michael@0: /*
michael@0: *******************************************************************************
michael@0: *
michael@0: *   Copyright (C) 2000-2013, International Business Machines
michael@0: *   Corporation and others.  All Rights Reserved.
michael@0: *
michael@0: *******************************************************************************
michael@0: *   file name:  genmbcs.cpp
michael@0: *   encoding:   US-ASCII
michael@0: *   tab size:   8 (not used)
michael@0: *   indentation:4
michael@0: *
michael@0: *   created on: 2000jul06
michael@0: *   created by: Markus W. Scherer
michael@0: */
michael@0: 
michael@0: #include <stdio.h>
michael@0: #include "unicode/utypes.h"
michael@0: #include "cstring.h"
michael@0: #include "cmemory.h"
michael@0: #include "unewdata.h"
michael@0: #include "ucnv_cnv.h"
michael@0: #include "ucnvmbcs.h"
michael@0: #include "ucm.h"
michael@0: #include "makeconv.h"
michael@0: #include "genmbcs.h"
michael@0: 
michael@0: /*
michael@0:  * TODO: Split this file into toUnicode, SBCSFromUnicode and MBCSFromUnicode files.
michael@0:  * Reduce tests for maxCharLength.
michael@0:  */
michael@0: 
michael@0: struct MBCSData {
michael@0:     NewConverter newConverter;
michael@0: 
michael@0:     UCMFile *ucm;
michael@0: 
michael@0:     /* toUnicode (state table in ucm->states) */
michael@0:     _MBCSToUFallback toUFallbacks[MBCS_MAX_FALLBACK_COUNT];
michael@0:     int32_t countToUFallbacks;
michael@0:     uint16_t *unicodeCodeUnits;
michael@0: 
michael@0:     /* fromUnicode */
michael@0:     uint16_t stage1[MBCS_STAGE_1_SIZE];
michael@0:     uint16_t stage2Single[MBCS_STAGE_2_SIZE]; /* stage 2 for single-byte codepages */
michael@0:     uint32_t stage2[MBCS_STAGE_2_SIZE]; /* stage 2 for MBCS */
michael@0:     uint8_t *fromUBytes;
michael@0:     uint32_t stage2Top, stage3Top;
michael@0: 
michael@0:     /* fromUTF8 */
michael@0:     uint16_t stageUTF8[0x10000>>MBCS_UTF8_STAGE_SHIFT];  /* allow for utf8Max=0xffff */
michael@0: 
michael@0:     /*
michael@0:      * Maximum UTF-8-friendly code point.
michael@0:      * 0 if !utf8Friendly, otherwise 0x01ff..0xffff in steps of 0x100.
michael@0:      * If utf8Friendly, utf8Max is normally either MBCS_UTF8_MAX or 0xffff.
michael@0:      */
michael@0:     uint16_t utf8Max;
michael@0: 
michael@0:     UBool utf8Friendly;
michael@0:     UBool omitFromU;
michael@0: };
michael@0: 
michael@0: /* prototypes */
michael@0: static void
michael@0: MBCSClose(NewConverter *cnvData);
michael@0: 
michael@0: static UBool
michael@0: MBCSStartMappings(MBCSData *mbcsData);
michael@0: 
michael@0: static UBool
michael@0: MBCSAddToUnicode(MBCSData *mbcsData,
michael@0:                  const uint8_t *bytes, int32_t length,
michael@0:                  UChar32 c,
michael@0:                  int8_t flag);
michael@0: 
michael@0: static UBool
michael@0: MBCSIsValid(NewConverter *cnvData,
michael@0:             const uint8_t *bytes, int32_t length);
michael@0: 
michael@0: static UBool
michael@0: MBCSSingleAddFromUnicode(MBCSData *mbcsData,
michael@0:                          const uint8_t *bytes, int32_t length,
michael@0:                          UChar32 c,
michael@0:                          int8_t flag);
michael@0: 
michael@0: static UBool
michael@0: MBCSAddFromUnicode(MBCSData *mbcsData,
michael@0:                    const uint8_t *bytes, int32_t length,
michael@0:                    UChar32 c,
michael@0:                    int8_t flag);
michael@0: 
michael@0: static void
michael@0: MBCSPostprocess(MBCSData *mbcsData, const UConverterStaticData *staticData);
michael@0: 
michael@0: static UBool
michael@0: MBCSAddTable(NewConverter *cnvData, UCMTable *table, UConverterStaticData *staticData);
michael@0: 
michael@0: static uint32_t
michael@0: MBCSWrite(NewConverter *cnvData, const UConverterStaticData *staticData,
michael@0:           UNewDataMemory *pData, int32_t tableType);
michael@0: 
michael@0: /* helper ------------------------------------------------------------------- */
michael@0: 
michael@0: static inline char
michael@0: hexDigit(uint8_t digit) {
michael@0:     return digit<=9 ? (char)('0'+digit) : (char)('a'-10+digit);
michael@0: }
michael@0: 
michael@0: static inline char *
michael@0: printBytes(char *buffer, const uint8_t *bytes, int32_t length) {
michael@0:     char *s=buffer;
michael@0:     while(length>0) {
michael@0:         *s++=hexDigit((uint8_t)(*bytes>>4));
michael@0:         *s++=hexDigit((uint8_t)(*bytes&0xf));
michael@0:         ++bytes;
michael@0:         --length;
michael@0:     }
michael@0: 
michael@0:     *s=0;
michael@0:     return buffer;
michael@0: }
michael@0: 
michael@0: /* implementation ----------------------------------------------------------- */
michael@0: 
michael@0: static MBCSData gDummy;
michael@0: 
michael@0: U_CFUNC const MBCSData *
michael@0: MBCSGetDummy() {
michael@0:     uprv_memset(&gDummy, 0, sizeof(MBCSData));
michael@0: 
michael@0:     /*
michael@0:      * Set "pessimistic" values which may sometimes move too many
michael@0:      * mappings to the extension table (but never too few).
michael@0:      * These values cause MBCSOkForBaseFromUnicode() to return FALSE for the
michael@0:      * largest set of mappings.
michael@0:      * Assume maxCharLength>1.
michael@0:      */
michael@0:     gDummy.utf8Friendly=TRUE;
michael@0:     if(SMALL) {
michael@0:         gDummy.utf8Max=0xffff;
michael@0:         gDummy.omitFromU=TRUE;
michael@0:     } else {
michael@0:         gDummy.utf8Max=MBCS_UTF8_MAX;
michael@0:     }
michael@0:     return &gDummy;
michael@0: }
michael@0: 
michael@0: static void
michael@0: MBCSInit(MBCSData *mbcsData, UCMFile *ucm) {
michael@0:     uprv_memset(mbcsData, 0, sizeof(MBCSData));
michael@0: 
michael@0:     mbcsData->ucm=ucm; /* aliased, not owned */
michael@0: 
michael@0:     mbcsData->newConverter.close=MBCSClose;
michael@0:     mbcsData->newConverter.isValid=MBCSIsValid;
michael@0:     mbcsData->newConverter.addTable=MBCSAddTable;
michael@0:     mbcsData->newConverter.write=MBCSWrite;
michael@0: }
michael@0: 
michael@0: NewConverter *
michael@0: MBCSOpen(UCMFile *ucm) {
michael@0:     MBCSData *mbcsData=(MBCSData *)uprv_malloc(sizeof(MBCSData));
michael@0:     if(mbcsData==NULL) {
michael@0:         printf("out of memory\n");
michael@0:         exit(U_MEMORY_ALLOCATION_ERROR);
michael@0:     }
michael@0: 
michael@0:     MBCSInit(mbcsData, ucm);
michael@0:     return &mbcsData->newConverter;
michael@0: }
michael@0: 
michael@0: static void
michael@0: MBCSDestruct(MBCSData *mbcsData) {
michael@0:     uprv_free(mbcsData->unicodeCodeUnits);
michael@0:     uprv_free(mbcsData->fromUBytes);
michael@0: }
michael@0: 
michael@0: static void
michael@0: MBCSClose(NewConverter *cnvData) {
michael@0:     MBCSData *mbcsData=(MBCSData *)cnvData;
michael@0:     if(mbcsData!=NULL) {
michael@0:         MBCSDestruct(mbcsData);
michael@0:         uprv_free(mbcsData);
michael@0:     }
michael@0: }
michael@0: 
michael@0: static UBool
michael@0: MBCSStartMappings(MBCSData *mbcsData) {
michael@0:     int32_t i, sum, maxCharLength,
michael@0:             stage2NullLength, stage2AllocLength,
michael@0:             stage3NullLength, stage3AllocLength;
michael@0: 
michael@0:     /* toUnicode */
michael@0: 
michael@0:     /* allocate the code unit array and prefill it with "unassigned" values */
michael@0:     sum=mbcsData->ucm->states.countToUCodeUnits;
michael@0:     if(VERBOSE) {
michael@0:         printf("the total number of offsets is 0x%lx=%ld\n", (long)sum, (long)sum);
michael@0:     }
michael@0: 
michael@0:     if(sum>0) {
michael@0:         mbcsData->unicodeCodeUnits=(uint16_t *)uprv_malloc(sum*sizeof(uint16_t));
michael@0:         if(mbcsData->unicodeCodeUnits==NULL) {
michael@0:             fprintf(stderr, "error: out of memory allocating %ld 16-bit code units\n",
michael@0:                 (long)sum);
michael@0:             return FALSE;
michael@0:         }
michael@0:         for(i=0; i<sum; ++i) {
michael@0:             mbcsData->unicodeCodeUnits[i]=0xfffe;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* fromUnicode */
michael@0:     maxCharLength=mbcsData->ucm->states.maxCharLength;
michael@0: 
michael@0:     /* allocate the codepage mappings and preset the first 16 characters to 0 */
michael@0:     if(maxCharLength==1) {
michael@0:         /* allocate 64k 16-bit results for single-byte codepages */
michael@0:         sum=0x20000;
michael@0:     } else {
michael@0:         /* allocate 1M * maxCharLength bytes for at most 1M mappings */
michael@0:         sum=0x100000*maxCharLength;
michael@0:     }
michael@0:     mbcsData->fromUBytes=(uint8_t *)uprv_malloc(sum);
michael@0:     if(mbcsData->fromUBytes==NULL) {
michael@0:         fprintf(stderr, "error: out of memory allocating %ld B for target mappings\n", (long)sum);
michael@0:         return FALSE;
michael@0:     }
michael@0:     uprv_memset(mbcsData->fromUBytes, 0, sum);
michael@0: 
michael@0:     /*
michael@0:      * UTF-8-friendly fromUnicode tries: allocate multiple blocks at a time.
michael@0:      * See ucnvmbcs.h for details.
michael@0:      *
michael@0:      * There is code, for example in ucnv_MBCSGetUnicodeSetForUnicode(), which
michael@0:      * assumes that the initial stage 2/3 blocks are the all-unassigned ones.
michael@0:      * Therefore, we refine the data structure while maintaining this placement
michael@0:      * even though it would be convenient to allocate the ASCII block at the
michael@0:      * beginning of stage 3, for example.
michael@0:      *
michael@0:      * UTF-8-friendly fromUnicode tries work from sorted tables and are built
michael@0:      * pre-compacted, overlapping adjacent stage 2/3 blocks.
michael@0:      * This is necessary because the block allocation and compaction changes
michael@0:      * at SBCS_UTF8_MAX or MBCS_UTF8_MAX, and for MBCS tables the additional
michael@0:      * stage table uses direct indexes into stage 3, without a multiplier and
michael@0:      * thus with a smaller reach.
michael@0:      *
michael@0:      * Non-UTF-8-friendly fromUnicode tries work from unsorted tables
michael@0:      * (because implicit precision is used), and are compacted
michael@0:      * in post-processing.
michael@0:      *
michael@0:      * Preallocation for UTF-8-friendly fromUnicode tries:
michael@0:      *
michael@0:      * Stage 3:
michael@0:      * 64-entry all-unassigned first block followed by ASCII (128 entries).
michael@0:      *
michael@0:      * Stage 2:
michael@0:      * 64-entry all-unassigned first block followed by preallocated
michael@0:      * 64-block for ASCII.
michael@0:      */
michael@0: 
michael@0:     /* Preallocate ASCII as a linear 128-entry stage 3 block. */
michael@0:     stage2NullLength=MBCS_STAGE_2_BLOCK_SIZE;
michael@0:     stage2AllocLength=MBCS_STAGE_2_BLOCK_SIZE;
michael@0: 
michael@0:     stage3NullLength=MBCS_UTF8_STAGE_3_BLOCK_SIZE;
michael@0:     stage3AllocLength=128; /* ASCII U+0000..U+007f */
michael@0: 
michael@0:     /* Initialize stage 1 for the preallocated blocks. */
michael@0:     sum=stage2NullLength;
michael@0:     for(i=0; i<(stage2AllocLength>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT); ++i) {
michael@0:         mbcsData->stage1[i]=sum;
michael@0:         sum+=MBCS_STAGE_2_BLOCK_SIZE;
michael@0:     }
michael@0:     mbcsData->stage2Top=stage2NullLength+stage2AllocLength; /* ==sum */
michael@0: 
michael@0:     /*
michael@0:      * Stage 2 indexes count 16-blocks in stage 3 as follows:
michael@0:      * SBCS: directly, indexes increment by 16
michael@0:      * MBCS: indexes need to be multiplied by 16*maxCharLength, indexes increment by 1
michael@0:      * MBCS UTF-8: directly, indexes increment by 16
michael@0:      */
michael@0:     if(maxCharLength==1) {
michael@0:         sum=stage3NullLength;
michael@0:         for(i=0; i<(stage3AllocLength/MBCS_STAGE_3_BLOCK_SIZE); ++i) {
michael@0:             mbcsData->stage2Single[mbcsData->stage1[0]+i]=sum;
michael@0:             sum+=MBCS_STAGE_3_BLOCK_SIZE;
michael@0:         }
michael@0:     } else {
michael@0:         sum=stage3NullLength/MBCS_STAGE_3_GRANULARITY;
michael@0:         for(i=0; i<(stage3AllocLength/MBCS_STAGE_3_BLOCK_SIZE); ++i) {
michael@0:             mbcsData->stage2[mbcsData->stage1[0]+i]=sum;
michael@0:             sum+=MBCS_STAGE_3_BLOCK_SIZE/MBCS_STAGE_3_GRANULARITY;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     sum=stage3NullLength;
michael@0:     for(i=0; i<(stage3AllocLength/MBCS_UTF8_STAGE_3_BLOCK_SIZE); ++i) {
michael@0:         mbcsData->stageUTF8[i]=sum;
michael@0:         sum+=MBCS_UTF8_STAGE_3_BLOCK_SIZE;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * Allocate a 64-entry all-unassigned first stage 3 block,
michael@0:      * for UTF-8-friendly lookup with a trail byte,
michael@0:      * plus 128 entries for ASCII.
michael@0:      */
michael@0:     mbcsData->stage3Top=(stage3NullLength+stage3AllocLength)*maxCharLength; /* ==sum*maxCharLength */
michael@0: 
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: /* return TRUE for success */
michael@0: static UBool
michael@0: setFallback(MBCSData *mbcsData, uint32_t offset, UChar32 c) {
michael@0:     int32_t i=ucm_findFallback(mbcsData->toUFallbacks, mbcsData->countToUFallbacks, offset);
michael@0:     if(i>=0) {
michael@0:         /* if there is already a fallback for this offset, then overwrite it */
michael@0:         mbcsData->toUFallbacks[i].codePoint=c;
michael@0:         return TRUE;
michael@0:     } else {
michael@0:         /* if there is no fallback for this offset, then add one */
michael@0:         i=mbcsData->countToUFallbacks;
michael@0:         if(i>=MBCS_MAX_FALLBACK_COUNT) {
michael@0:             fprintf(stderr, "error: too many toUnicode fallbacks, currently at: U+%x\n", (int)c);
michael@0:             return FALSE;
michael@0:         } else {
michael@0:             mbcsData->toUFallbacks[i].offset=offset;
michael@0:             mbcsData->toUFallbacks[i].codePoint=c;
michael@0:             mbcsData->countToUFallbacks=i+1;
michael@0:             return TRUE;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: /* remove fallback if there is one with this offset; return the code point if there was such a fallback, otherwise -1 */
michael@0: static int32_t
michael@0: removeFallback(MBCSData *mbcsData, uint32_t offset) {
michael@0:     int32_t i=ucm_findFallback(mbcsData->toUFallbacks, mbcsData->countToUFallbacks, offset);
michael@0:     if(i>=0) {
michael@0:         _MBCSToUFallback *toUFallbacks;
michael@0:         int32_t limit, old;
michael@0: 
michael@0:         toUFallbacks=mbcsData->toUFallbacks;
michael@0:         limit=mbcsData->countToUFallbacks;
michael@0:         old=(int32_t)toUFallbacks[i].codePoint;
michael@0: 
michael@0:         /* copy the last fallback entry here to keep the list contiguous */
michael@0:         toUFallbacks[i].offset=toUFallbacks[limit-1].offset;
michael@0:         toUFallbacks[i].codePoint=toUFallbacks[limit-1].codePoint;
michael@0:         mbcsData->countToUFallbacks=limit-1;
michael@0:         return old;
michael@0:     } else {
michael@0:         return -1;
michael@0:     }
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * isFallback is almost a boolean:
michael@0:  * 1 (TRUE)  this is a fallback mapping
michael@0:  * 0 (FALSE) this is a precise mapping
michael@0:  * -1        the precision of this mapping is not specified
michael@0:  */
michael@0: static UBool
michael@0: MBCSAddToUnicode(MBCSData *mbcsData,
michael@0:                  const uint8_t *bytes, int32_t length,
michael@0:                  UChar32 c,
michael@0:                  int8_t flag) {
michael@0:     char buffer[10];
michael@0:     uint32_t offset=0;
michael@0:     int32_t i=0, entry, old;
michael@0:     uint8_t state=0;
michael@0: 
michael@0:     if(mbcsData->ucm->states.countStates==0) {
michael@0:         fprintf(stderr, "error: there is no state information!\n");
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     /* for SI/SO (like EBCDIC-stateful), double-byte sequences start in state 1 */
michael@0:     if(length==2 && mbcsData->ucm->states.outputType==MBCS_OUTPUT_2_SISO) {
michael@0:         state=1;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * Walk down the state table like in conversion,
michael@0:      * much like getNextUChar().
michael@0:      * We assume that c<=0x10ffff.
michael@0:      */
michael@0:     for(i=0;;) {
michael@0:         entry=mbcsData->ucm->states.stateTable[state][bytes[i++]];
michael@0:         if(MBCS_ENTRY_IS_TRANSITION(entry)) {
michael@0:             if(i==length) {
michael@0:                 fprintf(stderr, "error: byte sequence too short, ends in non-final state %hu: 0x%s (U+%x)\n",
michael@0:                     (short)state, printBytes(buffer, bytes, length), (int)c);
michael@0:                 return FALSE;
michael@0:             }
michael@0:             state=(uint8_t)MBCS_ENTRY_TRANSITION_STATE(entry);
michael@0:             offset+=MBCS_ENTRY_TRANSITION_OFFSET(entry);
michael@0:         } else {
michael@0:             if(i<length) {
michael@0:                 fprintf(stderr, "error: byte sequence too long by %d bytes, final state %u: 0x%s (U+%x)\n",
michael@0:                     (int)(length-i), state, printBytes(buffer, bytes, length), (int)c);
michael@0:                 return FALSE;
michael@0:             }
michael@0:             switch(MBCS_ENTRY_FINAL_ACTION(entry)) {
michael@0:             case MBCS_STATE_ILLEGAL:
michael@0:                 fprintf(stderr, "error: byte sequence ends in illegal state at U+%04x<->0x%s\n",
michael@0:                     (int)c, printBytes(buffer, bytes, length));
michael@0:                 return FALSE;
michael@0:             case MBCS_STATE_CHANGE_ONLY:
michael@0:                 fprintf(stderr, "error: byte sequence ends in state-change-only at U+%04x<->0x%s\n",
michael@0:                     (int)c, printBytes(buffer, bytes, length));
michael@0:                 return FALSE;
michael@0:             case MBCS_STATE_UNASSIGNED:
michael@0:                 fprintf(stderr, "error: byte sequence ends in unassigned state at U+%04x<->0x%s\n",
michael@0:                     (int)c, printBytes(buffer, bytes, length));
michael@0:                 return FALSE;
michael@0:             case MBCS_STATE_FALLBACK_DIRECT_16:
michael@0:             case MBCS_STATE_VALID_DIRECT_16:
michael@0:             case MBCS_STATE_FALLBACK_DIRECT_20:
michael@0:             case MBCS_STATE_VALID_DIRECT_20:
michael@0:                 if(MBCS_ENTRY_SET_STATE(entry, 0)!=MBCS_ENTRY_FINAL(0, MBCS_STATE_VALID_DIRECT_16, 0xfffe)) {
michael@0:                     /* the "direct" action's value is not "valid-direct-16-unassigned" any more */
michael@0:                     if(MBCS_ENTRY_FINAL_ACTION(entry)==MBCS_STATE_VALID_DIRECT_16 || MBCS_ENTRY_FINAL_ACTION(entry)==MBCS_STATE_FALLBACK_DIRECT_16) {
michael@0:                         old=MBCS_ENTRY_FINAL_VALUE(entry);
michael@0:                     } else {
michael@0:                         old=0x10000+MBCS_ENTRY_FINAL_VALUE(entry);
michael@0:                     }
michael@0:                     if(flag>=0) {
michael@0:                         fprintf(stderr, "error: duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
michael@0:                             (int)c, printBytes(buffer, bytes, length), (int)old);
michael@0:                         return FALSE;
michael@0:                     } else if(VERBOSE) {
michael@0:                         fprintf(stderr, "duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
michael@0:                             (int)c, printBytes(buffer, bytes, length), (int)old);
michael@0:                     }
michael@0:                     /*
michael@0:                      * Continue after the above warning
michael@0:                      * if the precision of the mapping is unspecified.
michael@0:                      */
michael@0:                 }
michael@0:                 /* reassign the correct action code */
michael@0:                 entry=MBCS_ENTRY_FINAL_SET_ACTION(entry, (MBCS_STATE_VALID_DIRECT_16+(flag==3 ? 2 : 0)+(c>=0x10000 ? 1 : 0)));
michael@0: 
michael@0:                 /* put the code point into bits 22..7 for BMP, c-0x10000 into 26..7 for others */
michael@0:                 if(c<=0xffff) {
michael@0:                     entry=MBCS_ENTRY_FINAL_SET_VALUE(entry, c);
michael@0:                 } else {
michael@0:                     entry=MBCS_ENTRY_FINAL_SET_VALUE(entry, c-0x10000);
michael@0:                 }
michael@0:                 mbcsData->ucm->states.stateTable[state][bytes[i-1]]=entry;
michael@0:                 break;
michael@0:             case MBCS_STATE_VALID_16:
michael@0:                 /* bits 26..16 are not used, 0 */
michael@0:                 /* bits 15..7 contain the final offset delta to one 16-bit code unit */
michael@0:                 offset+=MBCS_ENTRY_FINAL_VALUE_16(entry);
michael@0:                 /* check that this byte sequence is still unassigned */
michael@0:                 if((old=mbcsData->unicodeCodeUnits[offset])!=0xfffe || (old=removeFallback(mbcsData, offset))!=-1) {
michael@0:                     if(flag>=0) {
michael@0:                         fprintf(stderr, "error: duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
michael@0:                             (int)c, printBytes(buffer, bytes, length), (int)old);
michael@0:                         return FALSE;
michael@0:                     } else if(VERBOSE) {
michael@0:                         fprintf(stderr, "duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
michael@0:                             (int)c, printBytes(buffer, bytes, length), (int)old);
michael@0:                     }
michael@0:                 }
michael@0:                 if(c>=0x10000) {
michael@0:                     fprintf(stderr, "error: code point does not fit into valid-16-bit state at U+%04x<->0x%s\n",
michael@0:                         (int)c, printBytes(buffer, bytes, length));
michael@0:                     return FALSE;
michael@0:                 }
michael@0:                 if(flag>0) {
michael@0:                     /* assign only if there is no precise mapping */
michael@0:                     if(mbcsData->unicodeCodeUnits[offset]==0xfffe) {
michael@0:                         return setFallback(mbcsData, offset, c);
michael@0:                     }
michael@0:                 } else {
michael@0:                     mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
michael@0:                 }
michael@0:                 break;
michael@0:             case MBCS_STATE_VALID_16_PAIR:
michael@0:                 /* bits 26..16 are not used, 0 */
michael@0:                 /* bits 15..7 contain the final offset delta to two 16-bit code units */
michael@0:                 offset+=MBCS_ENTRY_FINAL_VALUE_16(entry);
michael@0:                 /* check that this byte sequence is still unassigned */
michael@0:                 old=mbcsData->unicodeCodeUnits[offset];
michael@0:                 if(old<0xfffe) {
michael@0:                     int32_t real;
michael@0:                     if(old<0xd800) {
michael@0:                         real=old;
michael@0:                     } else if(old<=0xdfff) {
michael@0:                         real=0x10000+((old&0x3ff)<<10)+((mbcsData->unicodeCodeUnits[offset+1])&0x3ff);
michael@0:                     } else /* old<=0xe001 */ {
michael@0:                         real=mbcsData->unicodeCodeUnits[offset+1];
michael@0:                     }
michael@0:                     if(flag>=0) {
michael@0:                         fprintf(stderr, "error: duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
michael@0:                             (int)c, printBytes(buffer, bytes, length), (int)real);
michael@0:                         return FALSE;
michael@0:                     } else if(VERBOSE) {
michael@0:                         fprintf(stderr, "duplicate codepage byte sequence at U+%04x<->0x%s see U+%04x\n",
michael@0:                             (int)c, printBytes(buffer, bytes, length), (int)real);
michael@0:                     }
michael@0:                 }
michael@0:                 if(flag>0) {
michael@0:                     /* assign only if there is no precise mapping */
michael@0:                     if(old<=0xdbff || old==0xe000) {
michael@0:                         /* do nothing */
michael@0:                     } else if(c<=0xffff) {
michael@0:                         /* set a BMP fallback code point as a pair with 0xe001 */
michael@0:                         mbcsData->unicodeCodeUnits[offset++]=0xe001;
michael@0:                         mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
michael@0:                     } else {
michael@0:                         /* set a fallback surrogate pair with two second surrogates */
michael@0:                         mbcsData->unicodeCodeUnits[offset++]=(uint16_t)(0xdbc0+(c>>10));
michael@0:                         mbcsData->unicodeCodeUnits[offset]=(uint16_t)(0xdc00+(c&0x3ff));
michael@0:                     }
michael@0:                 } else {
michael@0:                     if(c<0xd800) {
michael@0:                         /* set a BMP code point */
michael@0:                         mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
michael@0:                     } else if(c<=0xffff) {
michael@0:                         /* set a BMP code point above 0xd800 as a pair with 0xe000 */
michael@0:                         mbcsData->unicodeCodeUnits[offset++]=0xe000;
michael@0:                         mbcsData->unicodeCodeUnits[offset]=(uint16_t)c;
michael@0:                     } else {
michael@0:                         /* set a surrogate pair */
michael@0:                         mbcsData->unicodeCodeUnits[offset++]=(uint16_t)(0xd7c0+(c>>10));
michael@0:                         mbcsData->unicodeCodeUnits[offset]=(uint16_t)(0xdc00+(c&0x3ff));
michael@0:                     }
michael@0:                 }
michael@0:                 break;
michael@0:             default:
michael@0:                 /* reserved, must never occur */
michael@0:                 fprintf(stderr, "internal error: byte sequence reached reserved action code, entry 0x%02x: 0x%s (U+%x)\n",
michael@0:                     (int)entry, printBytes(buffer, bytes, length), (int)c);
michael@0:                 return FALSE;
michael@0:             }
michael@0: 
michael@0:             return TRUE;
michael@0:         }
michael@0:     }
michael@0: }
michael@0: 
michael@0: /* is this byte sequence valid? (this is almost the same as MBCSAddToUnicode()) */
michael@0: static UBool
michael@0: MBCSIsValid(NewConverter *cnvData,
michael@0:             const uint8_t *bytes, int32_t length) {
michael@0:     MBCSData *mbcsData=(MBCSData *)cnvData;
michael@0: 
michael@0:     return (UBool)(1==ucm_countChars(&mbcsData->ucm->states, bytes, length));
michael@0: }
michael@0: 
michael@0: static UBool
michael@0: MBCSSingleAddFromUnicode(MBCSData *mbcsData,
michael@0:                          const uint8_t *bytes, int32_t /*length*/,
michael@0:                          UChar32 c,
michael@0:                          int8_t flag) {
michael@0:     uint16_t *stage3, *p;
michael@0:     uint32_t idx;
michael@0:     uint16_t old;
michael@0:     uint8_t b;
michael@0: 
michael@0:     uint32_t blockSize, newTop, i, nextOffset, newBlock, min;
michael@0: 
michael@0:     /* ignore |2 SUB mappings */
michael@0:     if(flag==2) {
michael@0:         return TRUE;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * Walk down the triple-stage compact array ("trie") and
michael@0:      * allocate parts as necessary.
michael@0:      * Note that the first stage 2 and 3 blocks are reserved for all-unassigned mappings.
michael@0:      * We assume that length<=maxCharLength and that c<=0x10ffff.
michael@0:      */
michael@0:     stage3=(uint16_t *)mbcsData->fromUBytes;
michael@0:     b=*bytes;
michael@0: 
michael@0:     /* inspect stage 1 */
michael@0:     idx=c>>MBCS_STAGE_1_SHIFT;
michael@0:     if(mbcsData->utf8Friendly && c<=SBCS_UTF8_MAX) {
michael@0:         nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK&~(MBCS_UTF8_STAGE_3_BLOCKS-1);
michael@0:     } else {
michael@0:         nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK;
michael@0:     }
michael@0:     if(mbcsData->stage1[idx]==MBCS_STAGE_2_ALL_UNASSIGNED_INDEX) {
michael@0:         /* allocate another block in stage 2 */
michael@0:         newBlock=mbcsData->stage2Top;
michael@0:         if(mbcsData->utf8Friendly) {
michael@0:             min=newBlock-nextOffset; /* minimum block start with overlap */
michael@0:             while(min<newBlock && mbcsData->stage2Single[newBlock-1]==0) {
michael@0:                 --newBlock;
michael@0:             }
michael@0:         }
michael@0:         newTop=newBlock+MBCS_STAGE_2_BLOCK_SIZE;
michael@0: 
michael@0:         if(newTop>MBCS_MAX_STAGE_2_TOP) {
michael@0:             fprintf(stderr, "error: too many stage 2 entries at U+%04x<->0x%02x\n", (int)c, b);
michael@0:             return FALSE;
michael@0:         }
michael@0: 
michael@0:         /*
michael@0:          * each stage 2 block contains 64 16-bit words:
michael@0:          * 6 code point bits 9..4 with 1 stage 3 index
michael@0:          */
michael@0:         mbcsData->stage1[idx]=(uint16_t)newBlock;
michael@0:         mbcsData->stage2Top=newTop;
michael@0:     }
michael@0: 
michael@0:     /* inspect stage 2 */
michael@0:     idx=mbcsData->stage1[idx]+nextOffset;
michael@0:     if(mbcsData->utf8Friendly && c<=SBCS_UTF8_MAX) {
michael@0:         /* allocate 64-entry blocks for UTF-8-friendly lookup */
michael@0:         blockSize=MBCS_UTF8_STAGE_3_BLOCK_SIZE;
michael@0:         nextOffset=c&MBCS_UTF8_STAGE_3_BLOCK_MASK;
michael@0:     } else {
michael@0:         blockSize=MBCS_STAGE_3_BLOCK_SIZE;
michael@0:         nextOffset=c&MBCS_STAGE_3_BLOCK_MASK;
michael@0:     }
michael@0:     if(mbcsData->stage2Single[idx]==0) {
michael@0:         /* allocate another block in stage 3 */
michael@0:         newBlock=mbcsData->stage3Top;
michael@0:         if(mbcsData->utf8Friendly) {
michael@0:             min=newBlock-nextOffset; /* minimum block start with overlap */
michael@0:             while(min<newBlock && stage3[newBlock-1]==0) {
michael@0:                 --newBlock;
michael@0:             }
michael@0:         }
michael@0:         newTop=newBlock+blockSize;
michael@0: 
michael@0:         if(newTop>MBCS_STAGE_3_SBCS_SIZE) {
michael@0:             fprintf(stderr, "error: too many code points at U+%04x<->0x%02x\n", (int)c, b);
michael@0:             return FALSE;
michael@0:         }
michael@0:         /* each block has 16 uint16_t entries */
michael@0:         i=idx;
michael@0:         while(newBlock<newTop) {
michael@0:             mbcsData->stage2Single[i++]=(uint16_t)newBlock;
michael@0:             newBlock+=MBCS_STAGE_3_BLOCK_SIZE;
michael@0:         }
michael@0:         mbcsData->stage3Top=newTop; /* ==newBlock */
michael@0:     }
michael@0: 
michael@0:     /* write the codepage entry into stage 3 and get the previous entry */
michael@0:     p=stage3+mbcsData->stage2Single[idx]+nextOffset;
michael@0:     old=*p;
michael@0:     if(flag<=0) {
michael@0:         *p=(uint16_t)(0xf00|b);
michael@0:     } else if(IS_PRIVATE_USE(c)) {
michael@0:         *p=(uint16_t)(0xc00|b);
michael@0:     } else {
michael@0:         *p=(uint16_t)(0x800|b);
michael@0:     }
michael@0: 
michael@0:     /* check that this Unicode code point was still unassigned */
michael@0:     if(old>=0x100) {
michael@0:         if(flag>=0) {
michael@0:             fprintf(stderr, "error: duplicate Unicode code point at U+%04x<->0x%02x see 0x%02x\n",
michael@0:                 (int)c, b, old&0xff);
michael@0:             return FALSE;
michael@0:         } else if(VERBOSE) {
michael@0:             fprintf(stderr, "duplicate Unicode code point at U+%04x<->0x%02x see 0x%02x\n",
michael@0:                 (int)c, b, old&0xff);
michael@0:         }
michael@0:         /* continue after the above warning if the precision of the mapping is unspecified */
michael@0:     }
michael@0: 
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: static UBool
michael@0: MBCSAddFromUnicode(MBCSData *mbcsData,
michael@0:                    const uint8_t *bytes, int32_t length,
michael@0:                    UChar32 c,
michael@0:                    int8_t flag) {
michael@0:     char buffer[10];
michael@0:     const uint8_t *pb;
michael@0:     uint8_t *stage3, *p;
michael@0:     uint32_t idx, b, old, stage3Index;
michael@0:     int32_t maxCharLength;
michael@0: 
michael@0:     uint32_t blockSize, newTop, i, nextOffset, newBlock, min, overlap, maxOverlap;
michael@0: 
michael@0:     maxCharLength=mbcsData->ucm->states.maxCharLength;
michael@0: 
michael@0:     if( mbcsData->ucm->states.outputType==MBCS_OUTPUT_2_SISO &&
michael@0:         (!IGNORE_SISO_CHECK && (*bytes==0xe || *bytes==0xf))
michael@0:     ) {
michael@0:         fprintf(stderr, "error: illegal mapping to SI or SO for SI/SO codepage: U+%04x<->0x%s\n",
michael@0:             (int)c, printBytes(buffer, bytes, length));
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     if(flag==1 && length==1 && *bytes==0) {
michael@0:         fprintf(stderr, "error: unable to encode a |1 fallback from U+%04x to 0x%02x\n",
michael@0:             (int)c, *bytes);
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * Walk down the triple-stage compact array ("trie") and
michael@0:      * allocate parts as necessary.
michael@0:      * Note that the first stage 2 and 3 blocks are reserved for
michael@0:      * all-unassigned mappings.
michael@0:      * We assume that length<=maxCharLength and that c<=0x10ffff.
michael@0:      */
michael@0:     stage3=mbcsData->fromUBytes;
michael@0: 
michael@0:     /* inspect stage 1 */
michael@0:     idx=c>>MBCS_STAGE_1_SHIFT;
michael@0:     if(mbcsData->utf8Friendly && c<=mbcsData->utf8Max) {
michael@0:         nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK&~(MBCS_UTF8_STAGE_3_BLOCKS-1);
michael@0:     } else {
michael@0:         nextOffset=(c>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK;
michael@0:     }
michael@0:     if(mbcsData->stage1[idx]==MBCS_STAGE_2_ALL_UNASSIGNED_INDEX) {
michael@0:         /* allocate another block in stage 2 */
michael@0:         newBlock=mbcsData->stage2Top;
michael@0:         if(mbcsData->utf8Friendly) {
michael@0:             min=newBlock-nextOffset; /* minimum block start with overlap */
michael@0:             while(min<newBlock && mbcsData->stage2[newBlock-1]==0) {
michael@0:                 --newBlock;
michael@0:             }
michael@0:         }
michael@0:         newTop=newBlock+MBCS_STAGE_2_BLOCK_SIZE;
michael@0: 
michael@0:         if(newTop>MBCS_MAX_STAGE_2_TOP) {
michael@0:             fprintf(stderr, "error: too many stage 2 entries at U+%04x<->0x%s\n",
michael@0:                 (int)c, printBytes(buffer, bytes, length));
michael@0:             return FALSE;
michael@0:         }
michael@0: 
michael@0:         /*
michael@0:          * each stage 2 block contains 64 32-bit words:
michael@0:          * 6 code point bits 9..4 with value with bits 31..16 "assigned" flags and bits 15..0 stage 3 index
michael@0:          */
michael@0:         i=idx;
michael@0:         while(newBlock<newTop) {
michael@0:             mbcsData->stage1[i++]=(uint16_t)newBlock;
michael@0:             newBlock+=MBCS_STAGE_2_BLOCK_SIZE;
michael@0:         }
michael@0:         mbcsData->stage2Top=newTop; /* ==newBlock */
michael@0:     }
michael@0: 
michael@0:     /* inspect stage 2 */
michael@0:     idx=mbcsData->stage1[idx]+nextOffset;
michael@0:     if(mbcsData->utf8Friendly && c<=mbcsData->utf8Max) {
michael@0:         /* allocate 64-entry blocks for UTF-8-friendly lookup */
michael@0:         blockSize=MBCS_UTF8_STAGE_3_BLOCK_SIZE*maxCharLength;
michael@0:         nextOffset=c&MBCS_UTF8_STAGE_3_BLOCK_MASK;
michael@0:     } else {
michael@0:         blockSize=MBCS_STAGE_3_BLOCK_SIZE*maxCharLength;
michael@0:         nextOffset=c&MBCS_STAGE_3_BLOCK_MASK;
michael@0:     }
michael@0:     if(mbcsData->stage2[idx]==0) {
michael@0:         /* allocate another block in stage 3 */
michael@0:         newBlock=mbcsData->stage3Top;
michael@0:         if(mbcsData->utf8Friendly && nextOffset>=MBCS_STAGE_3_GRANULARITY) {
michael@0:             /*
michael@0:              * Overlap stage 3 blocks only in multiples of 16-entry blocks
michael@0:              * because of the indexing granularity in stage 2.
michael@0:              */
michael@0:             maxOverlap=(nextOffset&~(MBCS_STAGE_3_GRANULARITY-1))*maxCharLength;
michael@0:             for(overlap=0;
michael@0:                 overlap<maxOverlap && stage3[newBlock-overlap-1]==0;
michael@0:                 ++overlap) {}
michael@0: 
michael@0:             overlap=(overlap/MBCS_STAGE_3_GRANULARITY)/maxCharLength;
michael@0:             overlap=(overlap*MBCS_STAGE_3_GRANULARITY)*maxCharLength;
michael@0: 
michael@0:             newBlock-=overlap;
michael@0:         }
michael@0:         newTop=newBlock+blockSize;
michael@0: 
michael@0:         if(newTop>MBCS_STAGE_3_MBCS_SIZE*(uint32_t)maxCharLength) {
michael@0:             fprintf(stderr, "error: too many code points at U+%04x<->0x%s\n",
michael@0:                 (int)c, printBytes(buffer, bytes, length));
michael@0:             return FALSE;
michael@0:         }
michael@0:         /* each block has 16*maxCharLength bytes */
michael@0:         i=idx;
michael@0:         while(newBlock<newTop) {
michael@0:             mbcsData->stage2[i++]=(newBlock/MBCS_STAGE_3_GRANULARITY)/maxCharLength;
michael@0:             newBlock+=MBCS_STAGE_3_BLOCK_SIZE*maxCharLength;
michael@0:         }
michael@0:         mbcsData->stage3Top=newTop; /* ==newBlock */
michael@0:     }
michael@0: 
michael@0:     stage3Index=MBCS_STAGE_3_GRANULARITY*(uint32_t)(uint16_t)mbcsData->stage2[idx];
michael@0: 
michael@0:     /* Build an alternate, UTF-8-friendly stage table as well. */
michael@0:     if(mbcsData->utf8Friendly && c<=mbcsData->utf8Max) {
michael@0:         /* Overflow for uint16_t entries in stageUTF8? */
michael@0:         if(stage3Index>0xffff) {
michael@0:             /*
michael@0:              * This can occur only if the mapping table is nearly perfectly filled and if
michael@0:              * utf8Max==0xffff.
michael@0:              * (There is no known charset like this. GB 18030 does not map
michael@0:              * surrogate code points and LMBCS does not map 256 PUA code points.)
michael@0:              *
michael@0:              * Otherwise, stage3Index<=MBCS_UTF8_LIMIT<0xffff
michael@0:              * (stage3Index can at most reach exactly MBCS_UTF8_LIMIT)
michael@0:              * because we have a sorted table and there are at most MBCS_UTF8_LIMIT
michael@0:              * mappings with 0<=c<MBCS_UTF8_LIMIT, and there is only also
michael@0:              * the initial all-unassigned block in stage3.
michael@0:              *
michael@0:              * Solution for the overflow: Reduce utf8Max to the next lower value, 0xfeff.
michael@0:              *
michael@0:              * (See svn revision 20866 of the markus/ucnvutf8 feature branch for
michael@0:              * code that causes MBCSAddTable() to rebuild the table not utf8Friendly
michael@0:              * in case of overflow. That code was not tested.)
michael@0:              */
michael@0:             mbcsData->utf8Max=0xfeff;
michael@0:         } else {
michael@0:             /*
michael@0:              * The stage 3 block has been assigned for the regular trie.
michael@0:              * Just copy its index into stageUTF8[], without the granularity.
michael@0:              */
michael@0:             mbcsData->stageUTF8[c>>MBCS_UTF8_STAGE_SHIFT]=(uint16_t)stage3Index;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* write the codepage bytes into stage 3 and get the previous bytes */
michael@0: 
michael@0:     /* assemble the bytes into a single integer */
michael@0:     pb=bytes;
michael@0:     b=0;
michael@0:     switch(length) {
michael@0:     case 4:
michael@0:         b=*pb++;
michael@0:     case 3:
michael@0:         b=(b<<8)|*pb++;
michael@0:     case 2:
michael@0:         b=(b<<8)|*pb++;
michael@0:     case 1:
michael@0:     default:
michael@0:         b=(b<<8)|*pb++;
michael@0:         break;
michael@0:     }
michael@0: 
michael@0:     old=0;
michael@0:     p=stage3+(stage3Index+nextOffset)*maxCharLength;
michael@0:     switch(maxCharLength) {
michael@0:     case 2:
michael@0:         old=*(uint16_t *)p;
michael@0:         *(uint16_t *)p=(uint16_t)b;
michael@0:         break;
michael@0:     case 3:
michael@0:         old=(uint32_t)*p<<16;
michael@0:         *p++=(uint8_t)(b>>16);
michael@0:         old|=(uint32_t)*p<<8;
michael@0:         *p++=(uint8_t)(b>>8);
michael@0:         old|=*p;
michael@0:         *p=(uint8_t)b;
michael@0:         break;
michael@0:     case 4:
michael@0:         old=*(uint32_t *)p;
michael@0:         *(uint32_t *)p=b;
michael@0:         break;
michael@0:     default:
michael@0:         /* will never occur */
michael@0:         break;
michael@0:     }
michael@0: 
michael@0:     /* check that this Unicode code point was still unassigned */
michael@0:     if((mbcsData->stage2[idx+(nextOffset>>MBCS_STAGE_2_SHIFT)]&(1UL<<(16+(c&0xf))))!=0 || old!=0) {
michael@0:         if(flag>=0) {
michael@0:             fprintf(stderr, "error: duplicate Unicode code point at U+%04x<->0x%s see 0x%02x\n",
michael@0:                 (int)c, printBytes(buffer, bytes, length), (int)old);
michael@0:             return FALSE;
michael@0:         } else if(VERBOSE) {
michael@0:             fprintf(stderr, "duplicate Unicode code point at U+%04x<->0x%s see 0x%02x\n",
michael@0:                 (int)c, printBytes(buffer, bytes, length), (int)old);
michael@0:         }
michael@0:         /* continue after the above warning if the precision of the mapping is
michael@0:            unspecified */
michael@0:     }
michael@0:     if(flag<=0) {
michael@0:         /* set the roundtrip flag */
michael@0:         mbcsData->stage2[idx+(nextOffset>>4)]|=(1UL<<(16+(c&0xf)));
michael@0:     }
michael@0: 
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: U_CFUNC UBool
michael@0: MBCSOkForBaseFromUnicode(const MBCSData *mbcsData,
michael@0:                          const uint8_t *bytes, int32_t length,
michael@0:                          UChar32 c, int8_t flag) {
michael@0:     /*
michael@0:      * A 1:1 mapping does not fit into the MBCS base table's fromUnicode table under
michael@0:      * the following conditions:
michael@0:      *
michael@0:      * - a |2 SUB mapping for <subchar1> (no base table data structure for them)
michael@0:      * - a |1 fallback to 0x00 (result value 0, indistinguishable from unmappable entry)
michael@0:      * - a multi-byte mapping with leading 0x00 bytes (no explicit length field)
michael@0:      *
michael@0:      * Some of these tests are redundant with ucm_mappingType().
michael@0:      */
michael@0:     if( (flag==2 && length==1) ||
michael@0:         (flag==1 && bytes[0]==0) || /* testing length==1 would be redundant with the next test */
michael@0:         (flag<=1 && length>1 && bytes[0]==0)
michael@0:     ) {
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * Additional restrictions for UTF-8-friendly fromUnicode tables,
michael@0:      * for code points up to the maximum optimized one:
michael@0:      *
michael@0:      * - any mapping to 0x00 (result value 0, indistinguishable from unmappable entry)
michael@0:      * - any |1 fallback (no roundtrip flags in the optimized table)
michael@0:      */
michael@0:     if(mbcsData->utf8Friendly && flag<=1 && c<=mbcsData->utf8Max && (bytes[0]==0 || flag==1)) {
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * If we omit the fromUnicode data, we can only store roundtrips there
michael@0:      * because only they are recoverable from the toUnicode data.
michael@0:      * Fallbacks must go into the extension table.
michael@0:      */
michael@0:     if(mbcsData->omitFromU && flag!=0) {
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     /* All other mappings do fit into the base table. */
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: /* we can assume that the table only contains 1:1 mappings with <=4 bytes each */
michael@0: static UBool
michael@0: MBCSAddTable(NewConverter *cnvData, UCMTable *table, UConverterStaticData *staticData) {
michael@0:     MBCSData *mbcsData;
michael@0:     UCMapping *m;
michael@0:     UChar32 c;
michael@0:     int32_t i, maxCharLength;
michael@0:     int8_t f;
michael@0:     UBool isOK, utf8Friendly;
michael@0: 
michael@0:     staticData->unicodeMask=table->unicodeMask;
michael@0:     if(staticData->unicodeMask==3) {
michael@0:         fprintf(stderr, "error: contains mappings for both supplementary and surrogate code points\n");
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     staticData->conversionType=UCNV_MBCS;
michael@0: 
michael@0:     mbcsData=(MBCSData *)cnvData;
michael@0:     maxCharLength=mbcsData->ucm->states.maxCharLength;
michael@0: 
michael@0:     /*
michael@0:      * Generation of UTF-8-friendly data requires
michael@0:      * a sorted table, which makeconv generates when explicit precision
michael@0:      * indicators are used.
michael@0:      */
michael@0:     mbcsData->utf8Friendly=utf8Friendly=(UBool)((table->flagsType&UCM_FLAGS_EXPLICIT)!=0);
michael@0:     if(utf8Friendly) {
michael@0:         mbcsData->utf8Max=MBCS_UTF8_MAX;
michael@0:         if(SMALL && maxCharLength>1) {
michael@0:             mbcsData->omitFromU=TRUE;
michael@0:         }
michael@0:     } else {
michael@0:         mbcsData->utf8Max=0;
michael@0:         if(SMALL && maxCharLength>1) {
michael@0:             fprintf(stderr,
michael@0:                 "makeconv warning: --small not available for .ucm files without |0 etc.\n");
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     if(!MBCSStartMappings(mbcsData)) {
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     staticData->hasFromUnicodeFallback=FALSE;
michael@0:     staticData->hasToUnicodeFallback=FALSE;
michael@0: 
michael@0:     isOK=TRUE;
michael@0: 
michael@0:     m=table->mappings;
michael@0:     for(i=0; i<table->mappingsLength; ++m, ++i) {
michael@0:         c=m->u;
michael@0:         f=m->f;
michael@0: 
michael@0:         /*
michael@0:          * Small optimization for --small .cnv files:
michael@0:          *
michael@0:          * If there are fromUnicode mappings above MBCS_UTF8_MAX,
michael@0:          * then the file size will be smaller if we make utf8Max larger
michael@0:          * because the size increase in stageUTF8 will be more than balanced by
michael@0:          * how much less of stage2 needs to be stored.
michael@0:          *
michael@0:          * There is no point in doing this incrementally because stageUTF8
michael@0:          * uses so much less space per block than stage2,
michael@0:          * so we immediately increase utf8Max to 0xffff.
michael@0:          *
michael@0:          * Do not increase utf8Max if it is already at 0xfeff because MBCSAddFromUnicode()
michael@0:          * sets it to that value when stageUTF8 overflows.
michael@0:          */
michael@0:         if( mbcsData->omitFromU && f<=1 &&
michael@0:             mbcsData->utf8Max<c && c<=0xffff &&
michael@0:             mbcsData->utf8Max<0xfeff
michael@0:         ) {
michael@0:             mbcsData->utf8Max=0xffff;
michael@0:         }
michael@0: 
michael@0:         switch(f) {
michael@0:         case -1:
michael@0:             /* there was no precision/fallback indicator */
michael@0:             /* fall through to set the mappings */
michael@0:         case 0:
michael@0:             /* set roundtrip mappings */
michael@0:             isOK&=MBCSAddToUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
michael@0: 
michael@0:             if(maxCharLength==1) {
michael@0:                 isOK&=MBCSSingleAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
michael@0:             } else if(MBCSOkForBaseFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f)) {
michael@0:                 isOK&=MBCSAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
michael@0:             } else {
michael@0:                 m->f|=MBCS_FROM_U_EXT_FLAG;
michael@0:                 m->moveFlag=UCM_MOVE_TO_EXT;
michael@0:             }
michael@0:             break;
michael@0:         case 1:
michael@0:             /* set only a fallback mapping from Unicode to codepage */
michael@0:             if(maxCharLength==1) {
michael@0:                 staticData->hasFromUnicodeFallback=TRUE;
michael@0:                 isOK&=MBCSSingleAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
michael@0:             } else if(MBCSOkForBaseFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f)) {
michael@0:                 staticData->hasFromUnicodeFallback=TRUE;
michael@0:                 isOK&=MBCSAddFromUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
michael@0:             } else {
michael@0:                 m->f|=MBCS_FROM_U_EXT_FLAG;
michael@0:                 m->moveFlag=UCM_MOVE_TO_EXT;
michael@0:             }
michael@0:             break;
michael@0:         case 2:
michael@0:             /* ignore |2 SUB mappings, except to move <subchar1> mappings to the extension table */
michael@0:             if(maxCharLength>1 && m->bLen==1) {
michael@0:                 m->f|=MBCS_FROM_U_EXT_FLAG;
michael@0:                 m->moveFlag=UCM_MOVE_TO_EXT;
michael@0:             }
michael@0:             break;
michael@0:         case 3:
michael@0:             /* set only a fallback mapping from codepage to Unicode */
michael@0:             staticData->hasToUnicodeFallback=TRUE;
michael@0:             isOK&=MBCSAddToUnicode(mbcsData, m->b.bytes, m->bLen, c, f);
michael@0:             break;
michael@0:         case 4:
michael@0:             /* move "good one-way" mappings to the extension table */
michael@0:             m->f|=MBCS_FROM_U_EXT_FLAG;
michael@0:             m->moveFlag=UCM_MOVE_TO_EXT;
michael@0:             break;
michael@0:         default:
michael@0:             /* will not occur because the parser checked it already */
michael@0:             fprintf(stderr, "error: illegal fallback indicator %d\n", f);
michael@0:             return FALSE;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     MBCSPostprocess(mbcsData, staticData);
michael@0: 
michael@0:     return isOK;
michael@0: }
michael@0: 
michael@0: static UBool
michael@0: transformEUC(MBCSData *mbcsData) {
michael@0:     uint8_t *p8;
michael@0:     uint32_t i, value, oldLength, old3Top;
michael@0:     uint8_t b;
michael@0: 
michael@0:     oldLength=mbcsData->ucm->states.maxCharLength;
michael@0:     if(oldLength<3) {
michael@0:         return FALSE;
michael@0:     }
michael@0: 
michael@0:     old3Top=mbcsData->stage3Top;
michael@0: 
michael@0:     /* careful: 2-byte and 4-byte codes are stored in platform endianness! */
michael@0: 
michael@0:     /* test if all first bytes are in {0, 0x8e, 0x8f} */
michael@0:     p8=mbcsData->fromUBytes;
michael@0: 
michael@0: #if !U_IS_BIG_ENDIAN
michael@0:     if(oldLength==4) {
michael@0:         p8+=3;
michael@0:     }
michael@0: #endif
michael@0: 
michael@0:     for(i=0; i<old3Top; i+=oldLength) {
michael@0:         b=p8[i];
michael@0:         if(b!=0 && b!=0x8e && b!=0x8f) {
michael@0:             /* some first byte does not fit the EUC pattern, nothing to be done */
michael@0:             return FALSE;
michael@0:         }
michael@0:     }
michael@0:     /* restore p if it was modified above */
michael@0:     p8=mbcsData->fromUBytes;
michael@0: 
michael@0:     /* modify outputType and adjust stage3Top */
michael@0:     mbcsData->ucm->states.outputType=(int8_t)(MBCS_OUTPUT_3_EUC+oldLength-3);
michael@0:     mbcsData->stage3Top=(old3Top*(oldLength-1))/oldLength;
michael@0: 
michael@0:     /*
michael@0:      * EUC-encode all byte sequences;
michael@0:      * see "CJKV Information Processing" (1st ed. 1999) from Ken Lunde, O'Reilly,
michael@0:      * p. 161 in chapter 4 "Encoding Methods"
michael@0:      *
michael@0:      * This also must reverse the byte order if the platform is little-endian!
michael@0:      */
michael@0:     if(oldLength==3) {
michael@0:         uint16_t *q=(uint16_t *)p8;
michael@0:         for(i=0; i<old3Top; i+=oldLength) {
michael@0:             b=*p8;
michael@0:             if(b==0) {
michael@0:                 /* short sequences are stored directly */
michael@0:                 /* code set 0 or 1 */
michael@0:                 (*q++)=(uint16_t)((p8[1]<<8)|p8[2]);
michael@0:             } else if(b==0x8e) {
michael@0:                 /* code set 2 */
michael@0:                 (*q++)=(uint16_t)(((p8[1]&0x7f)<<8)|p8[2]);
michael@0:             } else /* b==0x8f */ {
michael@0:                 /* code set 3 */
michael@0:                 (*q++)=(uint16_t)((p8[1]<<8)|(p8[2]&0x7f));
michael@0:             }
michael@0:             p8+=3;
michael@0:         }
michael@0:     } else /* oldLength==4 */ {
michael@0:         uint8_t *q=p8;
michael@0:         uint32_t *p32=(uint32_t *)p8;
michael@0:         for(i=0; i<old3Top; i+=4) {
michael@0:             value=(*p32++);
michael@0:             if(value<=0xffffff) {
michael@0:                 /* short sequences are stored directly */
michael@0:                 /* code set 0 or 1 */
michael@0:                 (*q++)=(uint8_t)(value>>16);
michael@0:                 (*q++)=(uint8_t)(value>>8);
michael@0:                 (*q++)=(uint8_t)value;
michael@0:             } else if(value<=0x8effffff) {
michael@0:                 /* code set 2 */
michael@0:                 (*q++)=(uint8_t)((value>>16)&0x7f);
michael@0:                 (*q++)=(uint8_t)(value>>8);
michael@0:                 (*q++)=(uint8_t)value;
michael@0:             } else /* first byte is 0x8f */ {
michael@0:                 /* code set 3 */
michael@0:                 (*q++)=(uint8_t)(value>>16);
michael@0:                 (*q++)=(uint8_t)((value>>8)&0x7f);
michael@0:                 (*q++)=(uint8_t)value;
michael@0:             }
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     return TRUE;
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Compact stage 2 for SBCS by overlapping adjacent stage 2 blocks as far
michael@0:  * as possible. Overlapping is done on unassigned head and tail
michael@0:  * parts of blocks in steps of MBCS_STAGE_2_MULTIPLIER.
michael@0:  * Stage 1 indexes need to be adjusted accordingly.
michael@0:  * This function is very similar to genprops/store.c/compactStage().
michael@0:  */
michael@0: static void
michael@0: singleCompactStage2(MBCSData *mbcsData) {
michael@0:     /* this array maps the ordinal number of a stage 2 block to its new stage 1 index */
michael@0:     uint16_t map[MBCS_STAGE_2_MAX_BLOCKS];
michael@0:     uint16_t i, start, prevEnd, newStart;
michael@0: 
michael@0:     /* enter the all-unassigned first stage 2 block into the map */
michael@0:     map[0]=MBCS_STAGE_2_ALL_UNASSIGNED_INDEX;
michael@0: 
michael@0:     /* begin with the first block after the all-unassigned one */
michael@0:     start=newStart=MBCS_STAGE_2_FIRST_ASSIGNED;
michael@0:     while(start<mbcsData->stage2Top) {
michael@0:         prevEnd=(uint16_t)(newStart-1);
michael@0: 
michael@0:         /* find the size of the overlap */
michael@0:         for(i=0; i<MBCS_STAGE_2_BLOCK_SIZE && mbcsData->stage2Single[start+i]==0 && mbcsData->stage2Single[prevEnd-i]==0; ++i) {}
michael@0: 
michael@0:         if(i>0) {
michael@0:             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=(uint16_t)(newStart-i);
michael@0: 
michael@0:             /* move the non-overlapping indexes to their new positions */
michael@0:             start+=i;
michael@0:             for(i=(uint16_t)(MBCS_STAGE_2_BLOCK_SIZE-i); i>0; --i) {
michael@0:                 mbcsData->stage2Single[newStart++]=mbcsData->stage2Single[start++];
michael@0:             }
michael@0:         } else if(newStart<start) {
michael@0:             /* move the indexes to their new positions */
michael@0:             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=newStart;
michael@0:             for(i=MBCS_STAGE_2_BLOCK_SIZE; i>0; --i) {
michael@0:                 mbcsData->stage2Single[newStart++]=mbcsData->stage2Single[start++];
michael@0:             }
michael@0:         } else /* no overlap && newStart==start */ {
michael@0:             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=start;
michael@0:             start=newStart+=MBCS_STAGE_2_BLOCK_SIZE;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* adjust stage2Top */
michael@0:     if(VERBOSE && newStart<mbcsData->stage2Top) {
michael@0:         printf("compacting stage 2 from stage2Top=0x%lx to 0x%lx, saving %ld bytes\n",
michael@0:                 (unsigned long)mbcsData->stage2Top, (unsigned long)newStart,
michael@0:                 (long)(mbcsData->stage2Top-newStart)*2);
michael@0:     }
michael@0:     mbcsData->stage2Top=newStart;
michael@0: 
michael@0:     /* now adjust stage 1 */
michael@0:     for(i=0; i<MBCS_STAGE_1_SIZE; ++i) {
michael@0:         mbcsData->stage1[i]=map[mbcsData->stage1[i]>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT];
michael@0:     }
michael@0: }
michael@0: 
michael@0: /* Compact stage 3 for SBCS - same algorithm as above. */
michael@0: static void
michael@0: singleCompactStage3(MBCSData *mbcsData) {
michael@0:     uint16_t *stage3=(uint16_t *)mbcsData->fromUBytes;
michael@0: 
michael@0:     /* this array maps the ordinal number of a stage 3 block to its new stage 2 index */
michael@0:     uint16_t map[0x1000];
michael@0:     uint16_t i, start, prevEnd, newStart;
michael@0: 
michael@0:     /* enter the all-unassigned first stage 3 block into the map */
michael@0:     map[0]=0;
michael@0: 
michael@0:     /* begin with the first block after the all-unassigned one */
michael@0:     start=newStart=16;
michael@0:     while(start<mbcsData->stage3Top) {
michael@0:         prevEnd=(uint16_t)(newStart-1);
michael@0: 
michael@0:         /* find the size of the overlap */
michael@0:         for(i=0; i<16 && stage3[start+i]==0 && stage3[prevEnd-i]==0; ++i) {}
michael@0: 
michael@0:         if(i>0) {
michael@0:             map[start>>4]=(uint16_t)(newStart-i);
michael@0: 
michael@0:             /* move the non-overlapping indexes to their new positions */
michael@0:             start+=i;
michael@0:             for(i=(uint16_t)(16-i); i>0; --i) {
michael@0:                 stage3[newStart++]=stage3[start++];
michael@0:             }
michael@0:         } else if(newStart<start) {
michael@0:             /* move the indexes to their new positions */
michael@0:             map[start>>4]=newStart;
michael@0:             for(i=16; i>0; --i) {
michael@0:                 stage3[newStart++]=stage3[start++];
michael@0:             }
michael@0:         } else /* no overlap && newStart==start */ {
michael@0:             map[start>>4]=start;
michael@0:             start=newStart+=16;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* adjust stage3Top */
michael@0:     if(VERBOSE && newStart<mbcsData->stage3Top) {
michael@0:         printf("compacting stage 3 from stage3Top=0x%lx to 0x%lx, saving %ld bytes\n",
michael@0:                 (unsigned long)mbcsData->stage3Top, (unsigned long)newStart,
michael@0:                 (long)(mbcsData->stage3Top-newStart)*2);
michael@0:     }
michael@0:     mbcsData->stage3Top=newStart;
michael@0: 
michael@0:     /* now adjust stage 2 */
michael@0:     for(i=0; i<mbcsData->stage2Top; ++i) {
michael@0:         mbcsData->stage2Single[i]=map[mbcsData->stage2Single[i]>>4];
michael@0:     }
michael@0: }
michael@0: 
michael@0: /*
michael@0:  * Compact stage 2 by overlapping adjacent stage 2 blocks as far
michael@0:  * as possible. Overlapping is done on unassigned head and tail
michael@0:  * parts of blocks in steps of MBCS_STAGE_2_MULTIPLIER.
michael@0:  * Stage 1 indexes need to be adjusted accordingly.
michael@0:  * This function is very similar to genprops/store.c/compactStage().
michael@0:  */
michael@0: static void
michael@0: compactStage2(MBCSData *mbcsData) {
michael@0:     /* this array maps the ordinal number of a stage 2 block to its new stage 1 index */
michael@0:     uint16_t map[MBCS_STAGE_2_MAX_BLOCKS];
michael@0:     uint16_t i, start, prevEnd, newStart;
michael@0: 
michael@0:     /* enter the all-unassigned first stage 2 block into the map */
michael@0:     map[0]=MBCS_STAGE_2_ALL_UNASSIGNED_INDEX;
michael@0: 
michael@0:     /* begin with the first block after the all-unassigned one */
michael@0:     start=newStart=MBCS_STAGE_2_FIRST_ASSIGNED;
michael@0:     while(start<mbcsData->stage2Top) {
michael@0:         prevEnd=(uint16_t)(newStart-1);
michael@0: 
michael@0:         /* find the size of the overlap */
michael@0:         for(i=0; i<MBCS_STAGE_2_BLOCK_SIZE && mbcsData->stage2[start+i]==0 && mbcsData->stage2[prevEnd-i]==0; ++i) {}
michael@0: 
michael@0:         if(i>0) {
michael@0:             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=(uint16_t)(newStart-i);
michael@0: 
michael@0:             /* move the non-overlapping indexes to their new positions */
michael@0:             start+=i;
michael@0:             for(i=(uint16_t)(MBCS_STAGE_2_BLOCK_SIZE-i); i>0; --i) {
michael@0:                 mbcsData->stage2[newStart++]=mbcsData->stage2[start++];
michael@0:             }
michael@0:         } else if(newStart<start) {
michael@0:             /* move the indexes to their new positions */
michael@0:             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=newStart;
michael@0:             for(i=MBCS_STAGE_2_BLOCK_SIZE; i>0; --i) {
michael@0:                 mbcsData->stage2[newStart++]=mbcsData->stage2[start++];
michael@0:             }
michael@0:         } else /* no overlap && newStart==start */ {
michael@0:             map[start>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT]=start;
michael@0:             start=newStart+=MBCS_STAGE_2_BLOCK_SIZE;
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     /* adjust stage2Top */
michael@0:     if(VERBOSE && newStart<mbcsData->stage2Top) {
michael@0:         printf("compacting stage 2 from stage2Top=0x%lx to 0x%lx, saving %ld bytes\n",
michael@0:                 (unsigned long)mbcsData->stage2Top, (unsigned long)newStart,
michael@0:                 (long)(mbcsData->stage2Top-newStart)*4);
michael@0:     }
michael@0:     mbcsData->stage2Top=newStart;
michael@0: 
michael@0:     /* now adjust stage 1 */
michael@0:     for(i=0; i<MBCS_STAGE_1_SIZE; ++i) {
michael@0:         mbcsData->stage1[i]=map[mbcsData->stage1[i]>>MBCS_STAGE_2_BLOCK_SIZE_SHIFT];
michael@0:     }
michael@0: }
michael@0: 
michael@0: static void
michael@0: MBCSPostprocess(MBCSData *mbcsData, const UConverterStaticData * /*staticData*/) {
michael@0:     UCMStates *states;
michael@0:     int32_t maxCharLength, stage3Width;
michael@0: 
michael@0:     states=&mbcsData->ucm->states;
michael@0:     stage3Width=maxCharLength=states->maxCharLength;
michael@0: 
michael@0:     ucm_optimizeStates(states,
michael@0:                        &mbcsData->unicodeCodeUnits,
michael@0:                        mbcsData->toUFallbacks, mbcsData->countToUFallbacks,
michael@0:                        VERBOSE);
michael@0: 
michael@0:     /* try to compact the fromUnicode tables */
michael@0:     if(transformEUC(mbcsData)) {
michael@0:         --stage3Width;
michael@0:     }
michael@0: 
michael@0:     /*
michael@0:      * UTF-8-friendly tries are built precompacted, to cope with variable
michael@0:      * stage 3 allocation block sizes.
michael@0:      *
michael@0:      * Tables without precision indicators cannot be built that way,
michael@0:      * because if a block was overlapped with a previous one, then a smaller
michael@0:      * code point for the same block would not fit.
michael@0:      * Therefore, such tables are not marked UTF-8-friendly and must be
michael@0:      * compacted after all mappings are entered.
michael@0:      */
michael@0:     if(!mbcsData->utf8Friendly) {
michael@0:         if(maxCharLength==1) {
michael@0:             singleCompactStage3(mbcsData);
michael@0:             singleCompactStage2(mbcsData);
michael@0:         } else {
michael@0:             compactStage2(mbcsData);
michael@0:         }
michael@0:     }
michael@0: 
michael@0:     if(VERBOSE) {
michael@0:         /*uint32_t c, i1, i2, i2Limit, i3;*/
michael@0: 
michael@0:         printf("fromUnicode number of uint%s_t in stage 2: 0x%lx=%lu\n",
michael@0:                maxCharLength==1 ? "16" : "32",
michael@0:                (unsigned long)mbcsData->stage2Top,
michael@0:                (unsigned long)mbcsData->stage2Top);
michael@0:         printf("fromUnicode number of %d-byte stage 3 mapping entries: 0x%lx=%lu\n",
michael@0:                (int)stage3Width,
michael@0:                (unsigned long)mbcsData->stage3Top/stage3Width,
michael@0:                (unsigned long)mbcsData->stage3Top/stage3Width);
michael@0: #if 0
michael@0:         c=0;
michael@0:         for(i1=0; i1<MBCS_STAGE_1_SIZE; ++i1) {
michael@0:             i2=mbcsData->stage1[i1];
michael@0:             if(i2==0) {
michael@0:                 c+=MBCS_STAGE_2_BLOCK_SIZE*MBCS_STAGE_3_BLOCK_SIZE;
michael@0:                 continue;
michael@0:             }
michael@0:             for(i2Limit=i2+MBCS_STAGE_2_BLOCK_SIZE; i2<i2Limit; ++i2) {
michael@0:                 if(maxCharLength==1) {
michael@0:                     i3=mbcsData->stage2Single[i2];
michael@0:                 } else {
michael@0:                     i3=(uint16_t)mbcsData->stage2[i2];
michael@0:                 }
michael@0:                 if(i3==0) {
michael@0:                     c+=MBCS_STAGE_3_BLOCK_SIZE;
michael@0:                     continue;
michael@0:                 }
michael@0:                 printf("U+%04lx i1=0x%02lx i2=0x%04lx i3=0x%04lx\n",
michael@0:                        (unsigned long)c,
michael@0:                        (unsigned long)i1,
michael@0:                        (unsigned long)i2,
michael@0:                        (unsigned long)i3);
michael@0:                 c+=MBCS_STAGE_3_BLOCK_SIZE;
michael@0:             }
michael@0:         }
michael@0: #endif
michael@0:     }
michael@0: }
michael@0: 
michael@0: static uint32_t
michael@0: MBCSWrite(NewConverter *cnvData, const UConverterStaticData *staticData,
michael@0:           UNewDataMemory *pData, int32_t tableType) {
michael@0:     MBCSData *mbcsData=(MBCSData *)cnvData;
michael@0:     uint32_t stage2Start, stage2Length;
michael@0:     uint32_t top, stageUTF8Length=0;
michael@0:     int32_t i, stage1Top;
michael@0:     uint32_t headerLength;
michael@0: 
michael@0:     _MBCSHeader header=UCNV_MBCS_HEADER_INITIALIZER;
michael@0: 
michael@0:     stage2Length=mbcsData->stage2Top;
michael@0:     if(mbcsData->omitFromU) {
michael@0:         /* find how much of stage2 can be omitted */
michael@0:         int32_t utf8Limit=(int32_t)mbcsData->utf8Max+1;
michael@0:         uint32_t st2=0; /*initialized it to avoid compiler warnings */
michael@0: 
michael@0:         i=utf8Limit>>MBCS_STAGE_1_SHIFT;
michael@0:         if((utf8Limit&((1<<MBCS_STAGE_1_SHIFT)-1))!=0 && (st2=mbcsData->stage1[i])!=0) {
michael@0:             /* utf8Limit is in the middle of an existing stage 2 block */
michael@0:             stage2Start=st2+((utf8Limit>>MBCS_STAGE_2_SHIFT)&MBCS_STAGE_2_BLOCK_MASK);
michael@0:         } else {
michael@0:             /* find the last stage2 block with mappings before utf8Limit */
michael@0:             while(i>0 && (st2=mbcsData->stage1[--i])==0) {}
michael@0:             /* stage2 up to the end of this block corresponds to stageUTF8 */
michael@0:             stage2Start=st2+MBCS_STAGE_2_BLOCK_SIZE;
michael@0:         }
michael@0:         header.options|=MBCS_OPT_NO_FROM_U;
michael@0:         header.fullStage2Length=stage2Length;
michael@0:         stage2Length-=stage2Start;
michael@0:         if(VERBOSE) {
michael@0:             printf("+ omitting %lu out of %lu stage2 entries and %lu fromUBytes\n",
michael@0:                     (unsigned long)stage2Start,
michael@0:                     (unsigned long)mbcsData->stage2Top,
michael@0:                     (unsigned long)mbcsData->stage3Top);
michael@0:             printf("+ total size savings: %lu bytes\n", (unsigned long)stage2Start*4+mbcsData->stage3Top);
michael@0:         }
michael@0:     } else {
michael@0:         stage2Start=0;
michael@0:     }
michael@0: 
michael@0:     if(staticData->unicodeMask&UCNV_HAS_SUPPLEMENTARY) {
michael@0:         stage1Top=MBCS_STAGE_1_SIZE; /* 0x440==1088 */
michael@0:     } else {
michael@0:         stage1Top=0x40; /* 0x40==64 */
michael@0:     }
michael@0: 
michael@0:     /* adjust stage 1 entries to include the size of stage 1 in the offsets to stage 2 */
michael@0:     if(mbcsData->ucm->states.maxCharLength==1) {
michael@0:         for(i=0; i<stage1Top; ++i) {
michael@0:             mbcsData->stage1[i]+=(uint16_t)stage1Top;
michael@0:         }
michael@0: 
michael@0:         /* stage2Top/Length have counted 16-bit results, now we need to count bytes */
michael@0:         /* also round up to a multiple of 4 bytes */
michael@0:         stage2Length=(stage2Length*2+1)&~1;
michael@0: 
michael@0:         /* stage3Top has counted 16-bit results, now we need to count bytes */
michael@0:         mbcsData->stage3Top*=2;
michael@0: 
michael@0:         if(mbcsData->utf8Friendly) {
michael@0:             header.version[2]=(uint8_t)(SBCS_UTF8_MAX>>8); /* store 0x1f for max==0x1fff */
michael@0:         }
michael@0:     } else {
michael@0:         for(i=0; i<stage1Top; ++i) {
michael@0:             mbcsData->stage1[i]+=(uint16_t)stage1Top/2; /* stage 2 contains 32-bit entries, stage 1 16-bit entries */
michael@0:         }
michael@0: 
michael@0:         /* stage2Top/Length have counted 32-bit results, now we need to count bytes */
michael@0:         stage2Length*=4;
michael@0:         /* leave stage2Start counting 32-bit units */
michael@0: 
michael@0:         if(mbcsData->utf8Friendly) {
michael@0:             stageUTF8Length=(mbcsData->utf8Max+1)>>MBCS_UTF8_STAGE_SHIFT;
michael@0:             header.version[2]=(uint8_t)(mbcsData->utf8Max>>8); /* store 0xd7 for max==0xd7ff */
michael@0:         }
michael@0: 
michael@0:         /* stage3Top has already counted bytes */
michael@0:     }
michael@0: 
michael@0:     /* round up stage3Top so that the sizes of all data blocks are multiples of 4 */
michael@0:     mbcsData->stage3Top=(mbcsData->stage3Top+3)&~3;
michael@0: 
michael@0:     /* fill the header */
michael@0:     if(header.options&MBCS_OPT_INCOMPATIBLE_MASK) {
michael@0:         header.version[0]=5;
michael@0:         if(header.options&MBCS_OPT_NO_FROM_U) {
michael@0:             headerLength=10;  /* include fullStage2Length */
michael@0:         } else {
michael@0:             headerLength=MBCS_HEADER_V5_MIN_LENGTH;  /* 9 */
michael@0:         }
michael@0:     } else {
michael@0:         header.version[0]=4;
michael@0:         headerLength=MBCS_HEADER_V4_LENGTH;  /* 8 */
michael@0:     }
michael@0:     header.version[1]=4;
michael@0:     /* header.version[2] set above for utf8Friendly data */
michael@0: 
michael@0:     header.options|=(uint32_t)headerLength;
michael@0: 
michael@0:     header.countStates=mbcsData->ucm->states.countStates;
michael@0:     header.countToUFallbacks=mbcsData->countToUFallbacks;
michael@0: 
michael@0:     header.offsetToUCodeUnits=
michael@0:         headerLength*4+
michael@0:         mbcsData->ucm->states.countStates*1024+
michael@0:         mbcsData->countToUFallbacks*sizeof(_MBCSToUFallback);
michael@0:     header.offsetFromUTable=
michael@0:         header.offsetToUCodeUnits+
michael@0:         mbcsData->ucm->states.countToUCodeUnits*2;
michael@0:     header.offsetFromUBytes=
michael@0:         header.offsetFromUTable+
michael@0:         stage1Top*2+
michael@0:         stage2Length;
michael@0:     header.fromUBytesLength=mbcsData->stage3Top;
michael@0: 
michael@0:     top=header.offsetFromUBytes+stageUTF8Length*2;
michael@0:     if(!(header.options&MBCS_OPT_NO_FROM_U)) {
michael@0:         top+=header.fromUBytesLength;
michael@0:     }
michael@0: 
michael@0:     header.flags=(uint8_t)(mbcsData->ucm->states.outputType);
michael@0: 
michael@0:     if(tableType&TABLE_EXT) {
michael@0:         if(top>0xffffff) {
michael@0:             fprintf(stderr, "error: offset 0x%lx to extension table exceeds 0xffffff\n", (long)top);
michael@0:             return 0;
michael@0:         }
michael@0: 
michael@0:         header.flags|=top<<8;
michael@0:     }
michael@0: 
michael@0:     /* write the MBCS data */
michael@0:     udata_writeBlock(pData, &header, headerLength*4);
michael@0:     udata_writeBlock(pData, mbcsData->ucm->states.stateTable, header.countStates*1024);
michael@0:     udata_writeBlock(pData, mbcsData->toUFallbacks, mbcsData->countToUFallbacks*sizeof(_MBCSToUFallback));
michael@0:     udata_writeBlock(pData, mbcsData->unicodeCodeUnits, mbcsData->ucm->states.countToUCodeUnits*2);
michael@0:     udata_writeBlock(pData, mbcsData->stage1, stage1Top*2);
michael@0:     if(mbcsData->ucm->states.maxCharLength==1) {
michael@0:         udata_writeBlock(pData, mbcsData->stage2Single+stage2Start, stage2Length);
michael@0:     } else {
michael@0:         udata_writeBlock(pData, mbcsData->stage2+stage2Start, stage2Length);
michael@0:     }
michael@0:     if(!(header.options&MBCS_OPT_NO_FROM_U)) {
michael@0:         udata_writeBlock(pData, mbcsData->fromUBytes, mbcsData->stage3Top);
michael@0:     }
michael@0: 
michael@0:     if(stageUTF8Length>0) {
michael@0:         udata_writeBlock(pData, mbcsData->stageUTF8, stageUTF8Length*2);
michael@0:     }
michael@0: 
michael@0:     /* return the number of bytes that should have been written */
michael@0:     return top;
michael@0: }