1 /*
2 *******************************************************************************
3 *
4 *   Copyright (C) 2000-2013, 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
hexDigit(uint8_t digit)106 hexDigit(uint8_t digit) {
107     return digit<=9 ? (char)('0'+digit) : (char)('a'-10+digit);
108 }
109 
110 static inline char *
printBytes(char * buffer,const uint8_t * bytes,int32_t length)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 *
MBCSGetDummy()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
MBCSInit(MBCSData * mbcsData,UCMFile * ucm)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 *
MBCSOpen(UCMFile * ucm)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
MBCSDestruct(MBCSData * mbcsData)174 MBCSDestruct(MBCSData *mbcsData) {
175     uprv_free(mbcsData->unicodeCodeUnits);
176     uprv_free(mbcsData->fromUBytes);
177 }
178 
179 static void
MBCSClose(NewConverter * cnvData)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
MBCSStartMappings(MBCSData * mbcsData)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
setFallback(MBCSData * mbcsData,uint32_t offset,UChar32 c)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
removeFallback(MBCSData * mbcsData,uint32_t offset)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
MBCSAddToUnicode(MBCSData * mbcsData,const uint8_t * bytes,int32_t length,UChar32 c,int8_t flag)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
MBCSIsValid(NewConverter * cnvData,const uint8_t * bytes,int32_t length)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
MBCSSingleAddFromUnicode(MBCSData * mbcsData,const uint8_t * bytes,int32_t,UChar32 c,int8_t flag)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
MBCSAddFromUnicode(MBCSData * mbcsData,const uint8_t * bytes,int32_t length,UChar32 c,int8_t flag)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     case 3:
835         b=(b<<8)|*pb++;
836     case 2:
837         b=(b<<8)|*pb++;
838     case 1:
839     default:
840         b=(b<<8)|*pb++;
841         break;
842     }
843 
844     old=0;
845     p=stage3+(stage3Index+nextOffset)*maxCharLength;
846     switch(maxCharLength) {
847     case 2:
848         old=*(uint16_t *)p;
849         *(uint16_t *)p=(uint16_t)b;
850         break;
851     case 3:
852         old=(uint32_t)*p<<16;
853         *p++=(uint8_t)(b>>16);
854         old|=(uint32_t)*p<<8;
855         *p++=(uint8_t)(b>>8);
856         old|=*p;
857         *p=(uint8_t)b;
858         break;
859     case 4:
860         old=*(uint32_t *)p;
861         *(uint32_t *)p=b;
862         break;
863     default:
864         /* will never occur */
865         break;
866     }
867 
868     /* check that this Unicode code point was still unassigned */
869     if((mbcsData->stage2[idx+(nextOffset>>MBCS_STAGE_2_SHIFT)]&(1UL<<(16+(c&0xf))))!=0 || old!=0) {
870         if(flag>=0) {
871             fprintf(stderr, "error: duplicate Unicode code point at U+%04x<->0x%s see 0x%02x\n",
872                 (int)c, printBytes(buffer, bytes, length), (int)old);
873             return FALSE;
874         } else if(VERBOSE) {
875             fprintf(stderr, "duplicate Unicode code point at U+%04x<->0x%s see 0x%02x\n",
876                 (int)c, printBytes(buffer, bytes, length), (int)old);
877         }
878         /* continue after the above warning if the precision of the mapping is
879            unspecified */
880     }
881     if(flag<=0) {
882         /* set the roundtrip flag */
883         mbcsData->stage2[idx+(nextOffset>>4)]|=(1UL<<(16+(c&0xf)));
884     }
885 
886     return TRUE;
887 }
888 
889 U_CFUNC UBool
MBCSOkForBaseFromUnicode(const MBCSData * mbcsData,const uint8_t * bytes,int32_t length,UChar32 c,int8_t flag)890 MBCSOkForBaseFromUnicode(const MBCSData *mbcsData,
891                          const uint8_t *bytes, int32_t length,
892                          UChar32 c, int8_t flag) {
893     /*
894      * A 1:1 mapping does not fit into the MBCS base table's fromUnicode table under
895      * the following conditions:
896      *
897      * - a |2 SUB mapping for <subchar1> (no base table data structure for them)
898      * - a |1 fallback to 0x00 (result value 0, indistinguishable from unmappable entry)
899      * - a multi-byte mapping with leading 0x00 bytes (no explicit length field)
900      *
901      * Some of these tests are redundant with ucm_mappingType().
902      */
903     if( (flag==2 && length==1) ||
904         (flag==1 && bytes[0]==0) || /* testing length==1 would be redundant with the next test */
905         (flag<=1 && length>1 && bytes[0]==0)
906     ) {
907         return FALSE;
908     }
909 
910     /*
911      * Additional restrictions for UTF-8-friendly fromUnicode tables,
912      * for code points up to the maximum optimized one:
913      *
914      * - any mapping to 0x00 (result value 0, indistinguishable from unmappable entry)
915      * - any |1 fallback (no roundtrip flags in the optimized table)
916      */
917     if(mbcsData->utf8Friendly && flag<=1 && c<=mbcsData->utf8Max && (bytes[0]==0 || flag==1)) {
918         return FALSE;
919     }
920 
921     /*
922      * If we omit the fromUnicode data, we can only store roundtrips there
923      * because only they are recoverable from the toUnicode data.
924      * Fallbacks must go into the extension table.
925      */
926     if(mbcsData->omitFromU && flag!=0) {
927         return FALSE;
928     }
929 
930     /* All other mappings do fit into the base table. */
931     return TRUE;
932 }
933 
934 /* we can assume that the table only contains 1:1 mappings with <=4 bytes each */
935 static UBool
MBCSAddTable(NewConverter * cnvData,UCMTable * table,UConverterStaticData * staticData)936 MBCSAddTable(NewConverter *cnvData, UCMTable *table, UConverterStaticData *staticData) {
937     MBCSData *mbcsData;
938     UCMapping *m;
939     UChar32 c;
940     int32_t i, maxCharLength;
941     int8_t f;
942     UBool isOK, utf8Friendly;
943 
944     staticData->unicodeMask=table->unicodeMask;
945     if(staticData->unicodeMask==3) {
946         fprintf(stderr, "error: contains mappings for both supplementary and surrogate code points\n");
947         return FALSE;
948     }
949 
950     staticData->conversionType=UCNV_MBCS;
951 
952     mbcsData=(MBCSData *)cnvData;
953     maxCharLength=mbcsData->ucm->states.maxCharLength;
954 
955     /*
956      * Generation of UTF-8-friendly data requires
957      * a sorted table, which makeconv generates when explicit precision
958      * indicators are used.
959      */
960     mbcsData->utf8Friendly=utf8Friendly=(UBool)((table->flagsType&UCM_FLAGS_EXPLICIT)!=0);
961     if(utf8Friendly) {
962         mbcsData->utf8Max=MBCS_UTF8_MAX;
963         if(SMALL && maxCharLength>1) {
964             mbcsData->omitFromU=TRUE;
965         }
966     } else {
967         mbcsData->utf8Max=0;
968         if(SMALL && maxCharLength>1) {
969             fprintf(stderr,
970                 "makeconv warning: --small not available for .ucm files without |0 etc.\n");
971         }
972     }
973 
974     if(!MBCSStartMappings(mbcsData)) {
975         return FALSE;
976     }
977 
978     staticData->hasFromUnicodeFallback=FALSE;
979     staticData->hasToUnicodeFallback=FALSE;
980 
981     isOK=TRUE;
982 
983     m=table->mappings;
984     for(i=0; i<table->mappingsLength; ++m, ++i) {
985         c=m->u;
986         f=m->f;
987 
988         /*
989          * Small optimization for --small .cnv files:
990          *
991          * If there are fromUnicode mappings above MBCS_UTF8_MAX,
992          * then the file size will be smaller if we make utf8Max larger
993          * because the size increase in stageUTF8 will be more than balanced by
994          * how much less of stage2 needs to be stored.
995          *
996          * There is no point in doing this incrementally because stageUTF8
997          * uses so much less space per block than stage2,
998          * so we immediately increase utf8Max to 0xffff.
999          *
1000          * Do not increase utf8Max if it is already at 0xfeff because MBCSAddFromUnicode()
1001          * sets it to that value when stageUTF8 overflows.
1002          */
1003         if( mbcsData->omitFromU && f<=1 &&
1004             mbcsData->utf8Max<c && c<=0xffff &&
1005             mbcsData->utf8Max<0xfeff
1006         ) {
1007             mbcsData->utf8Max=0xffff;
1008         }
1009 
1010         switch(f) {
1011         case -1:
1012             /* there was no precision/fallback indicator */
1013             /* fall through to set the mappings */
1014         case 0:
1015             /* set roundtrip mappings */
1016             isOK&=MBCSAddToUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1017 
1018             if(maxCharLength==1) {
1019                 isOK&=MBCSSingleAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1020             } else if(MBCSOkForBaseFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f)) {
1021                 isOK&=MBCSAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1022             } else {
1023                 m->f|=MBCS_FROM_U_EXT_FLAG;
1024                 m->moveFlag=UCM_MOVE_TO_EXT;
1025             }
1026             break;
1027         case 1:
1028             /* set only a fallback mapping from Unicode to codepage */
1029             if(maxCharLength==1) {
1030                 staticData->hasFromUnicodeFallback=TRUE;
1031                 isOK&=MBCSSingleAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1032             } else if(MBCSOkForBaseFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f)) {
1033                 staticData->hasFromUnicodeFallback=TRUE;
1034                 isOK&=MBCSAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1035             } else {
1036                 m->f|=MBCS_FROM_U_EXT_FLAG;
1037                 m->moveFlag=UCM_MOVE_TO_EXT;
1038             }
1039             break;
1040         case 2:
1041             /* ignore |2 SUB mappings, except to move <subchar1> mappings to the extension table */
1042             if(maxCharLength>1 && m->bLen==1) {
1043                 m->f|=MBCS_FROM_U_EXT_FLAG;
1044                 m->moveFlag=UCM_MOVE_TO_EXT;
1045             }
1046             break;
1047         case 3:
1048             /* set only a fallback mapping from codepage to Unicode */
1049             staticData->hasToUnicodeFallback=TRUE;
1050             isOK&=MBCSAddToUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
1051             break;
1052         case 4:
1053             /* move "good one-way" mappings to the extension table */
1054             m->f|=MBCS_FROM_U_EXT_FLAG;
1055             m->moveFlag=UCM_MOVE_TO_EXT;
1056             break;
1057         default:
1058             /* will not occur because the parser checked it already */
1059             fprintf(stderr, "error: illegal fallback indicator %d\n", f);
1060             return FALSE;
1061         }
1062     }
1063 
1064     MBCSPostprocess(mbcsData, staticData);
1065 
1066     return isOK;
1067 }
1068 
1069 static UBool
transformEUC(MBCSData * mbcsData)1070 transformEUC(MBCSData *mbcsData) {
1071     uint8_t *p8;
1072     uint32_t i, value, oldLength, old3Top;
1073     uint8_t b;
1074 
1075     oldLength=mbcsData->ucm->states.maxCharLength;
1076     if(oldLength<3) {
1077         return FALSE;
1078     }
1079 
1080     old3Top=mbcsData->stage3Top;
1081 
1082     /* careful: 2-byte and 4-byte codes are stored in platform endianness! */
1083 
1084     /* test if all first bytes are in {0, 0x8e, 0x8f} */
1085     p8=mbcsData->fromUBytes;
1086 
1087 #if !U_IS_BIG_ENDIAN
1088     if(oldLength==4) {
1089         p8+=3;
1090     }
1091 #endif
1092 
1093     for(i=0; i<old3Top; i+=oldLength) {
1094         b=p8[i];
1095         if(b!=0 && b!=0x8e && b!=0x8f) {
1096             /* some first byte does not fit the EUC pattern, nothing to be done */
1097             return FALSE;
1098         }
1099     }
1100     /* restore p if it was modified above */
1101     p8=mbcsData->fromUBytes;
1102 
1103     /* modify outputType and adjust stage3Top */
1104     mbcsData->ucm->states.outputType=(int8_t)(MBCS_OUTPUT_3_EUC+oldLength-3);
1105     mbcsData->stage3Top=(old3Top*(oldLength-1))/oldLength;
1106 
1107     /*
1108      * EUC-encode all byte sequences;
1109      * see "CJKV Information Processing" (1st ed. 1999) from Ken Lunde, O'Reilly,
1110      * p. 161 in chapter 4 "Encoding Methods"
1111      *
1112      * This also must reverse the byte order if the platform is little-endian!
1113      */
1114     if(oldLength==3) {
1115         uint16_t *q=(uint16_t *)p8;
1116         for(i=0; i<old3Top; i+=oldLength) {
1117             b=*p8;
1118             if(b==0) {
1119                 /* short sequences are stored directly */
1120                 /* code set 0 or 1 */
1121                 (*q++)=(uint16_t)((p8[1]<<8)|p8[2]);
1122             } else if(b==0x8e) {
1123                 /* code set 2 */
1124                 (*q++)=(uint16_t)(((p8[1]&0x7f)<<8)|p8[2]);
1125             } else /* b==0x8f */ {
1126                 /* code set 3 */
1127                 (*q++)=(uint16_t)((p8[1]<<8)|(p8[2]&0x7f));
1128             }
1129             p8+=3;
1130         }
1131     } else /* oldLength==4 */ {
1132         uint8_t *q=p8;
1133         uint32_t *p32=(uint32_t *)p8;
1134         for(i=0; i<old3Top; i+=4) {
1135             value=(*p32++);
1136             if(value<=0xffffff) {
1137                 /* short sequences are stored directly */
1138                 /* code set 0 or 1 */
1139                 (*q++)=(uint8_t)(value>>16);
1140                 (*q++)=(uint8_t)(value>>8);
1141                 (*q++)=(uint8_t)value;
1142             } else if(value<=0x8effffff) {
1143                 /* code set 2 */
1144                 (*q++)=(uint8_t)((value>>16)&0x7f);
1145                 (*q++)=(uint8_t)(value>>8);
1146                 (*q++)=(uint8_t)value;
1147             } else /* first byte is 0x8f */ {
1148                 /* code set 3 */
1149                 (*q++)=(uint8_t)(value>>16);
1150                 (*q++)=(uint8_t)((value>>8)&0x7f);
1151                 (*q++)=(uint8_t)value;
1152             }
1153         }
1154     }
1155 
1156     return TRUE;
1157 }
1158 
1159 /*
1160  * Compact stage 2 for SBCS by overlapping adjacent stage 2 blocks as far
1161  * as possible. Overlapping is done on unassigned head and tail
1162  * parts of blocks in steps of MBCS_STAGE_2_MULTIPLIER.
1163  * Stage 1 indexes need to be adjusted accordingly.
1164  * This function is very similar to genprops/store.c/compactStage().
1165  */
1166 static void
singleCompactStage2(MBCSData * mbcsData)1167 singleCompactStage2(MBCSData *mbcsData) {
1168     /* this array maps the ordinal number of a stage 2 block to its new stage 1 index */
1169     uint16_t map[MBCS_STAGE_2_MAX_BLOCKS];
1170     uint16_t i, start, prevEnd, newStart;
1171 
1172     /* enter the all-unassigned first stage 2 block into the map */
1173     map[0]=MBCS_STAGE_2_ALL_UNASSIGNED_INDEX;
1174 
1175     /* begin with the first block after the all-unassigned one */
1176     start=newStart=MBCS_STAGE_2_FIRST_ASSIGNED;
1177     while(start<mbcsData->stage2Top) {
1178         prevEnd=(uint16_t)(newStart-1);
1179 
1180         /* find the size of the overlap */
1181         for(i=0; i<MBCS_STAGE_2_BLOCK_SIZE && mbcsData->stage2Single[start+i]==0 && mbcsData->stage2Single[prevEnd-i]==0; ++i) {}
1182 
1183         if(i>0) {
1184             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=(uint16_t)(newStart-i);
1185 
1186             /* move the non-overlapping indexes to their new positions */
1187             start+=i;
1188             for(i=(uint16_t)(MBCS_STAGE_2_BLOCK_SIZE-i); i>0; --i) {
1189                 mbcsData->stage2Single[newStart++]=mbcsData->stage2Single[start++];
1190             }
1191         } else if(newStart<start) {
1192             /* move the indexes to their new positions */
1193             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=newStart;
1194             for(i=MBCS_STAGE_2_BLOCK_SIZE; i>0; --i) {
1195                 mbcsData->stage2Single[newStart++]=mbcsData->stage2Single[start++];
1196             }
1197         } else /* no overlap && newStart==start */ {
1198             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=start;
1199             start=newStart+=MBCS_STAGE_2_BLOCK_SIZE;
1200         }
1201     }
1202 
1203     /* adjust stage2Top */
1204     if(VERBOSE && newStart<mbcsData->stage2Top) {
1205         printf("compacting stage 2 from stage2Top=0x%lx to 0x%lx, saving %ld bytes\n",
1206                 (unsigned long)mbcsData->stage2Top, (unsigned long)newStart,
1207                 (long)(mbcsData->stage2Top-newStart)*2);
1208     }
1209     mbcsData->stage2Top=newStart;
1210 
1211     /* now adjust stage 1 */
1212     for(i=0; i<MBCS_STAGE_1_SIZE; ++i) {
1213         mbcsData->stage1[i]=map[mbcsData->stage1[i]>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT];
1214     }
1215 }
1216 
1217 /* Compact stage 3 for SBCS - same algorithm as above. */
1218 static void
singleCompactStage3(MBCSData * mbcsData)1219 singleCompactStage3(MBCSData *mbcsData) {
1220     uint16_t *stage3=(uint16_t *)mbcsData->fromUBytes;
1221 
1222     /* this array maps the ordinal number of a stage 3 block to its new stage 2 index */
1223     uint16_t map[0x1000];
1224     uint16_t i, start, prevEnd, newStart;
1225 
1226     /* enter the all-unassigned first stage 3 block into the map */
1227     map[0]=0;
1228 
1229     /* begin with the first block after the all-unassigned one */
1230     start=newStart=16;
1231     while(start<mbcsData->stage3Top) {
1232         prevEnd=(uint16_t)(newStart-1);
1233 
1234         /* find the size of the overlap */
1235         for(i=0; i<16 && stage3[start+i]==0 && stage3[prevEnd-i]==0; ++i) {}
1236 
1237         if(i>0) {
1238             map[start>>4]=(uint16_t)(newStart-i);
1239 
1240             /* move the non-overlapping indexes to their new positions */
1241             start+=i;
1242             for(i=(uint16_t)(16-i); i>0; --i) {
1243                 stage3[newStart++]=stage3[start++];
1244             }
1245         } else if(newStart<start) {
1246             /* move the indexes to their new positions */
1247             map[start>>4]=newStart;
1248             for(i=16; i>0; --i) {
1249                 stage3[newStart++]=stage3[start++];
1250             }
1251         } else /* no overlap && newStart==start */ {
1252             map[start>>4]=start;
1253             start=newStart+=16;
1254         }
1255     }
1256 
1257     /* adjust stage3Top */
1258     if(VERBOSE && newStart<mbcsData->stage3Top) {
1259         printf("compacting stage 3 from stage3Top=0x%lx to 0x%lx, saving %ld bytes\n",
1260                 (unsigned long)mbcsData->stage3Top, (unsigned long)newStart,
1261                 (long)(mbcsData->stage3Top-newStart)*2);
1262     }
1263     mbcsData->stage3Top=newStart;
1264 
1265     /* now adjust stage 2 */
1266     for(i=0; i<mbcsData->stage2Top; ++i) {
1267         mbcsData->stage2Single[i]=map[mbcsData->stage2Single[i]>>4];
1268     }
1269 }
1270 
1271 /*
1272  * Compact stage 2 by overlapping adjacent stage 2 blocks as far
1273  * as possible. Overlapping is done on unassigned head and tail
1274  * parts of blocks in steps of MBCS_STAGE_2_MULTIPLIER.
1275  * Stage 1 indexes need to be adjusted accordingly.
1276  * This function is very similar to genprops/store.c/compactStage().
1277  */
1278 static void
compactStage2(MBCSData * mbcsData)1279 compactStage2(MBCSData *mbcsData) {
1280     /* this array maps the ordinal number of a stage 2 block to its new stage 1 index */
1281     uint16_t map[MBCS_STAGE_2_MAX_BLOCKS];
1282     uint16_t i, start, prevEnd, newStart;
1283 
1284     /* enter the all-unassigned first stage 2 block into the map */
1285     map[0]=MBCS_STAGE_2_ALL_UNASSIGNED_INDEX;
1286 
1287     /* begin with the first block after the all-unassigned one */
1288     start=newStart=MBCS_STAGE_2_FIRST_ASSIGNED;
1289     while(start<mbcsData->stage2Top) {
1290         prevEnd=(uint16_t)(newStart-1);
1291 
1292         /* find the size of the overlap */
1293         for(i=0; i<MBCS_STAGE_2_BLOCK_SIZE && mbcsData->stage2[start+i]==0 && mbcsData->stage2[prevEnd-i]==0; ++i) {}
1294 
1295         if(i>0) {
1296             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=(uint16_t)(newStart-i);
1297 
1298             /* move the non-overlapping indexes to their new positions */
1299             start+=i;
1300             for(i=(uint16_t)(MBCS_STAGE_2_BLOCK_SIZE-i); i>0; --i) {
1301                 mbcsData->stage2[newStart++]=mbcsData->stage2[start++];
1302             }
1303         } else if(newStart<start) {
1304             /* move the indexes to their new positions */
1305             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=newStart;
1306             for(i=MBCS_STAGE_2_BLOCK_SIZE; i>0; --i) {
1307                 mbcsData->stage2[newStart++]=mbcsData->stage2[start++];
1308             }
1309         } else /* no overlap && newStart==start */ {
1310             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=start;
1311             start=newStart+=MBCS_STAGE_2_BLOCK_SIZE;
1312         }
1313     }
1314 
1315     /* adjust stage2Top */
1316     if(VERBOSE && newStart<mbcsData->stage2Top) {
1317         printf("compacting stage 2 from stage2Top=0x%lx to 0x%lx, saving %ld bytes\n",
1318                 (unsigned long)mbcsData->stage2Top, (unsigned long)newStart,
1319                 (long)(mbcsData->stage2Top-newStart)*4);
1320     }
1321     mbcsData->stage2Top=newStart;
1322 
1323     /* now adjust stage 1 */
1324     for(i=0; i<MBCS_STAGE_1_SIZE; ++i) {
1325         mbcsData->stage1[i]=map[mbcsData->stage1[i]>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT];
1326     }
1327 }
1328 
1329 static void
MBCSPostprocess(MBCSData * mbcsData,const UConverterStaticData *)1330 MBCSPostprocess(MBCSData *mbcsData, const UConverterStaticData * /*staticData*/) {
1331     UCMStates *states;
1332     int32_t maxCharLength, stage3Width;
1333 
1334     states=&mbcsData->ucm->states;
1335     stage3Width=maxCharLength=states->maxCharLength;
1336 
1337     ucm_optimizeStates(states,
1338                        &mbcsData->unicodeCodeUnits,
1339                        mbcsData->toUFallbacks, mbcsData->countToUFallbacks,
1340                        VERBOSE);
1341 
1342     /* try to compact the fromUnicode tables */
1343     if(transformEUC(mbcsData)) {
1344         --stage3Width;
1345     }
1346 
1347     /*
1348      * UTF-8-friendly tries are built precompacted, to cope with variable
1349      * stage 3 allocation block sizes.
1350      *
1351      * Tables without precision indicators cannot be built that way,
1352      * because if a block was overlapped with a previous one, then a smaller
1353      * code point for the same block would not fit.
1354      * Therefore, such tables are not marked UTF-8-friendly and must be
1355      * compacted after all mappings are entered.
1356      */
1357     if(!mbcsData->utf8Friendly) {
1358         if(maxCharLength==1) {
1359             singleCompactStage3(mbcsData);
1360             singleCompactStage2(mbcsData);
1361         } else {
1362             compactStage2(mbcsData);
1363         }
1364     }
1365 
1366     if(VERBOSE) {
1367         /*uint32_t c, i1, i2, i2Limit, i3;*/
1368 
1369         printf("fromUnicode number of uint%s_t in stage 2: 0x%lx=%lu\n",
1370                maxCharLength==1 ? "16" : "32",
1371                (unsigned long)mbcsData->stage2Top,
1372                (unsigned long)mbcsData->stage2Top);
1373         printf("fromUnicode number of %d-byte stage 3 mapping entries: 0x%lx=%lu\n",
1374                (int)stage3Width,
1375                (unsigned long)mbcsData->stage3Top/stage3Width,
1376                (unsigned long)mbcsData->stage3Top/stage3Width);
1377 #if 0
1378         c=0;
1379         for(i1=0; i1<MBCS_STAGE_1_SIZE; ++i1) {
1380             i2=mbcsData->stage1[i1];
1381             if(i2==0) {
1382                 c+=MBCS_STAGE_2_BLOCK_SIZE*MBCS_STAGE_3_BLOCK_SIZE;
1383                 continue;
1384             }
1385             for(i2Limit=i2+MBCS_STAGE_2_BLOCK_SIZE; i2<i2Limit; ++i2) {
1386                 if(maxCharLength==1) {
1387                     i3=mbcsData->stage2Single[i2];
1388                 } else {
1389                     i3=(uint16_t)mbcsData->stage2[i2];
1390                 }
1391                 if(i3==0) {
1392                     c+=MBCS_STAGE_3_BLOCK_SIZE;
1393                     continue;
1394                 }
1395                 printf("U+%04lx i1=0x%02lx i2=0x%04lx i3=0x%04lx\n",
1396                        (unsigned long)c,
1397                        (unsigned long)i1,
1398                        (unsigned long)i2,
1399                        (unsigned long)i3);
1400                 c+=MBCS_STAGE_3_BLOCK_SIZE;
1401             }
1402         }
1403 #endif
1404     }
1405 }
1406 
1407 static uint32_t
MBCSWrite(NewConverter * cnvData,const UConverterStaticData * staticData,UNewDataMemory * pData,int32_t tableType)1408 MBCSWrite(NewConverter *cnvData, const UConverterStaticData *staticData,
1409           UNewDataMemory *pData, int32_t tableType) {
1410     MBCSData *mbcsData=(MBCSData *)cnvData;
1411     uint32_t stage2Start, stage2Length;
1412     uint32_t top, stageUTF8Length=0;
1413     int32_t i, stage1Top;
1414     uint32_t headerLength;
1415 
1416     _MBCSHeader header=UCNV_MBCS_HEADER_INITIALIZER;
1417 
1418     stage2Length=mbcsData->stage2Top;
1419     if(mbcsData->omitFromU) {
1420         /* find how much of stage2 can be omitted */
1421         int32_t utf8Limit=(int32_t)mbcsData->utf8Max+1;
1422         uint32_t st2=0; /*initialized it to avoid compiler warnings */
1423 
1424         i=utf8Limit>>MBCS_STAGE_1_SHIFT;
1425         if((utf8Limit&((1<<MBCS_STAGE_1_SHIFT)-1))!=0 && (st2=mbcsData->stage1[i])!=0) {
1426             /* utf8Limit is in the middle of an existing stage 2 block */
1427             stage2Start=st2+((utf8Limit>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK);
1428         } else {
1429             /* find the last stage2 block with mappings before utf8Limit */
1430             while(i>0 && (st2=mbcsData->stage1[--i])==0) {}
1431             /* stage2 up to the end of this block corresponds to stageUTF8 */
1432             stage2Start=st2+MBCS_STAGE_2_BLOCK_SIZE;
1433         }
1434         header.options|=MBCS_OPT_NO_FROM_U;
1435         header.fullStage2Length=stage2Length;
1436         stage2Length-=stage2Start;
1437         if(VERBOSE) {
1438             printf("+ omitting %lu out of %lu stage2 entries and %lu fromUBytes\n",
1439                     (unsigned long)stage2Start,
1440                     (unsigned long)mbcsData->stage2Top,
1441                     (unsigned long)mbcsData->stage3Top);
1442             printf("+ total size savings: %lu bytes\n", (unsigned long)stage2Start*4+mbcsData->stage3Top);
1443         }
1444     } else {
1445         stage2Start=0;
1446     }
1447 
1448     if(staticData->unicodeMask&UCNV_HAS_SUPPLEMENTARY) {
1449         stage1Top=MBCS_STAGE_1_SIZE; /* 0x440==1088 */
1450     } else {
1451         stage1Top=0x40; /* 0x40==64 */
1452     }
1453 
1454     /* adjust stage 1 entries to include the size of stage 1 in the offsets to stage 2 */
1455     if(mbcsData->ucm->states.maxCharLength==1) {
1456         for(i=0; i<stage1Top; ++i) {
1457             mbcsData->stage1[i]+=(uint16_t)stage1Top;
1458         }
1459 
1460         /* stage2Top/Length have counted 16-bit results, now we need to count bytes */
1461         /* also round up to a multiple of 4 bytes */
1462         stage2Length=(stage2Length*2+1)&~1;
1463 
1464         /* stage3Top has counted 16-bit results, now we need to count bytes */
1465         mbcsData->stage3Top*=2;
1466 
1467         if(mbcsData->utf8Friendly) {
1468             header.version[2]=(uint8_t)(SBCS_UTF8_MAX>>8); /* store 0x1f for max==0x1fff */
1469         }
1470     } else {
1471         for(i=0; i<stage1Top; ++i) {
1472             mbcsData->stage1[i]+=(uint16_t)stage1Top/2; /* stage 2 contains 32-bit entries, stage 1 16-bit entries */
1473         }
1474 
1475         /* stage2Top/Length have counted 32-bit results, now we need to count bytes */
1476         stage2Length*=4;
1477         /* leave stage2Start counting 32-bit units */
1478 
1479         if(mbcsData->utf8Friendly) {
1480             stageUTF8Length=(mbcsData->utf8Max+1)>>MBCS_UTF8_STAGE_SHIFT;
1481             header.version[2]=(uint8_t)(mbcsData->utf8Max>>8); /* store 0xd7 for max==0xd7ff */
1482         }
1483 
1484         /* stage3Top has already counted bytes */
1485     }
1486 
1487     /* round up stage3Top so that the sizes of all data blocks are multiples of 4 */
1488     mbcsData->stage3Top=(mbcsData->stage3Top+3)&~3;
1489 
1490     /* fill the header */
1491     if(header.options&MBCS_OPT_INCOMPATIBLE_MASK) {
1492         header.version[0]=5;
1493         if(header.options&MBCS_OPT_NO_FROM_U) {
1494             headerLength=10;  /* include fullStage2Length */
1495         } else {
1496             headerLength=MBCS_HEADER_V5_MIN_LENGTH;  /* 9 */
1497         }
1498     } else {
1499         header.version[0]=4;
1500         headerLength=MBCS_HEADER_V4_LENGTH;  /* 8 */
1501     }
1502     header.version[1]=4;
1503     /* header.version[2] set above for utf8Friendly data */
1504 
1505     header.options|=(uint32_t)headerLength;
1506 
1507     header.countStates=mbcsData->ucm->states.countStates;
1508     header.countToUFallbacks=mbcsData->countToUFallbacks;
1509 
1510     header.offsetToUCodeUnits=
1511         headerLength*4+
1512         mbcsData->ucm->states.countStates*1024+
1513         mbcsData->countToUFallbacks*sizeof(_MBCSToUFallback);
1514     header.offsetFromUTable=
1515         header.offsetToUCodeUnits+
1516         mbcsData->ucm->states.countToUCodeUnits*2;
1517     header.offsetFromUBytes=
1518         header.offsetFromUTable+
1519         stage1Top*2+
1520         stage2Length;
1521     header.fromUBytesLength=mbcsData->stage3Top;
1522 
1523     top=header.offsetFromUBytes+stageUTF8Length*2;
1524     if(!(header.options&MBCS_OPT_NO_FROM_U)) {
1525         top+=header.fromUBytesLength;
1526     }
1527 
1528     header.flags=(uint8_t)(mbcsData->ucm->states.outputType);
1529 
1530     if(tableType&TABLE_EXT) {
1531         if(top>0xffffff) {
1532             fprintf(stderr, "error: offset 0x%lx to extension table exceeds 0xffffff\n", (long)top);
1533             return 0;
1534         }
1535 
1536         header.flags|=top<<8;
1537     }
1538 
1539     /* write the MBCS data */
1540     udata_writeBlock(pData, &header, headerLength*4);
1541     udata_writeBlock(pData, mbcsData->ucm->states.stateTable, header.countStates*1024);
1542     udata_writeBlock(pData, mbcsData->toUFallbacks, mbcsData->countToUFallbacks*sizeof(_MBCSToUFallback));
1543     udata_writeBlock(pData, mbcsData->unicodeCodeUnits, mbcsData->ucm->states.countToUCodeUnits*2);
1544     udata_writeBlock(pData, mbcsData->stage1, stage1Top*2);
1545     if(mbcsData->ucm->states.maxCharLength==1) {
1546         udata_writeBlock(pData, mbcsData->stage2Single+stage2Start, stage2Length);
1547     } else {
1548         udata_writeBlock(pData, mbcsData->stage2+stage2Start, stage2Length);
1549     }
1550     if(!(header.options&MBCS_OPT_NO_FROM_U)) {
1551         udata_writeBlock(pData, mbcsData->fromUBytes, mbcsData->stage3Top);
1552     }
1553 
1554     if(stageUTF8Length>0) {
1555         udata_writeBlock(pData, mbcsData->stageUTF8, stageUTF8Length*2);
1556     }
1557 
1558     /* return the number of bytes that should have been written */
1559     return top;
1560 }
1561