/*
*******************************************************************************
* Copyright (C) 2013-2015, International Business Machines
* Corporation and others. All Rights Reserved.
*******************************************************************************
* collationdatareader.cpp
*
* created on: 2013feb07
* created by: Markus W. Scherer
*/
#include "unicode/utypes.h"
#if !UCONFIG_NO_COLLATION
#include "unicode/ucol.h"
#include "unicode/udata.h"
#include "unicode/uscript.h"
#include "cmemory.h"
#include "collation.h"
#include "collationdata.h"
#include "collationdatareader.h"
#include "collationfastlatin.h"
#include "collationkeys.h"
#include "collationrootelements.h"
#include "collationsettings.h"
#include "collationtailoring.h"
#include "collunsafe.h"
#include "normalizer2impl.h"
#include "uassert.h"
#include "ucmndata.h"
#include "utrie2.h"
U_NAMESPACE_BEGIN
namespace {
int32_t getIndex(const int32_t *indexes, int32_t length, int32_t i) {
return (i < length) ? indexes[i] : -1;
}
} // namespace
void
CollationDataReader::read(const CollationTailoring *base, const uint8_t *inBytes, int32_t inLength,
CollationTailoring &tailoring, UErrorCode &errorCode) {
if(U_FAILURE(errorCode)) { return; }
if(base != NULL) {
if(inBytes == NULL || (0 <= inLength && inLength < 24)) {
errorCode = U_ILLEGAL_ARGUMENT_ERROR;
return;
}
const DataHeader *header = reinterpret_cast<const DataHeader *>(inBytes);
if(!(header->dataHeader.magic1 == 0xda && header->dataHeader.magic2 == 0x27 &&
isAcceptable(tailoring.version, NULL, NULL, &header->info))) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
if(base->getUCAVersion() != tailoring.getUCAVersion()) {
errorCode = U_COLLATOR_VERSION_MISMATCH;
return;
}
int32_t headerLength = header->dataHeader.headerSize;
inBytes += headerLength;
if(inLength >= 0) {
inLength -= headerLength;
}
}
if(inBytes == NULL || (0 <= inLength && inLength < 8)) {
errorCode = U_ILLEGAL_ARGUMENT_ERROR;
return;
}
const int32_t *inIndexes = reinterpret_cast<const int32_t *>(inBytes);
int32_t indexesLength = inIndexes[IX_INDEXES_LENGTH];
if(indexesLength < 2 || (0 <= inLength && inLength < indexesLength * 4)) {
errorCode = U_INVALID_FORMAT_ERROR; // Not enough indexes.
return;
}
// Assume that the tailoring data is in initial state,
// with NULL pointers and 0 lengths.
// Set pointers to non-empty data parts.
// Do this in order of their byte offsets. (Should help porting to Java.)
int32_t index; // one of the indexes[] slots
int32_t offset; // byte offset for the index part
int32_t length; // number of bytes in the index part
if(indexesLength > IX_TOTAL_SIZE) {
length = inIndexes[IX_TOTAL_SIZE];
} else if(indexesLength > IX_REORDER_CODES_OFFSET) {
length = inIndexes[indexesLength - 1];
} else {
length = 0; // only indexes, and inLength was already checked for them
}
if(0 <= inLength && inLength < length) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
const CollationData *baseData = base == NULL ? NULL : base->data;
const int32_t *reorderCodes = NULL;
int32_t reorderCodesLength = 0;
const uint32_t *reorderRanges = NULL;
int32_t reorderRangesLength = 0;
index = IX_REORDER_CODES_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 4) {
if(baseData == NULL) {
// We assume for collation settings that
// the base data does not have a reordering.
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
reorderCodes = reinterpret_cast<const int32_t *>(inBytes + offset);
reorderCodesLength = length / 4;
// The reorderRanges (if any) are the trailing reorderCodes entries.
// Split the array at the boundary.
// Script or reorder codes do not exceed 16-bit values.
// Range limits are stored in the upper 16 bits, and are never 0.
while(reorderRangesLength < reorderCodesLength &&
(reorderCodes[reorderCodesLength - reorderRangesLength - 1] & 0xffff0000) != 0) {
++reorderRangesLength;
}
U_ASSERT(reorderRangesLength < reorderCodesLength);
if(reorderRangesLength != 0) {
reorderCodesLength -= reorderRangesLength;
reorderRanges = reinterpret_cast<const uint32_t *>(reorderCodes + reorderCodesLength);
}
}
// There should be a reorder table only if there are reorder codes.
// However, when there are reorder codes the reorder table may be omitted to reduce
// the data size.
const uint8_t *reorderTable = NULL;
index = IX_REORDER_TABLE_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 256) {
if(reorderCodesLength == 0) {
errorCode = U_INVALID_FORMAT_ERROR; // Reordering table without reordering codes.
return;
}
reorderTable = inBytes + offset;
} else {
// If we have reorder codes, then build the reorderTable at the end,
// when the CollationData is otherwise complete.
}
if(baseData != NULL && baseData->numericPrimary != (inIndexes[IX_OPTIONS] & 0xff000000)) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
CollationData *data = NULL; // Remains NULL if there are no mappings.
index = IX_TRIE_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 8) {
if(!tailoring.ensureOwnedData(errorCode)) { return; }
data = tailoring.ownedData;
data->base = baseData;
data->numericPrimary = inIndexes[IX_OPTIONS] & 0xff000000;
data->trie = tailoring.trie = utrie2_openFromSerialized(
UTRIE2_32_VALUE_BITS, inBytes + offset, length, NULL,
&errorCode);
if(U_FAILURE(errorCode)) { return; }
} else if(baseData != NULL) {
// Use the base data. Only the settings are tailored.
tailoring.data = baseData;
} else {
errorCode = U_INVALID_FORMAT_ERROR; // No mappings.
return;
}
index = IX_CES_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 8) {
if(data == NULL) {
errorCode = U_INVALID_FORMAT_ERROR; // Tailored ces without tailored trie.
return;
}
data->ces = reinterpret_cast<const int64_t *>(inBytes + offset);
data->cesLength = length / 8;
}
index = IX_CE32S_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 4) {
if(data == NULL) {
errorCode = U_INVALID_FORMAT_ERROR; // Tailored ce32s without tailored trie.
return;
}
data->ce32s = reinterpret_cast<const uint32_t *>(inBytes + offset);
data->ce32sLength = length / 4;
}
int32_t jamoCE32sStart = getIndex(inIndexes, indexesLength, IX_JAMO_CE32S_START);
if(jamoCE32sStart >= 0) {
if(data == NULL || data->ce32s == NULL) {
errorCode = U_INVALID_FORMAT_ERROR; // Index into non-existent ce32s[].
return;
}
data->jamoCE32s = data->ce32s + jamoCE32sStart;
} else if(data == NULL) {
// Nothing to do.
} else if(baseData != NULL) {
data->jamoCE32s = baseData->jamoCE32s;
} else {
errorCode = U_INVALID_FORMAT_ERROR; // No Jamo CE32s for Hangul processing.
return;
}
index = IX_ROOT_ELEMENTS_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 4) {
length /= 4;
if(data == NULL || length <= CollationRootElements::IX_SEC_TER_BOUNDARIES) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
data->rootElements = reinterpret_cast<const uint32_t *>(inBytes + offset);
data->rootElementsLength = length;
uint32_t commonSecTer = data->rootElements[CollationRootElements::IX_COMMON_SEC_AND_TER_CE];
if(commonSecTer != Collation::COMMON_SEC_AND_TER_CE) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
uint32_t secTerBoundaries = data->rootElements[CollationRootElements::IX_SEC_TER_BOUNDARIES];
if((secTerBoundaries >> 24) < CollationKeys::SEC_COMMON_HIGH) {
// [fixed last secondary common byte] is too low,
// and secondary weights would collide with compressed common secondaries.
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
}
index = IX_CONTEXTS_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 2) {
if(data == NULL) {
errorCode = U_INVALID_FORMAT_ERROR; // Tailored contexts without tailored trie.
return;
}
data->contexts = reinterpret_cast<const UChar *>(inBytes + offset);
data->contextsLength = length / 2;
}
index = IX_UNSAFE_BWD_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 2) {
if(data == NULL) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
if(baseData == NULL) {
#if defined(COLLUNSAFE_COLL_VERSION) && defined (COLLUNSAFE_SERIALIZE)
tailoring.unsafeBackwardSet = new UnicodeSet(unsafe_serializedData, unsafe_serializedCount, UnicodeSet::kSerialized, errorCode);
if(tailoring.unsafeBackwardSet == NULL) {
errorCode = U_MEMORY_ALLOCATION_ERROR;
return;
} else if (U_FAILURE(errorCode)) {
return;
}
#else
// Create the unsafe-backward set for the root collator.
// Include all non-zero combining marks and trail surrogates.
// We do this at load time, rather than at build time,
// to simplify Unicode version bootstrapping:
// The root data builder only needs the new FractionalUCA.txt data,
// but it need not be built with a version of ICU already updated to
// the corresponding new Unicode Character Database.
//
// The following is an optimized version of
// new UnicodeSet("[[:^lccc=0:][\\udc00-\\udfff]]").
// It is faster and requires fewer code dependencies.
tailoring.unsafeBackwardSet = new UnicodeSet(0xdc00, 0xdfff); // trail surrogates
if(tailoring.unsafeBackwardSet == NULL) {
errorCode = U_MEMORY_ALLOCATION_ERROR;
return;
}
data->nfcImpl.addLcccChars(*tailoring.unsafeBackwardSet);
#endif // !COLLUNSAFE_SERIALIZE || !COLLUNSAFE_COLL_VERSION
} else {
// Clone the root collator's set contents.
tailoring.unsafeBackwardSet = static_cast<UnicodeSet *>(
baseData->unsafeBackwardSet->cloneAsThawed());
if(tailoring.unsafeBackwardSet == NULL) {
errorCode = U_MEMORY_ALLOCATION_ERROR;
return;
}
}
// Add the ranges from the data file to the unsafe-backward set.
USerializedSet sset;
const uint16_t *unsafeData = reinterpret_cast<const uint16_t *>(inBytes + offset);
if(!uset_getSerializedSet(&sset, unsafeData, length / 2)) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
int32_t count = uset_getSerializedRangeCount(&sset);
for(int32_t i = 0; i < count; ++i) {
UChar32 start, end;
uset_getSerializedRange(&sset, i, &start, &end);
tailoring.unsafeBackwardSet->add(start, end);
}
// Mark each lead surrogate as "unsafe"
// if any of its 1024 associated supplementary code points is "unsafe".
UChar32 c = 0x10000;
for(UChar lead = 0xd800; lead < 0xdc00; ++lead, c += 0x400) {
if(!tailoring.unsafeBackwardSet->containsNone(c, c + 0x3ff)) {
tailoring.unsafeBackwardSet->add(lead);
}
}
tailoring.unsafeBackwardSet->freeze();
data->unsafeBackwardSet = tailoring.unsafeBackwardSet;
} else if(data == NULL) {
// Nothing to do.
} else if(baseData != NULL) {
// No tailoring-specific data: Alias the root collator's set.
data->unsafeBackwardSet = baseData->unsafeBackwardSet;
} else {
errorCode = U_INVALID_FORMAT_ERROR; // No unsafeBackwardSet.
return;
}
// If the fast Latin format version is different,
// or the version is set to 0 for "no fast Latin table",
// then just always use the normal string comparison path.
if(data != NULL) {
data->fastLatinTable = NULL;
data->fastLatinTableLength = 0;
if(((inIndexes[IX_OPTIONS] >> 16) & 0xff) == CollationFastLatin::VERSION) {
index = IX_FAST_LATIN_TABLE_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 2) {
data->fastLatinTable = reinterpret_cast<const uint16_t *>(inBytes + offset);
data->fastLatinTableLength = length / 2;
if((*data->fastLatinTable >> 8) != CollationFastLatin::VERSION) {
errorCode = U_INVALID_FORMAT_ERROR; // header vs. table version mismatch
return;
}
} else if(baseData != NULL) {
data->fastLatinTable = baseData->fastLatinTable;
data->fastLatinTableLength = baseData->fastLatinTableLength;
}
}
}
index = IX_SCRIPTS_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 2) {
if(data == NULL) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
const uint16_t *scripts = reinterpret_cast<const uint16_t *>(inBytes + offset);
int32_t scriptsLength = length / 2;
data->numScripts = scripts[0];
// There must be enough entries for both arrays, including more than two range starts.
data->scriptStartsLength = scriptsLength - (1 + data->numScripts + 16);
if(data->scriptStartsLength <= 2 ||
CollationData::MAX_NUM_SCRIPT_RANGES < data->scriptStartsLength) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
data->scriptsIndex = scripts + 1;
data->scriptStarts = scripts + 1 + data->numScripts + 16;
if(!(data->scriptStarts[0] == 0 &&
data->scriptStarts[1] == ((Collation::MERGE_SEPARATOR_BYTE + 1) << 8) &&
data->scriptStarts[data->scriptStartsLength - 1] ==
(Collation::TRAIL_WEIGHT_BYTE << 8))) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
} else if(data == NULL) {
// Nothing to do.
} else if(baseData != NULL) {
data->numScripts = baseData->numScripts;
data->scriptsIndex = baseData->scriptsIndex;
data->scriptStarts = baseData->scriptStarts;
data->scriptStartsLength = baseData->scriptStartsLength;
}
index = IX_COMPRESSIBLE_BYTES_OFFSET;
offset = getIndex(inIndexes, indexesLength, index);
length = getIndex(inIndexes, indexesLength, index + 1) - offset;
if(length >= 256) {
if(data == NULL) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
data->compressibleBytes = reinterpret_cast<const UBool *>(inBytes + offset);
} else if(data == NULL) {
// Nothing to do.
} else if(baseData != NULL) {
data->compressibleBytes = baseData->compressibleBytes;
} else {
errorCode = U_INVALID_FORMAT_ERROR; // No compressibleBytes[].
return;
}
const CollationSettings &ts = *tailoring.settings;
int32_t options = inIndexes[IX_OPTIONS] & 0xffff;
uint16_t fastLatinPrimaries[CollationFastLatin::LATIN_LIMIT];
int32_t fastLatinOptions = CollationFastLatin::getOptions(
tailoring.data, ts, fastLatinPrimaries, UPRV_LENGTHOF(fastLatinPrimaries));
if(options == ts.options && ts.variableTop != 0 &&
reorderCodesLength == ts.reorderCodesLength &&
uprv_memcmp(reorderCodes, ts.reorderCodes, reorderCodesLength * 4) == 0 &&
fastLatinOptions == ts.fastLatinOptions &&
(fastLatinOptions < 0 ||
uprv_memcmp(fastLatinPrimaries, ts.fastLatinPrimaries,
sizeof(fastLatinPrimaries)) == 0)) {
return;
}
CollationSettings *settings = SharedObject::copyOnWrite(tailoring.settings);
if(settings == NULL) {
errorCode = U_MEMORY_ALLOCATION_ERROR;
return;
}
settings->options = options;
// Set variableTop from options and scripts data.
settings->variableTop = tailoring.data->getLastPrimaryForGroup(
UCOL_REORDER_CODE_FIRST + settings->getMaxVariable());
if(settings->variableTop == 0) {
errorCode = U_INVALID_FORMAT_ERROR;
return;
}
if(reorderCodesLength != 0) {
settings->aliasReordering(*baseData, reorderCodes, reorderCodesLength,
reorderRanges, reorderRangesLength,
reorderTable, errorCode);
}
settings->fastLatinOptions = CollationFastLatin::getOptions(
tailoring.data, *settings,
settings->fastLatinPrimaries, UPRV_LENGTHOF(settings->fastLatinPrimaries));
}
UBool U_CALLCONV
CollationDataReader::isAcceptable(void *context,
const char * /* type */, const char * /*name*/,
const UDataInfo *pInfo) {
if(
pInfo->size >= 20 &&
pInfo->isBigEndian == U_IS_BIG_ENDIAN &&
pInfo->charsetFamily == U_CHARSET_FAMILY &&
pInfo->dataFormat[0] == 0x55 && // dataFormat="UCol"
pInfo->dataFormat[1] == 0x43 &&
pInfo->dataFormat[2] == 0x6f &&
pInfo->dataFormat[3] == 0x6c &&
pInfo->formatVersion[0] == 5
) {
UVersionInfo *version = static_cast<UVersionInfo *>(context);
if(version != NULL) {
uprv_memcpy(version, pInfo->dataVersion, 4);
}
return TRUE;
} else {
return FALSE;
}
}
U_NAMESPACE_END
#endif // !UCONFIG_NO_COLLATION