icuSources/tools/makeconv/genmbcs.cpp

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