qzdoom/src/farchive.cpp

1583 lines
32 KiB
C++
Raw Normal View History

2016-03-01 15:47:10 +00:00
/*
** farchive.cpp
** Implements an archiver for DObject serialization.
**
**---------------------------------------------------------------------------
** Copyright 1998-2009 Randy Heit
** All rights reserved.
**
** Redistribution and use in source and binary forms, with or without
** modification, are permitted provided that the following conditions
** are met:
**
** 1. Redistributions of source code must retain the above copyright
** notice, this list of conditions and the following disclaimer.
** 2. Redistributions in binary form must reproduce the above copyright
** notice, this list of conditions and the following disclaimer in the
** documentation and/or other materials provided with the distribution.
** 3. The name of the author may not be used to endorse or promote products
** derived from this software without specific prior written permission.
**
** THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
** IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
** OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
** IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
** INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
** NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
** DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
** THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
** (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
** THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
**---------------------------------------------------------------------------
**
** The structure of the archive file generated is influenced heavily by the
** description of the MFC archive format published somewhere in the MSDN
** library.
**
** Two major shortcomings of the format I use are that there is no version
** control and no support for storing the non-default portions of objects.
** The latter would allow for easier extension of objects in future
** releases even without a versioning system.
*/
#include <stddef.h>
#include <string.h>
#include <zlib.h>
#include <stdlib.h>
#include "doomtype.h"
#include "farchive.h"
#include "m_swap.h"
#include "m_crc32.h"
#include "cmdlib.h"
#include "i_system.h"
#include "c_cvars.h"
#include "c_dispatch.h"
#include "d_player.h"
#include "m_misc.h"
#include "dobject.h"
// These are special tokens found in the data stream of an archive.
// Whenever a new object is encountered, it gets created using new and
// is then asked to serialize itself before processing of the previous
// object continues. This can result in some very deep recursion if
// you aren't careful about how you organize your data.
#define NEW_OBJ ((BYTE)1) // Data for a new object follows
#define NEW_CLS_OBJ ((BYTE)2) // Data for a new class and object follows
#define OLD_OBJ ((BYTE)3) // Reference to an old object follows
#define NULL_OBJ ((BYTE)4) // Load as NULL
#define M1_OBJ ((BYTE)44) // Load as (DObject*)-1
#define NEW_PLYR_OBJ ((BYTE)5) // Data for a new player follows
#define NEW_PLYR_CLS_OBJ ((BYTE)6) // Data for a new class and player follows
#define NEW_NAME ((BYTE)27) // A new name follows
#define OLD_NAME ((BYTE)28) // Reference to an old name follows
#define NIL_NAME ((BYTE)33) // Load as NULL
#define NEW_SPRITE ((BYTE)11) // A new sprite name follows
#define OLD_SPRITE ((BYTE)12) // Reference to an old sprite name follows
#ifdef __BIG_ENDIAN__
static inline WORD SWAP_WORD(WORD x) { return x; }
static inline DWORD SWAP_DWORD(DWORD x) { return x; }
static inline QWORD SWAP_QWORD(QWORD x) { return x; }
static inline void SWAP_FLOAT(float x) { }
static inline void SWAP_DOUBLE(double &dst, double src) { dst = src; }
#else
#ifdef _MSC_VER
static inline WORD SWAP_WORD(WORD x) { return _byteswap_ushort(x); }
static inline DWORD SWAP_DWORD(DWORD x) { return _byteswap_ulong(x); }
static inline QWORD SWAP_QWORD(QWORD x) { return _byteswap_uint64(x); }
static inline void SWAP_DOUBLE(double &dst, double &src)
{
union twiddle { QWORD q; double d; } tdst, tsrc;
tsrc.d = src;
tdst.q = _byteswap_uint64(tsrc.q);
dst = tdst.d;
}
#else
static inline WORD SWAP_WORD(WORD x) { return (((x)<<8) | ((x)>>8)); }
static inline DWORD SWAP_DWORD(DWORD x) { return x = (((x)>>24) | (((x)>>8)&0xff00) | (((x)<<8)&0xff0000) | ((x)<<24)); }
static inline QWORD SWAP_QWORD(QWORD x)
{
union { QWORD q; DWORD d[2]; } t, u;
t.q = x;
u.d[0] = SWAP_DWORD(t.d[1]);
u.d[1] = SWAP_DWORD(t.d[0]);
return u.q;
}
static inline void SWAP_DOUBLE(double &dst, double &src)
{
union twiddle { double f; DWORD d[2]; } tdst, tsrc;
DWORD t;
tsrc.f = src;
t = tsrc.d[0];
tdst.d[0] = SWAP_DWORD(tsrc.d[1]);
tdst.d[1] = SWAP_DWORD(t);
dst = tdst.f;
}
#endif
static inline void SWAP_FLOAT(float &x)
{
union twiddle { DWORD i; float f; } t;
t.f = x;
t.i = SWAP_DWORD(t.i);
x = t.f;
}
#endif
// Output buffer size for compression; need some extra space.
// I assume the description in zlib.h is accurate.
#define OUT_LEN(a) ((a) + (a) / 1000 + 12)
void FCompressedFile::BeEmpty ()
{
m_Pos = 0;
m_BufferSize = 0;
m_MaxBufferSize = 0;
m_Buffer = NULL;
m_File = NULL;
m_NoCompress = false;
m_Mode = ENotOpen;
}
static const char LZOSig[4] = { 'F', 'L', 'Z', 'O' };
static const char ZSig[4] = { 'F', 'L', 'Z', 'L' };
FCompressedFile::FCompressedFile ()
{
BeEmpty ();
}
FCompressedFile::FCompressedFile (const char *name, EOpenMode mode, bool dontCompress)
{
BeEmpty ();
Open (name, mode);
m_NoCompress = dontCompress;
}
FCompressedFile::FCompressedFile (FILE *file, EOpenMode mode, bool dontCompress, bool postopen)
{
BeEmpty ();
m_Mode = mode;
m_File = file;
m_NoCompress = dontCompress;
if (postopen)
{
PostOpen ();
}
}
FCompressedFile::~FCompressedFile ()
{
Close ();
}
bool FCompressedFile::Open (const char *name, EOpenMode mode)
{
Close ();
if (name == NULL)
return false;
m_Mode = mode;
m_File = fopen (name, mode == EReading ? "rb" : "wb");
PostOpen ();
return !!m_File;
}
void FCompressedFile::PostOpen ()
{
if (m_File && m_Mode == EReading)
{
char sig[4];
fread (sig, 4, 1, m_File);
if (sig[0] != ZSig[0] || sig[1] != ZSig[1] || sig[2] != ZSig[2] || sig[3] != ZSig[3])
{
fclose (m_File);
m_File = NULL;
if (sig[0] == LZOSig[0] && sig[1] == LZOSig[1] && sig[2] == LZOSig[2] && sig[3] == LZOSig[3])
{
Printf ("Compressed files from older ZDooms are not supported.\n");
}
return;
}
else
{
DWORD sizes[2];
fread (sizes, sizeof(DWORD), 2, m_File);
sizes[0] = SWAP_DWORD (sizes[0]);
sizes[1] = SWAP_DWORD (sizes[1]);
unsigned int len = sizes[0] == 0 ? sizes[1] : sizes[0];
m_Buffer = (BYTE *)M_Malloc (len+8);
fread (m_Buffer+8, len, 1, m_File);
sizes[0] = SWAP_DWORD (sizes[0]);
sizes[1] = SWAP_DWORD (sizes[1]);
((DWORD *)m_Buffer)[0] = sizes[0];
((DWORD *)m_Buffer)[1] = sizes[1];
Explode ();
}
}
}
void FCompressedFile::Close ()
{
if (m_File)
{
if (m_Mode == EWriting)
{
Implode ();
fwrite (ZSig, 4, 1, m_File);
fwrite (m_Buffer, m_BufferSize + 8, 1, m_File);
}
fclose (m_File);
m_File = NULL;
}
if (m_Buffer)
{
M_Free (m_Buffer);
m_Buffer = NULL;
}
BeEmpty ();
}
void FCompressedFile::Flush ()
{
}
FFile::EOpenMode FCompressedFile::Mode () const
{
return m_Mode;
}
bool FCompressedFile::IsOpen () const
{
return !!m_File;
}
FFile &FCompressedFile::Write (const void *mem, unsigned int len)
{
if (m_Mode == EWriting)
{
if (m_Pos + len > m_MaxBufferSize)
{
do
{
m_MaxBufferSize = m_MaxBufferSize ? m_MaxBufferSize * 2 : 16384;
}
while (m_Pos + len > m_MaxBufferSize);
m_Buffer = (BYTE *)M_Realloc (m_Buffer, m_MaxBufferSize);
}
if (len == 1)
m_Buffer[m_Pos] = *(BYTE *)mem;
else
memcpy (m_Buffer + m_Pos, mem, len);
m_Pos += len;
if (m_Pos > m_BufferSize)
m_BufferSize = m_Pos;
}
else
{
I_Error ("Tried to write to reading cfile");
}
return *this;
}
FFile &FCompressedFile::Read (void *mem, unsigned int len)
{
if (m_Mode == EReading)
{
if (m_Pos + len > m_BufferSize)
{
I_Error ("Attempt to read past end of cfile");
}
if (len == 1)
*(BYTE *)mem = m_Buffer[m_Pos];
else
memcpy (mem, m_Buffer + m_Pos, len);
m_Pos += len;
}
else
{
I_Error ("Tried to read from writing cfile");
}
return *this;
}
unsigned int FCompressedFile::Tell () const
{
return m_Pos;
}
FFile &FCompressedFile::Seek (int pos, ESeekPos ofs)
{
if (ofs == ESeekRelative)
pos += m_Pos;
else if (ofs == ESeekEnd)
pos = m_BufferSize - pos;
if (pos < 0)
m_Pos = 0;
else if ((unsigned)pos > m_BufferSize)
m_Pos = m_BufferSize;
else
m_Pos = pos;
return *this;
}
CVAR (Bool, nofilecompression, false, CVAR_ARCHIVE|CVAR_GLOBALCONFIG)
void FCompressedFile::Implode ()
{
uLong outlen;
uLong len = m_BufferSize;
Byte *compressed = NULL;
BYTE *oldbuf = m_Buffer;
int r;
if (!nofilecompression && !m_NoCompress)
{
outlen = OUT_LEN(len);
do
{
compressed = new Bytef[outlen];
r = compress (compressed, &outlen, m_Buffer, len);
if (r == Z_BUF_ERROR)
{
delete[] compressed;
outlen += 1024;
}
} while (r == Z_BUF_ERROR);
// If the data could not be compressed, store it as-is.
if (r != Z_OK || outlen >= len)
{
DPrintf ("cfile could not be compressed\n");
outlen = 0;
}
else
{
DPrintf ("cfile shrank from %lu to %lu bytes\n", len, outlen);
}
}
else
{
outlen = 0;
}
m_MaxBufferSize = m_BufferSize = ((outlen == 0) ? len : outlen);
m_Buffer = (BYTE *)M_Malloc (m_BufferSize + 8);
m_Pos = 0;
DWORD *lens = (DWORD *)(m_Buffer);
lens[0] = BigLong((unsigned int)outlen);
lens[1] = BigLong((unsigned int)len);
if (outlen == 0)
memcpy (m_Buffer + 8, oldbuf, len);
else
memcpy (m_Buffer + 8, compressed, outlen);
if (compressed)
delete[] compressed;
M_Free (oldbuf);
}
void FCompressedFile::Explode ()
{
uLong expandsize, cprlen;
unsigned char *expand;
if (m_Buffer)
{
unsigned int *ints = (unsigned int *)(m_Buffer);
cprlen = BigLong(ints[0]);
expandsize = BigLong(ints[1]);
expand = (unsigned char *)M_Malloc (expandsize);
if (cprlen)
{
int r;
uLong newlen;
newlen = expandsize;
r = uncompress (expand, &newlen, m_Buffer + 8, cprlen);
if (r != Z_OK || newlen != expandsize)
{
M_Free (expand);
I_Error ("Could not decompress buffer: %s", M_ZLibError(r).GetChars());
}
}
else
{
memcpy (expand, m_Buffer + 8, expandsize);
}
if (FreeOnExplode ())
M_Free (m_Buffer);
m_Buffer = expand;
m_BufferSize = expandsize;
}
}
FCompressedMemFile::FCompressedMemFile ()
{
m_SourceFromMem = false;
m_ImplodedBuffer = NULL;
}
/*
FCompressedMemFile::FCompressedMemFile (const char *name, EOpenMode mode)
: FCompressedFile (name, mode)
{
m_SourceFromMem = false;
m_ImplodedBuffer = NULL;
}
*/
FCompressedMemFile::~FCompressedMemFile ()
{
if (m_ImplodedBuffer != NULL)
{
M_Free (m_ImplodedBuffer);
}
}
bool FCompressedMemFile::Open (const char *name, EOpenMode mode)
{
if (mode == EWriting)
{
if (name)
{
I_Error ("FCompressedMemFile cannot write to disk");
}
else
{
return Open ();
}
}
else
{
bool res = FCompressedFile::Open (name, EReading);
if (res)
{
fclose (m_File);
m_File = NULL;
}
return res;
}
return false;
}
bool FCompressedMemFile::Open (void *memblock)
{
Close ();
m_Mode = EReading;
m_Buffer = (BYTE *)memblock;
m_SourceFromMem = true;
Explode ();
m_SourceFromMem = false;
return !!m_Buffer;
}
bool FCompressedMemFile::Open ()
{
Close ();
m_Mode = EWriting;
m_BufferSize = 0;
m_MaxBufferSize = 16384;
m_Buffer = (unsigned char *)M_Malloc (16384);
m_Pos = 0;
return true;
}
bool FCompressedMemFile::Reopen ()
{
if (m_Buffer == NULL && m_ImplodedBuffer)
{
m_Mode = EReading;
m_Buffer = m_ImplodedBuffer;
m_SourceFromMem = true;
try
{
Explode ();
}
catch(...)
{
// If we just leave things as they are, m_Buffer and m_ImplodedBuffer
// both point to the same memory block and both will try to free it.
m_Buffer = NULL;
m_SourceFromMem = false;
throw;
}
m_SourceFromMem = false;
return true;
}
return false;
}
void FCompressedMemFile::Close ()
{
if (m_Mode == EWriting)
{
Implode ();
m_ImplodedBuffer = m_Buffer;
m_Buffer = NULL;
}
}
void FCompressedMemFile::Serialize (FArchive &arc)
{
if (arc.IsStoring ())
{
if (m_ImplodedBuffer == NULL)
{
I_Error ("FCompressedMemFile must be compressed before storing");
}
arc.Write (ZSig, 4);
DWORD sizes[2];
sizes[0] = SWAP_DWORD (((DWORD *)m_ImplodedBuffer)[0]);
sizes[1] = SWAP_DWORD (((DWORD *)m_ImplodedBuffer)[1]);
arc.Write (m_ImplodedBuffer, (sizes[0] ? sizes[0] : sizes[1])+8);
}
else
{
Close ();
m_Mode = EReading;
char sig[4];
DWORD sizes[2] = { 0, 0 };
arc.Read (sig, 4);
if (sig[0] != ZSig[0] || sig[1] != ZSig[1] || sig[2] != ZSig[2] || sig[3] != ZSig[3])
I_Error ("Expected to extract a compressed file");
arc << sizes[0] << sizes[1];
DWORD len = sizes[0] == 0 ? sizes[1] : sizes[0];
m_Buffer = (BYTE *)M_Malloc (len+8);
((DWORD *)m_Buffer)[0] = SWAP_DWORD(sizes[0]);
((DWORD *)m_Buffer)[1] = SWAP_DWORD(sizes[1]);
arc.Read (m_Buffer+8, len);
m_ImplodedBuffer = m_Buffer;
m_Buffer = NULL;
m_Mode = EWriting;
}
}
bool FCompressedMemFile::IsOpen () const
{
return !!m_Buffer;
}
void FCompressedMemFile::GetSizes(unsigned int &compressed, unsigned int &uncompressed) const
{
if (m_ImplodedBuffer != NULL)
{
compressed = BigLong(*(unsigned int *)m_ImplodedBuffer);
uncompressed = BigLong(*(unsigned int *)(m_ImplodedBuffer + 4));
}
else
{
compressed = 0;
uncompressed = m_BufferSize;
}
}
FPNGChunkFile::FPNGChunkFile (FILE *file, DWORD id)
: FCompressedFile (file, EWriting, true, false), m_ChunkID (id)
{
}
FPNGChunkFile::FPNGChunkFile (FILE *file, DWORD id, size_t chunklen)
: FCompressedFile (file, EReading, true, false), m_ChunkID (id)
{
m_Buffer = (BYTE *)M_Malloc (chunklen);
m_BufferSize = (unsigned int)chunklen;
fread (m_Buffer, chunklen, 1, m_File);
// Skip the CRC for now. Maybe later it will be used.
fseek (m_File, 4, SEEK_CUR);
}
// Unlike FCompressedFile::Close, m_File is left open
void FPNGChunkFile::Close ()
{
DWORD data[2];
DWORD crc;
if (m_File)
{
if (m_Mode == EWriting)
{
crc = CalcCRC32 ((BYTE *)&m_ChunkID, 4);
crc = AddCRC32 (crc, (BYTE *)m_Buffer, m_BufferSize);
data[0] = BigLong(m_BufferSize);
data[1] = m_ChunkID;
fwrite (data, 8, 1, m_File);
fwrite (m_Buffer, m_BufferSize, 1, m_File);
crc = SWAP_DWORD (crc);
fwrite (&crc, 4, 1, m_File);
}
m_File = NULL;
}
FCompressedFile::Close ();
}
FPNGChunkArchive::FPNGChunkArchive (FILE *file, DWORD id)
: FArchive (), Chunk (file, id)
{
AttachToFile (Chunk);
}
FPNGChunkArchive::FPNGChunkArchive (FILE *file, DWORD id, size_t len)
: FArchive (), Chunk (file, id, len)
{
AttachToFile (Chunk);
}
FPNGChunkArchive::~FPNGChunkArchive ()
{
// Close before FArchive's destructor, because Chunk will be
// destroyed before the FArchive is destroyed.
Close ();
}
//============================================
//
// FArchive
//
//============================================
FArchive::FArchive ()
{
}
FArchive::FArchive (FFile &file)
{
AttachToFile (file);
}
void FArchive::AttachToFile (FFile &file)
{
m_HubTravel = false;
m_File = &file;
if (file.Mode() == FFile::EReading)
{
m_Loading = true;
m_Storing = false;
}
else
{
m_Loading = false;
m_Storing = true;
}
m_Persistent = file.IsPersistent();
ClassToArchive.Clear();
ArchiveToClass.Clear();
ObjectToArchive.Clear();
ArchiveToObject.Clear();
memset(m_NameHash, 0xFF, sizeof(m_NameHash));
m_Names.Clear();
m_NameStorage.Clear();
m_NumSprites = 0;
m_SpriteMap = new int[sprites.Size()];
for (size_t s = 0; s < sprites.Size(); ++s)
{
m_SpriteMap[s] = -1;
}
}
FArchive::~FArchive ()
{
Close ();
if (m_SpriteMap)
delete[] m_SpriteMap;
}
void FArchive::Write (const void *mem, unsigned int len)
{
m_File->Write (mem, len);
}
void FArchive::Read (void *mem, unsigned int len)
{
m_File->Read (mem, len);
}
void FArchive::Close ()
{
if (m_File)
{
m_File->Close ();
m_File = NULL;
DPrintf ("Processed %u objects\n", ArchiveToObject.Size());
}
}
void FArchive::WriteByte(BYTE val)
{
m_File->Write(&val, 1);
}
void FArchive::WriteInt16(WORD val)
{
WORD out = LittleShort(val);
m_File->Write(&out, 2);
}
void FArchive::WriteInt32(DWORD val)
{
int out = LittleLong(val);
m_File->Write(&out, 4);
}
void FArchive::WriteInt64(QWORD val)
{
long long out = SWAP_QWORD(val);
m_File->Write(&out, 8);
}
2016-03-01 15:47:10 +00:00
void FArchive::WriteCount (DWORD count)
{
BYTE out;
do
{
out = count & 0x7f;
if (count >= 0x80)
out |= 0x80;
Write (&out, sizeof(BYTE));
count >>= 7;
} while (count);
}
DWORD FArchive::ReadCount ()
{
BYTE in;
DWORD count = 0;
int ofs = 0;
do
{
Read (&in, sizeof(BYTE));
count |= (in & 0x7f) << ofs;
ofs += 7;
} while (in & 0x80);
return count;
}
void FArchive::WriteName (const char *name)
{
BYTE id;
if (name == NULL)
{
id = NIL_NAME;
Write (&id, 1);
}
else
{
DWORD index = FindName (name);
if (index != NameMap::NO_INDEX)
{
id = OLD_NAME;
Write (&id, 1);
WriteCount (index);
}
else
{
AddName (name);
id = NEW_NAME;
Write (&id, 1);
WriteString (name);
}
}
}
const char *FArchive::ReadName ()
{
BYTE id;
operator<< (id);
if (id == NIL_NAME)
{
return NULL;
}
else if (id == OLD_NAME)
{
DWORD index = ReadCount ();
if (index >= m_Names.Size())
{
I_Error ("Name %u has not been read yet\n", index);
}
return &m_NameStorage[m_Names[index].StringStart];
}
else if (id == NEW_NAME)
{
DWORD index;
DWORD size = ReadCount ();
char *str;
index = (DWORD)m_NameStorage.Reserve (size);
str = &m_NameStorage[index];
Read (str, size-1);
str[size-1] = 0;
AddName (index);
return str;
}
else
{
I_Error ("Expected a name but got something else\n");
return NULL;
}
}
void FArchive::WriteString (const char *str)
{
if (str == NULL)
{
WriteCount (0);
}
else
{
DWORD size = (DWORD)(strlen (str) + 1);
WriteCount (size);
Write (str, size - 1);
}
}
void FArchive::WriteString(const FString &str)
{
// The count includes the '\0' terminator, but we don't
// actually write it out.
WriteCount(str.Len() + 1);
Write(str, str.Len());
}
2016-03-01 15:47:10 +00:00
FArchive &FArchive::operator<< (char *&str)
{
if (m_Storing)
{
WriteString (str);
}
else
{
DWORD size = ReadCount ();
char *str2;
if (size == 0)
{
str2 = NULL;
}
else
{
str2 = new char[size];
size--;
Read (str2, size);
str2[size] = 0;
ReplaceString ((char **)&str, str2);
}
if (str)
{
delete[] str;
}
str = str2;
}
return *this;
}
FArchive &FArchive::operator<< (FString &str)
{
if (m_Storing)
{
WriteString (str);
2016-03-01 15:47:10 +00:00
}
else
{
DWORD size = ReadCount();
if (size == 0)
{
str = "";
}
else
{
char *str2 = (char *)alloca(size*sizeof(char));
size--;
Read (str2, size);
str = FString(str2, size);
2016-03-01 15:47:10 +00:00
}
}
return *this;
}
FArchive &FArchive::operator<< (BYTE &c)
{
if (m_Storing)
Write (&c, sizeof(BYTE));
else
Read (&c, sizeof(BYTE));
return *this;
}
FArchive &FArchive::operator<< (WORD &w)
{
if (m_Storing)
{
WORD temp = SWAP_WORD(w);
Write (&temp, sizeof(WORD));
}
else
{
Read (&w, sizeof(WORD));
w = SWAP_WORD(w);
}
return *this;
}
FArchive &FArchive::operator<< (DWORD &w)
{
if (m_Storing)
{
DWORD temp = SWAP_DWORD(w);
Write (&temp, sizeof(DWORD));
}
else
{
Read (&w, sizeof(DWORD));
w = SWAP_DWORD(w);
}
return *this;
}
FArchive &FArchive::operator<< (QWORD &w)
{
if (m_Storing)
{
QWORD temp = SWAP_QWORD(w);
Write (&temp, sizeof(QWORD));
}
else
{
Read (&w, sizeof(QWORD));
w = SWAP_QWORD(w);
}
return *this;
}
FArchive &FArchive::operator<< (float &w)
{
if (m_Storing)
{
float temp = w;
SWAP_FLOAT(temp);
Write (&temp, sizeof(float));
}
else
{
Read (&w, sizeof(float));
SWAP_FLOAT(w);
}
return *this;
}
FArchive &FArchive::operator<< (double &w)
{
if (m_Storing)
{
double temp;
SWAP_DOUBLE(temp,w);
Write (&temp, sizeof(double));
}
else
{
Read (&w, sizeof(double));
SWAP_DOUBLE(w,w);
}
return *this;
}
FArchive &FArchive::operator<< (FName &n)
{ // In an archive, a "name" is a string that might be stored multiple times,
// so it is only stored once. It is still treated as a normal string. In the
// rest of the game, a name is a unique identifier for a number.
if (m_Storing)
{
WriteName (n.GetChars());
}
else
{
n = FName(ReadName());
}
return *this;
}
FArchive &FArchive::SerializePointer (void *ptrbase, BYTE **ptr, DWORD elemSize)
{
DWORD w;
if (m_Storing)
{
if (*(void **)ptr)
{
w = DWORD(((size_t)*ptr - (size_t)ptrbase) / elemSize);
}
else
{
w = ~0u;
}
WriteCount (w);
}
else
{
w = ReadCount ();
if (w != ~0u)
{
*(void **)ptr = (BYTE *)ptrbase + w * elemSize;
}
else
{
*(void **)ptr = NULL;
}
}
return *this;
}
FArchive &FArchive::SerializeObject (DObject *&object, PClass *type)
{
if (!type->IsDescendantOf(RUNTIME_CLASS(PClass)))
{ // a regular object
if (IsStoring())
{
return WriteObject(object);
}
else
{
return ReadObject(object, type);
}
}
else
{ // a class object
if (IsStoring())
{
UserWriteClass((PClass *)object);
}
else
{
UserReadClass(object);
}
return *this;
}
}
FArchive &FArchive::WriteObject (DObject *obj)
{
player_t *player;
BYTE id[2];
if (obj == NULL)
{
id[0] = NULL_OBJ;
Write (id, 1);
}
else if (obj == (DObject*)~0)
{
id[0] = M1_OBJ;
Write (id, 1);
}
else if (obj->ObjectFlags & OF_EuthanizeMe)
{
// Objects that want to die are not saved to the archive, but
// we leave the pointers to them alone.
id[0] = NULL_OBJ;
Write (id, 1);
}
else
{
PClass *type = obj->GetClass();
DWORD *classarcid;
if (type == RUNTIME_CLASS(DObject))
{
//I_Error ("Tried to save an instance of DObject.\n"
// "This should not happen.\n");
id[0] = NULL_OBJ;
Write (id, 1);
}
else if (NULL == (classarcid = ClassToArchive.CheckKey(type)))
{
// No instances of this class have been written out yet.
// Write out the class, then write out the object. If this
// is an actor controlled by a player, make note of that
// so that it can be overridden when moving around in a hub.
if (obj->IsKindOf (RUNTIME_CLASS (AActor)) &&
(player = static_cast<AActor *>(obj)->player) &&
player->mo == obj)
{
id[0] = NEW_PLYR_CLS_OBJ;
id[1] = (BYTE)(player - players);
Write (id, 2);
}
else
{
id[0] = NEW_CLS_OBJ;
Write (id, 1);
}
WriteClass (type);
// Printf ("Make class %s (%u)\n", type->Name, m_File->Tell());
MapObject (obj);
obj->SerializeUserVars (*this);
obj->Serialize (*this);
obj->CheckIfSerialized ();
}
else
{
// An instance of this class has already been saved. If
// this object has already been written, save a reference
// to the saved object. Otherwise, save a reference to the
// class, then save the object. Again, if this is a player-
// controlled actor, remember that.
DWORD *objarcid = ObjectToArchive.CheckKey(obj);
if (objarcid == NULL)
{
if (obj->IsKindOf (RUNTIME_CLASS (AActor)) &&
(player = static_cast<AActor *>(obj)->player) &&
player->mo == obj)
{
id[0] = NEW_PLYR_OBJ;
id[1] = (BYTE)(player - players);
Write (id, 2);
}
else
{
id[0] = NEW_OBJ;
Write (id, 1);
}
WriteCount (*classarcid);
// Printf ("Reuse class %s (%u)\n", type->Name, m_File->Tell());
MapObject (obj);
obj->SerializeUserVars (*this);
obj->Serialize (*this);
obj->CheckIfSerialized ();
}
else
{
id[0] = OLD_OBJ;
Write (id, 1);
WriteCount (*objarcid);
}
}
}
return *this;
}
FArchive &FArchive::ReadObject (DObject* &obj, PClass *wanttype)
{
BYTE objHead;
const PClass *type;
BYTE playerNum;
DWORD index;
operator<< (objHead);
switch (objHead)
{
case NULL_OBJ:
obj = NULL;
break;
case M1_OBJ:
obj = (DObject *)~0;
break;
case OLD_OBJ:
index = ReadCount();
if (index >= ArchiveToObject.Size())
{
I_Error ("Object reference too high (%u; max is %u)\n", index, ArchiveToObject.Size());
}
obj = ArchiveToObject[index];
break;
case NEW_PLYR_CLS_OBJ:
operator<< (playerNum);
if (m_HubTravel)
{
// If travelling inside a hub, use the existing player actor
type = ReadClass (wanttype);
// Printf ("New player class: %s (%u)\n", type->Name, m_File->Tell());
obj = players[playerNum].mo;
// But also create a new one so that we can get past the one
// stored in the archive.
AActor *tempobj = static_cast<AActor *>(type->CreateNew ());
MapObject (obj != NULL ? obj : tempobj);
tempobj->SerializeUserVars (*this);
tempobj->Serialize (*this);
tempobj->CheckIfSerialized ();
// If this player is not present anymore, keep the new body
// around just so that the load will succeed.
if (obj != NULL)
{
// When the temporary player's inventory items were loaded,
// they became owned by the real player. Undo that now.
for (AInventory *item = tempobj->Inventory; item != NULL; item = item->Inventory)
{
item->Owner = tempobj;
}
tempobj->Destroy ();
}
else
{
obj = tempobj;
players[playerNum].mo = static_cast<APlayerPawn *>(obj);
}
break;
}
/* fallthrough when not travelling to a previous level */
case NEW_CLS_OBJ:
type = ReadClass (wanttype);
// Printf ("New class: %s (%u)\n", type->Name, m_File->Tell());
obj = type->CreateNew ();
MapObject (obj);
obj->SerializeUserVars (*this);
obj->Serialize (*this);
obj->CheckIfSerialized ();
break;
case NEW_PLYR_OBJ:
operator<< (playerNum);
if (m_HubTravel)
{
type = ReadStoredClass (wanttype);
// Printf ("Use player class: %s (%u)\n", type->Name, m_File->Tell());
obj = players[playerNum].mo;
AActor *tempobj = static_cast<AActor *>(type->CreateNew ());
MapObject (obj != NULL ? obj : tempobj);
tempobj->SerializeUserVars (*this);
tempobj->Serialize (*this);
tempobj->CheckIfSerialized ();
if (obj != NULL)
{
for (AInventory *item = tempobj->Inventory;
item != NULL; item = item->Inventory)
{
item->Owner = tempobj;
}
tempobj->Destroy ();
}
else
{
obj = tempobj;
players[playerNum].mo = static_cast<APlayerPawn *>(obj);
}
break;
}
/* fallthrough when not travelling to a previous level */
case NEW_OBJ:
type = ReadStoredClass (wanttype);
// Printf ("Use class: %s (%u)\n", type->Name, m_File->Tell());
obj = type->CreateNew ();
MapObject (obj);
obj->SerializeUserVars (*this);
obj->Serialize (*this);
obj->CheckIfSerialized ();
break;
default:
I_Error ("Unknown object code (%d) in archive\n", objHead);
}
return *this;
}
void FArchive::WriteSprite (int spritenum)
{
BYTE id;
if ((unsigned)spritenum >= (unsigned)sprites.Size())
{
spritenum = 0;
}
if (m_SpriteMap[spritenum] < 0)
{
m_SpriteMap[spritenum] = (int)(m_NumSprites++);
id = NEW_SPRITE;
Write (&id, 1);
Write (sprites[spritenum].name, 4);
// Write the current sprite number as a hint, because
// these will only change between different versions.
WriteCount (spritenum);
}
else
{
id = OLD_SPRITE;
Write (&id, 1);
WriteCount (m_SpriteMap[spritenum]);
}
}
int FArchive::ReadSprite ()
{
BYTE id;
Read (&id, 1);
if (id == OLD_SPRITE)
{
DWORD index = ReadCount ();
if (index >= m_NumSprites)
{
I_Error ("Sprite %u has not been read yet\n", index);
}
return m_SpriteMap[index];
}
else if (id == NEW_SPRITE)
{
DWORD name;
DWORD hint;
Read (&name, 4);
hint = ReadCount ();
if (hint >= NumStdSprites || sprites[hint].dwName != name)
{
for (hint = NumStdSprites; hint-- != 0; )
{
if (sprites[hint].dwName == name)
{
break;
}
}
if (hint >= sprites.Size())
{ // Don't know this sprite, so just use the first one
hint = 0;
}
}
m_SpriteMap[m_NumSprites++] = hint;
return hint;
}
else
{
I_Error ("Expected a sprite but got something else\n");
return 0;
}
}
DWORD FArchive::AddName (const char *name)
{
DWORD index;
unsigned int hash = MakeKey (name) % EObjectHashSize;
index = FindName (name, hash);
if (index == NameMap::NO_INDEX)
{
DWORD namelen = (DWORD)(strlen (name) + 1);
DWORD strpos = (DWORD)m_NameStorage.Reserve (namelen);
NameMap mapper = { strpos, (DWORD)m_NameHash[hash] };
memcpy (&m_NameStorage[strpos], name, namelen);
m_NameHash[hash] = index = (DWORD)m_Names.Push (mapper);
}
return index;
}
DWORD FArchive::AddName (unsigned int start)
{
DWORD hash = MakeKey (&m_NameStorage[start]) % EObjectHashSize;
NameMap mapper = { (DWORD)start, (DWORD)m_NameHash[hash] };
return (DWORD)(m_NameHash[hash] = m_Names.Push (mapper));
}
DWORD FArchive::FindName (const char *name) const
{
return FindName (name, MakeKey (name) % EObjectHashSize);
}
DWORD FArchive::FindName (const char *name, unsigned int bucket) const
{
unsigned int map = m_NameHash[bucket];
while (map != NameMap::NO_INDEX)
{
const NameMap *mapping = &m_Names[map];
if (strcmp (name, &m_NameStorage[mapping->StringStart]) == 0)
{
return (DWORD)map;
}
map = mapping->HashNext;
}
return (DWORD)map;
}
DWORD FArchive::WriteClass (PClass *info)
{
if (ClassToArchive.CheckKey(info) != NULL)
{
I_Error ("Attempt to write '%s' twice.\n", info->TypeName.GetChars());
}
DWORD index = ArchiveToClass.Push(info);
ClassToArchive[info] = index;
WriteString (info->TypeName.GetChars());
return index;
}
PClass *FArchive::ReadClass ()
{
struct String {
String() { val = NULL; }
~String() { if (val) delete[] val; }
char *val;
} typeName;
operator<< (typeName.val);
FName zaname(typeName.val, true);
if (zaname != NAME_None)
{
PClass *type = PClass::FindClass(zaname);
if (type != NULL)
{
ClassToArchive[type] = ArchiveToClass.Push(type);
return type;
}
}
I_Error ("Unknown class '%s'\n", typeName.val);
return NULL;
}
PClass *FArchive::ReadClass (const PClass *wanttype)
{
PClass *type = ReadClass ();
if (!type->IsDescendantOf (wanttype))
{
I_Error ("Expected to extract an object of type '%s'.\n"
"Found one of type '%s' instead.\n",
wanttype->TypeName.GetChars(), type->TypeName.GetChars());
}
return type;
}
PClass *FArchive::ReadStoredClass (const PClass *wanttype)
{
DWORD index = ReadCount ();
PClass *type = ArchiveToClass[index];
if (!type->IsDescendantOf (wanttype))
{
I_Error ("Expected to extract an object of type '%s'.\n"
"Found one of type '%s' instead.\n",
wanttype->TypeName.GetChars(), type->TypeName.GetChars());
}
return type;
}
DWORD FArchive::MapObject (DObject *obj)
{
DWORD i = ArchiveToObject.Push(obj);
ObjectToArchive[obj] = i;
return i;
}
void FArchive::UserWriteClass (PClass *type)
{
BYTE id;
if (type == NULL)
{
id = 2;
Write (&id, 1);
}
else
{
DWORD *arcid;
if (NULL == (arcid = ClassToArchive.CheckKey(type)))
{
id = 1;
Write (&id, 1);
WriteClass (type);
}
else
{
id = 0;
Write (&id, 1);
WriteCount (*arcid);
}
}
}
void FArchive::UserReadClass (PClass *&type)
{
BYTE newclass;
Read (&newclass, 1);
switch (newclass)
{
case 0:
type = ReadStoredClass (RUNTIME_CLASS(DObject));
break;
case 1:
type = ReadClass ();
break;
case 2:
type = NULL;
break;
default:
I_Error ("Unknown class type %d in archive.\n", newclass);
break;
}
}
FArchive &operator<< (FArchive &arc, sector_t *&sec)
{
return arc.SerializePointer (sectors, (BYTE **)&sec, sizeof(*sectors));
}
FArchive &operator<< (FArchive &arc, const sector_t *&sec)
{
return arc.SerializePointer (sectors, (BYTE **)&sec, sizeof(*sectors));
}
FArchive &operator<< (FArchive &arc, line_t *&line)
{
return arc.SerializePointer (lines, (BYTE **)&line, sizeof(*lines));
}
FArchive &operator<< (FArchive &arc, vertex_t *&vert)
{
return arc.SerializePointer (vertexes, (BYTE **)&vert, sizeof(*vertexes));
}
FArchive &operator<< (FArchive &arc, side_t *&side)
{
return arc.SerializePointer (sides, (BYTE **)&side, sizeof(*sides));
}
FArchive &operator<<(FArchive &arc, DAngle &ang)
{
arc << ang.Degrees;
return arc;
}
FArchive &operator<<(FArchive &arc, DVector3 &vec)
{
arc << vec.X << vec.Y << vec.Z;
return arc;
}
FArchive &operator<<(FArchive &arc, DVector2 &vec)
{
arc << vec.X << vec.Y;
return arc;
}