mirror of
https://github.com/xiaoyifang/goldendict-ng.git
synced 2024-11-24 16:54:08 +00:00
186 lines
4.5 KiB
C++
186 lines
4.5 KiB
C++
/* This file is (c) 2008-2012 Konstantin Isakov <ikm@goldendict.org>
|
|
* Part of GoldenDict. Licensed under GPLv3 or later, see the LICENSE file */
|
|
|
|
#include "chunkedstorage.hh"
|
|
#include <zlib.h>
|
|
#include <string.h>
|
|
#include <QDataStream>
|
|
#include <QScopeGuard>
|
|
|
|
namespace ChunkedStorage {
|
|
|
|
enum
|
|
{
|
|
ChunkMaxSize = 65536 // Can't be more since it would overflow the address
|
|
};
|
|
|
|
Writer::Writer( File::Class & f ):
|
|
file( f ), chunkStarted( false ), bufferUsed( 0 )
|
|
{
|
|
// Create a sratchpad at the beginning of file. We use it to write chunk
|
|
// table if it would fit, in order to save some seek times.
|
|
|
|
char zero[ 4096 ];
|
|
|
|
memset( zero, 0, sizeof( zero ) );
|
|
|
|
scratchPadOffset = file.tell();
|
|
scratchPadSize = sizeof( zero );
|
|
|
|
file.write( zero, sizeof( zero ) );
|
|
}
|
|
|
|
uint32_t Writer::startNewBlock()
|
|
{
|
|
if ( bufferUsed >= ChunkMaxSize )
|
|
{
|
|
// Need to flush first.
|
|
saveCurrentChunk();
|
|
}
|
|
|
|
chunkStarted = true;
|
|
|
|
// The address is comprised of the offset within the chunk (in lower
|
|
// 16 bits, always fits there since ChunkMaxSize-1 does) and the
|
|
// number of the chunk, which is therefore limited to be 65535 max.
|
|
return bufferUsed | ( (uint32_t)offsets.size() << 16 );
|
|
}
|
|
|
|
void Writer::addToBlock( void const * data, size_t size )
|
|
{
|
|
if ( !size )
|
|
return;
|
|
|
|
if ( buffer.size() - bufferUsed < size )
|
|
buffer.resize( bufferUsed + size );
|
|
|
|
memcpy( &buffer.front() + bufferUsed, data, size );
|
|
|
|
bufferUsed += size;
|
|
|
|
chunkStarted = false;
|
|
}
|
|
|
|
void Writer::saveCurrentChunk()
|
|
{
|
|
size_t maxCompressedSize = compressBound( bufferUsed );
|
|
|
|
if ( bufferCompressed.size() < maxCompressedSize )
|
|
bufferCompressed.resize( maxCompressedSize );
|
|
|
|
unsigned long compressedSize = bufferCompressed.size();
|
|
|
|
if ( compress( &bufferCompressed.front(), &compressedSize,
|
|
&buffer.front(), bufferUsed ) != Z_OK )
|
|
throw exFailedToCompressChunk();
|
|
|
|
offsets.push_back( file.tell() );
|
|
|
|
file.write( (uint32_t) bufferUsed );
|
|
file.write( (uint32_t) compressedSize );
|
|
file.write( &bufferCompressed.front(), compressedSize );
|
|
|
|
bufferUsed = 0;
|
|
|
|
chunkStarted = false;
|
|
}
|
|
|
|
uint32_t Writer::finish()
|
|
{
|
|
if ( bufferUsed || chunkStarted )
|
|
saveCurrentChunk();
|
|
|
|
bool useScratchPad = false;
|
|
uint32_t savedOffset = 0;
|
|
|
|
if ( scratchPadSize >= offsets.size() * sizeof( uint32_t ) + sizeof( uint32_t ) )
|
|
{
|
|
useScratchPad = true;
|
|
savedOffset = file.tell();
|
|
file.seek( scratchPadOffset );
|
|
}
|
|
|
|
uint32_t offset = file.tell();
|
|
|
|
file.write( (uint32_t) offsets.size() );
|
|
|
|
if ( offsets.size() )
|
|
file.write( &offsets.front(), offsets.size() * sizeof( uint32_t ) );
|
|
|
|
if ( useScratchPad )
|
|
file.seek( savedOffset );
|
|
|
|
offsets.clear();
|
|
chunkStarted = false;
|
|
|
|
return offset;
|
|
}
|
|
|
|
Reader::Reader( File::Class & f, uint32_t offset ): file( f )
|
|
{
|
|
file.seek( offset );
|
|
|
|
uint32_t size = file.read< uint32_t >();
|
|
if ( size == 0 )
|
|
return;
|
|
offsets.resize( size );
|
|
file.read( &offsets.front(), offsets.size() * sizeof( uint32_t ) );
|
|
}
|
|
|
|
char * Reader::getBlock( uint32_t address, vector< char > & chunk )
|
|
{
|
|
size_t chunkIdx = address >> 16;
|
|
|
|
if ( chunkIdx >= offsets.size() )
|
|
throw exAddressOutOfRange();
|
|
|
|
// Read and decompress the chunk
|
|
{
|
|
// file.seek( offsets[ chunkIdx ] );
|
|
QMutexLocker _( &file.lock );
|
|
auto bytes = file.map( offsets[ chunkIdx ], 8 );
|
|
if( bytes == nullptr )
|
|
throw mapFailed();
|
|
auto qBytes = QByteArray::fromRawData( reinterpret_cast< char * >(bytes), 8 );
|
|
QDataStream in( qBytes );
|
|
in.setByteOrder( QDataStream::LittleEndian );
|
|
|
|
uint32_t uncompressedSize;
|
|
uint32_t compressedSize;
|
|
|
|
in >> uncompressedSize >> compressedSize;
|
|
|
|
file.unmap( bytes );
|
|
chunk.resize( uncompressedSize );
|
|
|
|
// vector< unsigned char > compressedData( compressedSize );
|
|
auto chunkDataBytes = file.map( offsets[ chunkIdx ] + 8, compressedSize );
|
|
if( chunkDataBytes == nullptr )
|
|
throw mapFailed();
|
|
// file.read( &compressedData.front(), compressedData.size() );
|
|
auto autoUnmap = qScopeGuard(
|
|
[ & ] {
|
|
file.unmap( chunkDataBytes );
|
|
} );
|
|
Q_UNUSED( autoUnmap )
|
|
|
|
unsigned long decompressedLength = chunk.size();
|
|
|
|
if( uncompress( (unsigned char *)&chunk.front(), &decompressedLength, chunkDataBytes, compressedSize ) != Z_OK
|
|
|| decompressedLength != chunk.size() )
|
|
{
|
|
throw exFailedToDecompressChunk();
|
|
}
|
|
|
|
}
|
|
|
|
size_t offsetInChunk = address & 0xffFF;
|
|
|
|
if ( offsetInChunk > chunk.size() ) // It can be equal to for 0-sized blocks
|
|
throw exAddressOutOfRange();
|
|
|
|
return &chunk.front() + offsetInChunk;
|
|
}
|
|
|
|
}
|