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