1 // © 2016 and later: Unicode, Inc. and others.
2 // License & terms of use: http://www.unicode.org/copyright.html
3 /*
4 *******************************************************************************
5 *
6 * Copyright (C) 2000-2016, International Business Machines
7 * Corporation and others. All Rights Reserved.
8 *
9 *******************************************************************************
10 * file name: genmbcs.cpp
11 * encoding: UTF-8
12 * tab size: 8 (not used)
13 * indentation:4
14 *
15 * created on: 2000jul06
16 * created by: Markus W. Scherer
17 */
18
19 #include <stdio.h>
20 #include "unicode/utypes.h"
21 #include "cstring.h"
22 #include "cmemory.h"
23 #include "unewdata.h"
24 #include "ucnv_cnv.h"
25 #include "ucnvmbcs.h"
26 #include "ucm.h"
27 #include "makeconv.h"
28 #include "genmbcs.h"
29
30 /*
31 * TODO: Split this file into toUnicode, SBCSFromUnicode and MBCSFromUnicode files.
32 * Reduce tests for maxCharLength.
33 */
34
35 struct MBCSData {
36 NewConverter newConverter;
37
38 UCMFile *ucm;
39
40 /* toUnicode (state table in ucm->states) */
41 _MBCSToUFallback toUFallbacks[MBCS_MAX_FALLBACK_COUNT];
42 int32_t countToUFallbacks;
43 uint16_t *unicodeCodeUnits;
44
45 /* fromUnicode */
46 uint16_t stage1[MBCS_STAGE_1_SIZE];
47 uint16_t stage2Single[MBCS_STAGE_2_SIZE]; /* stage 2 for single-byte codepages */
48 uint32_t stage2[MBCS_STAGE_2_SIZE]; /* stage 2 for MBCS */
49 uint8_t *fromUBytes;
50 uint32_t stage2Top, stage3Top;
51
52 /* fromUTF8 */
53 uint16_t stageUTF8[0x10000>>MBCS_UTF8_STAGE_SHIFT]; /* allow for utf8Max=0xffff */
54
55 /*
56 * Maximum UTF-8-friendly code point.
57 * 0 if !utf8Friendly, otherwise 0x01ff..0xffff in steps of 0x100.
58 * If utf8Friendly, utf8Max is normally either MBCS_UTF8_MAX or 0xffff.
59 */
60 uint16_t utf8Max;
61
62 UBool utf8Friendly;
63 UBool omitFromU;
64 };
65
66 /* prototypes */
67 U_CDECL_BEGIN
68 static void
69 MBCSClose(NewConverter *cnvData);
70
71 static UBool
72 MBCSStartMappings(MBCSData *mbcsData);
73
74 static UBool
75 MBCSAddToUnicode(MBCSData *mbcsData,
76 const uint8_t *bytes, int32_t length,
77 UChar32 c,
78 int8_t flag);
79
80 static UBool
81 MBCSIsValid(NewConverter *cnvData,
82 const uint8_t *bytes, int32_t length);
83
84 static UBool
85 MBCSSingleAddFromUnicode(MBCSData *mbcsData,
86 const uint8_t *bytes, int32_t length,
87 UChar32 c,
88 int8_t flag);
89
90 static UBool
91 MBCSAddFromUnicode(MBCSData *mbcsData,
92 const uint8_t *bytes, int32_t length,
93 UChar32 c,
94 int8_t flag);
95
96 static void
97 MBCSPostprocess(MBCSData *mbcsData, const UConverterStaticData *staticData);
98
99 static UBool
100 MBCSAddTable(NewConverter *cnvData, UCMTable *table, UConverterStaticData *staticData);
101
102 static uint32_t
103 MBCSWrite(NewConverter *cnvData, const UConverterStaticData *staticData,
104 UNewDataMemory *pData, int32_t tableType);
105 U_CDECL_END
106
107 /* helper ------------------------------------------------------------------- */
108
109 static inline char
hexDigit(uint8_t digit)110 hexDigit(uint8_t digit) {
111 return digit<=9 ? (char)('0'+digit) : (char)('a'-10+digit);
112 }
113
114 static inline char *
printBytes(char * buffer,const uint8_t * bytes,int32_t length)115 printBytes(char *buffer, const uint8_t *bytes, int32_t length) {
116 char *s=buffer;
117 while(length>0) {
118 *s++=hexDigit((uint8_t)(*bytes>>4));
119 *s++=hexDigit((uint8_t)(*bytes&0xf));
120 ++bytes;
121 --length;
122 }
123
124 *s=0;
125 return buffer;
126 }
127
128 /* implementation ----------------------------------------------------------- */
129
130 static MBCSData gDummy;
131
132
133 U_CFUNC const MBCSData *
MBCSGetDummy()134 MBCSGetDummy() {
135 uprv_memset(&gDummy, 0, sizeof(MBCSData));
136
137 /*
138 * Set "pessimistic" values which may sometimes move too many
139 * mappings to the extension table (but never too few).
140 * These values cause MBCSOkForBaseFromUnicode() to return FALSE for the
141 * largest set of mappings.
142 * Assume maxCharLength>1.
143 */
144 gDummy.utf8Friendly=TRUE;
145 if(SMALL) {
146 gDummy.utf8Max=0xffff;
147 gDummy.omitFromU=TRUE;
148 } else {
149 gDummy.utf8Max=MBCS_UTF8_MAX;
150 }
151 return &gDummy;
152 }
153
154 static void
MBCSInit(MBCSData * mbcsData,UCMFile * ucm)155 MBCSInit(MBCSData *mbcsData, UCMFile *ucm) {
156 uprv_memset(mbcsData, 0, sizeof(MBCSData));
157
158 mbcsData->ucm=ucm; /* aliased, not owned */
159
160 mbcsData->newConverter.close=MBCSClose;
161 mbcsData->newConverter.isValid=MBCSIsValid;
162 mbcsData->newConverter.addTable=MBCSAddTable;
163 mbcsData->newConverter.write=MBCSWrite;
164 }
165
166 U_CFUNC NewConverter *
MBCSOpen(UCMFile * ucm)167 MBCSOpen(UCMFile *ucm) {
168 MBCSData *mbcsData=(MBCSData *)uprv_malloc(sizeof(MBCSData));
169 if(mbcsData==NULL) {
170 printf("out of memory\n");
171 exit(U_MEMORY_ALLOCATION_ERROR);
172 }
173
174 MBCSInit(mbcsData, ucm);
175 return &mbcsData->newConverter;
176 }
177
178 static void
MBCSDestruct(MBCSData * mbcsData)179 MBCSDestruct(MBCSData *mbcsData) {
180 uprv_free(mbcsData->unicodeCodeUnits);
181 uprv_free(mbcsData->fromUBytes);
182 }
183
184 U_CDECL_BEGIN
185 static void
MBCSClose(NewConverter * cnvData)186 MBCSClose(NewConverter *cnvData) {
187 MBCSData *mbcsData=(MBCSData *)cnvData;
188 if(mbcsData!=NULL) {
189 MBCSDestruct(mbcsData);
190 uprv_free(mbcsData);
191 }
192 }
193 U_CDECL_END
194
195 static UBool
MBCSStartMappings(MBCSData * mbcsData)196 MBCSStartMappings(MBCSData *mbcsData) {
197 int32_t i, sum, maxCharLength,
198 stage2NullLength, stage2AllocLength,
199 stage3NullLength, stage3AllocLength;
200
201 /* toUnicode */
202
203 /* allocate the code unit array and prefill it with "unassigned" values */
204 sum=mbcsData->ucm->states.countToUCodeUnits;
205 if(VERBOSE) {
206 printf("the total number of offsets is 0x%lx=%ld\n", (long)sum, (long)sum);
207 }
208
209 if(sum>0) {
210 mbcsData->unicodeCodeUnits=(uint16_t *)uprv_malloc(sum*sizeof(uint16_t));
211 if(mbcsData->unicodeCodeUnits==NULL) {
212 fprintf(stderr, "error: out of memory allocating %ld 16-bit code units\n",
213 (long)sum);
214 return FALSE;
215 }
216 for(i=0; i<sum; ++i) {
217 mbcsData->unicodeCodeUnits[i]=0xfffe;
218 }
219 }
220
221 /* fromUnicode */
222 maxCharLength=mbcsData->ucm->states.maxCharLength;
223
224 /* allocate the codepage mappings and preset the first 16 characters to 0 */
225 if(maxCharLength==1) {
226 /* allocate 64k 16-bit results for single-byte codepages */
227 sum=0x20000;
228 } else {
229 /* allocate 1M * maxCharLength bytes for at most 1M mappings */
230 sum=0x100000*maxCharLength;
231 }
232 mbcsData->fromUBytes=(uint8_t *)uprv_malloc(sum);
233 if(mbcsData->fromUBytes==NULL) {
234 fprintf(stderr, "error: out of memory allocating %ld B for target mappings\n", (long)sum);
235 return FALSE;
236 }
237 uprv_memset(mbcsData->fromUBytes, 0, sum);
238
239 /*
240 * UTF-8-friendly fromUnicode tries: allocate multiple blocks at a time.
241 * See ucnvmbcs.h for details.
242 *
243 * There is code, for example in ucnv_MBCSGetUnicodeSetForUnicode(), which
244 * assumes that the initial stage 2/3 blocks are the all-unassigned ones.
245 * Therefore, we refine the data structure while maintaining this placement
246 * even though it would be convenient to allocate the ASCII block at the
247 * beginning of stage 3, for example.
248 *
249 * UTF-8-friendly fromUnicode tries work from sorted tables and are built
250 * pre-compacted, overlapping adjacent stage 2/3 blocks.
251 * This is necessary because the block allocation and compaction changes
252 * at SBCS_UTF8_MAX or MBCS_UTF8_MAX, and for MBCS tables the additional
253 * stage table uses direct indexes into stage 3, without a multiplier and
254 * thus with a smaller reach.
255 *
256 * Non-UTF-8-friendly fromUnicode tries work from unsorted tables
257 * (because implicit precision is used), and are compacted
258 * in post-processing.
259 *
260 * Preallocation for UTF-8-friendly fromUnicode tries:
261 *
262 * Stage 3:
263 * 64-entry all-unassigned first block followed by ASCII (128 entries).
264 *
265 * Stage 2:
266 * 64-entry all-unassigned first block followed by preallocated
267 * 64-block for ASCII.
268 */
269
270 /* Preallocate ASCII as a linear 128-entry stage 3 block. */
271 stage2NullLength=MBCS_STAGE_2_BLOCK_SIZE;
272 stage2AllocLength=MBCS_STAGE_2_BLOCK_SIZE;
273
274 stage3NullLength=MBCS_UTF8_STAGE_3_BLOCK_SIZE;
275 stage3AllocLength=128; /* ASCII U+0000..U+007f */
276
277 /* Initialize stage 1 for the preallocated blocks. */
278 sum=stage2NullLength;
279 for(i=0; i<(stage2AllocLength>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT); ++i) {
280 mbcsData->stage1[i]=sum;
281 sum+=MBCS_STAGE_2_BLOCK_SIZE;
282 }
283 mbcsData->stage2Top=stage2NullLength+stage2AllocLength; /* ==sum */
284
285 /*
286 * Stage 2 indexes count 16-blocks in stage 3 as follows:
287 * SBCS: directly, indexes increment by 16
288 * MBCS: indexes need to be multiplied by 16*maxCharLength, indexes increment by 1
289 * MBCS UTF-8: directly, indexes increment by 16
290 */
291 if(maxCharLength==1) {
292 sum=stage3NullLength;
293 for(i=0; i<(stage3AllocLength/MBCS_STAGE_3_BLOCK_SIZE); ++i) {
294 mbcsData->stage2Single[mbcsData->stage1[0]+i]=sum;
295 sum+=MBCS_STAGE_3_BLOCK_SIZE;
296 }
297 } else {
298 sum=stage3NullLength/MBCS_STAGE_3_GRANULARITY;
299 for(i=0; i<(stage3AllocLength/MBCS_STAGE_3_BLOCK_SIZE); ++i) {
300 mbcsData->stage2[mbcsData->stage1[0]+i]=sum;
301 sum+=MBCS_STAGE_3_BLOCK_SIZE/MBCS_STAGE_3_GRANULARITY;
302 }
303 }
304
305 sum=stage3NullLength;
306 for(i=0; i<(stage3AllocLength/MBCS_UTF8_STAGE_3_BLOCK_SIZE); ++i) {
307 mbcsData->stageUTF8[i]=sum;
308 sum+=MBCS_UTF8_STAGE_3_BLOCK_SIZE;
309 }
310
311 /*
312 * Allocate a 64-entry all-unassigned first stage 3 block,
313 * for UTF-8-friendly lookup with a trail byte,
314 * plus 128 entries for ASCII.
315 */
316 mbcsData->stage3Top=(stage3NullLength+stage3AllocLength)*maxCharLength; /* ==sum*maxCharLength */
317
318 return TRUE;
319 }
320
321 /* return TRUE for success */
322 static UBool
setFallback(MBCSData * mbcsData,uint32_t offset,UChar32 c)323 setFallback(MBCSData *mbcsData, uint32_t offset, UChar32 c) {
324 int32_t i=ucm_findFallback(mbcsData->toUFallbacks, mbcsData->countToUFallbacks, offset);
325 if(i>=0) {
326 /* if there is already a fallback for this offset, then overwrite it */
327 mbcsData->toUFallbacks[i].codePoint=c;
328 return TRUE;
329 } else {
330 /* if there is no fallback for this offset, then add one */
331 i=mbcsData->countToUFallbacks;
332 if(i>=MBCS_MAX_FALLBACK_COUNT) {
333 fprintf(stderr, "error: too many toUnicode fallbacks, currently at: U+%x\n", (int)c);
334 return FALSE;
335 } else {
336 mbcsData->toUFallbacks[i].offset=offset;
337 mbcsData->toUFallbacks[i].codePoint=c;
338 mbcsData->countToUFallbacks=i+1;
339 return TRUE;
340 }
341 }
342 }
343
344 /* remove fallback if there is one with this offset; return the code point if there was such a fallback, otherwise -1 */
345 static int32_t
removeFallback(MBCSData * mbcsData,uint32_t offset)346 removeFallback(MBCSData *mbcsData, uint32_t offset) {
347 int32_t i=ucm_findFallback(mbcsData->toUFallbacks, mbcsData->countToUFallbacks, offset);
348 if(i>=0) {
349 _MBCSToUFallback *toUFallbacks;
350 int32_t limit, old;
351
352 toUFallbacks=mbcsData->toUFallbacks;
353 limit=mbcsData->countToUFallbacks;
354 old=(int32_t)toUFallbacks[i].codePoint;
355
356 /* copy the last fallback entry here to keep the list contiguous */
357 toUFallbacks[i].offset=toUFallbacks[limit-1].offset;
358 toUFallbacks[i].codePoint=toUFallbacks[limit-1].codePoint;
359 mbcsData->countToUFallbacks=limit-1;
360 return old;
361 } else {
362 return -1;
363 }
364 }
365
366 /*
367 * isFallback is almost a boolean:
368 * 1 (TRUE) this is a fallback mapping
369 * 0 (FALSE) this is a precise mapping
370 * -1 the precision of this mapping is not specified
371 */
372 static UBool
MBCSAddToUnicode(MBCSData * mbcsData,const uint8_t * bytes,int32_t length,UChar32 c,int8_t flag)373 MBCSAddToUnicode(MBCSData *mbcsData,
374 const uint8_t *bytes, int32_t length,
375 UChar32 c,
376 int8_t flag) {
377 char buffer[10];
378 uint32_t offset=0;
379 int32_t i=0, entry, old;
380 uint8_t state=0;
381
382 if(mbcsData->ucm->states.countStates==0) {
383 fprintf(stderr, "error: there is no state information!\n");
384 return FALSE;
385 }
386
387 /* for SI/SO (like EBCDIC-stateful), double-byte sequences start in state 1 */
388 if(length==2 && mbcsData->ucm->states.outputType==MBCS_OUTPUT_2_SISO) {
389 state=1;
390 }
391
392 /*
393 * Walk down the state table like in conversion,
394 * much like getNextUChar().
395 * We assume that c<=0x10ffff.
396 */
397 for(i=0;;) {
398 entry=mbcsData->ucm->states.stateTable[state][bytes[i++]];
399 if(MBCS_ENTRY_IS_TRANSITION(entry)) {
400 if(i==length) {
401 fprintf(stderr, "error: byte sequence too short, ends in non-final state %hu: 0x%s (U+%x)\n",
402 (short)state, printBytes(buffer, bytes, length), (int)c);
403 return FALSE;
404 }
405 state=(uint8_t)MBCS_ENTRY_TRANSITION_STATE(entry);
406 offset+=MBCS_ENTRY_TRANSITION_OFFSET(entry);
407 } else {
408 if(i<length) {
409 fprintf(stderr, "error: byte sequence too long by %d bytes, final state %u: 0x%s (U+%x)\n",
410 (int)(length-i), state, printBytes(buffer, bytes, length), (int)c);
411 return FALSE;
412 }
413 switch(MBCS_ENTRY_FINAL_ACTION(entry)) {
414 case MBCS_STATE_ILLEGAL:
415 fprintf(stderr, "error: byte sequence ends in illegal state at U+%04x<->0x%s\n",
416 (int)c, printBytes(buffer, bytes, length));
417 return FALSE;
418 case MBCS_STATE_CHANGE_ONLY:
419 fprintf(stderr, "error: byte sequence ends in state-change-only at U+%04x<->0x%s\n",
420 (int)c, printBytes(buffer, bytes, length));
421 return FALSE;
422 case MBCS_STATE_UNASSIGNED:
423 fprintf(stderr, "error: byte sequence ends in unassigned state at U+%04x<->0x%s\n",
424 (int)c, printBytes(buffer, bytes, length));
425 return FALSE;
426 case MBCS_STATE_FALLBACK_DIRECT_16:
427 case MBCS_STATE_VALID_DIRECT_16:
428 case MBCS_STATE_FALLBACK_DIRECT_20:
429 case MBCS_STATE_VALID_DIRECT_20:
430 if(MBCS_ENTRY_SET_STATE(entry, 0)!=MBCS_ENTRY_FINAL(0, MBCS_STATE_VALID_DIRECT_16, 0xfffe)) {
431 /* the "direct" action's value is not "valid-direct-16-unassigned" any more */
432 if(MBCS_ENTRY_FINAL_ACTION(entry)==MBCS_STATE_VALID_DIRECT_16 || MBCS_ENTRY_FINAL_ACTION(entry)==MBCS_STATE_FALLBACK_DIRECT_16) {
433 old=MBCS_ENTRY_FINAL_VALUE(entry);
434 } else {
435 old=0x10000+MBCS_ENTRY_FINAL_VALUE(entry);
436 }
437 if(flag>=0) {
438 fprintf(stderr, "error: duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
439 (int)c, printBytes(buffer, bytes, length), (int)old);
440 return FALSE;
441 } else if(VERBOSE) {
442 fprintf(stderr, "duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
443 (int)c, printBytes(buffer, bytes, length), (int)old);
444 }
445 /*
446 * Continue after the above warning
447 * if the precision of the mapping is unspecified.
448 */
449 }
450 /* reassign the correct action code */
451 entry=MBCS_ENTRY_FINAL_SET_ACTION(entry, (MBCS_STATE_VALID_DIRECT_16+(flag==3 ? 2 : 0)+(c>=0x10000 ? 1 : 0)));
452
453 /* put the code point into bits 22..7 for BMP, c-0x10000 into 26..7 for others */
454 if(c<=0xffff) {
455 entry=MBCS_ENTRY_FINAL_SET_VALUE(entry, c);
456 } else {
457 entry=MBCS_ENTRY_FINAL_SET_VALUE(entry, c-0x10000);
458 }
459 mbcsData->ucm->states.stateTable[state][bytes[i-1]]=entry;
460 break;
461 case MBCS_STATE_VALID_16:
462 /* bits 26..16 are not used, 0 */
463 /* bits 15..7 contain the final offset delta to one 16-bit code unit */
464 offset+=MBCS_ENTRY_FINAL_VALUE_16(entry);
465 /* check that this byte sequence is still unassigned */
466 if((old=mbcsData->unicodeCodeUnits[offset])!=0xfffe || (old=removeFallback(mbcsData, offset))!=-1) {
467 if(flag>=0) {
468 fprintf(stderr, "error: duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
469 (int)c, printBytes(buffer, bytes, length), (int)old);
470 return FALSE;
471 } else if(VERBOSE) {
472 fprintf(stderr, "duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
473 (int)c, printBytes(buffer, bytes, length), (int)old);
474 }
475 }
476 if(c>=0x10000) {
477 fprintf(stderr, "error: code point does not fit into valid-16-bit state at U+%04x<->0x%s\n",
478 (int)c, printBytes(buffer, bytes, length));
479 return FALSE;
480 }
481 if(flag>0) {
482 /* assign only if there is no precise mapping */
483 if(mbcsData->unicodeCodeUnits[offset]==0xfffe) {
484 return setFallback(mbcsData, offset, c);
485 }
486 } else {
487 mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
488 }
489 break;
490 case MBCS_STATE_VALID_16_PAIR:
491 /* bits 26..16 are not used, 0 */
492 /* bits 15..7 contain the final offset delta to two 16-bit code units */
493 offset+=MBCS_ENTRY_FINAL_VALUE_16(entry);
494 /* check that this byte sequence is still unassigned */
495 old=mbcsData->unicodeCodeUnits[offset];
496 if(old<0xfffe) {
497 int32_t real;
498 if(old<0xd800) {
499 real=old;
500 } else if(old<=0xdfff) {
501 real=0x10000+((old&0x3ff)<<10)+((mbcsData->unicodeCodeUnits[offset+1])&0x3ff);
502 } else /* old<=0xe001 */ {
503 real=mbcsData->unicodeCodeUnits[offset+1];
504 }
505 if(flag>=0) {
506 fprintf(stderr, "error: duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
507 (int)c, printBytes(buffer, bytes, length), (int)real);
508 return FALSE;
509 } else if(VERBOSE) {
510 fprintf(stderr, "duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
511 (int)c, printBytes(buffer, bytes, length), (int)real);
512 }
513 }
514 if(flag>0) {
515 /* assign only if there is no precise mapping */
516 if(old<=0xdbff || old==0xe000) {
517 /* do nothing */
518 } else if(c<=0xffff) {
519 /* set a BMP fallback code point as a pair with 0xe001 */
520 mbcsData->unicodeCodeUnits[offset++]=0xe001;
521 mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
522 } else {
523 /* set a fallback surrogate pair with two second surrogates */
524 mbcsData->unicodeCodeUnits[offset++]=(uint16_t)(0xdbc0+(c>>10));
525 mbcsData->unicodeCodeUnits[offset]=(uint16_t)(0xdc00+(c&0x3ff));
526 }
527 } else {
528 if(c<0xd800) {
529 /* set a BMP code point */
530 mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
531 } else if(c<=0xffff) {
532 /* set a BMP code point above 0xd800 as a pair with 0xe000 */
533 mbcsData->unicodeCodeUnits[offset++]=0xe000;
534 mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
535 } else {
536 /* set a surrogate pair */
537 mbcsData->unicodeCodeUnits[offset++]=(uint16_t)(0xd7c0+(c>>10));
538 mbcsData->unicodeCodeUnits[offset]=(uint16_t)(0xdc00+(c&0x3ff));
539 }
540 }
541 break;
542 default:
543 /* reserved, must never occur */
544 fprintf(stderr, "internal error: byte sequence reached reserved action code, entry 0x%02x: 0x%s (U+%x)\n",
545 (int)entry, printBytes(buffer, bytes, length), (int)c);
546 return FALSE;
547 }
548
549 return TRUE;
550 }
551 }
552 }
553
554 U_CDECL_BEGIN
555 /* is this byte sequence valid? (this is almost the same as MBCSAddToUnicode()) */
556 static UBool
MBCSIsValid(NewConverter * cnvData,const uint8_t * bytes,int32_t length)557 MBCSIsValid(NewConverter *cnvData,
558 const uint8_t *bytes, int32_t length) {
559 MBCSData *mbcsData=(MBCSData *)cnvData;
560
561 return (UBool)(1==ucm_countChars(&mbcsData->ucm->states, bytes, length));
562 }
563 U_CDECL_END
564 static UBool
MBCSSingleAddFromUnicode(MBCSData * mbcsData,const uint8_t * bytes,int32_t,UChar32 c,int8_t flag)565 MBCSSingleAddFromUnicode(MBCSData *mbcsData,
566 const uint8_t *bytes, int32_t /*length*/,
567 UChar32 c,
568 int8_t flag) {
569 uint16_t *stage3, *p;
570 uint32_t idx;
571 uint16_t old;
572 uint8_t b;
573
574 uint32_t blockSize, newTop, i, nextOffset, newBlock, min;
575
576 /* ignore |2 SUB mappings */
577 if(flag==2) {
578 return TRUE;
579 }
580
581 /*
582 * Walk down the triple-stage compact array ("trie") and
583 * allocate parts as necessary.
584 * Note that the first stage 2 and 3 blocks are reserved for all-unassigned mappings.
585 * We assume that length<=maxCharLength and that c<=0x10ffff.
586 */
587 stage3=(uint16_t *)mbcsData->fromUBytes;
588 b=*bytes;
589
590 /* inspect stage 1 */
591 idx=c>>MBCS_STAGE_1_SHIFT;
592 if(mbcsData->utf8Friendly && c<=SBCS_UTF8_MAX) {
593 nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK&~(MBCS_UTF8_STAGE_3_BLOCKS-1);
594 } else {
595 nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK;
596 }
597 if(mbcsData->stage1[idx]==MBCS_STAGE_2_ALL_UNASSIGNED_INDEX) {
598 /* allocate another block in stage 2 */
599 newBlock=mbcsData->stage2Top;
600 if(mbcsData->utf8Friendly) {
601 min=newBlock-nextOffset; /* minimum block start with overlap */
602 while(min<newBlock && mbcsData->stage2Single[newBlock-1]==0) {
603 --newBlock;
604 }
605 }
606 newTop=newBlock+MBCS_STAGE_2_BLOCK_SIZE;
607
608 if(newTop>MBCS_MAX_STAGE_2_TOP) {
609 fprintf(stderr, "error: too many stage 2 entries at U+%04x<->0x%02x\n", (int)c, b);
610 return FALSE;
611 }
612
613 /*
614 * each stage 2 block contains 64 16-bit words:
615 * 6 code point bits 9..4 with 1 stage 3 index
616 */
617 mbcsData->stage1[idx]=(uint16_t)newBlock;
618 mbcsData->stage2Top=newTop;
619 }
620
621 /* inspect stage 2 */
622 idx=mbcsData->stage1[idx]+nextOffset;
623 if(mbcsData->utf8Friendly && c<=SBCS_UTF8_MAX) {
624 /* allocate 64-entry blocks for UTF-8-friendly lookup */
625 blockSize=MBCS_UTF8_STAGE_3_BLOCK_SIZE;
626 nextOffset=c&MBCS_UTF8_STAGE_3_BLOCK_MASK;
627 } else {
628 blockSize=MBCS_STAGE_3_BLOCK_SIZE;
629 nextOffset=c&MBCS_STAGE_3_BLOCK_MASK;
630 }
631 if(mbcsData->stage2Single[idx]==0) {
632 /* allocate another block in stage 3 */
633 newBlock=mbcsData->stage3Top;
634 if(mbcsData->utf8Friendly) {
635 min=newBlock-nextOffset; /* minimum block start with overlap */
636 while(min<newBlock && stage3[newBlock-1]==0) {
637 --newBlock;
638 }
639 }
640 newTop=newBlock+blockSize;
641
642 if(newTop>MBCS_STAGE_3_SBCS_SIZE) {
643 fprintf(stderr, "error: too many code points at U+%04x<->0x%02x\n", (int)c, b);
644 return FALSE;
645 }
646 /* each block has 16 uint16_t entries */
647 i=idx;
648 while(newBlock<newTop) {
649 mbcsData->stage2Single[i++]=(uint16_t)newBlock;
650 newBlock+=MBCS_STAGE_3_BLOCK_SIZE;
651 }
652 mbcsData->stage3Top=newTop; /* ==newBlock */
653 }
654
655 /* write the codepage entry into stage 3 and get the previous entry */
656 p=stage3+mbcsData->stage2Single[idx]+nextOffset;
657 old=*p;
658 if(flag<=0) {
659 *p=(uint16_t)(0xf00|b);
660 } else if(IS_PRIVATE_USE(c)) {
661 *p=(uint16_t)(0xc00|b);
662 } else {
663 *p=(uint16_t)(0x800|b);
664 }
665
666 /* check that this Unicode code point was still unassigned */
667 if(old>=0x100) {
668 if(flag>=0) {
669 fprintf(stderr, "error: duplicate Unicode code point at U+%04x<->0x%02x see 0x%02x\n",
670 (int)c, b, old&0xff);
671 return FALSE;
672 } else if(VERBOSE) {
673 fprintf(stderr, "duplicate Unicode code point at U+%04x<->0x%02x see 0x%02x\n",
674 (int)c, b, old&0xff);
675 }
676 /* continue after the above warning if the precision of the mapping is unspecified */
677 }
678
679 return TRUE;
680 }
681
682 static UBool
MBCSAddFromUnicode(MBCSData * mbcsData,const uint8_t * bytes,int32_t length,UChar32 c,int8_t flag)683 MBCSAddFromUnicode(MBCSData *mbcsData,
684 const uint8_t *bytes, int32_t length,
685 UChar32 c,
686 int8_t flag) {
687 char buffer[10];
688 const uint8_t *pb;
689 uint8_t *stage3, *p;
690 uint32_t idx, b, old, stage3Index;
691 int32_t maxCharLength;
692
693 uint32_t blockSize, newTop, i, nextOffset, newBlock, min, overlap, maxOverlap;
694
695 maxCharLength=mbcsData->ucm->states.maxCharLength;
696
697 if( mbcsData->ucm->states.outputType==MBCS_OUTPUT_2_SISO &&
698 (!IGNORE_SISO_CHECK && (*bytes==0xe || *bytes==0xf))
699 ) {
700 fprintf(stderr, "error: illegal mapping to SI or SO for SI/SO codepage: U+%04x<->0x%s\n",
701 (int)c, printBytes(buffer, bytes, length));
702 return FALSE;
703 }
704
705 if(flag==1 && length==1 && *bytes==0) {
706 fprintf(stderr, "error: unable to encode a |1 fallback from U+%04x to 0x%02x\n",
707 (int)c, *bytes);
708 return FALSE;
709 }
710
711 /*
712 * Walk down the triple-stage compact array ("trie") and
713 * allocate parts as necessary.
714 * Note that the first stage 2 and 3 blocks are reserved for
715 * all-unassigned mappings.
716 * We assume that length<=maxCharLength and that c<=0x10ffff.
717 */
718 stage3=mbcsData->fromUBytes;
719
720 /* inspect stage 1 */
721 idx=c>>MBCS_STAGE_1_SHIFT;
722 if(mbcsData->utf8Friendly && c<=mbcsData->utf8Max) {
723 nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK&~(MBCS_UTF8_STAGE_3_BLOCKS-1);
724 } else {
725 nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK;
726 }
727 if(mbcsData->stage1[idx]==MBCS_STAGE_2_ALL_UNASSIGNED_INDEX) {
728 /* allocate another block in stage 2 */
729 newBlock=mbcsData->stage2Top;
730 if(mbcsData->utf8Friendly) {
731 min=newBlock-nextOffset; /* minimum block start with overlap */
732 while(min<newBlock && mbcsData->stage2[newBlock-1]==0) {
733 --newBlock;
734 }
735 }
736 newTop=newBlock+MBCS_STAGE_2_BLOCK_SIZE;
737
738 if(newTop>MBCS_MAX_STAGE_2_TOP) {
739 fprintf(stderr, "error: too many stage 2 entries at U+%04x<->0x%s\n",
740 (int)c, printBytes(buffer, bytes, length));
741 return FALSE;
742 }
743
744 /*
745 * each stage 2 block contains 64 32-bit words:
746 * 6 code point bits 9..4 with value with bits 31..16 "assigned" flags and bits 15..0 stage 3 index
747 */
748 i=idx;
749 while(newBlock<newTop) {
750 mbcsData->stage1[i++]=(uint16_t)newBlock;
751 newBlock+=MBCS_STAGE_2_BLOCK_SIZE;
752 }
753 mbcsData->stage2Top=newTop; /* ==newBlock */
754 }
755
756 /* inspect stage 2 */
757 idx=mbcsData->stage1[idx]+nextOffset;
758 if(mbcsData->utf8Friendly && c<=mbcsData->utf8Max) {
759 /* allocate 64-entry blocks for UTF-8-friendly lookup */
760 blockSize=MBCS_UTF8_STAGE_3_BLOCK_SIZE*maxCharLength;
761 nextOffset=c&MBCS_UTF8_STAGE_3_BLOCK_MASK;
762 } else {
763 blockSize=MBCS_STAGE_3_BLOCK_SIZE*maxCharLength;
764 nextOffset=c&MBCS_STAGE_3_BLOCK_MASK;
765 }
766 if(mbcsData->stage2[idx]==0) {
767 /* allocate another block in stage 3 */
768 newBlock=mbcsData->stage3Top;
769 if(mbcsData->utf8Friendly && nextOffset>=MBCS_STAGE_3_GRANULARITY) {
770 /*
771 * Overlap stage 3 blocks only in multiples of 16-entry blocks
772 * because of the indexing granularity in stage 2.
773 */
774 maxOverlap=(nextOffset&~(MBCS_STAGE_3_GRANULARITY-1))*maxCharLength;
775 for(overlap=0;
776 overlap<maxOverlap && stage3[newBlock-overlap-1]==0;
777 ++overlap) {}
778
779 overlap=(overlap/MBCS_STAGE_3_GRANULARITY)/maxCharLength;
780 overlap=(overlap*MBCS_STAGE_3_GRANULARITY)*maxCharLength;
781
782 newBlock-=overlap;
783 }
784 newTop=newBlock+blockSize;
785
786 if(newTop>MBCS_STAGE_3_MBCS_SIZE*(uint32_t)maxCharLength) {
787 fprintf(stderr, "error: too many code points at U+%04x<->0x%s\n",
788 (int)c, printBytes(buffer, bytes, length));
789 return FALSE;
790 }
791 /* each block has 16*maxCharLength bytes */
792 i=idx;
793 while(newBlock<newTop) {
794 mbcsData->stage2[i++]=(newBlock/MBCS_STAGE_3_GRANULARITY)/maxCharLength;
795 newBlock+=MBCS_STAGE_3_BLOCK_SIZE*maxCharLength;
796 }
797 mbcsData->stage3Top=newTop; /* ==newBlock */
798 }
799
800 stage3Index=MBCS_STAGE_3_GRANULARITY*(uint32_t)(uint16_t)mbcsData->stage2[idx];
801
802 /* Build an alternate, UTF-8-friendly stage table as well. */
803 if(mbcsData->utf8Friendly && c<=mbcsData->utf8Max) {
804 /* Overflow for uint16_t entries in stageUTF8? */
805 if(stage3Index>0xffff) {
806 /*
807 * This can occur only if the mapping table is nearly perfectly filled and if
808 * utf8Max==0xffff.
809 * (There is no known charset like this. GB 18030 does not map
810 * surrogate code points and LMBCS does not map 256 PUA code points.)
811 *
812 * Otherwise, stage3Index<=MBCS_UTF8_LIMIT<0xffff
813 * (stage3Index can at most reach exactly MBCS_UTF8_LIMIT)
814 * because we have a sorted table and there are at most MBCS_UTF8_LIMIT
815 * mappings with 0<=c<MBCS_UTF8_LIMIT, and there is only also
816 * the initial all-unassigned block in stage3.
817 *
818 * Solution for the overflow: Reduce utf8Max to the next lower value, 0xfeff.
819 *
820 * (See svn revision 20866 of the markus/ucnvutf8 feature branch for
821 * code that causes MBCSAddTable() to rebuild the table not utf8Friendly
822 * in case of overflow. That code was not tested.)
823 */
824 mbcsData->utf8Max=0xfeff;
825 } else {
826 /*
827 * The stage 3 block has been assigned for the regular trie.
828 * Just copy its index into stageUTF8[], without the granularity.
829 */
830 mbcsData->stageUTF8[c>>MBCS_UTF8_STAGE_SHIFT]=(uint16_t)stage3Index;
831 }
832 }
833
834 /* write the codepage bytes into stage 3 and get the previous bytes */
835
836 /* assemble the bytes into a single integer */
837 pb=bytes;
838 b=0;
839 switch(length) {
840 case 4:
841 b=*pb++;
842 U_FALLTHROUGH;
843 case 3:
844 b=(b<<8)|*pb++;
845 U_FALLTHROUGH;
846 case 2:
847 b=(b<<8)|*pb++;
848 U_FALLTHROUGH;
849 case 1:
850 default:
851 b=(b<<8)|*pb++;
852 break;
853 }
854
855 old=0;
856 p=stage3+(stage3Index+nextOffset)*maxCharLength;
857 switch(maxCharLength) {
858 case 2:
859 old=*(uint16_t *)p;
860 *(uint16_t *)p=(uint16_t)b;
861 break;
862 case 3:
863 old=(uint32_t)*p<<16;
864 *p++=(uint8_t)(b>>16);
865 old|=(uint32_t)*p<<8;
866 *p++=(uint8_t)(b>>8);
867 old|=*p;
868 *p=(uint8_t)b;
869 break;
870 case 4:
871 old=*(uint32_t *)p;
872 *(uint32_t *)p=b;
873 break;
874 default:
875 /* will never occur */
876 break;
877 }
878
879 /* check that this Unicode code point was still unassigned */
880 if((mbcsData->stage2[idx+(nextOffset>>MBCS_STAGE_2_SHIFT)]&(1UL<<(16+(c&0xf))))!=0 || old!=0) {
881 if(flag>=0) {
882 fprintf(stderr, "error: duplicate Unicode code point at U+%04x<->0x%s see 0x%02x\n",
883 (int)c, printBytes(buffer, bytes, length), (int)old);
884 return FALSE;
885 } else if(VERBOSE) {
886 fprintf(stderr, "duplicate Unicode code point at U+%04x<->0x%s see 0x%02x\n",
887 (int)c, printBytes(buffer, bytes, length), (int)old);
888 }
889 /* continue after the above warning if the precision of the mapping is
890 unspecified */
891 }
892 if(flag<=0) {
893 /* set the roundtrip flag */
894 mbcsData->stage2[idx+(nextOffset>>4)]|=(1UL<<(16+(c&0xf)));
895 }
896
897 return TRUE;
898 }
899
900 U_CFUNC UBool
MBCSOkForBaseFromUnicode(const MBCSData * mbcsData,const uint8_t * bytes,int32_t length,UChar32 c,int8_t flag)901 MBCSOkForBaseFromUnicode(const MBCSData *mbcsData,
902 const uint8_t *bytes, int32_t length,
903 UChar32 c, int8_t flag) {
904 /*
905 * A 1:1 mapping does not fit into the MBCS base table's fromUnicode table under
906 * the following conditions:
907 *
908 * - a |2 SUB mapping for <subchar1> (no base table data structure for them)
909 * - a |1 fallback to 0x00 (result value 0, indistinguishable from unmappable entry)
910 * - a multi-byte mapping with leading 0x00 bytes (no explicit length field)
911 *
912 * Some of these tests are redundant with ucm_mappingType().
913 */
914 if( (flag==2 && length==1) ||
915 (flag==1 && bytes[0]==0) || /* testing length==1 would be redundant with the next test */
916 (flag<=1 && length>1 && bytes[0]==0)
917 ) {
918 return FALSE;
919 }
920
921 /*
922 * Additional restrictions for UTF-8-friendly fromUnicode tables,
923 * for code points up to the maximum optimized one:
924 *
925 * - any mapping to 0x00 (result value 0, indistinguishable from unmappable entry)
926 * - any |1 fallback (no roundtrip flags in the optimized table)
927 */
928 if(mbcsData->utf8Friendly && flag<=1 && c<=mbcsData->utf8Max && (bytes[0]==0 || flag==1)) {
929 return FALSE;
930 }
931
932 /*
933 * If we omit the fromUnicode data, we can only store roundtrips there
934 * because only they are recoverable from the toUnicode data.
935 * Fallbacks must go into the extension table.
936 */
937 if(mbcsData->omitFromU && flag!=0) {
938 return FALSE;
939 }
940
941 /* All other mappings do fit into the base table. */
942 return TRUE;
943 }
944
945 U_CDECL_BEGIN
946 /* we can assume that the table only contains 1:1 mappings with <=4 bytes each */
947 static UBool
MBCSAddTable(NewConverter * cnvData,UCMTable * table,UConverterStaticData * staticData)948 MBCSAddTable(NewConverter *cnvData, UCMTable *table, UConverterStaticData *staticData) {
949 MBCSData *mbcsData;
950 UCMapping *m;
951 UChar32 c;
952 int32_t i, maxCharLength;
953 int8_t f;
954 UBool isOK, utf8Friendly;
955
956 staticData->unicodeMask=table->unicodeMask;
957 if(staticData->unicodeMask==3) {
958 fprintf(stderr, "error: contains mappings for both supplementary and surrogate code points\n");
959 return FALSE;
960 }
961
962 staticData->conversionType=UCNV_MBCS;
963
964 mbcsData=(MBCSData *)cnvData;
965 maxCharLength=mbcsData->ucm->states.maxCharLength;
966
967 /*
968 * Generation of UTF-8-friendly data requires
969 * a sorted table, which makeconv generates when explicit precision
970 * indicators are used.
971 */
972 mbcsData->utf8Friendly=utf8Friendly=(UBool)((table->flagsType&UCM_FLAGS_EXPLICIT)!=0);
973 if(utf8Friendly) {
974 mbcsData->utf8Max=MBCS_UTF8_MAX;
975 if(SMALL && maxCharLength>1) {
976 mbcsData->omitFromU=TRUE;
977 }
978 } else {
979 mbcsData->utf8Max=0;
980 if(SMALL && maxCharLength>1) {
981 fprintf(stderr,
982 "makeconv warning: --small not available for .ucm files without |0 etc.\n");
983 }
984 }
985
986 if(!MBCSStartMappings(mbcsData)) {
987 return FALSE;
988 }
989
990 staticData->hasFromUnicodeFallback=FALSE;
991 staticData->hasToUnicodeFallback=FALSE;
992
993 isOK=TRUE;
994
995 m=table->mappings;
996 for(i=0; i<table->mappingsLength; ++m, ++i) {
997 c=m->u;
998 f=m->f;
999
1000 /*
1001 * Small optimization for --small .cnv files:
1002 *
1003 * If there are fromUnicode mappings above MBCS_UTF8_MAX,
1004 * then the file size will be smaller if we make utf8Max larger
1005 * because the size increase in stageUTF8 will be more than balanced by
1006 * how much less of stage2 needs to be stored.
1007 *
1008 * There is no point in doing this incrementally because stageUTF8
1009 * uses so much less space per block than stage2,
1010 * so we immediately increase utf8Max to 0xffff.
1011 *
1012 * Do not increase utf8Max if it is already at 0xfeff because MBCSAddFromUnicode()
1013 * sets it to that value when stageUTF8 overflows.
1014 */
1015 if( mbcsData->omitFromU && f<=1 &&
1016 mbcsData->utf8Max<c && c<=0xffff &&
1017 mbcsData->utf8Max<0xfeff
1018 ) {
1019 mbcsData->utf8Max=0xffff;
1020 }
1021
1022 switch(f) {
1023 case -1:
1024 /* there was no precision/fallback indicator */
1025 /* fall through to set the mappings */
1026 U_FALLTHROUGH;
1027 case 0:
1028 /* set roundtrip mappings */
1029 isOK&=MBCSAddToUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1030
1031 if(maxCharLength==1) {
1032 isOK&=MBCSSingleAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1033 } else if(MBCSOkForBaseFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f)) {
1034 isOK&=MBCSAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1035 } else {
1036 m->f|=MBCS_FROM_U_EXT_FLAG;
1037 m->moveFlag=UCM_MOVE_TO_EXT;
1038 }
1039 break;
1040 case 1:
1041 /* set only a fallback mapping from Unicode to codepage */
1042 if(maxCharLength==1) {
1043 staticData->hasFromUnicodeFallback=TRUE;
1044 isOK&=MBCSSingleAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1045 } else if(MBCSOkForBaseFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f)) {
1046 staticData->hasFromUnicodeFallback=TRUE;
1047 isOK&=MBCSAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1048 } else {
1049 m->f|=MBCS_FROM_U_EXT_FLAG;
1050 m->moveFlag=UCM_MOVE_TO_EXT;
1051 }
1052 break;
1053 case 2:
1054 /* ignore |2 SUB mappings, except to move <subchar1> mappings to the extension table */
1055 if(maxCharLength>1 && m->bLen==1) {
1056 m->f|=MBCS_FROM_U_EXT_FLAG;
1057 m->moveFlag=UCM_MOVE_TO_EXT;
1058 }
1059 break;
1060 case 3:
1061 /* set only a fallback mapping from codepage to Unicode */
1062 staticData->hasToUnicodeFallback=TRUE;
1063 isOK&=MBCSAddToUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1064 break;
1065 case 4:
1066 /* move "good one-way" mappings to the extension table */
1067 m->f|=MBCS_FROM_U_EXT_FLAG;
1068 m->moveFlag=UCM_MOVE_TO_EXT;
1069 break;
1070 default:
1071 /* will not occur because the parser checked it already */
1072 fprintf(stderr, "error: illegal fallback indicator %d\n", f);
1073 return FALSE;
1074 }
1075 }
1076
1077 MBCSPostprocess(mbcsData, staticData);
1078
1079 return isOK;
1080 }
1081 U_CDECL_END
1082 static UBool
transformEUC(MBCSData * mbcsData)1083 transformEUC(MBCSData *mbcsData) {
1084 uint8_t *p8;
1085 uint32_t i, value, oldLength, old3Top;
1086 uint8_t b;
1087
1088 oldLength=mbcsData->ucm->states.maxCharLength;
1089 if(oldLength<3) {
1090 return FALSE;
1091 }
1092
1093 old3Top=mbcsData->stage3Top;
1094
1095 /* careful: 2-byte and 4-byte codes are stored in platform endianness! */
1096
1097 /* test if all first bytes are in {0, 0x8e, 0x8f} */
1098 p8=mbcsData->fromUBytes;
1099
1100 #if !U_IS_BIG_ENDIAN
1101 if(oldLength==4) {
1102 p8+=3;
1103 }
1104 #endif
1105
1106 for(i=0; i<old3Top; i+=oldLength) {
1107 b=p8[i];
1108 if(b!=0 && b!=0x8e && b!=0x8f) {
1109 /* some first byte does not fit the EUC pattern, nothing to be done */
1110 return FALSE;
1111 }
1112 }
1113 /* restore p if it was modified above */
1114 p8=mbcsData->fromUBytes;
1115
1116 /* modify outputType and adjust stage3Top */
1117 mbcsData->ucm->states.outputType=(int8_t)(MBCS_OUTPUT_3_EUC+oldLength-3);
1118 mbcsData->stage3Top=(old3Top*(oldLength-1))/oldLength;
1119
1120 /*
1121 * EUC-encode all byte sequences;
1122 * see "CJKV Information Processing" (1st ed. 1999) from Ken Lunde, O'Reilly,
1123 * p. 161 in chapter 4 "Encoding Methods"
1124 *
1125 * This also must reverse the byte order if the platform is little-endian!
1126 */
1127 if(oldLength==3) {
1128 uint16_t *q=(uint16_t *)p8;
1129 for(i=0; i<old3Top; i+=oldLength) {
1130 b=*p8;
1131 if(b==0) {
1132 /* short sequences are stored directly */
1133 /* code set 0 or 1 */
1134 (*q++)=(uint16_t)((p8[1]<<8)|p8[2]);
1135 } else if(b==0x8e) {
1136 /* code set 2 */
1137 (*q++)=(uint16_t)(((p8[1]&0x7f)<<8)|p8[2]);
1138 } else /* b==0x8f */ {
1139 /* code set 3 */
1140 (*q++)=(uint16_t)((p8[1]<<8)|(p8[2]&0x7f));
1141 }
1142 p8+=3;
1143 }
1144 } else /* oldLength==4 */ {
1145 uint8_t *q=p8;
1146 uint32_t *p32=(uint32_t *)p8;
1147 for(i=0; i<old3Top; i+=4) {
1148 value=(*p32++);
1149 if(value<=0xffffff) {
1150 /* short sequences are stored directly */
1151 /* code set 0 or 1 */
1152 (*q++)=(uint8_t)(value>>16);
1153 (*q++)=(uint8_t)(value>>8);
1154 (*q++)=(uint8_t)value;
1155 } else if(value<=0x8effffff) {
1156 /* code set 2 */
1157 (*q++)=(uint8_t)((value>>16)&0x7f);
1158 (*q++)=(uint8_t)(value>>8);
1159 (*q++)=(uint8_t)value;
1160 } else /* first byte is 0x8f */ {
1161 /* code set 3 */
1162 (*q++)=(uint8_t)(value>>16);
1163 (*q++)=(uint8_t)((value>>8)&0x7f);
1164 (*q++)=(uint8_t)value;
1165 }
1166 }
1167 }
1168
1169 return TRUE;
1170 }
1171
1172 /*
1173 * Compact stage 2 for SBCS by overlapping adjacent stage 2 blocks as far
1174 * as possible. Overlapping is done on unassigned head and tail
1175 * parts of blocks in steps of MBCS_STAGE_2_MULTIPLIER.
1176 * Stage 1 indexes need to be adjusted accordingly.
1177 * This function is very similar to genprops/store.c/compactStage().
1178 */
1179 static void
singleCompactStage2(MBCSData * mbcsData)1180 singleCompactStage2(MBCSData *mbcsData) {
1181 /* this array maps the ordinal number of a stage 2 block to its new stage 1 index */
1182 uint16_t map[MBCS_STAGE_2_MAX_BLOCKS];
1183 uint16_t i, start, prevEnd, newStart;
1184
1185 /* enter the all-unassigned first stage 2 block into the map */
1186 map[0]=MBCS_STAGE_2_ALL_UNASSIGNED_INDEX;
1187
1188 /* begin with the first block after the all-unassigned one */
1189 start=newStart=MBCS_STAGE_2_FIRST_ASSIGNED;
1190 while(start<mbcsData->stage2Top) {
1191 prevEnd=(uint16_t)(newStart-1);
1192
1193 /* find the size of the overlap */
1194 for(i=0; i<MBCS_STAGE_2_BLOCK_SIZE && mbcsData->stage2Single[start+i]==0 && mbcsData->stage2Single[prevEnd-i]==0; ++i) {}
1195
1196 if(i>0) {
1197 map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=(uint16_t)(newStart-i);
1198
1199 /* move the non-overlapping indexes to their new positions */
1200 start+=i;
1201 for(i=(uint16_t)(MBCS_STAGE_2_BLOCK_SIZE-i); i>0; --i) {
1202 mbcsData->stage2Single[newStart++]=mbcsData->stage2Single[start++];
1203 }
1204 } else if(newStart<start) {
1205 /* move the indexes to their new positions */
1206 map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=newStart;
1207 for(i=MBCS_STAGE_2_BLOCK_SIZE; i>0; --i) {
1208 mbcsData->stage2Single[newStart++]=mbcsData->stage2Single[start++];
1209 }
1210 } else /* no overlap && newStart==start */ {
1211 map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=start;
1212 start=newStart+=MBCS_STAGE_2_BLOCK_SIZE;
1213 }
1214 }
1215
1216 /* adjust stage2Top */
1217 if(VERBOSE && newStart<mbcsData->stage2Top) {
1218 printf("compacting stage 2 from stage2Top=0x%lx to 0x%lx, saving %ld bytes\n",
1219 (unsigned long)mbcsData->stage2Top, (unsigned long)newStart,
1220 (long)(mbcsData->stage2Top-newStart)*2);
1221 }
1222 mbcsData->stage2Top=newStart;
1223
1224 /* now adjust stage 1 */
1225 for(i=0; i<MBCS_STAGE_1_SIZE; ++i) {
1226 mbcsData->stage1[i]=map[mbcsData->stage1[i]>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT];
1227 }
1228 }
1229
1230 /* Compact stage 3 for SBCS - same algorithm as above. */
1231 static void
singleCompactStage3(MBCSData * mbcsData)1232 singleCompactStage3(MBCSData *mbcsData) {
1233 uint16_t *stage3=(uint16_t *)mbcsData->fromUBytes;
1234
1235 /* this array maps the ordinal number of a stage 3 block to its new stage 2 index */
1236 uint16_t map[0x1000];
1237 uint16_t i, start, prevEnd, newStart;
1238
1239 /* enter the all-unassigned first stage 3 block into the map */
1240 map[0]=0;
1241
1242 /* begin with the first block after the all-unassigned one */
1243 start=newStart=16;
1244 while(start<mbcsData->stage3Top) {
1245 prevEnd=(uint16_t)(newStart-1);
1246
1247 /* find the size of the overlap */
1248 for(i=0; i<16 && stage3[start+i]==0 && stage3[prevEnd-i]==0; ++i) {}
1249
1250 if(i>0) {
1251 map[start>>4]=(uint16_t)(newStart-i);
1252
1253 /* move the non-overlapping indexes to their new positions */
1254 start+=i;
1255 for(i=(uint16_t)(16-i); i>0; --i) {
1256 stage3[newStart++]=stage3[start++];
1257 }
1258 } else if(newStart<start) {
1259 /* move the indexes to their new positions */
1260 map[start>>4]=newStart;
1261 for(i=16; i>0; --i) {
1262 stage3[newStart++]=stage3[start++];
1263 }
1264 } else /* no overlap && newStart==start */ {
1265 map[start>>4]=start;
1266 start=newStart+=16;
1267 }
1268 }
1269
1270 /* adjust stage3Top */
1271 if(VERBOSE && newStart<mbcsData->stage3Top) {
1272 printf("compacting stage 3 from stage3Top=0x%lx to 0x%lx, saving %ld bytes\n",
1273 (unsigned long)mbcsData->stage3Top, (unsigned long)newStart,
1274 (long)(mbcsData->stage3Top-newStart)*2);
1275 }
1276 mbcsData->stage3Top=newStart;
1277
1278 /* now adjust stage 2 */
1279 for(i=0; i<mbcsData->stage2Top; ++i) {
1280 mbcsData->stage2Single[i]=map[mbcsData->stage2Single[i]>>4];
1281 }
1282 }
1283
1284 /*
1285 * Compact stage 2 by overlapping adjacent stage 2 blocks as far
1286 * as possible. Overlapping is done on unassigned head and tail
1287 * parts of blocks in steps of MBCS_STAGE_2_MULTIPLIER.
1288 * Stage 1 indexes need to be adjusted accordingly.
1289 * This function is very similar to genprops/store.c/compactStage().
1290 */
1291 static void
compactStage2(MBCSData * mbcsData)1292 compactStage2(MBCSData *mbcsData) {
1293 /* this array maps the ordinal number of a stage 2 block to its new stage 1 index */
1294 uint16_t map[MBCS_STAGE_2_MAX_BLOCKS];
1295 uint16_t i, start, prevEnd, newStart;
1296
1297 /* enter the all-unassigned first stage 2 block into the map */
1298 map[0]=MBCS_STAGE_2_ALL_UNASSIGNED_INDEX;
1299
1300 /* begin with the first block after the all-unassigned one */
1301 start=newStart=MBCS_STAGE_2_FIRST_ASSIGNED;
1302 while(start<mbcsData->stage2Top) {
1303 prevEnd=(uint16_t)(newStart-1);
1304
1305 /* find the size of the overlap */
1306 for(i=0; i<MBCS_STAGE_2_BLOCK_SIZE && mbcsData->stage2[start+i]==0 && mbcsData->stage2[prevEnd-i]==0; ++i) {}
1307
1308 if(i>0) {
1309 map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=(uint16_t)(newStart-i);
1310
1311 /* move the non-overlapping indexes to their new positions */
1312 start+=i;
1313 for(i=(uint16_t)(MBCS_STAGE_2_BLOCK_SIZE-i); i>0; --i) {
1314 mbcsData->stage2[newStart++]=mbcsData->stage2[start++];
1315 }
1316 } else if(newStart<start) {
1317 /* move the indexes to their new positions */
1318 map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=newStart;
1319 for(i=MBCS_STAGE_2_BLOCK_SIZE; i>0; --i) {
1320 mbcsData->stage2[newStart++]=mbcsData->stage2[start++];
1321 }
1322 } else /* no overlap && newStart==start */ {
1323 map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=start;
1324 start=newStart+=MBCS_STAGE_2_BLOCK_SIZE;
1325 }
1326 }
1327
1328 /* adjust stage2Top */
1329 if(VERBOSE && newStart<mbcsData->stage2Top) {
1330 printf("compacting stage 2 from stage2Top=0x%lx to 0x%lx, saving %ld bytes\n",
1331 (unsigned long)mbcsData->stage2Top, (unsigned long)newStart,
1332 (long)(mbcsData->stage2Top-newStart)*4);
1333 }
1334 mbcsData->stage2Top=newStart;
1335
1336 /* now adjust stage 1 */
1337 for(i=0; i<MBCS_STAGE_1_SIZE; ++i) {
1338 mbcsData->stage1[i]=map[mbcsData->stage1[i]>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT];
1339 }
1340 }
1341
1342 static void
MBCSPostprocess(MBCSData * mbcsData,const UConverterStaticData *)1343 MBCSPostprocess(MBCSData *mbcsData, const UConverterStaticData * /*staticData*/) {
1344 UCMStates *states;
1345 int32_t maxCharLength, stage3Width;
1346
1347 states=&mbcsData->ucm->states;
1348 stage3Width=maxCharLength=states->maxCharLength;
1349
1350 ucm_optimizeStates(states,
1351 &mbcsData->unicodeCodeUnits,
1352 mbcsData->toUFallbacks, mbcsData->countToUFallbacks,
1353 VERBOSE);
1354
1355 /* try to compact the fromUnicode tables */
1356 if(transformEUC(mbcsData)) {
1357 --stage3Width;
1358 }
1359
1360 /*
1361 * UTF-8-friendly tries are built precompacted, to cope with variable
1362 * stage 3 allocation block sizes.
1363 *
1364 * Tables without precision indicators cannot be built that way,
1365 * because if a block was overlapped with a previous one, then a smaller
1366 * code point for the same block would not fit.
1367 * Therefore, such tables are not marked UTF-8-friendly and must be
1368 * compacted after all mappings are entered.
1369 */
1370 if(!mbcsData->utf8Friendly) {
1371 if(maxCharLength==1) {
1372 singleCompactStage3(mbcsData);
1373 singleCompactStage2(mbcsData);
1374 } else {
1375 compactStage2(mbcsData);
1376 }
1377 }
1378
1379 if(VERBOSE) {
1380 /*uint32_t c, i1, i2, i2Limit, i3;*/
1381
1382 printf("fromUnicode number of uint%s_t in stage 2: 0x%lx=%lu\n",
1383 maxCharLength==1 ? "16" : "32",
1384 (unsigned long)mbcsData->stage2Top,
1385 (unsigned long)mbcsData->stage2Top);
1386 printf("fromUnicode number of %d-byte stage 3 mapping entries: 0x%lx=%lu\n",
1387 (int)stage3Width,
1388 (unsigned long)mbcsData->stage3Top/stage3Width,
1389 (unsigned long)mbcsData->stage3Top/stage3Width);
1390 #if 0
1391 c=0;
1392 for(i1=0; i1<MBCS_STAGE_1_SIZE; ++i1) {
1393 i2=mbcsData->stage1[i1];
1394 if(i2==0) {
1395 c+=MBCS_STAGE_2_BLOCK_SIZE*MBCS_STAGE_3_BLOCK_SIZE;
1396 continue;
1397 }
1398 for(i2Limit=i2+MBCS_STAGE_2_BLOCK_SIZE; i2<i2Limit; ++i2) {
1399 if(maxCharLength==1) {
1400 i3=mbcsData->stage2Single[i2];
1401 } else {
1402 i3=(uint16_t)mbcsData->stage2[i2];
1403 }
1404 if(i3==0) {
1405 c+=MBCS_STAGE_3_BLOCK_SIZE;
1406 continue;
1407 }
1408 printf("U+%04lx i1=0x%02lx i2=0x%04lx i3=0x%04lx\n",
1409 (unsigned long)c,
1410 (unsigned long)i1,
1411 (unsigned long)i2,
1412 (unsigned long)i3);
1413 c+=MBCS_STAGE_3_BLOCK_SIZE;
1414 }
1415 }
1416 #endif
1417 }
1418 }
1419
1420 U_CDECL_BEGIN
1421 static uint32_t
MBCSWrite(NewConverter * cnvData,const UConverterStaticData * staticData,UNewDataMemory * pData,int32_t tableType)1422 MBCSWrite(NewConverter *cnvData, const UConverterStaticData *staticData,
1423 UNewDataMemory *pData, int32_t tableType) {
1424 MBCSData *mbcsData=(MBCSData *)cnvData;
1425 uint32_t stage2Start, stage2Length;
1426 uint32_t top, stageUTF8Length=0;
1427 int32_t i, stage1Top;
1428 uint32_t headerLength;
1429
1430 _MBCSHeader header=UCNV_MBCS_HEADER_INITIALIZER;
1431
1432 stage2Length=mbcsData->stage2Top;
1433 if(mbcsData->omitFromU) {
1434 /* find how much of stage2 can be omitted */
1435 int32_t utf8Limit=(int32_t)mbcsData->utf8Max+1;
1436 uint32_t st2=0; /*initialized it to avoid compiler warnings */
1437
1438 i=utf8Limit>>MBCS_STAGE_1_SHIFT;
1439 if((utf8Limit&((1<<MBCS_STAGE_1_SHIFT)-1))!=0 && (st2=mbcsData->stage1[i])!=0) {
1440 /* utf8Limit is in the middle of an existing stage 2 block */
1441 stage2Start=st2+((utf8Limit>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK);
1442 } else {
1443 /* find the last stage2 block with mappings before utf8Limit */
1444 while(i>0 && (st2=mbcsData->stage1[--i])==0) {}
1445 /* stage2 up to the end of this block corresponds to stageUTF8 */
1446 stage2Start=st2+MBCS_STAGE_2_BLOCK_SIZE;
1447 }
1448 header.options|=MBCS_OPT_NO_FROM_U;
1449 header.fullStage2Length=stage2Length;
1450 stage2Length-=stage2Start;
1451 if(VERBOSE) {
1452 printf("+ omitting %lu out of %lu stage2 entries and %lu fromUBytes\n",
1453 (unsigned long)stage2Start,
1454 (unsigned long)mbcsData->stage2Top,
1455 (unsigned long)mbcsData->stage3Top);
1456 printf("+ total size savings: %lu bytes\n", (unsigned long)stage2Start*4+mbcsData->stage3Top);
1457 }
1458 } else {
1459 stage2Start=0;
1460 }
1461
1462 if(staticData->unicodeMask&UCNV_HAS_SUPPLEMENTARY) {
1463 stage1Top=MBCS_STAGE_1_SIZE; /* 0x440==1088 */
1464 } else {
1465 stage1Top=0x40; /* 0x40==64 */
1466 }
1467
1468 /* adjust stage 1 entries to include the size of stage 1 in the offsets to stage 2 */
1469 if(mbcsData->ucm->states.maxCharLength==1) {
1470 for(i=0; i<stage1Top; ++i) {
1471 mbcsData->stage1[i]+=(uint16_t)stage1Top;
1472 }
1473
1474 /* stage2Top/Length have counted 16-bit results, now we need to count bytes */
1475 /* also round up to a multiple of 4 bytes */
1476 stage2Length=(stage2Length*2+1)&~1;
1477
1478 /* stage3Top has counted 16-bit results, now we need to count bytes */
1479 mbcsData->stage3Top*=2;
1480
1481 if(mbcsData->utf8Friendly) {
1482 header.version[2]=(uint8_t)(SBCS_UTF8_MAX>>8); /* store 0x1f for max==0x1fff */
1483 }
1484 } else {
1485 for(i=0; i<stage1Top; ++i) {
1486 mbcsData->stage1[i]+=(uint16_t)stage1Top/2; /* stage 2 contains 32-bit entries, stage 1 16-bit entries */
1487 }
1488
1489 /* stage2Top/Length have counted 32-bit results, now we need to count bytes */
1490 stage2Length*=4;
1491 /* leave stage2Start counting 32-bit units */
1492
1493 if(mbcsData->utf8Friendly) {
1494 stageUTF8Length=(mbcsData->utf8Max+1)>>MBCS_UTF8_STAGE_SHIFT;
1495 header.version[2]=(uint8_t)(mbcsData->utf8Max>>8); /* store 0xd7 for max==0xd7ff */
1496 }
1497
1498 /* stage3Top has already counted bytes */
1499 }
1500
1501 /* round up stage3Top so that the sizes of all data blocks are multiples of 4 */
1502 mbcsData->stage3Top=(mbcsData->stage3Top+3)&~3;
1503
1504 /* fill the header */
1505 if(header.options&MBCS_OPT_INCOMPATIBLE_MASK) {
1506 header.version[0]=5;
1507 if(header.options&MBCS_OPT_NO_FROM_U) {
1508 headerLength=10; /* include fullStage2Length */
1509 } else {
1510 headerLength=MBCS_HEADER_V5_MIN_LENGTH; /* 9 */
1511 }
1512 } else {
1513 header.version[0]=4;
1514 headerLength=MBCS_HEADER_V4_LENGTH; /* 8 */
1515 }
1516 header.version[1]=4;
1517 /* header.version[2] set above for utf8Friendly data */
1518
1519 header.options|=(uint32_t)headerLength;
1520
1521 header.countStates=mbcsData->ucm->states.countStates;
1522 header.countToUFallbacks=mbcsData->countToUFallbacks;
1523
1524 header.offsetToUCodeUnits=
1525 headerLength*4+
1526 mbcsData->ucm->states.countStates*1024+
1527 mbcsData->countToUFallbacks*sizeof(_MBCSToUFallback);
1528 header.offsetFromUTable=
1529 header.offsetToUCodeUnits+
1530 mbcsData->ucm->states.countToUCodeUnits*2;
1531 header.offsetFromUBytes=
1532 header.offsetFromUTable+
1533 stage1Top*2+
1534 stage2Length;
1535 header.fromUBytesLength=mbcsData->stage3Top;
1536
1537 top=header.offsetFromUBytes+stageUTF8Length*2;
1538 if(!(header.options&MBCS_OPT_NO_FROM_U)) {
1539 top+=header.fromUBytesLength;
1540 }
1541
1542 header.flags=(uint8_t)(mbcsData->ucm->states.outputType);
1543
1544 if(tableType&TABLE_EXT) {
1545 if(top>0xffffff) {
1546 fprintf(stderr, "error: offset 0x%lx to extension table exceeds 0xffffff\n", (long)top);
1547 return 0;
1548 }
1549
1550 header.flags|=top<<8;
1551 }
1552
1553 /* write the MBCS data */
1554 udata_writeBlock(pData, &header, headerLength*4);
1555 udata_writeBlock(pData, mbcsData->ucm->states.stateTable, header.countStates*1024);
1556 udata_writeBlock(pData, mbcsData->toUFallbacks, mbcsData->countToUFallbacks*sizeof(_MBCSToUFallback));
1557 udata_writeBlock(pData, mbcsData->unicodeCodeUnits, mbcsData->ucm->states.countToUCodeUnits*2);
1558 udata_writeBlock(pData, mbcsData->stage1, stage1Top*2);
1559 if(mbcsData->ucm->states.maxCharLength==1) {
1560 udata_writeBlock(pData, mbcsData->stage2Single+stage2Start, stage2Length);
1561 } else {
1562 udata_writeBlock(pData, mbcsData->stage2+stage2Start, stage2Length);
1563 }
1564 if(!(header.options&MBCS_OPT_NO_FROM_U)) {
1565 udata_writeBlock(pData, mbcsData->fromUBytes, mbcsData->stage3Top);
1566 }
1567
1568 if(stageUTF8Length>0) {
1569 udata_writeBlock(pData, mbcsData->stageUTF8, stageUTF8Length*2);
1570 }
1571
1572 /* return the number of bytes that should have been written */
1573 return top;
1574 }
1575 U_CDECL_END
1576