michael@0: /*
michael@0: *******************************************************************************
michael@0: *
michael@0: *   Copyright (C) 2008-2011, International Business Machines
michael@0: *   Corporation, Google and others.  All Rights Reserved.
michael@0: *
michael@0: *******************************************************************************
michael@0: */
michael@0: // Author : eldawy@google.com (Mohamed Eldawy)
michael@0: // ucnvsel.cpp
michael@0: //
michael@0: // Purpose: To generate a list of encodings capable of handling
michael@0: // a given Unicode text
michael@0: //
michael@0: // Started 09-April-2008
michael@0: 
michael@0: /**
michael@0:  * \file
michael@0:  *
michael@0:  * This is an implementation of an encoding selector.
michael@0:  * The goal is, given a unicode string, find the encodings
michael@0:  * this string can be mapped to. To make processing faster
michael@0:  * a trie is built when you call ucnvsel_open() that
michael@0:  * stores all encodings a codepoint can map to
michael@0:  */
michael@0: 
michael@0: #include "unicode/ucnvsel.h"
michael@0: 
michael@0: #if !UCONFIG_NO_CONVERSION
michael@0: 
michael@0: #include <string.h>
michael@0: 
michael@0: #include "unicode/uchar.h"
michael@0: #include "unicode/uniset.h"
michael@0: #include "unicode/ucnv.h"
michael@0: #include "unicode/ustring.h"
michael@0: #include "unicode/uchriter.h"
michael@0: #include "utrie2.h"
michael@0: #include "propsvec.h"
michael@0: #include "uassert.h"
michael@0: #include "ucmndata.h"
michael@0: #include "uenumimp.h"
michael@0: #include "cmemory.h"
michael@0: #include "cstring.h"
michael@0: 
michael@0: U_NAMESPACE_USE
michael@0: 
michael@0: struct UConverterSelector {
michael@0:   UTrie2 *trie;              // 16 bit trie containing offsets into pv
michael@0:   uint32_t* pv;              // table of bits!
michael@0:   int32_t pvCount;
michael@0:   char** encodings;          // which encodings did user ask to use?
michael@0:   int32_t encodingsCount;
michael@0:   int32_t encodingStrLength;
michael@0:   uint8_t* swapped;
michael@0:   UBool ownPv, ownEncodingStrings;
michael@0: };
michael@0: 
michael@0: static void generateSelectorData(UConverterSelector* result,
michael@0:                                  UPropsVectors *upvec,
michael@0:                                  const USet* excludedCodePoints,
michael@0:                                  const UConverterUnicodeSet whichSet,
michael@0:                                  UErrorCode* status) {
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return;
michael@0:   }
michael@0: 
michael@0:   int32_t columns = (result->encodingsCount+31)/32;
michael@0: 
michael@0:   // set errorValue to all-ones
michael@0:   for (int32_t col = 0; col < columns; col++) {
michael@0:     upvec_setValue(upvec, UPVEC_ERROR_VALUE_CP, UPVEC_ERROR_VALUE_CP,
michael@0:                    col, ~0, ~0, status);
michael@0:   }
michael@0: 
michael@0:   for (int32_t i = 0; i < result->encodingsCount; ++i) {
michael@0:     uint32_t mask;
michael@0:     uint32_t column;
michael@0:     int32_t item_count;
michael@0:     int32_t j;
michael@0:     UConverter* test_converter = ucnv_open(result->encodings[i], status);
michael@0:     if (U_FAILURE(*status)) {
michael@0:       return;
michael@0:     }
michael@0:     USet* unicode_point_set;
michael@0:     unicode_point_set = uset_open(1, 0);  // empty set
michael@0: 
michael@0:     ucnv_getUnicodeSet(test_converter, unicode_point_set,
michael@0:                        whichSet, status);
michael@0:     if (U_FAILURE(*status)) {
michael@0:       ucnv_close(test_converter);
michael@0:       return;
michael@0:     }
michael@0: 
michael@0:     column = i / 32;
michael@0:     mask = 1 << (i%32);
michael@0:     // now iterate over intervals on set i!
michael@0:     item_count = uset_getItemCount(unicode_point_set);
michael@0: 
michael@0:     for (j = 0; j < item_count; ++j) {
michael@0:       UChar32 start_char;
michael@0:       UChar32 end_char;
michael@0:       UErrorCode smallStatus = U_ZERO_ERROR;
michael@0:       uset_getItem(unicode_point_set, j, &start_char, &end_char, NULL, 0,
michael@0:                    &smallStatus);
michael@0:       if (U_FAILURE(smallStatus)) {
michael@0:         // this will be reached for the converters that fill the set with
michael@0:         // strings. Those should be ignored by our system
michael@0:       } else {
michael@0:         upvec_setValue(upvec, start_char, end_char, column, ~0, mask,
michael@0:                        status);
michael@0:       }
michael@0:     }
michael@0:     ucnv_close(test_converter);
michael@0:     uset_close(unicode_point_set);
michael@0:     if (U_FAILURE(*status)) {
michael@0:       return;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   // handle excluded encodings! Simply set their values to all 1's in the upvec
michael@0:   if (excludedCodePoints) {
michael@0:     int32_t item_count = uset_getItemCount(excludedCodePoints);
michael@0:     for (int32_t j = 0; j < item_count; ++j) {
michael@0:       UChar32 start_char;
michael@0:       UChar32 end_char;
michael@0: 
michael@0:       uset_getItem(excludedCodePoints, j, &start_char, &end_char, NULL, 0,
michael@0:                    status);
michael@0:       for (int32_t col = 0; col < columns; col++) {
michael@0:         upvec_setValue(upvec, start_char, end_char, col, ~0, ~0,
michael@0:                       status);
michael@0:       }
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   // alright. Now, let's put things in the same exact form you'd get when you
michael@0:   // unserialize things.
michael@0:   result->trie = upvec_compactToUTrie2WithRowIndexes(upvec, status);
michael@0:   result->pv = upvec_cloneArray(upvec, &result->pvCount, NULL, status);
michael@0:   result->pvCount *= columns;  // number of uint32_t = rows * columns
michael@0:   result->ownPv = TRUE;
michael@0: }
michael@0: 
michael@0: /* open a selector. If converterListSize is 0, build for all converters.
michael@0:    If excludedCodePoints is NULL, don't exclude any codepoints */
michael@0: U_CAPI UConverterSelector* U_EXPORT2
michael@0: ucnvsel_open(const char* const*  converterList, int32_t converterListSize,
michael@0:              const USet* excludedCodePoints,
michael@0:              const UConverterUnicodeSet whichSet, UErrorCode* status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return NULL;
michael@0:   }
michael@0:   // ensure args make sense!
michael@0:   if (converterListSize < 0 || (converterList == NULL && converterListSize != 0)) {
michael@0:     *status = U_ILLEGAL_ARGUMENT_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0: 
michael@0:   // allocate a new converter
michael@0:   LocalUConverterSelectorPointer newSelector(
michael@0:     (UConverterSelector*)uprv_malloc(sizeof(UConverterSelector)));
michael@0:   if (newSelector.isNull()) {
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   uprv_memset(newSelector.getAlias(), 0, sizeof(UConverterSelector));
michael@0: 
michael@0:   if (converterListSize == 0) {
michael@0:     converterList = NULL;
michael@0:     converterListSize = ucnv_countAvailable();
michael@0:   }
michael@0:   newSelector->encodings =
michael@0:     (char**)uprv_malloc(converterListSize * sizeof(char*));
michael@0:   if (!newSelector->encodings) {
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   newSelector->encodings[0] = NULL;  // now we can call ucnvsel_close()
michael@0: 
michael@0:   // make a backup copy of the list of converters
michael@0:   int32_t totalSize = 0;
michael@0:   int32_t i;
michael@0:   for (i = 0; i < converterListSize; i++) {
michael@0:     totalSize +=
michael@0:       (int32_t)uprv_strlen(converterList != NULL ? converterList[i] : ucnv_getAvailableName(i)) + 1;
michael@0:   }
michael@0:   // 4-align the totalSize to 4-align the size of the serialized form
michael@0:   int32_t encodingStrPadding = totalSize & 3;
michael@0:   if (encodingStrPadding != 0) {
michael@0:     encodingStrPadding = 4 - encodingStrPadding;
michael@0:   }
michael@0:   newSelector->encodingStrLength = totalSize += encodingStrPadding;
michael@0:   char* allStrings = (char*) uprv_malloc(totalSize);
michael@0:   if (!allStrings) {
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0: 
michael@0:   for (i = 0; i < converterListSize; i++) {
michael@0:     newSelector->encodings[i] = allStrings;
michael@0:     uprv_strcpy(newSelector->encodings[i],
michael@0:                 converterList != NULL ? converterList[i] : ucnv_getAvailableName(i));
michael@0:     allStrings += uprv_strlen(newSelector->encodings[i]) + 1;
michael@0:   }
michael@0:   while (encodingStrPadding > 0) {
michael@0:     *allStrings++ = 0;
michael@0:     --encodingStrPadding;
michael@0:   }
michael@0: 
michael@0:   newSelector->ownEncodingStrings = TRUE;
michael@0:   newSelector->encodingsCount = converterListSize;
michael@0:   UPropsVectors *upvec = upvec_open((converterListSize+31)/32, status);
michael@0:   generateSelectorData(newSelector.getAlias(), upvec, excludedCodePoints, whichSet, status);
michael@0:   upvec_close(upvec);
michael@0: 
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return NULL;
michael@0:   }
michael@0: 
michael@0:   return newSelector.orphan();
michael@0: }
michael@0: 
michael@0: /* close opened selector */
michael@0: U_CAPI void U_EXPORT2
michael@0: ucnvsel_close(UConverterSelector *sel) {
michael@0:   if (!sel) {
michael@0:     return;
michael@0:   }
michael@0:   if (sel->ownEncodingStrings) {
michael@0:     uprv_free(sel->encodings[0]);
michael@0:   }
michael@0:   uprv_free(sel->encodings);
michael@0:   if (sel->ownPv) {
michael@0:     uprv_free(sel->pv);
michael@0:   }
michael@0:   utrie2_close(sel->trie);
michael@0:   uprv_free(sel->swapped);
michael@0:   uprv_free(sel);
michael@0: }
michael@0: 
michael@0: static const UDataInfo dataInfo = {
michael@0:   sizeof(UDataInfo),
michael@0:   0,
michael@0: 
michael@0:   U_IS_BIG_ENDIAN,
michael@0:   U_CHARSET_FAMILY,
michael@0:   U_SIZEOF_UCHAR,
michael@0:   0,
michael@0: 
michael@0:   { 0x43, 0x53, 0x65, 0x6c },   /* dataFormat="CSel" */
michael@0:   { 1, 0, 0, 0 },               /* formatVersion */
michael@0:   { 0, 0, 0, 0 }                /* dataVersion */
michael@0: };
michael@0: 
michael@0: enum {
michael@0:   UCNVSEL_INDEX_TRIE_SIZE,      // trie size in bytes
michael@0:   UCNVSEL_INDEX_PV_COUNT,       // number of uint32_t in the bit vectors
michael@0:   UCNVSEL_INDEX_NAMES_COUNT,    // number of encoding names
michael@0:   UCNVSEL_INDEX_NAMES_LENGTH,   // number of encoding name bytes including padding
michael@0:   UCNVSEL_INDEX_SIZE = 15,      // bytes following the DataHeader
michael@0:   UCNVSEL_INDEX_COUNT = 16
michael@0: };
michael@0: 
michael@0: /*
michael@0:  * Serialized form of a UConverterSelector, formatVersion 1:
michael@0:  *
michael@0:  * The serialized form begins with a standard ICU DataHeader with a UDataInfo
michael@0:  * as the template above.
michael@0:  * This is followed by:
michael@0:  *   int32_t indexes[UCNVSEL_INDEX_COUNT];          // see index entry constants above
michael@0:  *   serialized UTrie2;                             // indexes[UCNVSEL_INDEX_TRIE_SIZE] bytes
michael@0:  *   uint32_t pv[indexes[UCNVSEL_INDEX_PV_COUNT]];  // bit vectors
michael@0:  *   char* encodingNames[indexes[UCNVSEL_INDEX_NAMES_LENGTH]];  // NUL-terminated strings + padding
michael@0:  */
michael@0: 
michael@0: /* serialize a selector */
michael@0: U_CAPI int32_t U_EXPORT2
michael@0: ucnvsel_serialize(const UConverterSelector* sel,
michael@0:                   void* buffer, int32_t bufferCapacity, UErrorCode* status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return 0;
michael@0:   }
michael@0:   // ensure args make sense!
michael@0:   uint8_t *p = (uint8_t *)buffer;
michael@0:   if (bufferCapacity < 0 ||
michael@0:       (bufferCapacity > 0 && (p == NULL || (U_POINTER_MASK_LSB(p, 3) != 0)))
michael@0:   ) {
michael@0:     *status = U_ILLEGAL_ARGUMENT_ERROR;
michael@0:     return 0;
michael@0:   }
michael@0:   // add up the size of the serialized form
michael@0:   int32_t serializedTrieSize = utrie2_serialize(sel->trie, NULL, 0, status);
michael@0:   if (*status != U_BUFFER_OVERFLOW_ERROR && U_FAILURE(*status)) {
michael@0:     return 0;
michael@0:   }
michael@0:   *status = U_ZERO_ERROR;
michael@0: 
michael@0:   DataHeader header;
michael@0:   uprv_memset(&header, 0, sizeof(header));
michael@0:   header.dataHeader.headerSize = (uint16_t)((sizeof(header) + 15) & ~15);
michael@0:   header.dataHeader.magic1 = 0xda;
michael@0:   header.dataHeader.magic2 = 0x27;
michael@0:   uprv_memcpy(&header.info, &dataInfo, sizeof(dataInfo));
michael@0: 
michael@0:   int32_t indexes[UCNVSEL_INDEX_COUNT] = {
michael@0:     serializedTrieSize,
michael@0:     sel->pvCount,
michael@0:     sel->encodingsCount,
michael@0:     sel->encodingStrLength
michael@0:   };
michael@0: 
michael@0:   int32_t totalSize =
michael@0:     header.dataHeader.headerSize +
michael@0:     (int32_t)sizeof(indexes) +
michael@0:     serializedTrieSize +
michael@0:     sel->pvCount * 4 +
michael@0:     sel->encodingStrLength;
michael@0:   indexes[UCNVSEL_INDEX_SIZE] = totalSize - header.dataHeader.headerSize;
michael@0:   if (totalSize > bufferCapacity) {
michael@0:     *status = U_BUFFER_OVERFLOW_ERROR;
michael@0:     return totalSize;
michael@0:   }
michael@0:   // ok, save!
michael@0:   int32_t length = header.dataHeader.headerSize;
michael@0:   uprv_memcpy(p, &header, sizeof(header));
michael@0:   uprv_memset(p + sizeof(header), 0, length - sizeof(header));
michael@0:   p += length;
michael@0: 
michael@0:   length = (int32_t)sizeof(indexes);
michael@0:   uprv_memcpy(p, indexes, length);
michael@0:   p += length;
michael@0: 
michael@0:   utrie2_serialize(sel->trie, p, serializedTrieSize, status);
michael@0:   p += serializedTrieSize;
michael@0: 
michael@0:   length = sel->pvCount * 4;
michael@0:   uprv_memcpy(p, sel->pv, length);
michael@0:   p += length;
michael@0: 
michael@0:   uprv_memcpy(p, sel->encodings[0], sel->encodingStrLength);
michael@0:   p += sel->encodingStrLength;
michael@0: 
michael@0:   return totalSize;
michael@0: }
michael@0: 
michael@0: /**
michael@0:  * swap a selector into the desired Endianness and Asciiness of
michael@0:  * the system. Just as FYI, selectors are always saved in the format
michael@0:  * of the system that created them. They are only converted if used
michael@0:  * on another system. In other words, selectors created on different
michael@0:  * system can be different even if the params are identical (endianness
michael@0:  * and Asciiness differences only)
michael@0:  *
michael@0:  * @param ds pointer to data swapper containing swapping info
michael@0:  * @param inData pointer to incoming data
michael@0:  * @param length length of inData in bytes
michael@0:  * @param outData pointer to output data. Capacity should
michael@0:  *                be at least equal to capacity of inData
michael@0:  * @param status an in/out ICU UErrorCode
michael@0:  * @return 0 on failure, number of bytes swapped on success
michael@0:  *         number of bytes swapped can be smaller than length
michael@0:  */
michael@0: static int32_t
michael@0: ucnvsel_swap(const UDataSwapper *ds,
michael@0:              const void *inData, int32_t length,
michael@0:              void *outData, UErrorCode *status) {
michael@0:   /* udata_swapDataHeader checks the arguments */
michael@0:   int32_t headerSize = udata_swapDataHeader(ds, inData, length, outData, status);
michael@0:   if(U_FAILURE(*status)) {
michael@0:     return 0;
michael@0:   }
michael@0: 
michael@0:   /* check data format and format version */
michael@0:   const UDataInfo *pInfo = (const UDataInfo *)((const char *)inData + 4);
michael@0:   if(!(
michael@0:     pInfo->dataFormat[0] == 0x43 &&  /* dataFormat="CSel" */
michael@0:     pInfo->dataFormat[1] == 0x53 &&
michael@0:     pInfo->dataFormat[2] == 0x65 &&
michael@0:     pInfo->dataFormat[3] == 0x6c
michael@0:   )) {
michael@0:     udata_printError(ds, "ucnvsel_swap(): data format %02x.%02x.%02x.%02x is not recognized as UConverterSelector data\n",
michael@0:                      pInfo->dataFormat[0], pInfo->dataFormat[1],
michael@0:                      pInfo->dataFormat[2], pInfo->dataFormat[3]);
michael@0:     *status = U_INVALID_FORMAT_ERROR;
michael@0:     return 0;
michael@0:   }
michael@0:   if(pInfo->formatVersion[0] != 1) {
michael@0:     udata_printError(ds, "ucnvsel_swap(): format version %02x is not supported\n",
michael@0:                      pInfo->formatVersion[0]);
michael@0:     *status = U_UNSUPPORTED_ERROR;
michael@0:     return 0;
michael@0:   }
michael@0: 
michael@0:   if(length >= 0) {
michael@0:     length -= headerSize;
michael@0:     if(length < 16*4) {
michael@0:       udata_printError(ds, "ucnvsel_swap(): too few bytes (%d after header) for UConverterSelector data\n",
michael@0:                        length);
michael@0:       *status = U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:       return 0;
michael@0:     }
michael@0:   }
michael@0: 
michael@0:   const uint8_t *inBytes = (const uint8_t *)inData + headerSize;
michael@0:   uint8_t *outBytes = (uint8_t *)outData + headerSize;
michael@0: 
michael@0:   /* read the indexes */
michael@0:   const int32_t *inIndexes = (const int32_t *)inBytes;
michael@0:   int32_t indexes[16];
michael@0:   int32_t i;
michael@0:   for(i = 0; i < 16; ++i) {
michael@0:     indexes[i] = udata_readInt32(ds, inIndexes[i]);
michael@0:   }
michael@0: 
michael@0:   /* get the total length of the data */
michael@0:   int32_t size = indexes[UCNVSEL_INDEX_SIZE];
michael@0:   if(length >= 0) {
michael@0:     if(length < size) {
michael@0:       udata_printError(ds, "ucnvsel_swap(): too few bytes (%d after header) for all of UConverterSelector data\n",
michael@0:                        length);
michael@0:       *status = U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:       return 0;
michael@0:     }
michael@0: 
michael@0:     /* copy the data for inaccessible bytes */
michael@0:     if(inBytes != outBytes) {
michael@0:       uprv_memcpy(outBytes, inBytes, size);
michael@0:     }
michael@0: 
michael@0:     int32_t offset = 0, count;
michael@0: 
michael@0:     /* swap the int32_t indexes[] */
michael@0:     count = UCNVSEL_INDEX_COUNT*4;
michael@0:     ds->swapArray32(ds, inBytes, count, outBytes, status);
michael@0:     offset += count;
michael@0: 
michael@0:     /* swap the UTrie2 */
michael@0:     count = indexes[UCNVSEL_INDEX_TRIE_SIZE];
michael@0:     utrie2_swap(ds, inBytes + offset, count, outBytes + offset, status);
michael@0:     offset += count;
michael@0: 
michael@0:     /* swap the uint32_t pv[] */
michael@0:     count = indexes[UCNVSEL_INDEX_PV_COUNT]*4;
michael@0:     ds->swapArray32(ds, inBytes + offset, count, outBytes + offset, status);
michael@0:     offset += count;
michael@0: 
michael@0:     /* swap the encoding names */
michael@0:     count = indexes[UCNVSEL_INDEX_NAMES_LENGTH];
michael@0:     ds->swapInvChars(ds, inBytes + offset, count, outBytes + offset, status);
michael@0:     offset += count;
michael@0: 
michael@0:     U_ASSERT(offset == size);
michael@0:   }
michael@0: 
michael@0:   return headerSize + size;
michael@0: }
michael@0: 
michael@0: /* unserialize a selector */
michael@0: U_CAPI UConverterSelector* U_EXPORT2
michael@0: ucnvsel_openFromSerialized(const void* buffer, int32_t length, UErrorCode* status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return NULL;
michael@0:   }
michael@0:   // ensure args make sense!
michael@0:   const uint8_t *p = (const uint8_t *)buffer;
michael@0:   if (length <= 0 ||
michael@0:       (length > 0 && (p == NULL || (U_POINTER_MASK_LSB(p, 3) != 0)))
michael@0:   ) {
michael@0:     *status = U_ILLEGAL_ARGUMENT_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   // header
michael@0:   if (length < 32) {
michael@0:     // not even enough space for a minimal header
michael@0:     *status = U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   const DataHeader *pHeader = (const DataHeader *)p;
michael@0:   if (!(
michael@0:     pHeader->dataHeader.magic1==0xda &&
michael@0:     pHeader->dataHeader.magic2==0x27 &&
michael@0:     pHeader->info.dataFormat[0] == 0x43 &&
michael@0:     pHeader->info.dataFormat[1] == 0x53 &&
michael@0:     pHeader->info.dataFormat[2] == 0x65 &&
michael@0:     pHeader->info.dataFormat[3] == 0x6c
michael@0:   )) {
michael@0:     /* header not valid or dataFormat not recognized */
michael@0:     *status = U_INVALID_FORMAT_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   if (pHeader->info.formatVersion[0] != 1) {
michael@0:     *status = U_UNSUPPORTED_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   uint8_t* swapped = NULL;
michael@0:   if (pHeader->info.isBigEndian != U_IS_BIG_ENDIAN ||
michael@0:       pHeader->info.charsetFamily != U_CHARSET_FAMILY
michael@0:   ) {
michael@0:     // swap the data
michael@0:     UDataSwapper *ds =
michael@0:       udata_openSwapperForInputData(p, length, U_IS_BIG_ENDIAN, U_CHARSET_FAMILY, status);
michael@0:     int32_t totalSize = ucnvsel_swap(ds, p, -1, NULL, status);
michael@0:     if (U_FAILURE(*status)) {
michael@0:       udata_closeSwapper(ds);
michael@0:       return NULL;
michael@0:     }
michael@0:     if (length < totalSize) {
michael@0:       udata_closeSwapper(ds);
michael@0:       *status = U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:       return NULL;
michael@0:     }
michael@0:     swapped = (uint8_t*)uprv_malloc(totalSize);
michael@0:     if (swapped == NULL) {
michael@0:       udata_closeSwapper(ds);
michael@0:       *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:       return NULL;
michael@0:     }
michael@0:     ucnvsel_swap(ds, p, length, swapped, status);
michael@0:     udata_closeSwapper(ds);
michael@0:     if (U_FAILURE(*status)) {
michael@0:       uprv_free(swapped);
michael@0:       return NULL;
michael@0:     }
michael@0:     p = swapped;
michael@0:     pHeader = (const DataHeader *)p;
michael@0:   }
michael@0:   if (length < (pHeader->dataHeader.headerSize + 16 * 4)) {
michael@0:     // not even enough space for the header and the indexes
michael@0:     uprv_free(swapped);
michael@0:     *status = U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   p += pHeader->dataHeader.headerSize;
michael@0:   length -= pHeader->dataHeader.headerSize;
michael@0:   // indexes
michael@0:   const int32_t *indexes = (const int32_t *)p;
michael@0:   if (length < indexes[UCNVSEL_INDEX_SIZE]) {
michael@0:     uprv_free(swapped);
michael@0:     *status = U_INDEX_OUTOFBOUNDS_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   p += UCNVSEL_INDEX_COUNT * 4;
michael@0:   // create and populate the selector object
michael@0:   UConverterSelector* sel = (UConverterSelector*)uprv_malloc(sizeof(UConverterSelector));
michael@0:   char **encodings =
michael@0:     (char **)uprv_malloc(
michael@0:       indexes[UCNVSEL_INDEX_NAMES_COUNT] * sizeof(char *));
michael@0:   if (sel == NULL || encodings == NULL) {
michael@0:     uprv_free(swapped);
michael@0:     uprv_free(sel);
michael@0:     uprv_free(encodings);
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   uprv_memset(sel, 0, sizeof(UConverterSelector));
michael@0:   sel->pvCount = indexes[UCNVSEL_INDEX_PV_COUNT];
michael@0:   sel->encodings = encodings;
michael@0:   sel->encodingsCount = indexes[UCNVSEL_INDEX_NAMES_COUNT];
michael@0:   sel->encodingStrLength = indexes[UCNVSEL_INDEX_NAMES_LENGTH];
michael@0:   sel->swapped = swapped;
michael@0:   // trie
michael@0:   sel->trie = utrie2_openFromSerialized(UTRIE2_16_VALUE_BITS,
michael@0:                                         p, indexes[UCNVSEL_INDEX_TRIE_SIZE], NULL,
michael@0:                                         status);
michael@0:   p += indexes[UCNVSEL_INDEX_TRIE_SIZE];
michael@0:   if (U_FAILURE(*status)) {
michael@0:     ucnvsel_close(sel);
michael@0:     return NULL;
michael@0:   }
michael@0:   // bit vectors
michael@0:   sel->pv = (uint32_t *)p;
michael@0:   p += sel->pvCount * 4;
michael@0:   // encoding names
michael@0:   char* s = (char*)p;
michael@0:   for (int32_t i = 0; i < sel->encodingsCount; ++i) {
michael@0:     sel->encodings[i] = s;
michael@0:     s += uprv_strlen(s) + 1;
michael@0:   }
michael@0:   p += sel->encodingStrLength;
michael@0: 
michael@0:   return sel;
michael@0: }
michael@0: 
michael@0: // a bunch of functions for the enumeration thingie! Nothing fancy here. Just
michael@0: // iterate over the selected encodings
michael@0: struct Enumerator {
michael@0:   int16_t* index;
michael@0:   int16_t length;
michael@0:   int16_t cur;
michael@0:   const UConverterSelector* sel;
michael@0: };
michael@0: 
michael@0: U_CDECL_BEGIN
michael@0: 
michael@0: static void U_CALLCONV
michael@0: ucnvsel_close_selector_iterator(UEnumeration *enumerator) {
michael@0:   uprv_free(((Enumerator*)(enumerator->context))->index);
michael@0:   uprv_free(enumerator->context);
michael@0:   uprv_free(enumerator);
michael@0: }
michael@0: 
michael@0: 
michael@0: static int32_t U_CALLCONV
michael@0: ucnvsel_count_encodings(UEnumeration *enumerator, UErrorCode *status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return 0;
michael@0:   }
michael@0:   return ((Enumerator*)(enumerator->context))->length;
michael@0: }
michael@0: 
michael@0: 
michael@0: static const char* U_CALLCONV ucnvsel_next_encoding(UEnumeration* enumerator,
michael@0:                                                  int32_t* resultLength,
michael@0:                                                  UErrorCode* status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return NULL;
michael@0:   }
michael@0: 
michael@0:   int16_t cur = ((Enumerator*)(enumerator->context))->cur;
michael@0:   const UConverterSelector* sel;
michael@0:   const char* result;
michael@0:   if (cur >= ((Enumerator*)(enumerator->context))->length) {
michael@0:     return NULL;
michael@0:   }
michael@0:   sel = ((Enumerator*)(enumerator->context))->sel;
michael@0:   result = sel->encodings[((Enumerator*)(enumerator->context))->index[cur] ];
michael@0:   ((Enumerator*)(enumerator->context))->cur++;
michael@0:   if (resultLength) {
michael@0:     *resultLength = (int32_t)uprv_strlen(result);
michael@0:   }
michael@0:   return result;
michael@0: }
michael@0: 
michael@0: static void U_CALLCONV ucnvsel_reset_iterator(UEnumeration* enumerator,
michael@0:                                            UErrorCode* status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return ;
michael@0:   }
michael@0:   ((Enumerator*)(enumerator->context))->cur = 0;
michael@0: }
michael@0: 
michael@0: U_CDECL_END
michael@0: 
michael@0: 
michael@0: static const UEnumeration defaultEncodings = {
michael@0:   NULL,
michael@0:     NULL,
michael@0:     ucnvsel_close_selector_iterator,
michael@0:     ucnvsel_count_encodings,
michael@0:     uenum_unextDefault,
michael@0:     ucnvsel_next_encoding, 
michael@0:     ucnvsel_reset_iterator
michael@0: };
michael@0: 
michael@0: 
michael@0: // internal fn to intersect two sets of masks
michael@0: // returns whether the mask has reduced to all zeros
michael@0: static UBool intersectMasks(uint32_t* dest, const uint32_t* source1, int32_t len) {
michael@0:   int32_t i;
michael@0:   uint32_t oredDest = 0;
michael@0:   for (i = 0 ; i < len ; ++i) {
michael@0:     oredDest |= (dest[i] &= source1[i]);
michael@0:   }
michael@0:   return oredDest == 0;
michael@0: }
michael@0: 
michael@0: // internal fn to count how many 1's are there in a mask
michael@0: // algorithm taken from  http://graphics.stanford.edu/~seander/bithacks.html
michael@0: static int16_t countOnes(uint32_t* mask, int32_t len) {
michael@0:   int32_t i, totalOnes = 0;
michael@0:   for (i = 0 ; i < len ; ++i) {
michael@0:     uint32_t ent = mask[i];
michael@0:     for (; ent; totalOnes++)
michael@0:     {
michael@0:       ent &= ent - 1; // clear the least significant bit set
michael@0:     }
michael@0:   }
michael@0:   return totalOnes;
michael@0: }
michael@0: 
michael@0: 
michael@0: /* internal function! */
michael@0: static UEnumeration *selectForMask(const UConverterSelector* sel,
michael@0:                                    uint32_t *mask, UErrorCode *status) {
michael@0:   // this is the context we will use. Store a table of indices to which
michael@0:   // encodings are legit.
michael@0:   struct Enumerator* result = (Enumerator*)uprv_malloc(sizeof(Enumerator));
michael@0:   if (result == NULL) {
michael@0:     uprv_free(mask);
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   result->index = NULL;  // this will be allocated later!
michael@0:   result->length = result->cur = 0;
michael@0:   result->sel = sel;
michael@0: 
michael@0:   UEnumeration *en = (UEnumeration *)uprv_malloc(sizeof(UEnumeration));
michael@0:   if (en == NULL) {
michael@0:     // TODO(markus): Combine Enumerator and UEnumeration into one struct.
michael@0:     uprv_free(mask);
michael@0:     uprv_free(result);
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   memcpy(en, &defaultEncodings, sizeof(UEnumeration));
michael@0:   en->context = result;
michael@0: 
michael@0:   int32_t columns = (sel->encodingsCount+31)/32;
michael@0:   int16_t numOnes = countOnes(mask, columns);
michael@0:   // now, we know the exact space we need for index
michael@0:   if (numOnes > 0) {
michael@0:     result->index = (int16_t*) uprv_malloc(numOnes * sizeof(int16_t));
michael@0: 
michael@0:     int32_t i, j;
michael@0:     int16_t k = 0;
michael@0:     for (j = 0 ; j < columns; j++) {
michael@0:       uint32_t v = mask[j];
michael@0:       for (i = 0 ; i < 32 && k < sel->encodingsCount; i++, k++) {
michael@0:         if ((v & 1) != 0) {
michael@0:           result->index[result->length++] = k;
michael@0:         }
michael@0:         v >>= 1;
michael@0:       }
michael@0:     }
michael@0:   } //otherwise, index will remain NULL (and will never be touched by
michael@0:     //the enumerator code anyway)
michael@0:   uprv_free(mask);
michael@0:   return en;
michael@0: }
michael@0: 
michael@0: /* check a string against the selector - UTF16 version */
michael@0: U_CAPI UEnumeration * U_EXPORT2
michael@0: ucnvsel_selectForString(const UConverterSelector* sel,
michael@0:                         const UChar *s, int32_t length, UErrorCode *status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return NULL;
michael@0:   }
michael@0:   // ensure args make sense!
michael@0:   if (sel == NULL || (s == NULL && length != 0)) {
michael@0:     *status = U_ILLEGAL_ARGUMENT_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0: 
michael@0:   int32_t columns = (sel->encodingsCount+31)/32;
michael@0:   uint32_t* mask = (uint32_t*) uprv_malloc(columns * 4);
michael@0:   if (mask == NULL) {
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   uprv_memset(mask, ~0, columns *4);
michael@0: 
michael@0:   if(s!=NULL) {
michael@0:     const UChar *limit;
michael@0:     if (length >= 0) {
michael@0:       limit = s + length;
michael@0:     } else {
michael@0:       limit = NULL;
michael@0:     }
michael@0:     
michael@0:     while (limit == NULL ? *s != 0 : s != limit) {
michael@0:       UChar32 c;
michael@0:       uint16_t pvIndex;
michael@0:       UTRIE2_U16_NEXT16(sel->trie, s, limit, c, pvIndex);
michael@0:       if (intersectMasks(mask, sel->pv+pvIndex, columns)) {
michael@0:         break;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0:   return selectForMask(sel, mask, status);
michael@0: }
michael@0: 
michael@0: /* check a string against the selector - UTF8 version */
michael@0: U_CAPI UEnumeration * U_EXPORT2
michael@0: ucnvsel_selectForUTF8(const UConverterSelector* sel,
michael@0:                       const char *s, int32_t length, UErrorCode *status) {
michael@0:   // check if already failed
michael@0:   if (U_FAILURE(*status)) {
michael@0:     return NULL;
michael@0:   }
michael@0:   // ensure args make sense!
michael@0:   if (sel == NULL || (s == NULL && length != 0)) {
michael@0:     *status = U_ILLEGAL_ARGUMENT_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0: 
michael@0:   int32_t columns = (sel->encodingsCount+31)/32;
michael@0:   uint32_t* mask = (uint32_t*) uprv_malloc(columns * 4);
michael@0:   if (mask == NULL) {
michael@0:     *status = U_MEMORY_ALLOCATION_ERROR;
michael@0:     return NULL;
michael@0:   }
michael@0:   uprv_memset(mask, ~0, columns *4);
michael@0: 
michael@0:   if (length < 0) {
michael@0:     length = (int32_t)uprv_strlen(s);
michael@0:   }
michael@0: 
michael@0:   if(s!=NULL) {
michael@0:     const char *limit = s + length;
michael@0:     
michael@0:     while (s != limit) {
michael@0:       uint16_t pvIndex;
michael@0:       UTRIE2_U8_NEXT16(sel->trie, s, limit, pvIndex);
michael@0:       if (intersectMasks(mask, sel->pv+pvIndex, columns)) {
michael@0:         break;
michael@0:       }
michael@0:     }
michael@0:   }
michael@0:   return selectForMask(sel, mask, status);
michael@0: }
michael@0: 
michael@0: #endif  // !UCONFIG_NO_CONVERSION