xref: /sqlite-3.40.0/src/utf.c (revision 74e4352a)
1 /*
2 ** 2004 April 13
3 **
4 ** The author disclaims copyright to this source code.  In place of
5 ** a legal notice, here is a blessing:
6 **
7 **    May you do good and not evil.
8 **    May you find forgiveness for yourself and forgive others.
9 **    May you share freely, never taking more than you give.
10 **
11 *************************************************************************
12 ** This file contains routines used to translate between UTF-8,
13 ** UTF-16, UTF-16BE, and UTF-16LE.
14 **
15 ** $Id: utf.c,v 1.43 2006/10/19 01:58:44 drh Exp $
16 **
17 ** Notes on UTF-8:
18 **
19 **   Byte-0    Byte-1    Byte-2    Byte-3    Value
20 **  0xxxxxxx                                 00000000 00000000 0xxxxxxx
21 **  110yyyyy  10xxxxxx                       00000000 00000yyy yyxxxxxx
22 **  1110zzzz  10yyyyyy  10xxxxxx             00000000 zzzzyyyy yyxxxxxx
23 **  11110uuu  10uuzzzz  10yyyyyy  10xxxxxx   000uuuuu zzzzyyyy yyxxxxxx
24 **
25 **
26 ** Notes on UTF-16:  (with wwww+1==uuuuu)
27 **
28 **      Word-0               Word-1          Value
29 **  110110ww wwzzzzyy   110111yy yyxxxxxx    000uuuuu zzzzyyyy yyxxxxxx
30 **  zzzzyyyy yyxxxxxx                        00000000 zzzzyyyy yyxxxxxx
31 **
32 **
33 ** BOM or Byte Order Mark:
34 **     0xff 0xfe   little-endian utf-16 follows
35 **     0xfe 0xff   big-endian utf-16 follows
36 **
37 **
38 ** Handling of malformed strings:
39 **
40 ** SQLite accepts and processes malformed strings without an error wherever
41 ** possible. However this is not possible when converting between UTF-8 and
42 ** UTF-16.
43 **
44 ** When converting malformed UTF-8 strings to UTF-16, one instance of the
45 ** replacement character U+FFFD for each byte that cannot be interpeted as
46 ** part of a valid unicode character.
47 **
48 ** When converting malformed UTF-16 strings to UTF-8, one instance of the
49 ** replacement character U+FFFD for each pair of bytes that cannot be
50 ** interpeted as part of a valid unicode character.
51 **
52 ** This file contains the following public routines:
53 **
54 ** sqlite3VdbeMemTranslate() - Translate the encoding used by a Mem* string.
55 ** sqlite3VdbeMemHandleBom() - Handle byte-order-marks in UTF16 Mem* strings.
56 ** sqlite3utf16ByteLen()     - Calculate byte-length of a void* UTF16 string.
57 ** sqlite3utf8CharLen()      - Calculate char-length of a char* UTF8 string.
58 ** sqlite3utf8LikeCompare()  - Do a LIKE match given two UTF8 char* strings.
59 **
60 */
61 #include "sqliteInt.h"
62 #include <assert.h>
63 #include "vdbeInt.h"
64 
65 /*
66 ** This table maps from the first byte of a UTF-8 character to the number
67 ** of trailing bytes expected. A value '4' indicates that the table key
68 ** is not a legal first byte for a UTF-8 character.
69 */
70 static const u8 xtra_utf8_bytes[256]  = {
71 /* 0xxxxxxx */
72 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
73 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
74 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
75 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
76 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
77 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
78 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
79 0, 0, 0, 0, 0, 0, 0, 0,     0, 0, 0, 0, 0, 0, 0, 0,
80 
81 /* 10wwwwww */
82 4, 4, 4, 4, 4, 4, 4, 4,     4, 4, 4, 4, 4, 4, 4, 4,
83 4, 4, 4, 4, 4, 4, 4, 4,     4, 4, 4, 4, 4, 4, 4, 4,
84 4, 4, 4, 4, 4, 4, 4, 4,     4, 4, 4, 4, 4, 4, 4, 4,
85 4, 4, 4, 4, 4, 4, 4, 4,     4, 4, 4, 4, 4, 4, 4, 4,
86 
87 /* 110yyyyy */
88 1, 1, 1, 1, 1, 1, 1, 1,     1, 1, 1, 1, 1, 1, 1, 1,
89 1, 1, 1, 1, 1, 1, 1, 1,     1, 1, 1, 1, 1, 1, 1, 1,
90 
91 /* 1110zzzz */
92 2, 2, 2, 2, 2, 2, 2, 2,     2, 2, 2, 2, 2, 2, 2, 2,
93 
94 /* 11110yyy */
95 3, 3, 3, 3, 3, 3, 3, 3,     4, 4, 4, 4, 4, 4, 4, 4,
96 };
97 
98 /*
99 ** This table maps from the number of trailing bytes in a UTF-8 character
100 ** to an integer constant that is effectively calculated for each character
101 ** read by a naive implementation of a UTF-8 character reader. The code
102 ** in the READ_UTF8 macro explains things best.
103 */
104 static const int xtra_utf8_bits[] =  {
105   0,
106   12416,          /* (0xC0 << 6) + (0x80) */
107   925824,         /* (0xE0 << 12) + (0x80 << 6) + (0x80) */
108   63447168        /* (0xF0 << 18) + (0x80 << 12) + (0x80 << 6) + 0x80 */
109 };
110 
111 /*
112 ** If a UTF-8 character contains N bytes extra bytes (N bytes follow
113 ** the initial byte so that the total character length is N+1) then
114 ** masking the character with utf8_mask[N] must produce a non-zero
115 ** result.  Otherwise, we have an (illegal) overlong encoding.
116 */
117 static const int utf_mask[] = {
118   0x00000000,
119   0xffffff80,
120   0xfffff800,
121   0xffff0000,
122 };
123 
124 #define READ_UTF8(zIn, c) { \
125   int xtra;                                            \
126   c = *(zIn)++;                                        \
127   xtra = xtra_utf8_bytes[c];                           \
128   switch( xtra ){                                      \
129     case 4: c = (int)0xFFFD; break;                    \
130     case 3: c = (c<<6) + *(zIn)++;                     \
131     case 2: c = (c<<6) + *(zIn)++;                     \
132     case 1: c = (c<<6) + *(zIn)++;                     \
133     c -= xtra_utf8_bits[xtra];                         \
134     if( (utf_mask[xtra]&c)==0                          \
135         || (c&0xFFFFF800)==0xD800                      \
136         || (c&0xFFFFFFFE)==0xFFFE ){  c = 0xFFFD; }    \
137   }                                                    \
138 }
139 int sqlite3ReadUtf8(const unsigned char *z){
140   int c;
141   READ_UTF8(z, c);
142   return c;
143 }
144 
145 #define SKIP_UTF8(zIn) {                               \
146   zIn += (xtra_utf8_bytes[*(u8 *)zIn] + 1);            \
147 }
148 
149 #define WRITE_UTF8(zOut, c) {                          \
150   if( c<0x00080 ){                                     \
151     *zOut++ = (c&0xFF);                                \
152   }                                                    \
153   else if( c<0x00800 ){                                \
154     *zOut++ = 0xC0 + ((c>>6)&0x1F);                    \
155     *zOut++ = 0x80 + (c & 0x3F);                       \
156   }                                                    \
157   else if( c<0x10000 ){                                \
158     *zOut++ = 0xE0 + ((c>>12)&0x0F);                   \
159     *zOut++ = 0x80 + ((c>>6) & 0x3F);                  \
160     *zOut++ = 0x80 + (c & 0x3F);                       \
161   }else{                                               \
162     *zOut++ = 0xF0 + ((c>>18) & 0x07);                 \
163     *zOut++ = 0x80 + ((c>>12) & 0x3F);                 \
164     *zOut++ = 0x80 + ((c>>6) & 0x3F);                  \
165     *zOut++ = 0x80 + (c & 0x3F);                       \
166   }                                                    \
167 }
168 
169 #define WRITE_UTF16LE(zOut, c) {                                \
170   if( c<=0xFFFF ){                                              \
171     *zOut++ = (c&0x00FF);                                       \
172     *zOut++ = ((c>>8)&0x00FF);                                  \
173   }else{                                                        \
174     *zOut++ = (((c>>10)&0x003F) + (((c-0x10000)>>10)&0x00C0));  \
175     *zOut++ = (0x00D8 + (((c-0x10000)>>18)&0x03));              \
176     *zOut++ = (c&0x00FF);                                       \
177     *zOut++ = (0x00DC + ((c>>8)&0x03));                         \
178   }                                                             \
179 }
180 
181 #define WRITE_UTF16BE(zOut, c) {                                \
182   if( c<=0xFFFF ){                                              \
183     *zOut++ = ((c>>8)&0x00FF);                                  \
184     *zOut++ = (c&0x00FF);                                       \
185   }else{                                                        \
186     *zOut++ = (0x00D8 + (((c-0x10000)>>18)&0x03));              \
187     *zOut++ = (((c>>10)&0x003F) + (((c-0x10000)>>10)&0x00C0));  \
188     *zOut++ = (0x00DC + ((c>>8)&0x03));                         \
189     *zOut++ = (c&0x00FF);                                       \
190   }                                                             \
191 }
192 
193 #define READ_UTF16LE(zIn, c){                                         \
194   c = (*zIn++);                                                       \
195   c += ((*zIn++)<<8);                                                 \
196   if( c>=0xD800 && c<=0xE000 ){                                       \
197     int c2 = (*zIn++);                                                \
198     c2 += ((*zIn++)<<8);                                              \
199     c = (c2&0x03FF) + ((c&0x003F)<<10) + (((c&0x03C0)+0x0040)<<10);   \
200     if( (c & 0xFFFF0000)==0 ) c = 0xFFFD;                             \
201   }                                                                   \
202 }
203 
204 #define READ_UTF16BE(zIn, c){                                         \
205   c = ((*zIn++)<<8);                                                  \
206   c += (*zIn++);                                                      \
207   if( c>=0xD800 && c<=0xE000 ){                                       \
208     int c2 = ((*zIn++)<<8);                                           \
209     c2 += (*zIn++);                                                   \
210     c = (c2&0x03FF) + ((c&0x003F)<<10) + (((c&0x03C0)+0x0040)<<10);   \
211     if( (c & 0xFFFF0000)==0 ) c = 0xFFFD;                             \
212   }                                                                   \
213 }
214 
215 #define SKIP_UTF16BE(zIn){                                            \
216   if( *zIn>=0xD8 && (*zIn<0xE0 || (*zIn==0xE0 && *(zIn+1)==0x00)) ){  \
217     zIn += 4;                                                         \
218   }else{                                                              \
219     zIn += 2;                                                         \
220   }                                                                   \
221 }
222 #define SKIP_UTF16LE(zIn){                                            \
223   zIn++;                                                              \
224   if( *zIn>=0xD8 && (*zIn<0xE0 || (*zIn==0xE0 && *(zIn-1)==0x00)) ){  \
225     zIn += 3;                                                         \
226   }else{                                                              \
227     zIn += 1;                                                         \
228   }                                                                   \
229 }
230 
231 #define RSKIP_UTF16LE(zIn){                                            \
232   if( *zIn>=0xD8 && (*zIn<0xE0 || (*zIn==0xE0 && *(zIn-1)==0x00)) ){  \
233     zIn -= 4;                                                         \
234   }else{                                                              \
235     zIn -= 2;                                                         \
236   }                                                                   \
237 }
238 #define RSKIP_UTF16BE(zIn){                                            \
239   zIn--;                                                              \
240   if( *zIn>=0xD8 && (*zIn<0xE0 || (*zIn==0xE0 && *(zIn+1)==0x00)) ){  \
241     zIn -= 3;                                                         \
242   }else{                                                              \
243     zIn -= 1;                                                         \
244   }                                                                   \
245 }
246 
247 /*
248 ** If the TRANSLATE_TRACE macro is defined, the value of each Mem is
249 ** printed on stderr on the way into and out of sqlite3VdbeMemTranslate().
250 */
251 /* #define TRANSLATE_TRACE 1 */
252 
253 #ifndef SQLITE_OMIT_UTF16
254 /*
255 ** This routine transforms the internal text encoding used by pMem to
256 ** desiredEnc. It is an error if the string is already of the desired
257 ** encoding, or if *pMem does not contain a string value.
258 */
259 int sqlite3VdbeMemTranslate(Mem *pMem, u8 desiredEnc){
260   unsigned char zShort[NBFS]; /* Temporary short output buffer */
261   int len;                    /* Maximum length of output string in bytes */
262   unsigned char *zOut;                  /* Output buffer */
263   unsigned char *zIn;                   /* Input iterator */
264   unsigned char *zTerm;                 /* End of input */
265   unsigned char *z;                     /* Output iterator */
266   unsigned int c;
267 
268   assert( pMem->flags&MEM_Str );
269   assert( pMem->enc!=desiredEnc );
270   assert( pMem->enc!=0 );
271   assert( pMem->n>=0 );
272 
273 #if defined(TRANSLATE_TRACE) && defined(SQLITE_DEBUG)
274   {
275     char zBuf[100];
276     sqlite3VdbeMemPrettyPrint(pMem, zBuf);
277     fprintf(stderr, "INPUT:  %s\n", zBuf);
278   }
279 #endif
280 
281   /* If the translation is between UTF-16 little and big endian, then
282   ** all that is required is to swap the byte order. This case is handled
283   ** differently from the others.
284   */
285   if( pMem->enc!=SQLITE_UTF8 && desiredEnc!=SQLITE_UTF8 ){
286     u8 temp;
287     int rc;
288     rc = sqlite3VdbeMemMakeWriteable(pMem);
289     if( rc!=SQLITE_OK ){
290       assert( rc==SQLITE_NOMEM );
291       return SQLITE_NOMEM;
292     }
293     zIn = (u8*)pMem->z;
294     zTerm = &zIn[pMem->n];
295     while( zIn<zTerm ){
296       temp = *zIn;
297       *zIn = *(zIn+1);
298       zIn++;
299       *zIn++ = temp;
300     }
301     pMem->enc = desiredEnc;
302     goto translate_out;
303   }
304 
305   /* Set len to the maximum number of bytes required in the output buffer. */
306   if( desiredEnc==SQLITE_UTF8 ){
307     /* When converting from UTF-16, the maximum growth results from
308     ** translating a 2-byte character to a 4-byte UTF-8 character.
309     ** A single byte is required for the output string
310     ** nul-terminator.
311     */
312     len = pMem->n * 2 + 1;
313   }else{
314     /* When converting from UTF-8 to UTF-16 the maximum growth is caused
315     ** when a 1-byte UTF-8 character is translated into a 2-byte UTF-16
316     ** character. Two bytes are required in the output buffer for the
317     ** nul-terminator.
318     */
319     len = pMem->n * 2 + 2;
320   }
321 
322   /* Set zIn to point at the start of the input buffer and zTerm to point 1
323   ** byte past the end.
324   **
325   ** Variable zOut is set to point at the output buffer. This may be space
326   ** obtained from malloc(), or Mem.zShort, if it large enough and not in
327   ** use, or the zShort array on the stack (see above).
328   */
329   zIn = (u8*)pMem->z;
330   zTerm = &zIn[pMem->n];
331   if( len>NBFS ){
332     zOut = sqliteMallocRaw(len);
333     if( !zOut ) return SQLITE_NOMEM;
334   }else{
335     zOut = zShort;
336   }
337   z = zOut;
338 
339   if( pMem->enc==SQLITE_UTF8 ){
340     if( desiredEnc==SQLITE_UTF16LE ){
341       /* UTF-8 -> UTF-16 Little-endian */
342       while( zIn<zTerm ){
343         READ_UTF8(zIn, c);
344         WRITE_UTF16LE(z, c);
345       }
346     }else{
347       assert( desiredEnc==SQLITE_UTF16BE );
348       /* UTF-8 -> UTF-16 Big-endian */
349       while( zIn<zTerm ){
350         READ_UTF8(zIn, c);
351         WRITE_UTF16BE(z, c);
352       }
353     }
354     pMem->n = z - zOut;
355     *z++ = 0;
356   }else{
357     assert( desiredEnc==SQLITE_UTF8 );
358     if( pMem->enc==SQLITE_UTF16LE ){
359       /* UTF-16 Little-endian -> UTF-8 */
360       while( zIn<zTerm ){
361         READ_UTF16LE(zIn, c);
362         WRITE_UTF8(z, c);
363       }
364     }else{
365       /* UTF-16 Little-endian -> UTF-8 */
366       while( zIn<zTerm ){
367         READ_UTF16BE(zIn, c);
368         WRITE_UTF8(z, c);
369       }
370     }
371     pMem->n = z - zOut;
372   }
373   *z = 0;
374   assert( (pMem->n+(desiredEnc==SQLITE_UTF8?1:2))<=len );
375 
376   sqlite3VdbeMemRelease(pMem);
377   pMem->flags &= ~(MEM_Static|MEM_Dyn|MEM_Ephem|MEM_Short);
378   pMem->enc = desiredEnc;
379   if( zOut==zShort ){
380     memcpy(pMem->zShort, zOut, len);
381     zOut = (u8*)pMem->zShort;
382     pMem->flags |= (MEM_Term|MEM_Short);
383   }else{
384     pMem->flags |= (MEM_Term|MEM_Dyn);
385   }
386   pMem->z = (char*)zOut;
387 
388 translate_out:
389 #if defined(TRANSLATE_TRACE) && defined(SQLITE_DEBUG)
390   {
391     char zBuf[100];
392     sqlite3VdbeMemPrettyPrint(pMem, zBuf);
393     fprintf(stderr, "OUTPUT: %s\n", zBuf);
394   }
395 #endif
396   return SQLITE_OK;
397 }
398 
399 /*
400 ** This routine checks for a byte-order mark at the beginning of the
401 ** UTF-16 string stored in *pMem. If one is present, it is removed and
402 ** the encoding of the Mem adjusted. This routine does not do any
403 ** byte-swapping, it just sets Mem.enc appropriately.
404 **
405 ** The allocation (static, dynamic etc.) and encoding of the Mem may be
406 ** changed by this function.
407 */
408 int sqlite3VdbeMemHandleBom(Mem *pMem){
409   int rc = SQLITE_OK;
410   u8 bom = 0;
411 
412   if( pMem->n<0 || pMem->n>1 ){
413     u8 b1 = *(u8 *)pMem->z;
414     u8 b2 = *(((u8 *)pMem->z) + 1);
415     if( b1==0xFE && b2==0xFF ){
416       bom = SQLITE_UTF16BE;
417     }
418     if( b1==0xFF && b2==0xFE ){
419       bom = SQLITE_UTF16LE;
420     }
421   }
422 
423   if( bom ){
424     /* This function is called as soon as a string is stored in a Mem*,
425     ** from within sqlite3VdbeMemSetStr(). At that point it is not possible
426     ** for the string to be stored in Mem.zShort, or for it to be stored
427     ** in dynamic memory with no destructor.
428     */
429     assert( !(pMem->flags&MEM_Short) );
430     assert( !(pMem->flags&MEM_Dyn) || pMem->xDel );
431     if( pMem->flags & MEM_Dyn ){
432       void (*xDel)(void*) = pMem->xDel;
433       char *z = pMem->z;
434       pMem->z = 0;
435       pMem->xDel = 0;
436       rc = sqlite3VdbeMemSetStr(pMem, &z[2], pMem->n-2, bom, SQLITE_TRANSIENT);
437       xDel(z);
438     }else{
439       rc = sqlite3VdbeMemSetStr(pMem, &pMem->z[2], pMem->n-2, bom,
440           SQLITE_TRANSIENT);
441     }
442   }
443   return rc;
444 }
445 #endif /* SQLITE_OMIT_UTF16 */
446 
447 /*
448 ** pZ is a UTF-8 encoded unicode string. If nByte is less than zero,
449 ** return the number of unicode characters in pZ up to (but not including)
450 ** the first 0x00 byte. If nByte is not less than zero, return the
451 ** number of unicode characters in the first nByte of pZ (or up to
452 ** the first 0x00, whichever comes first).
453 */
454 int sqlite3utf8CharLen(const char *z, int nByte){
455   int r = 0;
456   const char *zTerm;
457   if( nByte>=0 ){
458     zTerm = &z[nByte];
459   }else{
460     zTerm = (const char *)(-1);
461   }
462   assert( z<=zTerm );
463   while( *z!=0 && z<zTerm ){
464     SKIP_UTF8(z);
465     r++;
466   }
467   return r;
468 }
469 
470 #ifndef SQLITE_OMIT_UTF16
471 /*
472 ** Convert a UTF-16 string in the native encoding into a UTF-8 string.
473 ** Memory to hold the UTF-8 string is obtained from malloc and must be
474 ** freed by the calling function.
475 **
476 ** NULL is returned if there is an allocation error.
477 */
478 char *sqlite3utf16to8(const void *z, int nByte){
479   Mem m;
480   memset(&m, 0, sizeof(m));
481   sqlite3VdbeMemSetStr(&m, z, nByte, SQLITE_UTF16NATIVE, SQLITE_STATIC);
482   sqlite3VdbeChangeEncoding(&m, SQLITE_UTF8);
483   assert( (m.flags & MEM_Term)!=0 || sqlite3MallocFailed() );
484   assert( (m.flags & MEM_Str)!=0 || sqlite3MallocFailed() );
485   return (m.flags & MEM_Dyn)!=0 ? m.z : sqliteStrDup(m.z);
486 }
487 
488 /*
489 ** pZ is a UTF-16 encoded unicode string. If nChar is less than zero,
490 ** return the number of bytes up to (but not including), the first pair
491 ** of consecutive 0x00 bytes in pZ. If nChar is not less than zero,
492 ** then return the number of bytes in the first nChar unicode characters
493 ** in pZ (or up until the first pair of 0x00 bytes, whichever comes first).
494 */
495 int sqlite3utf16ByteLen(const void *zIn, int nChar){
496   unsigned int c = 1;
497   char const *z = zIn;
498   int n = 0;
499   if( SQLITE_UTF16NATIVE==SQLITE_UTF16BE ){
500     /* Using an "if (SQLITE_UTF16NATIVE==SQLITE_UTF16BE)" construct here
501     ** and in other parts of this file means that at one branch will
502     ** not be covered by coverage testing on any single host. But coverage
503     ** will be complete if the tests are run on both a little-endian and
504     ** big-endian host. Because both the UTF16NATIVE and SQLITE_UTF16BE
505     ** macros are constant at compile time the compiler can determine
506     ** which branch will be followed. It is therefore assumed that no runtime
507     ** penalty is paid for this "if" statement.
508     */
509     while( c && ((nChar<0) || n<nChar) ){
510       READ_UTF16BE(z, c);
511       n++;
512     }
513   }else{
514     while( c && ((nChar<0) || n<nChar) ){
515       READ_UTF16LE(z, c);
516       n++;
517     }
518   }
519   return (z-(char const *)zIn)-((c==0)?2:0);
520 }
521 
522 /*
523 ** UTF-16 implementation of the substr()
524 */
525 void sqlite3utf16Substr(
526   sqlite3_context *context,
527   int argc,
528   sqlite3_value **argv
529 ){
530   int y, z;
531   unsigned char const *zStr;
532   unsigned char const *zStrEnd;
533   unsigned char const *zStart;
534   unsigned char const *zEnd;
535   int i;
536 
537   zStr = (unsigned char const *)sqlite3_value_text16(argv[0]);
538   zStrEnd = &zStr[sqlite3_value_bytes16(argv[0])];
539   y = sqlite3_value_int(argv[1]);
540   z = sqlite3_value_int(argv[2]);
541 
542   if( y>0 ){
543     y = y-1;
544     zStart = zStr;
545     if( SQLITE_UTF16BE==SQLITE_UTF16NATIVE ){
546       for(i=0; i<y && zStart<zStrEnd; i++) SKIP_UTF16BE(zStart);
547     }else{
548       for(i=0; i<y && zStart<zStrEnd; i++) SKIP_UTF16LE(zStart);
549     }
550   }else{
551     zStart = zStrEnd;
552     if( SQLITE_UTF16BE==SQLITE_UTF16NATIVE ){
553       for(i=y; i<0 && zStart>zStr; i++) RSKIP_UTF16BE(zStart);
554     }else{
555       for(i=y; i<0 && zStart>zStr; i++) RSKIP_UTF16LE(zStart);
556     }
557     for(; i<0; i++) z -= 1;
558   }
559 
560   zEnd = zStart;
561   if( SQLITE_UTF16BE==SQLITE_UTF16NATIVE ){
562     for(i=0; i<z && zEnd<zStrEnd; i++) SKIP_UTF16BE(zEnd);
563   }else{
564     for(i=0; i<z && zEnd<zStrEnd; i++) SKIP_UTF16LE(zEnd);
565   }
566 
567   sqlite3_result_text16(context, zStart, zEnd-zStart, SQLITE_TRANSIENT);
568 }
569 
570 #if defined(SQLITE_TEST)
571 /*
572 ** This routine is called from the TCL test function "translate_selftest".
573 ** It checks that the primitives for serializing and deserializing
574 ** characters in each encoding are inverses of each other.
575 */
576 void sqlite3utfSelfTest(){
577   unsigned int i, t;
578   unsigned char zBuf[20];
579   unsigned char *z;
580   int n;
581   unsigned int c;
582 
583   for(i=0; i<0x00110000; i++){
584     z = zBuf;
585     WRITE_UTF8(z, i);
586     n = z-zBuf;
587     z = zBuf;
588     READ_UTF8(z, c);
589     t = i;
590     if( i>=0xD800 && i<=0xDFFF ) t = 0xFFFD;
591     if( (i&0xFFFFFFFE)==0xFFFE ) t = 0xFFFD;
592     assert( c==t );
593     assert( (z-zBuf)==n );
594   }
595   for(i=0; i<0x00110000; i++){
596     if( i>=0xD800 && i<=0xE000 ) continue;
597     z = zBuf;
598     WRITE_UTF16LE(z, i);
599     n = z-zBuf;
600     z = zBuf;
601     READ_UTF16LE(z, c);
602     assert( c==i );
603     assert( (z-zBuf)==n );
604   }
605   for(i=0; i<0x00110000; i++){
606     if( i>=0xD800 && i<=0xE000 ) continue;
607     z = zBuf;
608     WRITE_UTF16BE(z, i);
609     n = z-zBuf;
610     z = zBuf;
611     READ_UTF16BE(z, c);
612     assert( c==i );
613     assert( (z-zBuf)==n );
614   }
615 }
616 #endif /* SQLITE_TEST */
617 #endif /* SQLITE_OMIT_UTF16 */
618