UNICODE_STRING_NOT_REPRESENTABLE,
337                                  ERROR("Failed to convert UTF-8 string to "
338                                        "multibyte string!");
339                                  ERROR("This may be because the UTF-8 data "
340                                        "could not be represented\n"
341                                        "        in your locale's character encoding."),
342                                  )
343 #endif
344
345 int
346 tstr_to_utf8_simple(const tchar *tstr, char **out)
347 {
348         size_t out_nbytes;
349         return tstr_to_utf8(tstr, tstrlen(tstr) * sizeof(tchar),
350                             out, &out_nbytes);
351 }
352
353 int
354 utf8_to_tstr_simple(const char *utf8str, tchar **out)
355 {
356         size_t out_nbytes;
357         return utf8_to_tstr(utf8str, strlen(utf8str), out, &out_nbytes);
358 }
359
360 static void
361 iconv_cleanup(struct iconv_list_head *head)
362 {
363         pthread_mutex_destroy(&head->mutex);
364         while (!list_empty(&head->list)) {
365                 struct iconv_node *i;
366
367                 i = container_of(head->list.next, struct iconv_node, list);
368                 list_del(&i->list);
369                 iconv_close(i->cd);
370                 FREE(i);
371         }
372 }
373
374 void
375 iconv_global_cleanup(void)
376 {
377         iconv_cleanup(&iconv_utf8_to_tstr);
378         iconv_cleanup(&iconv_tstr_to_utf8);
379 #if !TCHAR_IS_UTF16LE
380         iconv_cleanup(&iconv_utf16le_to_tstr);
381         iconv_cleanup(&iconv_tstr_to_utf16le);
382         iconv_cleanup(&iconv_utf16le_to_utf8);
383         iconv_cleanup(&iconv_utf8_to_utf16le);
384 #endif
385 }
386
387 /* A table that maps from UCS-2 characters to their upper case equivalents.
388  * Index and array values are both CPU endian.
389  * Note: this is only an *approximation* of real UTF-16 case folding.
390  */
391 static u16 upcase[65536];
392
393 /*
394  * Initialize the 'upcase' table.
395  *
396  * Code modified from NTFS-3g, original copyright notices below:
397  *
398  * Copyright (c) 2000-2004 Anton Altaparmakov
399  * Copyright (c) 2002-2009 Szabolcs Szakacsits
400  * Copyright (c) 2008-2011 Jean-Pierre Andre
401  * Copyright (c) 2008      Bernhard Kaindl
402  *
403  * License GPLv2 or later.  (Or may actually be LGPL since it's in libntfs-3g.)
404  *
405  * The expanded table itself is the same as that used by Windows Vista for
406  * NTFS case insensitivity.
407  */
408 void
409 init_upcase(void)
410 {
411         /*
412          *      This is the table as defined by Vista
413          */
414         /*
415          * "Start" is inclusive and "End" is exclusive, every value has the
416          * value of "Add" added to it.
417          */
418         static const u16 uc_run_table[][3] = { /* Start, End, Add */
419         {0x0061, 0x007b,   -32}, {0x00e0, 0x00f7,  -32}, {0x00f8, 0x00ff, -32},
420         {0x0256, 0x0258,  -205}, {0x028a, 0x028c, -217}, {0x037b, 0x037e, 130},
421         {0x03ac, 0x03ad,   -38}, {0x03ad, 0x03b0,  -37}, {0x03b1, 0x03c2, -32},
422         {0x03c2, 0x03c3,   -31}, {0x03c3, 0x03cc,  -32}, {0x03cc, 0x03cd, -64},
423         {0x03cd, 0x03cf,   -63}, {0x0430, 0x0450,  -32}, {0x0450, 0x0460, -80},
424         {0x0561, 0x0587,   -48}, {0x1f00, 0x1f08,    8}, {0x1f10, 0x1f16,   8},
425         {0x1f20, 0x1f28,     8}, {0x1f30, 0x1f38,    8}, {0x1f40, 0x1f46,   8},
426         {0x1f51, 0x1f52,     8}, {0x1f53, 0x1f54,    8}, {0x1f55, 0x1f56,   8},
427         {0x1f57, 0x1f58,     8}, {0x1f60, 0x1f68,    8}, {0x1f70, 0x1f72,  74},
428         {0x1f72, 0x1f76,    86}, {0x1f76, 0x1f78,  100}, {0x1f78, 0x1f7a, 128},
429         {0x1f7a, 0x1f7c,   112}, {0x1f7c, 0x1f7e,  126}, {0x1f80, 0x1f88,   8},
430         {0x1f90, 0x1f98,     8}, {0x1fa0, 0x1fa8,    8}, {0x1fb0, 0x1fb2,   8},
431         {0x1fb3, 0x1fb4,     9}, {0x1fcc, 0x1fcd,   -9}, {0x1fd0, 0x1fd2,   8},
432         {0x1fe0, 0x1fe2,     8}, {0x1fe5, 0x1fe6,    7}, {0x1ffc, 0x1ffd,  -9},
433         {0x2170, 0x2180,   -16}, {0x24d0, 0x24ea,  -26}, {0x2c30, 0x2c5f, -48},
434         {0x2d00, 0x2d26, -7264}, {0xff41, 0xff5b,  -32},
435         };
436         /*
437          * "Start" is exclusive and "End" is inclusive, every second value is
438          * decremented by one.
439          */
440         static const u16 uc_dup_table[][2] = { /* Start, End */
441         {0x0100, 0x012f}, {0x0132, 0x0137}, {0x0139, 0x0149}, {0x014a, 0x0178},
442         {0x0179, 0x017e}, {0x01a0, 0x01a6}, {0x01b3, 0x01b7}, {0x01cd, 0x01dd},
443         {0x01de, 0x01ef}, {0x01f4, 0x01f5}, {0x01f8, 0x01f9}, {0x01fa, 0x0220},
444         {0x0222, 0x0234}, {0x023b, 0x023c}, {0x0241, 0x0242}, {0x0246, 0x024f},
445         {0x03d8, 0x03ef}, {0x03f7, 0x03f8}, {0x03fa, 0x03fb}, {0x0460, 0x0481},
446         {0x048a, 0x04bf}, {0x04c1, 0x04c4}, {0x04c5, 0x04c8}, {0x04c9, 0x04ce},
447         {0x04ec, 0x04ed}, {0x04d0, 0x04eb}, {0x04ee, 0x04f5}, {0x04f6, 0x0513},
448         {0x1e00, 0x1e95}, {0x1ea0, 0x1ef9}, {0x2183, 0x2184}, {0x2c60, 0x2c61},
449         {0x2c67, 0x2c6c}, {0x2c75, 0x2c76}, {0x2c80, 0x2ce3},
450         };
451         /*
452          * Set the Unicode character at offset "Offset" to "Value".  Note,
453          * "Value" is host endian.
454          */
455         static const u16 uc_byte_table[][2] = { /* Offset, Value */
456         {0x00ff, 0x0178}, {0x0180, 0x0243}, {0x0183, 0x0182}, {0x0185, 0x0184},
457         {0x0188, 0x0187}, {0x018c, 0x018b}, {0x0192, 0x0191}, {0x0195, 0x01f6},
458         {0x0199, 0x0198}, {0x019a, 0x023d}, {0x019e, 0x0220}, {0x01a8, 0x01a7},
459         {0x01ad, 0x01ac}, {0x01b0, 0x01af}, {0x01b9, 0x01b8}, {0x01bd, 0x01bc},
460         {0x01bf, 0x01f7}, {0x01c6, 0x01c4}, {0x01c9, 0x01c7}, {0x01cc, 0x01ca},
461         {0x01dd, 0x018e}, {0x01f3, 0x01f1}, {0x023a, 0x2c65}, {0x023e, 0x2c66},
462         {0x0253, 0x0181}, {0x0254, 0x0186}, {0x0259, 0x018f}, {0x025b, 0x0190},
463         {0x0260, 0x0193}, {0x0263, 0x0194}, {0x0268, 0x0197}, {0x0269, 0x0196},
464         {0x026b, 0x2c62}, {0x026f, 0x019c}, {0x0272, 0x019d}, {0x0275, 0x019f},
465         {0x027d, 0x2c64}, {0x0280, 0x01a6}, {0x0283, 0x01a9}, {0x0288, 0x01ae},
466         {0x0289, 0x0244}, {0x028c, 0x0245}, {0x0292, 0x01b7}, {0x03f2, 0x03f9},
467         {0x04cf, 0x04c0}, {0x1d7d, 0x2c63}, {0x214e, 0x2132},
468         };
469
470         for (u32 i = 0; i < 65536; i++)
471                 upcase[i] = i;
472
473         for (u32 r = 0; r < ARRAY_LEN(uc_run_table); r++)
474                 for (u32 i = uc_run_table[r][0]; i < uc_run_table[r][1]; i++)
475                         upcase[i] = i + uc_run_table[r][2];
476
477         for (u32 r = 0; r < ARRAY_LEN(uc_dup_table); r++)
478                 for (u32 i = uc_dup_table[r][0]; i < uc_dup_table[r][1]; i += 2)
479                         upcase[i + 1] = i;
480
481         for (u32 r = 0; r < ARRAY_LEN(uc_byte_table); r++)
482                 upcase[uc_byte_table[r][0]] = uc_byte_table[r][1];
483
484 #if 0
485         /* Sanity checks  */
486         wimlib_assert(upcase['a'] == 'A');
487         wimlib_assert(upcase['A'] == 'A');
488         wimlib_assert(upcase['z'] == 'Z');
489         wimlib_assert(upcase['Z'] == 'Z');
490         wimlib_assert(upcase['1'] == '1');
491         wimlib_assert(upcase[0x00e9] == 0x00c9); /* Latin letter e, with acute accent  */
492         wimlib_assert(upcase[0x00c9] == 0x00c9);
493         wimlib_assert(upcase[0x03c1] == 0x03a1); /* Greek letter rho  */
494         wimlib_assert(upcase[0x03a1] == 0x03a1);
495         wimlib_assert(upcase[0x0436] == 0x0416); /* Cyrillic letter zhe  */
496         wimlib_assert(upcase[0x0416] == 0x0416);
497         wimlib_assert(upcase[0x0567] == 0x0537); /* Armenian letter eh  */
498         wimlib_assert(upcase[0x0537] == 0x0537);
499         wimlib_assert(upcase[0x24d0] == 0x24b6); /* Circled Latin letter A
500                                                     (is that a real character???)  */
501         wimlib_assert(upcase[0x24b6] == 0x24b6);
502         wimlib_assert(upcase[0x2603] == 0x2603); /* Note to self: Upper case
503                                                     snowman symbol does not
504                                                     exist.  */
505 #endif
506 }
507
508 /* Compare UTF-16LE strings case-sensitively (%ignore_case == false) or
509  * case-insensitively (%ignore_case == true).
510  *
511  * This is implemented using the default upper-case table used by NTFS.  It does
512  * not handle all possible cases allowed by UTF-16LE.  For example, different
513  * normalizations of the same sequence of "characters" are not considered equal.
514  * It hopefully does the right thing most of the time though.  */
515 int
516 cmp_utf16le_strings(const utf16lechar *s1, size_t n1,
517                     const utf16lechar *s2, size_t n2,
518                     bool ignore_case)
519 {
520         size_t n = min(n1, n2);
521
522         if (ignore_case) {
523                 for (size_t i = 0; i < n; i++) {
524                         u16 c1 = upcase[le16_to_cpu(s1[i])];
525                         u16 c2 = upcase[le16_to_cpu(s2[i])];
526                         if (c1 != c2)
527                                 return (c1 < c2) ? -1 : 1;
528                 }
529         } else {
530                 for (size_t i = 0; i < n; i++) {
531                         u16 c1 = le16_to_cpu(s1[i]);
532                         u16 c2 = le16_to_cpu(s2[i]);
533                         if (c1 != c2)
534                                 return (c1 < c2) ? -1 : 1;
535                 }
536         }
537         if (n1 == n2)
538                 return 0;
539         return (n1 < n2) ? -1 : 1;
540 }
541
542 /* Duplicates a string of system-dependent encoding into a UTF-16LE string and
543  * returns the string and its length, in bytes, in the pointer arguments.  Frees
544  * any existing string at the return location before overwriting it. */
545 int
546 get_utf16le_string(const tchar *name, utf16lechar **name_utf16le_ret,
547                    u16 *name_utf16le_nbytes_ret)
548 {
549         utf16lechar *name_utf16le;
550         size_t name_utf16le_nbytes;
551         int ret;
552 #if TCHAR_IS_UTF16LE
553         name_utf16le_nbytes = tstrlen(name) * sizeof(utf16lechar);
554         name_utf16le = MALLOC(name_utf16le_nbytes + sizeof(utf16lechar));
555         if (name_utf16le == NULL)
556                 return WIMLIB_ERR_NOMEM;
557         memcpy(name_utf16le, name, name_utf16le_nbytes + sizeof(utf16lechar));
558         ret = 0;
559 #else
560
561         ret = tstr_to_utf16le(name, tstrlen(name), &name_utf16le,
562                               &name_utf16le_nbytes);
563         if (ret == 0) {
564                 if (name_utf16le_nbytes > 0xffff) {
565                         FREE(name_utf16le);
566                         ERROR("Multibyte string \"%"TS"\" is too long!", name);
567                         ret = WIMLIB_ERR_INVALID_UTF8_STRING;
568                 }
569         }
570 #endif
571         if (ret == 0) {
572                 FREE(*name_utf16le_ret);
573                 *name_utf16le_ret = name_utf16le;
574                 *name_utf16le_nbytes_ret = name_utf16le_nbytes;
575         }
576         return ret;
577 }
578