summaryrefslogtreecommitdiffstats
path: root/converter/main.cpp
diff options
context:
space:
mode:
authoradmin@omencraft.com <admin@omencraft.com@0a769ca7-a7f5-676a-18bf-c427514a06d6>2011-11-04 17:27:11 +0100
committeradmin@omencraft.com <admin@omencraft.com@0a769ca7-a7f5-676a-18bf-c427514a06d6>2011-11-04 17:27:11 +0100
commit563028f6dbaad33152a873ca68f4b619839a9589 (patch)
tree37de244a7e14441b58f55ad904e502e51f2ab163 /converter/main.cpp
parentAdded cRedstone to project file (diff)
downloadcuberite-563028f6dbaad33152a873ca68f4b619839a9589.tar
cuberite-563028f6dbaad33152a873ca68f4b619839a9589.tar.gz
cuberite-563028f6dbaad33152a873ca68f4b619839a9589.tar.bz2
cuberite-563028f6dbaad33152a873ca68f4b619839a9589.tar.lz
cuberite-563028f6dbaad33152a873ca68f4b619839a9589.tar.xz
cuberite-563028f6dbaad33152a873ca68f4b619839a9589.tar.zst
cuberite-563028f6dbaad33152a873ca68f4b619839a9589.zip
Diffstat (limited to '')
-rw-r--r--converter/main.cpp (renamed from converter/cConvert.cpp)77
1 files changed, 19 insertions, 58 deletions
diff --git a/converter/cConvert.cpp b/converter/main.cpp
index 4ee5b6b74..08a480dd1 100644
--- a/converter/cConvert.cpp
+++ b/converter/main.cpp
@@ -1,13 +1,14 @@
#include <iostream>
#include <fstream>
#include <string>
+#include <cstring>
#include <stdio.h>
//#include <string.h>
#include <ctype.h>
#include "zlib.h"
#include "cNBTData.h"
-#include "timer.h"
-#include "quicksort.h"
+#include "cTimer.h"
+#include "cQuicksort.h"
#include <dirent.h>
//#include "dircont.h"
@@ -112,7 +113,9 @@ int main () {
if( fread( &trash, sizeof(byte4), 1, f) != 1 ) { cout << "ERROR 2jkd READING FROM FILE " << SourceFile; fclose(f); return false; }
}
frloc = 8192; //current location of fread is at 4096+ 4096 since we read through and collected important info from the header.
- quicksort(toffarr, 0, 1023); //sort the array from smallest to larget offset locations so we only have to read through the file once.
+
+ cQuicksort Quick;
+ Quick.cQuicksort::quicksort(toffarr, 0, 1023); //sort the array from smallest to larget offset locations so we only have to read through the file once.
for ( short ia = 0; ia < 1024; ia++ ) {//a region file can hold a maximum of 1024 chunks (32*32)
if (ia < 3500 ) { //only run chunk # 3
@@ -129,20 +132,13 @@ int main () {
compdlength = ((byte1*256*256*256) + (byte2*256*256) + (byte3*256) + byte4 - 0); //length of compressed chunk data
if( fread( &byte5, sizeof(byte5), 1, f) != 1 ) { cout << "ERROR 2341 READING FROM FILE " << SourceFile; fclose(f); return false; } //compression type, 1 = GZip (RFC1952) (unused in practice) , 2 = Zlib (RFC1950)
- //printf("byte1: %i\n", byte1);
- //printf("byte2: %i\n", byte2);
- //printf("byte3: %i\n", byte3);
- //printf("byte4: %i\n", byte4);
- //printf("byte5: %i\n", byte5);
-
frloc += 5; //moved ahead 5 bytes while reading data.
- // TODO - delete [] temparr after you're done with it, now it's a memory leak
- char* compBlockData = new char[compdlength]; //can't get fread to read more than one char at a time into a char array... so that's what I'll do. :( At least it works.
+ char* compBlockData = new char[compdlength];
if( fread( compBlockData, compdlength, 1, f) != 1 ) { cout << "ERROR rf22 READING FROM FILE " << SourceFile; fclose(f); return false; }
frloc = frloc + compdlength;
- uLongf DestSize = 98576;// uncompressed chunks should never be larger than this
+ uLongf DestSize = 128576;// uncompressed chunks should never be larger than this
char* BlockData = new char[ DestSize ];
@@ -168,7 +164,6 @@ int main () {
- //testing of nbtparser.
NumChunks++;
cNBTData* NBTData = new cNBTData(BlockData, (int)DestSize);
NBTData->ParseData();
@@ -200,37 +195,18 @@ int main () {
char* CompressedChunk = new char[ CompressedSize ];
int UnChunkArrLoc = 0;
int xPos = NBTData->GetInteger("xPos");
- char* xtemppos = (char *)&xPos;
int zPos = NBTData->GetInteger("zPos");
- char* ztemppos = (char *)&zPos;
+ memcpy( t_FakeHeader + t_FakeHeaderSz + 1, &xPos, sizeof(int) );t_FakeHeaderSz += sizeof(int);
+ memcpy( t_FakeHeader + t_FakeHeaderSz + 1, &zPos, sizeof(int) );t_FakeHeaderSz += sizeof(int);
- for(unsigned int i = 0; i < sizeof(int); i++) {
- t_FakeHeader[t_FakeHeaderSz+1] = xtemppos[i];
- t_FakeHeaderSz ++;
- }
- for(unsigned int i = 0; i < sizeof(int); i++) {
- t_FakeHeader[t_FakeHeaderSz+1] = ztemppos[i];
- t_FakeHeaderSz ++;
- }
//todo: inserert json code and add it to chunk data
- for(unsigned int i = 0; i < 32768; i++) {
- UncompressedChunk[ UnChunkArrLoc ] = NBTData->GetByteArray("Blocks")[i];
- UnChunkArrLoc ++;
- }
- for(unsigned int i = 0; i < 16384; i++) {
- UncompressedChunk[ UnChunkArrLoc ] = NBTData->GetByteArray("Data")[i];
- UnChunkArrLoc ++;
- }
- for(unsigned int i = 0; i < 16384; i++) {
- UncompressedChunk[ UnChunkArrLoc ] = NBTData->GetByteArray("BlockLight")[i];
- UnChunkArrLoc ++;
- }
- for(unsigned int i = 0; i < 16384; i++) {
- UncompressedChunk[ UnChunkArrLoc ] = NBTData->GetByteArray("SkyLight")[i];
- UnChunkArrLoc ++;
- }
+
+ memcpy( UncompressedChunk + UnChunkArrLoc, NBTData->GetByteArray("Blocks"), 32768 );UnChunkArrLoc += 32768;
+ memcpy( UncompressedChunk + UnChunkArrLoc, NBTData->GetByteArray("Data"), 16384 );UnChunkArrLoc += 16384;
+ memcpy( UncompressedChunk + UnChunkArrLoc, NBTData->GetByteArray("BlockLight"), 16384 );UnChunkArrLoc += 16384;
+ memcpy( UncompressedChunk + UnChunkArrLoc, NBTData->GetByteArray("SkyLight"), 16384 );UnChunkArrLoc += 16384;
errorcode = compress2( (Bytef*)CompressedChunk, &CompressedSize, (const Bytef*)UncompressedChunk, UncompressedChunkSz, Z_DEFAULT_COMPRESSION);
if( errorcode != Z_OK )
@@ -239,34 +215,19 @@ int main () {
break;
}
- char* c_sizetemppos = (char *)&CompressedSize;
- char* uc_sizetemppos = (char *)&UncompressedChunkSz;
-
- for(unsigned int i = 0; i < sizeof(int); i++) {
- t_FakeHeader[t_FakeHeaderSz+1] = c_sizetemppos[i];
- t_FakeHeaderSz ++;
- }
- for(unsigned int i = 0; i < sizeof(int); i++) {
- t_FakeHeader[t_FakeHeaderSz+1] = uc_sizetemppos[i];
- t_FakeHeaderSz ++;
- }
- for(unsigned int i = 0; i < CompressedSize; i++) {
- t_CompChunk[t_CompChunkSz+1] = CompressedChunk[i];
- t_CompChunkSz ++;
- }
+ memcpy( t_FakeHeader + t_FakeHeaderSz + 1, &CompressedSize, sizeof(int) );t_FakeHeaderSz += sizeof(int);
+ memcpy( t_FakeHeader + t_FakeHeaderSz + 1, &UncompressedChunkSz, sizeof(int) );t_FakeHeaderSz += sizeof(int);
+ memcpy( t_CompChunk + t_CompChunkSz + 1, CompressedChunk, CompressedSize );t_CompChunkSz += CompressedSize;
- //printf("Coord(X,Z):ChunkSize: %i,%i:%i\n", NBTData->GetInteger("xPos"), NBTData->GetInteger("zPos"), (int)CompressedSize );
-
NBTData->CloseCompound();// Close the compounds after you're done
NBTData->CloseCompound();
- //fwrite( BlockData, DestSize, 1, wf );
-
delete [] UncompressedChunk;
delete [] CompressedChunk;
delete [] compBlockData;
delete [] BlockData;
+ delete [] NBTData;
while ( (frloc < toffarr[ia+1]) && (ia<1023) ) { //loop through Notch's junk data until we get to another chunk offset possition to start the loop again
if( fread( &trash, sizeof(byte4), 1, f) != 1 ) { cout << "ERROR 2nkd READING FROM FILE " << SourceFile; fclose(f); return false; }