+iconv_global_init(void)
+{
+ iconv_init(&iconv_utf8_to_tstr);
+ iconv_init(&iconv_tstr_to_utf8);
+#if !TCHAR_IS_UTF16LE
+ iconv_init(&iconv_utf16le_to_tstr);
+ iconv_init(&iconv_tstr_to_utf16le);
+ iconv_init(&iconv_utf16le_to_utf8);
+ iconv_init(&iconv_utf8_to_utf16le);
+#endif
+}
+
+void
+iconv_global_cleanup(void)
+{
+ iconv_cleanup(&iconv_utf8_to_tstr);
+ iconv_cleanup(&iconv_tstr_to_utf8);
+#if !TCHAR_IS_UTF16LE
+ iconv_cleanup(&iconv_utf16le_to_tstr);
+ iconv_cleanup(&iconv_tstr_to_utf16le);
+ iconv_cleanup(&iconv_utf16le_to_utf8);
+ iconv_cleanup(&iconv_utf8_to_utf16le);
+#endif
+}
+
+/* A table that maps from UCS-2 characters to their upper case equivalents.
+ * Index and array values are both CPU endian.
+ * Note: this is only an *approximation* of real UTF-16 case folding.
+ */
+static u16 upcase[65536];
+
+void
+init_upcase(void)
+{
+ /* This is the table used in NTFS volumes formatted by Windows 10.
+ * It was compressed by tools/compress_upcase_table.c. */
+ static const u16 upcase_compressed[] = {
+ 0x0000, 0x0000, 0x0060, 0x0000, 0x0000, 0xffe0, 0x0019, 0x0061,
+ 0x0061, 0x0000, 0x001b, 0x005d, 0x0008, 0x0060, 0x0000, 0x0079,
+ 0x0000, 0x0000, 0x0000, 0xffff, 0x002f, 0x0100, 0x0002, 0x0000,
+ 0x0007, 0x012b, 0x0011, 0x0121, 0x002f, 0x0103, 0x0006, 0x0101,
+ 0x0000, 0x00c3, 0x0006, 0x0131, 0x0007, 0x012e, 0x0004, 0x0000,
+ 0x0003, 0x012f, 0x0000, 0x0061, 0x0004, 0x0130, 0x0000, 0x00a3,
+ 0x0003, 0x0000, 0x0000, 0x0082, 0x000b, 0x0131, 0x0006, 0x0189,
+ 0x0008, 0x012f, 0x0007, 0x012e, 0x0000, 0x0038, 0x0006, 0x0000,
+ 0x0000, 0xfffe, 0x0007, 0x01c4, 0x000f, 0x0101, 0x0000, 0xffb1,
+ 0x0015, 0x011e, 0x0004, 0x01cc, 0x002a, 0x0149, 0x0014, 0x0149,
+ 0x0007, 0x0000, 0x0009, 0x018c, 0x000b, 0x0138, 0x0000, 0x2a1f,
+ 0x0000, 0x2a1c, 0x0000, 0x0000, 0x0000, 0xff2e, 0x0000, 0xff32,
+ 0x0000, 0x0000, 0x0000, 0xff33, 0x0000, 0xff33, 0x0000, 0x0000,
+ 0x0000, 0xff36, 0x0000, 0x0000, 0x0000, 0xff35, 0x0004, 0x0000,
+ 0x0002, 0x0257, 0x0000, 0x0000, 0x0000, 0xff31, 0x0004, 0x0000,
+ 0x0000, 0xff2f, 0x0000, 0xff2d, 0x0000, 0x0000, 0x0000, 0x29f7,
+ 0x0003, 0x0000, 0x0002, 0x0269, 0x0000, 0x29fd, 0x0000, 0xff2b,
+ 0x0002, 0x0000, 0x0000, 0xff2a, 0x0007, 0x0000, 0x0000, 0x29e7,
+ 0x0002, 0x0000, 0x0000, 0xff26, 0x0005, 0x027e, 0x0003, 0x027e,
+ 0x0000, 0xffbb, 0x0000, 0xff27, 0x0000, 0xff27, 0x0000, 0xffb9,
+ 0x0005, 0x0000, 0x0000, 0xff25, 0x0065, 0x007b, 0x0079, 0x0293,
+ 0x0008, 0x012d, 0x0003, 0x019c, 0x0002, 0x037b, 0x002e, 0x0000,
+ 0x0000, 0xffda, 0x0000, 0xffdb, 0x0002, 0x03ad, 0x0012, 0x0060,
+ 0x000a, 0x0060, 0x0000, 0xffc0, 0x0000, 0xffc1, 0x0000, 0xffc1,
+ 0x0008, 0x0000, 0x0000, 0xfff8, 0x001a, 0x0118, 0x0000, 0x0007,
+ 0x0008, 0x018d, 0x0009, 0x0233, 0x0046, 0x0035, 0x0006, 0x0061,
+ 0x0000, 0xffb0, 0x000f, 0x0450, 0x0025, 0x010e, 0x000a, 0x036b,
+ 0x0032, 0x048b, 0x000e, 0x0100, 0x0000, 0xfff1, 0x0037, 0x048a,
+ 0x0026, 0x0465, 0x0034, 0x0000, 0x0000, 0xffd0, 0x0025, 0x0561,
+ 0x00de, 0x0293, 0x1714, 0x0587, 0x0000, 0x8a04, 0x0003, 0x0000,
+ 0x0000, 0x0ee6, 0x0087, 0x02ee, 0x0092, 0x1e01, 0x0069, 0x1df7,
+ 0x0000, 0x0008, 0x0007, 0x1f00, 0x0008, 0x0000, 0x000e, 0x1f02,
+ 0x0008, 0x1f0e, 0x0010, 0x1f06, 0x001a, 0x1f06, 0x0002, 0x1f0f,
+ 0x0007, 0x1f50, 0x0017, 0x1f19, 0x0000, 0x004a, 0x0000, 0x004a,
+ 0x0000, 0x0056, 0x0003, 0x1f72, 0x0000, 0x0064, 0x0000, 0x0064,
+ 0x0000, 0x0080, 0x0000, 0x0080, 0x0000, 0x0070, 0x0000, 0x0070,
+ 0x0000, 0x007e, 0x0000, 0x007e, 0x0028, 0x1f1e, 0x000c, 0x1f06,
+ 0x0000, 0x0000, 0x0000, 0x0009, 0x000f, 0x0000, 0x000d, 0x1fb3,
+ 0x000d, 0x1f44, 0x0008, 0x1fcd, 0x0006, 0x03f2, 0x0015, 0x1fbb,
+ 0x014e, 0x0587, 0x0000, 0xffe4, 0x0021, 0x0000, 0x0000, 0xfff0,
+ 0x000f, 0x2170, 0x000a, 0x0238, 0x0346, 0x0587, 0x0000, 0xffe6,
+ 0x0019, 0x24d0, 0x0746, 0x0587, 0x0026, 0x0561, 0x000b, 0x057e,
+ 0x0004, 0x012f, 0x0000, 0xd5d5, 0x0000, 0xd5d8, 0x000c, 0x022e,
+ 0x000e, 0x03f8, 0x006e, 0x1e33, 0x0011, 0x0000, 0x0000, 0xe3a0,
+ 0x0025, 0x2d00, 0x17f2, 0x0587, 0x6129, 0x2d26, 0x002e, 0x0201,
+ 0x002a, 0x1def, 0x0098, 0xa5b7, 0x0040, 0x1dff, 0x000e, 0x0368,
+ 0x000d, 0x022b, 0x034c, 0x2184, 0x5469, 0x2d26, 0x007f, 0x0061,
+ 0x0040, 0x0000,
+ };
+
+ /* Simple LZ decoder */
+ const u16 *in_next = upcase_compressed;
+ for (u32 i = 0; i < ARRAY_LEN(upcase); ) {
+ u16 length = *in_next++;
+ u16 src_pos = *in_next++;
+ if (length == 0) {
+ /* Literal */
+ upcase[i++] = src_pos;
+ } else {
+ /* Match */
+ do {
+ upcase[i++] = upcase[src_pos++];
+ } while (--length);
+ }
+ }
+
+ /* Delta filter */
+ for (u32 i = 0; i < ARRAY_LEN(upcase); i++)
+ upcase[i] += i;
+
+#if 0
+ /* Sanity checks */
+ wimlib_assert(upcase['a'] == 'A');
+ wimlib_assert(upcase['A'] == 'A');
+ wimlib_assert(upcase['z'] == 'Z');
+ wimlib_assert(upcase['Z'] == 'Z');
+ wimlib_assert(upcase['1'] == '1');
+ wimlib_assert(upcase[0x00e9] == 0x00c9); /* Latin letter e, with acute accent */
+ wimlib_assert(upcase[0x00c9] == 0x00c9);
+ wimlib_assert(upcase[0x03c1] == 0x03a1); /* Greek letter rho */
+ wimlib_assert(upcase[0x03a1] == 0x03a1);
+ wimlib_assert(upcase[0x0436] == 0x0416); /* Cyrillic letter zhe */
+ wimlib_assert(upcase[0x0416] == 0x0416);
+ wimlib_assert(upcase[0x0567] == 0x0537); /* Armenian letter eh */
+ wimlib_assert(upcase[0x0537] == 0x0537);
+ wimlib_assert(upcase[0x24d0] == 0x24b6); /* Circled Latin letter A
+ (is that a real character???) */
+ wimlib_assert(upcase[0x24b6] == 0x24b6);
+ wimlib_assert(upcase[0x2603] == 0x2603); /* Note to self: Upper case
+ snowman symbol does not
+ exist. */
+#endif
+}
+
+/* Compare UTF-16LE strings case-sensitively (%ignore_case == false) or
+ * case-insensitively (%ignore_case == true).
+ *
+ * This is implemented using the default upper-case table used by NTFS. It does
+ * not handle all possible cases allowed by UTF-16LE. For example, different
+ * normalizations of the same sequence of "characters" are not considered equal.
+ * It hopefully does the right thing most of the time though. */
+int
+cmp_utf16le_strings(const utf16lechar *s1, size_t n1,
+ const utf16lechar *s2, size_t n2,
+ bool ignore_case)
+{
+ size_t n = min(n1, n2);
+
+ if (ignore_case) {
+ for (size_t i = 0; i < n; i++) {
+ u16 c1 = upcase[le16_to_cpu(s1[i])];
+ u16 c2 = upcase[le16_to_cpu(s2[i])];
+ if (c1 != c2)
+ return (c1 < c2) ? -1 : 1;
+ }
+ } else {
+ for (size_t i = 0; i < n; i++) {
+ u16 c1 = le16_to_cpu(s1[i]);
+ u16 c2 = le16_to_cpu(s2[i]);
+ if (c1 != c2)
+ return (c1 < c2) ? -1 : 1;
+ }
+ }
+ if (n1 == n2)
+ return 0;
+ return (n1 < n2) ? -1 : 1;
+}
+
+/* Like cmp_utf16le_strings(), but assumes the strings are null terminated. */
+int
+cmp_utf16le_strings_z(const utf16lechar *s1, const utf16lechar *s2,
+ bool ignore_case)
+{
+ if (ignore_case) {
+ for (;;) {
+ u16 c1 = upcase[le16_to_cpu(*s1)];
+ u16 c2 = upcase[le16_to_cpu(*s2)];
+ if (c1 != c2)
+ return (c1 < c2) ? -1 : 1;
+ if (c1 == 0)
+ return 0;
+ s1++, s2++;
+ }
+ } else {
+ while (*s1 && *s1 == *s2)
+ s1++, s2++;
+ if (*s1 == *s2)
+ return 0;
+ return (le16_to_cpu(*s1) < le16_to_cpu(*s2)) ? -1 : 1;
+ }
+}
+
+/* Duplicate a UTF-16LE string. The input string might not be null terminated
+ * and might be misaligned, but the returned string is guaranteed to be null
+ * terminated and properly aligned. */
+utf16lechar *
+utf16le_dupz(const void *ustr, size_t usize)
+{
+ utf16lechar *dup = MALLOC(usize + sizeof(utf16lechar));
+ if (dup) {
+ memcpy(dup, ustr, usize);
+ dup[usize / sizeof(utf16lechar)] = 0;
+ }
+ return dup;
+}
+
+/* Duplicate a null-terminated UTF-16LE string. */
+utf16lechar *
+utf16le_dup(const utf16lechar *ustr)
+{
+ const utf16lechar *p = ustr;
+ while (*p++)
+ ;
+ return memdup(ustr, (const u8 *)p - (const u8 *)ustr);
+}
+
+/* Return the length, in bytes, of a UTF-null terminated UTF-16 string,
+ * excluding the null terminator. */
+size_t
+utf16le_len_bytes(const utf16lechar *s)