summaryrefslogblamecommitdiffstats
path: root/src/WorldStorage/WSSCompact.cpp
blob: 885099713550b9de51c1fe7f9f86f0a3bd8f7d72 (plain) (tree)
1
2
3
4
5
6
7
8
9






                                                                                            
                     
                      
                      
                                 
                                         
                                                
                                             
                                             

                                           
                                           

                                        




















                                                   
                                                  

































                                                                                                                       


                                                                              
                                                                              





                                                                              
                                                                              
 


























































































                                                                                                                       

                                                                



















                                                                                                           
                                                                                   




































































                                                                                                                                             
                       




                                                                                                   

                                                                                                    
                         
                                                                              


                            
                                                                                 



                                           
                           
                                                                               
                                                                                                  
         

                                                                                                        
                 






                                                                             
 














                                                                                                        
                         
                                                                           
                                                                                              
         


                                                                                                                      
                 






                                                                           
 
                      
                                                                     
                                                                                        
         

                                                                                                            
                 






                                                                        
 
                            
                                                                     
                                                                                        
         

                                                                                         
                 






                                                                           
 
                          
                                                                             
                                                                                                
         

                                                                                                  
                 






                                                                           
 
                               
                                                                                     
                                                                                                        
         

                                                                                                                 
                 






                                                                                














                                                                                                  















                                                                                                                                    
                                                                                                                  
                               
                                                 






















































































































































































































































                                                                                                                                                       
                                                                                                                               








































































































































                                                                                                                                                                              
                                                                                                                                



















































































                                                                                                                                     


                                                                             

                              
                                                   










































                                                                                                      
                                                                                   

















                                                                                                                                                                
                                                                                                      


































































                                                                                                                                                       

// WSSCompact.cpp

// Interfaces to the cWSSCompact class representing the "compact" storage schema (PAK-files)

#include "Globals.h"
#include "WSSCompact.h"
#include "../World.h"
#include "zlib/zlib.h"
#include "json/json.h"
#include "../StringCompression.h"
#include "../BlockEntities/ChestEntity.h"
#include "../BlockEntities/CommandBlockEntity.h"
#include "../BlockEntities/DispenserEntity.h"
#include "../BlockEntities/FlowerPotEntity.h"
#include "../BlockEntities/FurnaceEntity.h"
#include "../BlockEntities/JukeboxEntity.h"
#include "../BlockEntities/MobHeadEntity.h"
#include "../BlockEntities/NoteEntity.h"
#include "../BlockEntities/SignEntity.h"





#pragma pack(push, 1)
/// The chunk header, as stored in the file:
struct cWSSCompact::sChunkHeader
{
	int m_ChunkX;
	int m_ChunkZ;
	int m_CompressedSize;
	int m_UncompressedSize;
} ;
#pragma pack(pop)





/// The maximum number of PAK files that are cached
const int MAX_PAK_FILES = static_cast<size_t>(16);

/// The maximum number of unsaved chunks before the cPAKFile saves them to disk
const int MAX_DIRTY_CHUNKS = 16;





///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// cJsonChunkSerializer:

cJsonChunkSerializer::cJsonChunkSerializer(void) :
	m_HasJsonData(false)
{
}





void cJsonChunkSerializer::Entity(cEntity * a_Entity)
{
	// TODO: a_Entity->SaveToJson(m_Root);
}





void cJsonChunkSerializer::BlockEntity(cBlockEntity * a_BlockEntity)
{
	const char * SaveInto = NULL;
	switch (a_BlockEntity->GetBlockType())
	{
		case E_BLOCK_CHEST:         SaveInto = "Chests";        break;
		case E_BLOCK_DISPENSER:     SaveInto = "Dispensers";    break;
		case E_BLOCK_DROPPER:       SaveInto = "Droppers";      break;
		case E_BLOCK_FLOWER_POT:    SaveInto = "FlowerPots";    break;
		case E_BLOCK_FURNACE:       SaveInto = "Furnaces";      break;
		case E_BLOCK_SIGN_POST:     SaveInto = "Signs";         break;
		case E_BLOCK_WALLSIGN:      SaveInto = "Signs";         break;
		case E_BLOCK_NOTE_BLOCK:    SaveInto = "Notes";         break;
		case E_BLOCK_JUKEBOX:       SaveInto = "Jukeboxes";     break;
		case E_BLOCK_COMMAND_BLOCK: SaveInto = "CommandBlocks"; break;
		case E_BLOCK_HEAD:          SaveInto = "MobHeads";      break;

		default:
		{
			ASSERT(!"Unhandled blocktype in BlockEntities list while saving to JSON");
			break;
		}
	}  // switch (BlockEntity->GetBlockType())
	if (SaveInto == NULL)
	{
		return;
	}
	
	Json::Value val;
	a_BlockEntity->SaveToJson(val);
	m_Root[SaveInto].append(val);
	m_HasJsonData = true;
}





bool cJsonChunkSerializer::LightIsValid(bool a_IsLightValid)
{
	if (!a_IsLightValid)
	{
		return false;
	}
	m_Root["IsLightValid"] = true;
	m_HasJsonData = true;
	return true;
}





///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// cWSSCompact:

cWSSCompact::~cWSSCompact()
{
	for (cPAKFiles::iterator itr = m_PAKFiles.begin(); itr != m_PAKFiles.end(); ++itr)
	{
		delete *itr;
	}
}





bool cWSSCompact::LoadChunk(const cChunkCoords & a_Chunk)
{
	AString ChunkData;
	int UncompressedSize = 0;
	if (!GetChunkData(a_Chunk, UncompressedSize, ChunkData))
	{
		// The reason for failure is already printed in GetChunkData()
		return false;
	}
	
	return LoadChunkFromData(a_Chunk, UncompressedSize, ChunkData, m_World);
}





bool cWSSCompact::SaveChunk(const cChunkCoords & a_Chunk)
{
	cCSLock Lock(m_CS);
	
	cPAKFile * f = LoadPAKFile(a_Chunk);
	if (f == NULL)
	{
		// For some reason we couldn't locate the file
		LOG("Cannot locate a proper PAK file for chunk [%d, %d]", a_Chunk.m_ChunkX, a_Chunk.m_ChunkZ);
		return false;
	}
	return f->SaveChunk(a_Chunk, m_World);
}





cWSSCompact::cPAKFile * cWSSCompact::LoadPAKFile(const cChunkCoords & a_Chunk)
{
	// ASSUMES that m_CS has been locked
	
	// We need to retain this weird conversion code, because some edge chunks are in the wrong PAK file
	const int LayerX = FAST_FLOOR_DIV(a_Chunk.m_ChunkX, 32);
	const int LayerZ = FAST_FLOOR_DIV(a_Chunk.m_ChunkZ, 32);
	
	// Is it already cached?
	for (cPAKFiles::iterator itr = m_PAKFiles.begin(); itr != m_PAKFiles.end(); ++itr)
	{
		if (((*itr) != NULL) && ((*itr)->GetLayerX() == LayerX) && ((*itr)->GetLayerZ() == LayerZ))
		{
			// Move the file to front and return it:
			cPAKFile * f = *itr;
			if (itr != m_PAKFiles.begin())
			{
				m_PAKFiles.erase(itr);
				m_PAKFiles.push_front(f);
			}
			return f;
		}
	}
	
	// Load it anew:
	AString FileName;
	Printf(FileName, "%s/X%i_Z%i.pak", m_World->GetName().c_str(), LayerX, LayerZ );
	cPAKFile * f = new cPAKFile(FileName, LayerX, LayerZ, m_CompressionFactor);
	if (f == NULL)
	{
		return NULL;
	}
	m_PAKFiles.push_front(f);
	
	// If there are too many PAK files cached, delete the last one used:
	if (m_PAKFiles.size() > MAX_PAK_FILES)
	{
		delete m_PAKFiles.back();
		m_PAKFiles.pop_back();
	}
	return f;
}





bool cWSSCompact::GetChunkData(const cChunkCoords & a_Chunk, int & a_UncompressedSize, AString & a_Data)
{
	cCSLock Lock(m_CS);
	cPAKFile * f = LoadPAKFile(a_Chunk);
	if (f == NULL)
	{
		return false;
	}
	return f->GetChunkData(a_Chunk, a_UncompressedSize, a_Data);
}





/*
// TODO: Rewrite saving to use the same principles as loading
bool cWSSCompact::SetChunkData(const cChunkCoords & a_Chunk, int a_UncompressedSize, const AString & a_Data)
{
	cCSLock Lock(m_CS);
	cPAKFile * f = LoadPAKFile(a_Chunk);
	if (f == NULL)
	{
		return false;
	}
	return f->SetChunkData(a_Chunk, a_UncompressedSize, a_Data);
}
*/





bool cWSSCompact::EraseChunkData(const cChunkCoords & a_Chunk)
{
	cCSLock Lock(m_CS);
	cPAKFile * f = LoadPAKFile(a_Chunk);
	if (f == NULL)
	{
		return false;
	}
	return f->EraseChunkData(a_Chunk);
}





void cWSSCompact::LoadEntitiesFromJson(Json::Value & a_Value, cEntityList & a_Entities, cBlockEntityList & a_BlockEntities, cWorld * a_World)
{
	// Load chests:
	Json::Value AllChests = a_Value.get("Chests", Json::nullValue);
	if (!AllChests.empty())
	{
		for (Json::Value::iterator itr = AllChests.begin(); itr != AllChests.end(); ++itr )
		{
			std::auto_ptr<cChestEntity> ChestEntity(new cChestEntity(0, 0, 0, a_World));
			if (!ChestEntity->LoadFromJson(*itr))
			{
				LOGWARNING("ERROR READING CHEST FROM JSON!" );
			}
			else
			{
				a_BlockEntities.push_back(ChestEntity.release());
			}
		}  // for itr - AllChests[]
	}

	// Load dispensers:
	Json::Value AllDispensers = a_Value.get("Dispensers", Json::nullValue);
	for (Json::Value::iterator itr = AllDispensers.begin(); itr != AllDispensers.end(); ++itr)
	{
		std::auto_ptr<cDispenserEntity> DispenserEntity(new cDispenserEntity(0, 0, 0, a_World));
		if (!DispenserEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING DISPENSER FROM JSON!" );
		}
		else
		{
			a_BlockEntities.push_back(DispenserEntity.release());
		}
	}  // for itr - AllDispensers[]

	// Load Flowerpots:
	Json::Value AllFlowerPots = a_Value.get("FlowerPots", Json::nullValue);
	for (Json::Value::iterator itr = AllFlowerPots.begin(); itr != AllFlowerPots.end(); ++itr)
	{
		std::auto_ptr<cFlowerPotEntity> FlowerPotEntity(new cFlowerPotEntity(0, 0, 0, a_World));
		if (!FlowerPotEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING FLOWERPOT FROM JSON!" );
		}
		else
		{
			a_BlockEntities.push_back(FlowerPotEntity.release());
		}
	}  // for itr - AllFlowerPots[]

	// Load furnaces:
	Json::Value AllFurnaces = a_Value.get("Furnaces", Json::nullValue);
	for (Json::Value::iterator itr = AllFurnaces.begin(); itr != AllFurnaces.end(); ++itr)
	{
		// TODO: The block type and meta aren't correct, there's no way to get them here
		std::auto_ptr<cFurnaceEntity> FurnaceEntity(new cFurnaceEntity(0, 0, 0, E_BLOCK_FURNACE, 0, a_World));
		if (!FurnaceEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING FURNACE FROM JSON!" );
		}
		else
		{
			a_BlockEntities.push_back(FurnaceEntity.release());
		}
	}  // for itr - AllFurnaces[]

	// Load signs:
	Json::Value AllSigns = a_Value.get("Signs", Json::nullValue);
	for (Json::Value::iterator itr = AllSigns.begin(); itr != AllSigns.end(); ++itr)
	{
		std::auto_ptr<cSignEntity> SignEntity(new cSignEntity(E_BLOCK_SIGN_POST, 0, 0, 0, a_World));
		if (!SignEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING SIGN FROM JSON!");
		}
		else
		{
			a_BlockEntities.push_back(SignEntity.release());
		}
	}  // for itr - AllSigns[]

	// Load note blocks:
	Json::Value AllNotes = a_Value.get("Notes", Json::nullValue);
	for (Json::Value::iterator itr = AllNotes.begin(); itr != AllNotes.end(); ++itr)
	{
		std::auto_ptr<cNoteEntity> NoteEntity(new cNoteEntity(0, 0, 0, a_World));
		if (!NoteEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING NOTE BLOCK FROM JSON!" );
		}
		else
		{
			a_BlockEntities.push_back(NoteEntity.release());
		}
	}  // for itr - AllNotes[]

	// Load jukeboxes:
	Json::Value AllJukeboxes = a_Value.get("Jukeboxes", Json::nullValue);
	for (Json::Value::iterator itr = AllJukeboxes.begin(); itr != AllJukeboxes.end(); ++itr)
	{
		std::auto_ptr<cJukeboxEntity> JukeboxEntity(new cJukeboxEntity(0, 0, 0, a_World));
		if (!JukeboxEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING JUKEBOX FROM JSON!" );
		}
		else
		{
			a_BlockEntities.push_back(JukeboxEntity.release());
		}
	}  // for itr - AllJukeboxes[]

	// Load command blocks:
	Json::Value AllCommandBlocks = a_Value.get("CommandBlocks", Json::nullValue);
	for (Json::Value::iterator itr = AllCommandBlocks.begin(); itr != AllCommandBlocks.end(); ++itr)
	{
		std::auto_ptr<cCommandBlockEntity> CommandBlockEntity(new cCommandBlockEntity(0, 0, 0, a_World));
		if (!CommandBlockEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING COMMAND BLOCK FROM JSON!" );
		}
		else
		{
			a_BlockEntities.push_back(CommandBlockEntity.release());
		}
	}  // for itr - AllCommandBlocks[]

	// Load mob heads:
	Json::Value AllMobHeads = a_Value.get("MobHeads", Json::nullValue);
	for (Json::Value::iterator itr = AllMobHeads.begin(); itr != AllMobHeads.end(); ++itr)
	{
		std::auto_ptr<cMobHeadEntity> MobHeadEntity(new cMobHeadEntity(0, 0, 0, a_World));
		if (!MobHeadEntity->LoadFromJson(*itr))
		{
			LOGWARNING("ERROR READING MOB HEAD FROM JSON!" );
		}
		else
		{
			a_BlockEntities.push_back(MobHeadEntity.release());
		}
	}  // for itr - AllMobHeads[]
}





///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
// cWSSCompact::cPAKFile

#define READ(Var) \
	if (f.Read(&Var, sizeof(Var)) != sizeof(Var)) \
	{ \
		LOGERROR("ERROR READING %s FROM FILE %s (line %d); file offset %d", #Var, m_FileName.c_str(), __LINE__, f.Tell()); \
		return; \
	}

cWSSCompact::cPAKFile::cPAKFile(const AString & a_FileName, int a_LayerX, int a_LayerZ, int a_CompressionFactor) :
	m_FileName(a_FileName),
	m_CompressionFactor(a_CompressionFactor),
	m_LayerX(a_LayerX),
	m_LayerZ(a_LayerZ),
	m_NumDirty(0),
	m_ChunkVersion( CHUNK_VERSION ), // Init with latest version
	m_PakVersion( PAK_VERSION )
{
	cFile f;
	if (!f.Open(m_FileName, cFile::fmRead))
	{
		return;
	}
	
	// Read headers:
	READ(m_PakVersion);
	if (m_PakVersion != 1)
	{
		LOGERROR("File \"%s\" is in an unknown pak format (%d)", m_FileName.c_str(), m_PakVersion);
		return;
	}
	
	READ(m_ChunkVersion);
	switch( m_ChunkVersion )
	{
	case 1:
		m_ChunkSize.Set(16, 128, 16);
		break;
	case 2:
	case 3:
		m_ChunkSize.Set(16, 256, 16);
		break;
	default:
		LOGERROR("File \"%s\" is in an unknown chunk format (%d)", m_FileName.c_str(), m_ChunkVersion);
		return;
	};
	
	short NumChunks = 0;
	READ(NumChunks);
	
	// Read chunk headers:
	for (int i = 0; i < NumChunks; i++)
	{
		sChunkHeader * Header = new sChunkHeader;
		READ(*Header);
		m_ChunkHeaders.push_back(Header);
	}  // for i - chunk headers

	// Read chunk data:
	if (f.ReadRestOfFile(m_DataContents) == -1)
	{
		LOGERROR("Cannot read file \"%s\" contents", m_FileName.c_str());
		return;
	}

	if( m_ChunkVersion == 1 ) // Convert chunks to version 2
	{
		UpdateChunk1To2();
	}
#if AXIS_ORDER == AXIS_ORDER_XZY
	if( m_ChunkVersion == 2 ) // Convert chunks to version 3
	{
		UpdateChunk2To3();
	}
#endif
}





cWSSCompact::cPAKFile::~cPAKFile()
{
	if (m_NumDirty > 0)
	{
		SynchronizeFile();
	}
	for (sChunkHeaders::iterator itr = m_ChunkHeaders.begin(); itr != m_ChunkHeaders.end(); ++itr)
	{
		delete *itr;
	}
}





bool cWSSCompact::cPAKFile::GetChunkData(const cChunkCoords & a_Chunk, int & a_UncompressedSize, AString & a_Data)
{
	int ChunkX = a_Chunk.m_ChunkX;
	int ChunkZ = a_Chunk.m_ChunkZ;
	sChunkHeader * Header = NULL;
	int Offset = 0;
	for (sChunkHeaders::iterator itr = m_ChunkHeaders.begin(); itr != m_ChunkHeaders.end(); ++itr)
	{
		if (((*itr)->m_ChunkX == ChunkX) && ((*itr)->m_ChunkZ == ChunkZ))
		{
			Header = *itr;
			break;
		}
		Offset += (*itr)->m_CompressedSize;
	}
	if ((Header == NULL) || (Offset + Header->m_CompressedSize > (int)m_DataContents.size()))
	{
		// Chunk not found / data invalid
		return false;
	}

	a_UncompressedSize = Header->m_UncompressedSize;
	a_Data.assign(m_DataContents, Offset, Header->m_CompressedSize);
	return true;
}





bool cWSSCompact::cPAKFile::SaveChunk(const cChunkCoords & a_Chunk, cWorld * a_World)
{
	if (!SaveChunkToData(a_Chunk, a_World))
	{
		return false;
	}
	if (m_NumDirty > MAX_DIRTY_CHUNKS)
	{
		SynchronizeFile();
	}
	return true;
}





void cWSSCompact::cPAKFile::UpdateChunk1To2()
{
	int Offset = 0;
	AString NewDataContents;
	int ChunksConverted = 0;
	for (sChunkHeaders::iterator itr = m_ChunkHeaders.begin(); itr != m_ChunkHeaders.end(); ++itr)
	{
		sChunkHeader * Header = *itr;

		if( ChunksConverted % 32 == 0 )
		{
			LOGINFO("Updating \"%s\" version 1 to version 2: %d %%", m_FileName.c_str(), (ChunksConverted * 100) / m_ChunkHeaders.size() );
		}
		ChunksConverted++;

		AString Data;
		int UncompressedSize = Header->m_UncompressedSize;
		Data.assign(m_DataContents, Offset, Header->m_CompressedSize);
		Offset += Header->m_CompressedSize;

		// Crude data integrity check:
		int ExpectedSize = (16*128*16)*2 + (16*128*16)/2; // For version 1
		if (UncompressedSize < ExpectedSize)
		{
			LOGWARNING("Chunk [%d, %d] has too short decompressed data (%d bytes out of %d needed), erasing",
				Header->m_ChunkX, Header->m_ChunkZ,
				UncompressedSize, ExpectedSize
				);
			Offset += Header->m_CompressedSize;
			continue;
		}

		// Decompress the data:
		AString UncompressedData;
		{
			int errorcode = UncompressString(Data.data(), Data.size(), UncompressedData, UncompressedSize);
			if (errorcode != Z_OK)
			{
				LOGERROR("Error %d decompressing data for chunk [%d, %d]", 
					errorcode,
					Header->m_ChunkX, Header->m_ChunkZ
					);
				Offset += Header->m_CompressedSize;
				continue;
			}
		}

		if (UncompressedSize != (int)UncompressedData.size())
		{
			LOGWARNING("Uncompressed data size differs (exp %d bytes, got %d) for chunk [%d, %d]",
				UncompressedSize, UncompressedData.size(),
				Header->m_ChunkX, Header->m_ChunkZ
				);
			Offset += Header->m_CompressedSize;
			continue;
		}


		// Old version is 128 blocks high with YZX axis order
		char ConvertedData[cChunkDef::BlockDataSize];
		int Index = 0;
		unsigned int InChunkOffset = 0;
		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) 
		{
			for( int y = 0; y < 128; ++y )
			{
				ConvertedData[Index++] = UncompressedData[y + z * 128 + x * 128 * 16 + InChunkOffset];
			}
			// Add 128 empty blocks after an old y column
			memset(ConvertedData + Index, E_BLOCK_AIR, 128);
			Index += 128;
		}
		InChunkOffset += (16 * 128 * 16);
		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) // Metadata
		{
			for( int y = 0; y < 64; ++y )
			{
				ConvertedData[Index++] = UncompressedData[y + z * 64 + x * 64 * 16 + InChunkOffset];
			}
			memset(ConvertedData + Index, 0, 64);
			Index += 64;
		}
		InChunkOffset += (16 * 128 * 16) / 2;
		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) // Block light
		{
			for( int y = 0; y < 64; ++y )
			{
				ConvertedData[Index++] = UncompressedData[y + z * 64 + x * 64 * 16 + InChunkOffset];
			}
			memset(ConvertedData + Index, 0, 64);
			Index += 64;
		}
		InChunkOffset += (16*128*16)/2;
		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) // Sky light
		{
			for( int y = 0; y < 64; ++y )
			{
				ConvertedData[Index++] = UncompressedData[y + z * 64 + x * 64 * 16 + InChunkOffset];
			}
			memset(ConvertedData + Index, 0, 64);
			Index += 64;
		}
		InChunkOffset += (16 * 128 * 16) / 2;

		AString Converted(ConvertedData, ARRAYCOUNT(ConvertedData));
		
		// Add JSON data afterwards
		if (UncompressedData.size() > InChunkOffset)
		{
			Converted.append( UncompressedData.begin() + InChunkOffset, UncompressedData.end() );
		}

		// Re-compress data
		AString CompressedData;
		{
			int errorcode = CompressString(Converted.data(), Converted.size(), CompressedData,m_CompressionFactor);
			if (errorcode != Z_OK)
			{
				LOGERROR("Error %d compressing data for chunk [%d, %d]", 
					errorcode,
					Header->m_ChunkX, Header->m_ChunkZ
				);
				continue;
			}
		}

		// Save into file's cache
		Header->m_UncompressedSize = Converted.size();
		Header->m_CompressedSize = CompressedData.size();
		NewDataContents.append( CompressedData );
	}

	// Done converting
	m_DataContents = NewDataContents;
	m_ChunkVersion = 2;
	SynchronizeFile();
	
	LOGINFO("Updated \"%s\" version 1 to version 2", m_FileName.c_str() );
}





void cWSSCompact::cPAKFile::UpdateChunk2To3()
{
	int Offset = 0;
	AString NewDataContents;
	int ChunksConverted = 0;
	for (sChunkHeaders::iterator itr = m_ChunkHeaders.begin(); itr != m_ChunkHeaders.end(); ++itr)
	{
		sChunkHeader * Header = *itr;

		if( ChunksConverted % 32 == 0 )
		{
			LOGINFO("Updating \"%s\" version 2 to version 3: %d %%", m_FileName.c_str(), (ChunksConverted * 100) / m_ChunkHeaders.size() );
		}
		ChunksConverted++;

		AString Data;
		int UncompressedSize = Header->m_UncompressedSize;
		Data.assign(m_DataContents, Offset, Header->m_CompressedSize);
		Offset += Header->m_CompressedSize;

		// Crude data integrity check:
		const int ExpectedSize = (16*256*16)*2 + (16*256*16)/2; // For version 2
		if (UncompressedSize < ExpectedSize)
		{
			LOGWARNING("Chunk [%d, %d] has too short decompressed data (%d bytes out of %d needed), erasing",
				Header->m_ChunkX, Header->m_ChunkZ,
				UncompressedSize, ExpectedSize
				);
			Offset += Header->m_CompressedSize;
			continue;
		}

		// Decompress the data:
		AString UncompressedData;
		{
			int errorcode = UncompressString(Data.data(), Data.size(), UncompressedData, UncompressedSize);
			if (errorcode != Z_OK)
			{
				LOGERROR("Error %d decompressing data for chunk [%d, %d]", 
					errorcode,
					Header->m_ChunkX, Header->m_ChunkZ
					);
				Offset += Header->m_CompressedSize;
				continue;
			}
		}

		if (UncompressedSize != (int)UncompressedData.size())
		{
			LOGWARNING("Uncompressed data size differs (exp %d bytes, got %d) for chunk [%d, %d]",
				UncompressedSize, UncompressedData.size(),
				Header->m_ChunkX, Header->m_ChunkZ
				);
			Offset += Header->m_CompressedSize;
			continue;
		}

		char ConvertedData[ExpectedSize];
		memset(ConvertedData, 0, ExpectedSize);

		// Cannot use cChunk::MakeIndex because it might change again?????????
		// For compatibility, use what we know is current
		#define MAKE_2_INDEX( x, y, z ) ( y + (z * 256) + (x * 256 * 16) )
		#define MAKE_3_INDEX( x, y, z ) ( x + (z * 16) + (y * 16 * 16) )

		unsigned int InChunkOffset = 0;
		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) for( int y = 0; y < 256; ++y ) // YZX Loop order is important, in 1.1 Y was first then Z then X
		{
			ConvertedData[ MAKE_3_INDEX(x, y, z) ] = UncompressedData[InChunkOffset];
			++InChunkOffset;
		}  // for y, z, x

		
		unsigned int index2 = 0;
		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) for( int y = 0; y < 256; ++y )
		{
			ConvertedData[ InChunkOffset + MAKE_3_INDEX(x, y, z)/2 ] |= ( (UncompressedData[ InChunkOffset + index2/2 ] >> ((index2&1)*4) ) & 0x0f ) << ((x&1)*4);
			++index2;
		}
		InChunkOffset += index2 / 2;
		index2 = 0;

		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) for( int y = 0; y < 256; ++y )
		{
			ConvertedData[ InChunkOffset + MAKE_3_INDEX(x, y, z)/2 ] |= ( (UncompressedData[ InChunkOffset + index2/2 ] >> ((index2&1)*4) ) & 0x0f ) << ((x&1)*4);
			++index2;
		}
		InChunkOffset += index2 / 2;
		index2 = 0;

		for( int x = 0; x < 16; ++x ) for( int z = 0; z < 16; ++z ) for( int y = 0; y < 256; ++y )
		{
			ConvertedData[ InChunkOffset + MAKE_3_INDEX(x, y, z)/2 ] |= ( (UncompressedData[ InChunkOffset + index2/2 ] >> ((index2&1)*4) ) & 0x0f ) << ((x&1)*4);
			++index2;
		}
		InChunkOffset += index2 / 2;
		index2 = 0;

		AString Converted(ConvertedData, ExpectedSize);

		// Add JSON data afterwards
		if (UncompressedData.size() > InChunkOffset)
		{
			Converted.append( UncompressedData.begin() + InChunkOffset, UncompressedData.end() );
		}

		// Re-compress data
		AString CompressedData;
		{
			int errorcode = CompressString(Converted.data(), Converted.size(), CompressedData, m_CompressionFactor);
			if (errorcode != Z_OK)
			{
				LOGERROR("Error %d compressing data for chunk [%d, %d]", 
					errorcode,
					Header->m_ChunkX, Header->m_ChunkZ
					);
				continue;
			}
		}

		// Save into file's cache
		Header->m_UncompressedSize = Converted.size();
		Header->m_CompressedSize = CompressedData.size();
		NewDataContents.append( CompressedData );
	}

	// Done converting
	m_DataContents = NewDataContents;
	m_ChunkVersion = 3;
	SynchronizeFile();

	LOGINFO("Updated \"%s\" version 2 to version 3", m_FileName.c_str() );
}





bool cWSSCompact::LoadChunkFromData(const cChunkCoords & a_Chunk, int & a_UncompressedSize, const AString & a_Data, cWorld * a_World)
{
	// Crude data integrity check:
	if (a_UncompressedSize < cChunkDef::BlockDataSize)
	{
		LOGWARNING("Chunk [%d, %d] has too short decompressed data (%d bytes out of %d needed), erasing",
			a_Chunk.m_ChunkX, a_Chunk.m_ChunkZ,
			a_UncompressedSize, cChunkDef::BlockDataSize
		);
		EraseChunkData(a_Chunk);
		return false;
	}
	
	// Decompress the data:
	AString UncompressedData;
	int errorcode = UncompressString(a_Data.data(), a_Data.size(), UncompressedData, a_UncompressedSize);
	if (errorcode != Z_OK)
	{
		LOGERROR("Error %d decompressing data for chunk [%d, %d]", 
			errorcode,
			a_Chunk.m_ChunkX, a_Chunk.m_ChunkZ
		);
		return false;
	}
	
	if (a_UncompressedSize != (int)UncompressedData.size())
	{
		LOGWARNING("Uncompressed data size differs (exp %d bytes, got %d) for chunk [%d, %d]",
			a_UncompressedSize, UncompressedData.size(),
			a_Chunk.m_ChunkX, a_Chunk.m_ChunkZ
		);
		return false;
	}

	cEntityList      Entities;
	cBlockEntityList BlockEntities;
	bool IsLightValid = false;
	
	if (a_UncompressedSize > cChunkDef::BlockDataSize)
	{
		Json::Value root;   // will contain the root value after parsing.
		Json::Reader reader;
		if ( !reader.parse( UncompressedData.data() + cChunkDef::BlockDataSize, root, false ) )
		{
			LOGERROR("Failed to parse trailing JSON in chunk [%d, %d]!",
				a_Chunk.m_ChunkX, a_Chunk.m_ChunkZ
			);
		}
		else
		{
			LoadEntitiesFromJson(root, Entities, BlockEntities, a_World);
			IsLightValid = root.get("IsLightValid", false).asBool();
		}
	}

	BLOCKTYPE *  BlockData  = (BLOCKTYPE *)UncompressedData.data();
	NIBBLETYPE * MetaData   = (NIBBLETYPE *)(BlockData + MetaOffset);
	NIBBLETYPE * BlockLight = (NIBBLETYPE *)(BlockData + LightOffset);
	NIBBLETYPE * SkyLight   = (NIBBLETYPE *)(BlockData + SkyLightOffset);
	
	a_World->SetChunkData(
		a_Chunk.m_ChunkX, a_Chunk.m_ChunkZ,
		BlockData, MetaData,
		IsLightValid ? BlockLight : NULL,
		IsLightValid ? SkyLight : NULL,
		NULL, NULL,
		Entities, BlockEntities,
		false
	);

	return true;
}





bool cWSSCompact::cPAKFile::EraseChunkData(const cChunkCoords & a_Chunk)
{
	int ChunkX = a_Chunk.m_ChunkX;
	int ChunkZ = a_Chunk.m_ChunkZ;
	int Offset = 0;
	for (sChunkHeaders::iterator itr = m_ChunkHeaders.begin(); itr != m_ChunkHeaders.end(); ++itr)
	{
		if (((*itr)->m_ChunkX == ChunkX) && ((*itr)->m_ChunkZ == ChunkZ))
		{
			m_DataContents.erase(Offset, (*itr)->m_CompressedSize);
			delete *itr;
			itr = m_ChunkHeaders.erase(itr);
			return true;
		}
		Offset += (*itr)->m_CompressedSize;
	}
	
	return false;
}





bool cWSSCompact::cPAKFile::SaveChunkToData(const cChunkCoords & a_Chunk, cWorld * a_World)
{
	// Serialize the chunk:
	cJsonChunkSerializer Serializer;
	if (!a_World->GetChunkData(a_Chunk.m_ChunkX, a_Chunk.m_ChunkZ, Serializer))
	{
		// Chunk not valid
		LOG("cWSSCompact: Trying to save chunk [%d, %d, %d] that has no data, ignoring request.", a_Chunk.m_ChunkX, a_Chunk.m_ChunkY, a_Chunk.m_ChunkZ);
		return false;
	}

	AString Data;
	Data.assign((const char *)Serializer.GetBlockData(), cChunkDef::BlockDataSize);
	if (Serializer.HasJsonData())
	{
		AString JsonData;
		Json::StyledWriter writer;
		JsonData = writer.write(Serializer.GetRoot());
		Data.append(JsonData);
	}
	
	// Compress the data:
	AString CompressedData;
	int errorcode = CompressString(Data.data(), Data.size(), CompressedData, m_CompressionFactor);
	if ( errorcode != Z_OK )
	{
		LOGERROR("Error %i compressing data for chunk [%d, %d, %d]", errorcode, a_Chunk.m_ChunkX, a_Chunk.m_ChunkY, a_Chunk.m_ChunkZ);
		return false;
	}
	
	// Erase any existing data for the chunk:
	EraseChunkData(a_Chunk);
	
	// Save the header:
	sChunkHeader * Header = new sChunkHeader;
	if (Header == NULL)
	{
		LOGWARNING("Cannot create a new chunk header to save chunk [%d, %d, %d]", a_Chunk.m_ChunkX, a_Chunk.m_ChunkY, a_Chunk.m_ChunkZ);
		return false;
	}
	Header->m_CompressedSize = (int)CompressedData.size();
	Header->m_ChunkX = a_Chunk.m_ChunkX;
	Header->m_ChunkZ = a_Chunk.m_ChunkZ;
	Header->m_UncompressedSize = (int)Data.size();
	m_ChunkHeaders.push_back(Header);
	
	m_DataContents.append(CompressedData.data(), CompressedData.size());
	
	m_NumDirty++;
	return true;
}





#define WRITE(Var) \
	if (f.Write(&Var, sizeof(Var)) != sizeof(Var)) \
	{ \
		LOGERROR("cWSSCompact: ERROR writing %s to file \"%s\" (line %d); file offset %d", #Var, m_FileName.c_str(), __LINE__, f.Tell()); \
		return; \
	}

void cWSSCompact::cPAKFile::SynchronizeFile(void)
{
	cFile f;
	if (!f.Open(m_FileName, cFile::fmWrite))
	{
		LOGERROR("Cannot open PAK file \"%s\" for writing", m_FileName.c_str());
		return;
	}
	
	WRITE(m_PakVersion);
	WRITE(m_ChunkVersion);
	short NumChunks = (short)m_ChunkHeaders.size();
	WRITE(NumChunks);
	for (sChunkHeaders::iterator itr = m_ChunkHeaders.begin(); itr != m_ChunkHeaders.end(); ++itr)
	{
		WRITE(**itr);
	}
	if (f.Write(m_DataContents.data(), m_DataContents.size()) != (int)m_DataContents.size())
	{
		LOGERROR("cWSSCompact: ERROR writing chunk contents to file \"%s\" (line %d); file offset %d", m_FileName.c_str(), __LINE__, f.Tell());
		return;
	}
	m_NumDirty = 0;
}