goldendict-ng/indexedzip.cc
2012-02-21 01:47:14 +04:00

211 lines
4.8 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 "indexedzip.hh"
#include "zipfile.hh"
#include <zlib.h>
#include "dprintf.hh"
#include "utf8.hh"
#include "iconv.hh"
using namespace BtreeIndexing;
using std::vector;
bool IndexedZip::openZipFile( QString const & name )
{
zip.setFileName( name );
zipIsOpen = zip.open( QFile::ReadOnly );
return zipIsOpen;
}
bool IndexedZip::hasFile( gd::wstring const & name )
{
if ( !zipIsOpen )
return false;
vector< WordArticleLink > links = findArticles( name );
return !links.empty();
}
bool IndexedZip::loadFile( gd::wstring const & name, vector< char > & data )
{
if ( !zipIsOpen )
return false;
vector< WordArticleLink > links = findArticles( name );
if ( links.empty() )
return false;
// Now seek into the zip file and read its header
if ( !zip.seek( links[ 0 ].articleOffset ) )
return false;
ZipFile::LocalFileHeader header;
if ( !ZipFile::readLocalHeader( zip, header ) )
{
DPRINTF( "Failed to load header\n" );
return false;
}
// Which algorithm was used?
switch( header.compressionMethod )
{
case ZipFile::Uncompressed:
DPRINTF( "Uncompressed\n" );
data.resize( header.uncompressedSize );
return zip.read( &data.front(), data.size() ) == data.size();
case ZipFile::Deflated:
{
DPRINTF( "Deflated\n" );
// Now do the deflation
QByteArray compressedData = zip.read( header.compressedSize );
if ( compressedData.size() != (int)header.compressedSize )
return false;
data.resize( header.uncompressedSize );
z_stream stream;
memset( &stream, 0, sizeof( stream ) );
stream.next_in = ( Bytef * ) compressedData.data();
stream.avail_in = compressedData.size();
stream.next_out = ( Bytef * ) &data.front();
stream.avail_out = data.size();
if ( inflateInit2( &stream, -MAX_WBITS ) != Z_OK )
{
data.clear();
return false;
}
if ( inflate( &stream, Z_FINISH ) != Z_STREAM_END )
{
DPRINTF( "Not zstream end!" );
data.clear();
inflateEnd( &stream );
return false;
}
inflateEnd( &stream );
return true;
}
default:
return false;
}
}
bool IndexedZip::indexFile( BtreeIndexing::IndexedWords &zipFileNames )
{
if ( !zipIsOpen )
return false;
if ( !ZipFile::positionAtCentralDir( zip ) )
return false;
// File seems to be a valid zip file
ZipFile::CentralDirEntry entry;
while( ZipFile::readNextEntry( zip, entry ) )
{
if ( entry.compressionMethod == ZipFile::Unsupported )
{
DPRINTF( "Warning: compression method unsupported -- skipping file %s\n",
entry.fileName.data() );
continue;
}
// Check if the file name has some non-ascii letters.
unsigned char const * ptr = ( unsigned char const * )
entry.fileName.constData();
bool hasNonAscii = false;
for( ; ; )
{
if ( *ptr & 0x80 )
{
hasNonAscii = true;
break;
}
else
if ( !*ptr++ )
break;
}
if ( !hasNonAscii )
{
// Add entry as is
zipFileNames.addSingleWord( Utf8::decode( entry.fileName.data() ),
entry.localHeaderOffset );
}
else
{
// Try assuming different encodings. Those are UTF8 and two
// Russian ones (Windows and Windows OEM). Unfortunately, zip
// files do not say which encoding they utilize.
// Utf8
try
{
wstring decoded = Utf8::decode( entry.fileName.constData() );
zipFileNames.addSingleWord( decoded,
entry.localHeaderOffset );
}
catch( Utf8::exCantDecode )
{
// Failed to decode
}
// CP866
try
{
wstring decoded = Iconv::toWstring( "CP866", entry.fileName.constData(),
entry.fileName.size() );
zipFileNames.addSingleWord( decoded,
entry.localHeaderOffset );
}
catch( Iconv::Ex )
{
// Failed to decode
}
// CP1251
try
{
wstring decoded = Iconv::toWstring( "CP1251", entry.fileName.constData(),
entry.fileName.size() );
zipFileNames.addSingleWord( decoded,
entry.localHeaderOffset );
}
catch( Iconv::Ex )
{
// Failed to decode
}
}
}
return true;
}