mirror of
https://github.com/ZDoom/gzdoom-gles.git
synced 2024-11-15 00:51:24 +00:00
1491 lines
38 KiB
C++
1491 lines
38 KiB
C++
|
|
//**************************************************************************
|
|
//**
|
|
//** w_wad.c : Heretic 2 : Raven Software, Corp.
|
|
//**
|
|
//** $RCSfile: w_wad.c,v $
|
|
//** $Revision: 1.6 $
|
|
//** $Date: 95/10/06 20:56:47 $
|
|
//** $Author: cjr $
|
|
//**
|
|
//**************************************************************************
|
|
|
|
// HEADER FILES ------------------------------------------------------------
|
|
|
|
#include <stdlib.h>
|
|
#include <ctype.h>
|
|
#include <sys/types.h>
|
|
#include <sys/stat.h>
|
|
#include <string.h>
|
|
|
|
#include "m_alloc.h"
|
|
#include "doomtype.h"
|
|
#include "m_argv.h"
|
|
#include "i_system.h"
|
|
#include "cmdlib.h"
|
|
#include "w_wad.h"
|
|
#include "m_crc32.h"
|
|
#include "v_text.h"
|
|
#include "templates.h"
|
|
#include "gi.h"
|
|
|
|
// MACROS ------------------------------------------------------------------
|
|
|
|
#define NULL_INDEX (0xffff)
|
|
|
|
// TYPES -------------------------------------------------------------------
|
|
|
|
struct rffinfo_t
|
|
{
|
|
// Should be "RFF\x18"
|
|
DWORD Magic;
|
|
DWORD Version;
|
|
DWORD DirOfs;
|
|
DWORD NumLumps;
|
|
};
|
|
|
|
union MergedHeader
|
|
{
|
|
DWORD magic;
|
|
wadinfo_t wad;
|
|
rffinfo_t rff;
|
|
};
|
|
|
|
struct rfflump_t
|
|
{
|
|
BYTE IDontKnow[16];
|
|
DWORD FilePos;
|
|
DWORD Size;
|
|
BYTE IStillDontKnow[8];
|
|
BYTE Flags;
|
|
char Extension[3];
|
|
char Name[8];
|
|
BYTE WhatIsThis[4];
|
|
};
|
|
|
|
//
|
|
// WADFILE I/O related stuff.
|
|
//
|
|
struct FWadCollection::LumpRecord
|
|
{
|
|
char name[8];
|
|
int position;
|
|
int size;
|
|
int namespc;
|
|
short wadnum;
|
|
WORD flags;
|
|
};
|
|
|
|
enum
|
|
{
|
|
LUMPF_BLOODCRYPT = 1 // Lump uses Blood-style encryption
|
|
};
|
|
|
|
class FWadCollection::WadFileRecord : public FileReader
|
|
{
|
|
public:
|
|
WadFileRecord (FILE *file);
|
|
~WadFileRecord ();
|
|
|
|
char *Name;
|
|
DWORD FirstLump;
|
|
DWORD LastLump;
|
|
};
|
|
|
|
|
|
// EXTERNAL FUNCTION PROTOTYPES --------------------------------------------
|
|
|
|
void W_SysWadInit ();
|
|
|
|
// PUBLIC FUNCTION PROTOTYPES ----------------------------------------------
|
|
|
|
// PRIVATE FUNCTION PROTOTYPES ---------------------------------------------
|
|
|
|
static void BloodCrypt (void *data, int key, int len);
|
|
static void PrintLastError ();
|
|
|
|
// EXTERNAL DATA DECLARATIONS ----------------------------------------------
|
|
|
|
// PUBLIC DATA DEFINITIONS -------------------------------------------------
|
|
|
|
FWadCollection Wads;
|
|
|
|
// PRIVATE DATA DEFINITIONS ------------------------------------------------
|
|
|
|
// CODE --------------------------------------------------------------------
|
|
|
|
//==========================================================================
|
|
//
|
|
// strupr
|
|
//
|
|
// Suprisingly, glibc lacks this
|
|
//==========================================================================
|
|
|
|
#ifndef HAVE_STRUPR
|
|
void strupr (char *s)
|
|
{
|
|
while (*s)
|
|
{
|
|
*s = toupper (*s);
|
|
s++;
|
|
}
|
|
}
|
|
#endif
|
|
|
|
//==========================================================================
|
|
//
|
|
// filelength
|
|
//
|
|
// Suprisingly, glibc lacks this
|
|
//==========================================================================
|
|
|
|
#ifndef HAVE_FILELENGTH
|
|
int filelength (int handle)
|
|
{
|
|
struct stat fileinfo;
|
|
|
|
if (fstat (handle, &fileinfo) == -1)
|
|
{
|
|
close (handle);
|
|
I_Error ("Error fstating");
|
|
}
|
|
return fileinfo.st_size;
|
|
}
|
|
#endif
|
|
|
|
//==========================================================================
|
|
//
|
|
// uppercoppy
|
|
//
|
|
// [RH] Copy up to 8 chars, upper-casing them in the process
|
|
//==========================================================================
|
|
|
|
void uppercopy (char *to, const char *from)
|
|
{
|
|
int i;
|
|
|
|
for (i = 0; i < 8 && from[i]; i++)
|
|
to[i] = toupper (from[i]);
|
|
for (; i < 8; i++)
|
|
to[i] = 0;
|
|
}
|
|
|
|
FWadCollection::FWadCollection ()
|
|
: FirstLumpIndex(NULL), NextLumpIndex(NULL), LumpInfo(NULL), Wads(NULL),
|
|
NumLumps(0), NumWads(0)
|
|
{
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_InitMultipleFiles
|
|
//
|
|
// Pass a null terminated list of files to use. All files are optional,
|
|
// but at least one file must be found. Lump names can appear multiple
|
|
// times. The name searcher looks backwards, so a later file can
|
|
// override an earlier one.
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::InitMultipleFiles (wadlist_t **filenames)
|
|
{
|
|
int numfiles;
|
|
|
|
// open all the files, load headers, and count lumps
|
|
numfiles = 0;
|
|
NumWads = 0;
|
|
NumLumps = 0;
|
|
LumpInfo = NULL; // will be realloced as lumps are added
|
|
|
|
while (*filenames)
|
|
{
|
|
wadlist_t *next = (*filenames)->next;
|
|
int baselump = NumLumps;
|
|
char name[PATH_MAX];
|
|
|
|
// [RH] Automatically append .wad extension if none is specified.
|
|
strcpy (name, (*filenames)->name);
|
|
FixPathSeperator (name);
|
|
DefaultExtension (name, ".wad");
|
|
|
|
AddFile (name);
|
|
free (*filenames);
|
|
*filenames = next;
|
|
|
|
// The first two files are always zdoom.wad and the IWAD, which
|
|
// do not contain skins.
|
|
if (++numfiles > 2)
|
|
SkinHack (baselump);
|
|
}
|
|
|
|
if (NumLumps == 0)
|
|
{
|
|
I_FatalError ("W_InitMultipleFiles: no files found");
|
|
}
|
|
|
|
// [RH] Merge sprite and flat groups.
|
|
// (We don't need to bother with patches, since
|
|
// Doom doesn't use markers to identify them.)
|
|
RenameSprites (MergeLumps ("S_START", "S_END", ns_sprites));
|
|
MergeLumps ("F_START", "F_END", ns_flats);
|
|
MergeLumps ("C_START", "C_END", ns_colormaps);
|
|
MergeLumps ("A_START", "A_END", ns_acslibrary);
|
|
MergeLumps ("TX_START", "TX_END", ns_newtextures);
|
|
MergeLumps ("V_START", "V_END", ns_strifevoices);
|
|
|
|
// [RH] Set up hash table
|
|
FirstLumpIndex = new WORD[NumLumps];
|
|
NextLumpIndex = new WORD[NumLumps];
|
|
InitHashChains ();
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_AddFile
|
|
//
|
|
// Files with a .wad extension are wadlink files with multiple lumps,
|
|
// other files are single lumps with the base filename for the lump name.
|
|
//
|
|
// [RH] Removed reload hack
|
|
//==========================================================================
|
|
|
|
void FWadCollection::AddFile (const char *filename)
|
|
{
|
|
WadFileRecord *wadinfo;
|
|
MergedHeader header;
|
|
LumpRecord* lump_p;
|
|
unsigned i;
|
|
FILE* handle;
|
|
int startlump;
|
|
wadlump_t* fileinfo = NULL, *fileinfo2free = NULL;
|
|
wadlump_t singleinfo;
|
|
|
|
// open the file and add to directory
|
|
handle = fopen (filename, "rb");
|
|
if (handle == NULL)
|
|
{ // Didn't find file
|
|
Printf (TEXTCOLOR_RED " couldn't open %s\n", filename);
|
|
PrintLastError ();
|
|
return;
|
|
}
|
|
|
|
Printf (" adding %s", filename);
|
|
startlump = NumLumps;
|
|
|
|
wadinfo = new WadFileRecord (handle);
|
|
|
|
// [RH] Determine if file is a WAD based on its signature, not its name.
|
|
if (wadinfo->Read (&header, sizeof(header)) == 0)
|
|
{
|
|
Printf (TEXTCOLOR_RED " couldn't read %s\n", filename);
|
|
PrintLastError ();
|
|
delete wadinfo;
|
|
return;
|
|
}
|
|
|
|
wadinfo->Name = copystring (filename);
|
|
|
|
if (header.magic == IWAD_ID || header.magic == PWAD_ID)
|
|
{ // This is a WAD file
|
|
|
|
header.wad.NumLumps = LittleLong(header.wad.NumLumps);
|
|
header.wad.InfoTableOfs = LittleLong(header.wad.InfoTableOfs);
|
|
fileinfo = fileinfo2free = new wadlump_t[header.wad.NumLumps];
|
|
wadinfo->Seek (header.wad.InfoTableOfs, SEEK_SET);
|
|
wadinfo->Read (fileinfo, header.wad.NumLumps * sizeof(wadlump_t));
|
|
NumLumps += header.wad.NumLumps;
|
|
Printf (" (%ld lumps)", header.wad.NumLumps);
|
|
}
|
|
else if (header.magic == RFF_ID)
|
|
{ // This is a Blood RFF file
|
|
|
|
rfflump_t *lumps, *rff_p;
|
|
int skipped = 0;
|
|
|
|
header.rff.NumLumps = LittleLong(header.rff.NumLumps);
|
|
header.rff.DirOfs = LittleLong(header.rff.DirOfs);
|
|
lumps = new rfflump_t[header.rff.NumLumps];
|
|
wadinfo->Seek (header.rff.DirOfs, SEEK_SET);
|
|
wadinfo->Read (lumps, header.rff.NumLumps * sizeof(rfflump_t));
|
|
BloodCrypt (lumps, header.rff.DirOfs, header.rff.NumLumps * sizeof(rfflump_t));
|
|
|
|
NumLumps += header.rff.NumLumps;
|
|
LumpInfo = (LumpRecord *)Realloc (LumpInfo, NumLumps*sizeof(LumpRecord));
|
|
lump_p = &LumpInfo[startlump];
|
|
|
|
for (i = 0, rff_p = lumps; i < header.rff.NumLumps; ++i, ++rff_p)
|
|
{
|
|
if (rff_p->Extension[0] == 'S' && rff_p->Extension[1] == 'F' &&
|
|
rff_p->Extension[2] == 'X')
|
|
{
|
|
lump_p->namespc = ns_bloodsfx;
|
|
}
|
|
else if (rff_p->Extension[0] == 'R' && rff_p->Extension[1] == 'A' &&
|
|
rff_p->Extension[2] == 'W')
|
|
{
|
|
lump_p->namespc = ns_bloodraw;
|
|
}
|
|
else
|
|
{
|
|
//lump_p->namespc = ns_bloodmisc;
|
|
--NumLumps;
|
|
++skipped;
|
|
continue;
|
|
}
|
|
uppercopy (lump_p->name, rff_p->Name);
|
|
lump_p->wadnum = (WORD)NumWads;
|
|
lump_p->position = LittleLong(rff_p->FilePos);
|
|
lump_p->size = LittleLong(rff_p->Size);
|
|
lump_p->flags = (rff_p->Flags & 0x10) >> 4;
|
|
lump_p++;
|
|
}
|
|
delete[] lumps;
|
|
if (skipped != 0)
|
|
{
|
|
LumpInfo = (LumpRecord *)Realloc (LumpInfo, NumLumps*sizeof(LumpRecord));
|
|
}
|
|
}
|
|
else
|
|
{ // This is just a single lump file
|
|
char name[PATH_MAX];
|
|
|
|
fileinfo2free = NULL;
|
|
fileinfo = &singleinfo;
|
|
singleinfo.FilePos = 0;
|
|
singleinfo.Size = LittleLong(wadinfo->GetLength());
|
|
ExtractFileBase (filename, name);
|
|
strupr (name);
|
|
strncpy (singleinfo.Name, name, 8);
|
|
NumLumps++;
|
|
}
|
|
Printf ("\n");
|
|
|
|
// Fill in lumpinfo
|
|
if (header.magic != RFF_ID)
|
|
{
|
|
LumpInfo = (LumpRecord *)Realloc (LumpInfo, NumLumps*sizeof(LumpRecord));
|
|
lump_p = &LumpInfo[startlump];
|
|
for (i = startlump; i < (unsigned)NumLumps; i++, lump_p++, fileinfo++)
|
|
{
|
|
// [RH] Convert name to uppercase during copy
|
|
uppercopy (lump_p->name, fileinfo->Name);
|
|
lump_p->wadnum = (WORD)NumWads;
|
|
lump_p->position = LittleLong(fileinfo->FilePos);
|
|
lump_p->size = LittleLong(fileinfo->Size);
|
|
lump_p->namespc = ns_global;
|
|
lump_p->flags = 0;
|
|
}
|
|
|
|
if (fileinfo2free)
|
|
{
|
|
delete[] fileinfo2free;
|
|
}
|
|
|
|
ScanForFlatHack (startlump);
|
|
}
|
|
|
|
wadinfo->FirstLump = startlump;
|
|
wadinfo->LastLump = NumLumps - 1;
|
|
Wads = (WadFileRecord **)Realloc (Wads, (++NumWads)*sizeof(WadFileRecord*));
|
|
Wads[NumWads-1] = wadinfo;
|
|
|
|
// [RH] Put the Strife Teaser voices into the voices namespace
|
|
if (NumWads == IWAD_FILENUM+1 && gameinfo.gametype == GAME_Strife && gameinfo.flags & GI_SHAREWARE)
|
|
{
|
|
FindStrifeTeaserVoices ();
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_CheckIfWadLoaded
|
|
//
|
|
// Returns true if the specified wad is loaded, false otherwise.
|
|
// If a fully-qualified path is specified, then the wad must match exactly.
|
|
// Otherwise, any wad with that name will work, whatever its path.
|
|
//
|
|
//==========================================================================
|
|
|
|
bool FWadCollection::CheckIfWadLoaded (const char *name)
|
|
{
|
|
unsigned int i;
|
|
|
|
if (strrchr (name, '/') != NULL)
|
|
{
|
|
for (i = 0; i < NumWads; ++i)
|
|
{
|
|
if (stricmp (GetWadFullName (i), name) == 0)
|
|
{
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
for (i = 0; i < NumWads; ++i)
|
|
{
|
|
if (stricmp (GetWadName (i), name) == 0)
|
|
{
|
|
return true;
|
|
}
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_NumLumps
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::GetNumLumps () const
|
|
{
|
|
return NumLumps;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_CheckNumForName
|
|
//
|
|
// Returns -1 if name not found. The version with a third parameter will
|
|
// look exclusively in the specified wad for the lump.
|
|
//
|
|
// [RH] Changed to use hash lookup ala BOOM instead of a linear search
|
|
// and namespace parameter
|
|
//==========================================================================
|
|
|
|
int FWadCollection::CheckNumForName (const char *name, int space)
|
|
{
|
|
char uname[8];
|
|
WORD i;
|
|
|
|
if (name == NULL)
|
|
{
|
|
return -1;
|
|
}
|
|
|
|
uppercopy (uname, name);
|
|
i = FirstLumpIndex[LumpNameHash (uname) % NumLumps];
|
|
|
|
while (i != NULL_INDEX &&
|
|
(*(__int64 *)&LumpInfo[i].name != *(__int64 *)&uname ||
|
|
LumpInfo[i].namespc != space))
|
|
{
|
|
i = NextLumpIndex[i];
|
|
}
|
|
|
|
return i != NULL_INDEX ? i : -1;
|
|
}
|
|
|
|
int FWadCollection::CheckNumForName (const char *name, int space, int wadnum)
|
|
{
|
|
char uname[8];
|
|
WORD i;
|
|
|
|
if (wadnum < 0)
|
|
{
|
|
return CheckNumForName (name, space);
|
|
}
|
|
|
|
uppercopy (uname, name);
|
|
i = FirstLumpIndex[LumpNameHash (uname) % NumLumps];
|
|
|
|
while (i != NULL_INDEX &&
|
|
(*(__int64 *)&LumpInfo[i].name != *(__int64 *)&uname ||
|
|
LumpInfo[i].namespc != space ||
|
|
LumpInfo[i].wadnum != wadnum))
|
|
{
|
|
i = NextLumpIndex[i];
|
|
}
|
|
|
|
return i != NULL_INDEX ? i : -1;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_GetNumForName
|
|
//
|
|
// Calls W_CheckNumForName, but bombs out if not found.
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::GetNumForName (const char *name)
|
|
{
|
|
int i;
|
|
|
|
i = CheckNumForName (name);
|
|
|
|
if (i == -1)
|
|
I_Error ("W_GetNumForName: %s not found!", name);
|
|
|
|
return i;
|
|
}
|
|
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_LumpLength
|
|
//
|
|
// Returns the buffer size needed to load the given lump.
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::LumpLength (int lump) const
|
|
{
|
|
if ((size_t)lump >= NumLumps)
|
|
{
|
|
I_Error ("W_LumpLength: %i >= NumLumps",lump);
|
|
}
|
|
|
|
return LumpInfo[lump].size;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// GetLumpOffset
|
|
//
|
|
// Returns the offset from the beginning of the file to the lump.
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::GetLumpOffset (int lump) const
|
|
{
|
|
if ((size_t)lump >= NumLumps)
|
|
{
|
|
I_Error ("W_LumpLength: %i >= NumLumps",lump);
|
|
}
|
|
|
|
return LumpInfo[lump].position;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_LumpNameHash
|
|
//
|
|
// NOTE: s should already be uppercase, in contrast to the BOOM version.
|
|
//
|
|
// Hash function used for lump names.
|
|
// Must be mod'ed with table size.
|
|
// Can be used for any 8-character names.
|
|
//
|
|
//==========================================================================
|
|
|
|
DWORD FWadCollection::LumpNameHash (const char *s)
|
|
{
|
|
const DWORD *table = GetCRCTable ();;
|
|
DWORD hash = 0xffffffff;
|
|
int i;
|
|
|
|
for (i = 8; i > 0 && *s; --i, ++s)
|
|
{
|
|
hash = CRC1 (hash, *s, table);
|
|
}
|
|
return hash ^ 0xffffffff;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_InitHashChains
|
|
//
|
|
// Prepares the lumpinfos for hashing.
|
|
// (Hey! This looks suspiciously like something from Boom! :-)
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::InitHashChains (void)
|
|
{
|
|
char name[8];
|
|
unsigned int i, j;
|
|
|
|
// Mark all buckets as empty
|
|
memset (FirstLumpIndex, 255, NumLumps*sizeof(FirstLumpIndex[0]));
|
|
memset (NextLumpIndex, 255, NumLumps*sizeof(FirstLumpIndex[0]));
|
|
|
|
// Now set up the chains
|
|
for (i = 0; i < (unsigned)NumLumps; i++)
|
|
{
|
|
uppercopy (name, LumpInfo[i].name);
|
|
j = LumpNameHash (name) % NumLumps;
|
|
NextLumpIndex[i] = FirstLumpIndex[j];
|
|
FirstLumpIndex[j] = i;
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// IsMarker
|
|
//
|
|
// (from BOOM)
|
|
//
|
|
//==========================================================================
|
|
|
|
bool FWadCollection::IsMarker (const FWadCollection::LumpRecord *lump, const char *marker) const
|
|
{
|
|
if (lump->namespc != ns_global)
|
|
{
|
|
return false;
|
|
}
|
|
if (strncmp (lump->name, marker, 8) == 0)
|
|
{
|
|
// If the previous lump was of the form FF_END and this one is
|
|
// of the form F_END, ignore this as a marker
|
|
if (marker[2] == 'E' && lump > LumpInfo)
|
|
{
|
|
if ((lump - 1)->name[0] == *marker &&
|
|
strncmp ((lump - 1)->name + 1, marker, 7) == 0)
|
|
{
|
|
return false;
|
|
}
|
|
}
|
|
return true;
|
|
}
|
|
// Treat double-character markers the same as single-character markers.
|
|
// (So if FF_START appears in the wad, it will be treated as if it is F_START.
|
|
// However, TTX_STAR will not be treated the same as TX_START because it
|
|
// is not a single-character marker.)
|
|
if (marker[1] == '_' &&
|
|
lump->name[0] == *marker &&
|
|
strncmp (lump->name + 1, marker, 7) == 0)
|
|
{
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// ScanForFlatHack
|
|
//
|
|
// Try to detect wads that add extra flats by sticking an extra F_END
|
|
// at the end of the flat list without any corresponding FF_START.
|
|
// In other words, fix gothic2.wad.
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::ScanForFlatHack (int startlump)
|
|
{
|
|
if (Args.CheckParm ("-noflathack"))
|
|
{
|
|
return;
|
|
}
|
|
|
|
for (int i = startlump; (DWORD)i < NumLumps; ++i)
|
|
{
|
|
if (LumpInfo[i].name[0] == 'F')
|
|
{
|
|
int j;
|
|
|
|
if (strcmp (LumpInfo[i].name + 1, "_START") == 0 ||
|
|
strncmp (LumpInfo[i].name + 1, "F_START", 7) == 0)
|
|
{
|
|
// If the wad has a F_START/FF_START marker, check for
|
|
// a FF_START-flats-FF_END-flats-F_END pattern as seen
|
|
// in darkhour.wad. At what point do I stop making hacks
|
|
// for wads that are incorrect?
|
|
for (i = i + 1; (DWORD)i < NumLumps; ++i)
|
|
{
|
|
if (LumpInfo[i].name[0] == 'F' && strcmp (LumpInfo[i].name + 1, "F_END") == 0)
|
|
{
|
|
// Found FF_END
|
|
break;
|
|
}
|
|
if (LumpInfo[i].size != 4096)
|
|
{
|
|
return;
|
|
}
|
|
}
|
|
if (i < (int)NumLumps)
|
|
{
|
|
// Look for flats-F_END
|
|
for (j = ++i; (DWORD)j < NumLumps; ++j)
|
|
{
|
|
if (LumpInfo[j].name[0] == 'F' && strcmp (LumpInfo[j].name + 1, "_END") == 0)
|
|
{
|
|
// Found F_END, so bump all the flats between FF_END/F_END up and move the
|
|
// FF_END so it immediately precedes F_END.
|
|
if (i != j - 1)
|
|
{
|
|
for (; i < j; ++i)
|
|
{
|
|
LumpInfo[i - 1] = LumpInfo[i];
|
|
}
|
|
--i;
|
|
strcpy (LumpInfo[i].name, "FF_END");
|
|
LumpInfo[i].size = 0;
|
|
LumpInfo[i].namespc = ns_global;
|
|
LumpInfo[i].flags = 0;
|
|
}
|
|
return;
|
|
}
|
|
if (LumpInfo[j].size != 4096)
|
|
{
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
return;
|
|
}
|
|
|
|
// No need to look for FF_END, because Doom doesn't. One minor
|
|
// nitpick: Doom will look for the last F_END; this finds the first
|
|
// one if there is more than one in the file. Too bad. If there's
|
|
// more than one F_END, this algorithm won't be able to properly
|
|
// determine where to put the F_START anyway.
|
|
if (strcmp (LumpInfo[i].name + 1, "_END") == 0)
|
|
{
|
|
// When F_END is found, back up past any lumps of length
|
|
// 4096, then insert an F_START marker.
|
|
for (j = i - 1; j >= startlump && LumpInfo[j].size == 4096; --j)
|
|
{
|
|
}
|
|
if (j == i - 1)
|
|
{
|
|
// Oh no! There are no flats immediately before F_END. Maybe they are
|
|
// at the beginning of the wad (e.g. slipgate.wad).
|
|
for (j = startlump; LumpInfo[j].size == 4096; ++j)
|
|
{
|
|
}
|
|
if (j > startlump)
|
|
{
|
|
// Okay, there are probably flats at the beginning of the wad.
|
|
// Move the F_END marker so it immediately follows them, and
|
|
// then add an F_START marker at the start of the wad.
|
|
for (; i > j; --i)
|
|
{
|
|
LumpInfo[i] = LumpInfo[i-1];
|
|
}
|
|
strcpy (LumpInfo[j].name, "F_END");
|
|
LumpInfo[j].size = 0;
|
|
LumpInfo[j].namespc = ns_global;
|
|
LumpInfo[j].flags = 0;
|
|
j = startlump - 1;
|
|
}
|
|
else
|
|
{
|
|
// Oh well. There won't be any flats loaded from this wad, I guess.
|
|
j = i - 1;
|
|
}
|
|
}
|
|
++NumLumps;
|
|
LumpInfo = (LumpRecord *)Realloc (LumpInfo, NumLumps*sizeof(LumpRecord));
|
|
for (; i > j; --i)
|
|
{
|
|
LumpInfo[i+1] = LumpInfo[i];
|
|
}
|
|
++i;
|
|
strcpy (LumpInfo[i].name, "F_START");
|
|
LumpInfo[i].size = 0;
|
|
LumpInfo[i].namespc = ns_global;
|
|
LumpInfo[i].flags = 0;
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// RenameSprites
|
|
//
|
|
// Renames sprites in IWADs so that unique actors can have unique sprites,
|
|
// making it possible to import any actor from any game into any other
|
|
// game without jumping through hoops to resolve duplicate sprite names.
|
|
// You just need to know what the sprite's new name is.
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::RenameSprites (int startlump)
|
|
{
|
|
bool renameAll;
|
|
|
|
static const DWORD HereticRenames[] =
|
|
{ MAKE_ID('H','E','A','D'), MAKE_ID('L','I','C','H'), // Ironlich
|
|
};
|
|
|
|
static const DWORD HexenRenames[] =
|
|
{ MAKE_ID('B','A','R','L'), MAKE_ID('Z','B','A','R'), // ZBarrel
|
|
MAKE_ID('A','R','M','1'), MAKE_ID('A','R','_','1'), // MeshArmor
|
|
MAKE_ID('A','R','M','2'), MAKE_ID('A','R','_','2'), // FalconShield
|
|
MAKE_ID('A','R','M','3'), MAKE_ID('A','R','_','3'), // PlatinumHelm
|
|
MAKE_ID('A','R','M','4'), MAKE_ID('A','R','_','4'), // AmuletOfWarding
|
|
MAKE_ID('S','U','I','T'), MAKE_ID('Z','S','U','I'), // ZSuitOfArmor and ZArmorChunk
|
|
MAKE_ID('T','R','E','1'), MAKE_ID('Z','T','R','E'), // ZTree and ZTreeDead
|
|
MAKE_ID('T','R','E','2'), MAKE_ID('T','R','E','S'), // ZTreeSwamp150
|
|
MAKE_ID('C','A','N','D'), MAKE_ID('B','C','A','N'), // ZBlueCandle
|
|
MAKE_ID('R','O','C','K'), MAKE_ID('R','O','K','K'), // rocks and dirt in a_debris.cpp
|
|
MAKE_ID('W','A','T','R'), MAKE_ID('H','W','A','T'), // Strife also has WATR
|
|
MAKE_ID('G','I','B','S'), MAKE_ID('P','O','L','5'), // RealGibs
|
|
MAKE_ID('E','G','G','M'), MAKE_ID('P','R','K','M'), // PorkFX
|
|
};
|
|
|
|
static const DWORD StrifeRenames[] =
|
|
{ MAKE_ID('M','I','S','L'), MAKE_ID('S','M','I','S'), // lots of places
|
|
MAKE_ID('A','R','M','1'), MAKE_ID('A','R','M','3'), // MetalArmor
|
|
MAKE_ID('A','R','M','2'), MAKE_ID('A','R','M','4'), // LeatherArmor
|
|
MAKE_ID('P','M','A','P'), MAKE_ID('S','M','A','P'), // StrifeMap
|
|
MAKE_ID('T','L','M','P'), MAKE_ID('T','E','C','H'), // TechLampSilver and TechLampBrass
|
|
MAKE_ID('T','R','E','1'), MAKE_ID('T','R','E','T'), // TreeStub
|
|
MAKE_ID('B','A','R','1'), MAKE_ID('B','A','R','C'), // BarricadeColumn
|
|
MAKE_ID('S','H','T','2'), MAKE_ID('M','P','U','F'), // MaulerPuff
|
|
MAKE_ID('B','A','R','L'), MAKE_ID('B','B','A','R'), // StrifeBurningBarrel
|
|
MAKE_ID('T','R','C','H'), MAKE_ID('T','R','H','L'), // SmallTorchLit
|
|
MAKE_ID('S','H','R','D'), MAKE_ID('S','H','A','R'), // glass shards
|
|
MAKE_ID('B','L','S','T'), MAKE_ID('M','A','U','L'), // Mauler
|
|
MAKE_ID('L','O','G','G'), MAKE_ID('L','O','G','W'), // StickInWater
|
|
MAKE_ID('V','A','S','E'), MAKE_ID('V','A','Z','E'), // Pot and Pitcher
|
|
MAKE_ID('C','N','D','L'), MAKE_ID('K','N','D','L'), // Candle
|
|
MAKE_ID('P','O','T','1'), MAKE_ID('M','P','O','T'), // MetalPot
|
|
MAKE_ID('S','P','I','D'), MAKE_ID('S','T','L','K'), // Stalker
|
|
};
|
|
|
|
const DWORD *renames;
|
|
int numrenames;
|
|
|
|
switch (gameinfo.gametype)
|
|
{
|
|
case GAME_Doom:
|
|
default:
|
|
// Doom's sprites don't get renamed.
|
|
return;
|
|
|
|
case GAME_Heretic:
|
|
renames = HereticRenames;
|
|
numrenames = sizeof(HereticRenames)/8;
|
|
break;
|
|
|
|
case GAME_Hexen:
|
|
renames = HexenRenames;
|
|
numrenames = sizeof(HexenRenames)/8;
|
|
break;
|
|
|
|
case GAME_Strife:
|
|
renames = StrifeRenames;
|
|
numrenames = sizeof(StrifeRenames)/8;
|
|
break;
|
|
}
|
|
|
|
renameAll = !!Args.CheckParm ("-oldsprites");
|
|
|
|
for (DWORD i = startlump + 1;
|
|
i < NumLumps &&
|
|
*(DWORD *)LumpInfo[i].name != MAKE_ID('S','_','E','N') &&
|
|
*(((DWORD *)LumpInfo[i].name) + 1) != MAKE_ID('D',0,0,0);
|
|
++i)
|
|
{
|
|
// Only sprites in the IWAD normally get renamed
|
|
if (renameAll || LumpInfo[i].wadnum == IWAD_FILENUM)
|
|
{
|
|
for (int j = 0; j < numrenames; ++j)
|
|
{
|
|
if (*(DWORD *)LumpInfo[i].name == renames[j*2])
|
|
{
|
|
*(DWORD *)LumpInfo[i].name = renames[j*2+1];
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// MergeLumps
|
|
//
|
|
// Merge multiple tagged groups into one
|
|
// Basically from BOOM, too, although I tried to write it independently.
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::MergeLumps (const char *start, const char *end, int space)
|
|
{
|
|
char ustart[8], uend[8];
|
|
LumpRecord *newlumpinfos;
|
|
int newlumps, oldlumps;
|
|
int markerpos = -1;
|
|
unsigned int i;
|
|
BOOL insideBlock;
|
|
|
|
uppercopy (ustart, start);
|
|
uppercopy (uend, end);
|
|
|
|
newlumpinfos = new LumpRecord[NumLumps];
|
|
|
|
newlumps = 0;
|
|
oldlumps = 0;
|
|
insideBlock = false;
|
|
|
|
for (i = 0; i < NumLumps; i++)
|
|
{
|
|
if (!insideBlock)
|
|
{
|
|
// Check if this is the start of a block
|
|
if (IsMarker (LumpInfo + i, ustart))
|
|
{
|
|
insideBlock = true;
|
|
markerpos = i;
|
|
|
|
// Create start marker if we haven't already
|
|
if (!newlumps)
|
|
{
|
|
newlumps++;
|
|
strncpy (newlumpinfos[0].name, ustart, 8);
|
|
newlumpinfos[0].wadnum = -1;
|
|
newlumpinfos[0].position =
|
|
newlumpinfos[0].size = 0;
|
|
newlumpinfos[0].namespc = ns_global;
|
|
}
|
|
}
|
|
else
|
|
{
|
|
// Copy lumpinfo down this list
|
|
LumpInfo[oldlumps++] = LumpInfo[i];
|
|
}
|
|
}
|
|
else
|
|
{
|
|
if (i && LumpInfo[i].wadnum != LumpInfo[i-1].wadnum)
|
|
{
|
|
// Blocks cannot span multiple files
|
|
insideBlock = false;
|
|
LumpInfo[oldlumps++] = LumpInfo[i];
|
|
}
|
|
else if (IsMarker (LumpInfo + i, uend))
|
|
{
|
|
// It is the end of a block. We'll add the end marker once
|
|
// we've processed everything.
|
|
insideBlock = false;
|
|
}
|
|
else
|
|
{
|
|
newlumpinfos[newlumps] = LumpInfo[i];
|
|
newlumpinfos[newlumps++].namespc = space;
|
|
}
|
|
}
|
|
}
|
|
|
|
// Now copy the merged lumps to the end of the old list
|
|
// and create the end marker entry.
|
|
|
|
if (newlumps)
|
|
{
|
|
if (size_t(oldlumps + newlumps) >= NumLumps)
|
|
LumpInfo = (LumpRecord *)Realloc (LumpInfo, (oldlumps + newlumps + 1) * sizeof(LumpRecord));
|
|
|
|
memcpy (LumpInfo + oldlumps, newlumpinfos, sizeof(LumpRecord) * newlumps);
|
|
markerpos = oldlumps;
|
|
NumLumps = oldlumps + newlumps;
|
|
|
|
strncpy (LumpInfo[NumLumps].name, uend, 8);
|
|
LumpInfo[NumLumps].wadnum = -1;
|
|
LumpInfo[NumLumps].position =
|
|
LumpInfo[NumLumps].size = 0;
|
|
LumpInfo[NumLumps].namespc = ns_global;
|
|
NumLumps++;
|
|
}
|
|
|
|
delete[] newlumpinfos;
|
|
return markerpos;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// FindStrifeTeaserVoices
|
|
//
|
|
// Strife0.wad does not have the voices between V_START/V_END markers, so
|
|
// figure out which lumps are voices based on their names.
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::FindStrifeTeaserVoices ()
|
|
{
|
|
for (DWORD i = Wads[IWAD_FILENUM]->FirstLump; i <= Wads[IWAD_FILENUM]->LastLump; ++i)
|
|
{
|
|
if (LumpInfo[i].name[0] == 'V' &&
|
|
LumpInfo[i].name[1] == 'O' &&
|
|
LumpInfo[i].name[2] == 'C')
|
|
{
|
|
int j;
|
|
|
|
for (j = 3; j < 8; ++j)
|
|
{
|
|
if (LumpInfo[i].name[j] != 0 && !isdigit(LumpInfo[i].name[j]))
|
|
break;
|
|
}
|
|
if (j == 8)
|
|
{
|
|
LumpInfo[i].namespc = ns_strifevoices;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_FindLump
|
|
//
|
|
// Find a named lump. Specifically allows duplicates for merging of e.g.
|
|
// SNDINFO lumps.
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::FindLump (const char *name, int *lastlump, bool anyns)
|
|
{
|
|
char name8[8];
|
|
LumpRecord *lump_p;
|
|
|
|
uppercopy (name8, name);
|
|
|
|
lump_p = LumpInfo + *lastlump;
|
|
while (lump_p < LumpInfo + NumLumps)
|
|
{
|
|
if ((anyns || lump_p->namespc == ns_global) &&
|
|
*(__int64 *)&lump_p->name == *(__int64 *)&name8)
|
|
{
|
|
int lump = lump_p - LumpInfo;
|
|
*lastlump = lump + 1;
|
|
return lump;
|
|
}
|
|
lump_p++;
|
|
}
|
|
|
|
*lastlump = NumLumps;
|
|
return -1;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_CheckLumpName
|
|
//
|
|
//==========================================================================
|
|
|
|
bool FWadCollection::CheckLumpName (int lump, const char *name)
|
|
{
|
|
if ((size_t)lump >= NumLumps)
|
|
return false;
|
|
|
|
return !strnicmp (LumpInfo[lump].name, name, 8);
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_GetLumpName
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::GetLumpName (char *to, int lump) const
|
|
{
|
|
if ((size_t)lump >= NumLumps)
|
|
*to = 0;
|
|
else
|
|
uppercopy (to, LumpInfo[lump].name);
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// GetLumpNamespace
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::GetLumpNamespace (int lump) const
|
|
{
|
|
if ((size_t)lump >= NumLumps)
|
|
return ns_global;
|
|
else
|
|
return LumpInfo[lump].namespc;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_GetLumpFile
|
|
//
|
|
//==========================================================================
|
|
|
|
int FWadCollection::GetLumpFile (int lump) const
|
|
{
|
|
if ((size_t)lump >= NumLumps)
|
|
return -1;
|
|
return LumpInfo[lump].wadnum;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_ReadLump
|
|
//
|
|
// Loads the lump into the given buffer, which must be >= W_LumpLength().
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::ReadLump (int lump, void *dest)
|
|
{
|
|
FWadLump lumpr = OpenLumpNum (lump);
|
|
long size = lumpr.GetLength ();
|
|
long numread = lumpr.Read (dest, size);
|
|
|
|
if (numread != size)
|
|
{
|
|
I_Error ("W_ReadLump: only read %ld of %ld on lump %i\n",
|
|
numread, size, lump);
|
|
}
|
|
if (LumpInfo[lump].flags & LUMPF_BLOODCRYPT)
|
|
{
|
|
BloodCrypt (dest, 0, MIN<int> (size, 256));
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// ReadLump - variant 2
|
|
//
|
|
// Loads the lump into a newly created buffer and returns it.
|
|
//
|
|
//==========================================================================
|
|
|
|
FMemLump FWadCollection::ReadLump (int lump)
|
|
{
|
|
FWadLump lumpr = OpenLumpNum (lump);
|
|
long size = lumpr.GetLength ();
|
|
BYTE *dest = new BYTE[size];
|
|
long numread = lumpr.Read (dest, size);
|
|
|
|
if (numread != size)
|
|
{
|
|
I_Error ("W_ReadLump: only read %ld of %ld on lump %i\n",
|
|
numread, size, lump);
|
|
}
|
|
if (LumpInfo[lump].flags & LUMPF_BLOODCRYPT)
|
|
{
|
|
BloodCrypt (dest, 0, MIN<int> (size, 256));
|
|
}
|
|
return FMemLump (dest);
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// OpenLumpNum
|
|
//
|
|
// Returns a copy of the file object for a lump's wad and positions its
|
|
// file pointer at the beginning of the lump.
|
|
//
|
|
//==========================================================================
|
|
|
|
FWadLump FWadCollection::OpenLumpNum (int lump)
|
|
{
|
|
LumpRecord *l;
|
|
WadFileRecord *wad;
|
|
|
|
if ((unsigned)lump >= (unsigned)NumLumps)
|
|
{
|
|
I_Error ("W_MapLumpNum: %u >= NumLumps",lump);
|
|
}
|
|
|
|
l = &LumpInfo[lump];
|
|
wad = Wads[l->wadnum];
|
|
|
|
#if 0
|
|
// Blood encryption?
|
|
if (writeable || l->flags & LUMPF_BLOODCRYPT)
|
|
{
|
|
ptr = Malloc (l->size);
|
|
NonMappedPointers.insert (ptr);
|
|
W_ReadLump (lump, ptr);
|
|
return ptr;
|
|
}
|
|
#endif
|
|
|
|
wad->Seek (l->position, SEEK_SET);
|
|
return FWadLump (*wad, l->size);
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// ReopenLumpNum
|
|
//
|
|
// Similar to OpenLumpNum, except a new, independant file object is created
|
|
// for the lump's wad. Use this when you won't read the lump's data all at
|
|
// once (e.g. for streaming an Ogg Vorbis stream from a wad as music).
|
|
//
|
|
//==========================================================================
|
|
|
|
FWadLump *FWadCollection::ReopenLumpNum (int lump)
|
|
{
|
|
LumpRecord *l;
|
|
WadFileRecord *wad;
|
|
FILE *f;
|
|
|
|
if ((unsigned)lump >= (unsigned)NumLumps)
|
|
{
|
|
I_Error ("W_MapLumpNum: %u >= NumLumps",lump);
|
|
}
|
|
|
|
l = &LumpInfo[lump];
|
|
wad = Wads[l->wadnum];
|
|
|
|
f = fopen (wad->Name, "rb");
|
|
if (f == NULL)
|
|
{
|
|
I_Error ("Could not reopen %s\n", wad->Name);
|
|
}
|
|
|
|
fseek (f, l->position, SEEK_SET);
|
|
return new FWadLump (f, l->size);
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_GetWadName
|
|
//
|
|
// Returns the name of the given wad.
|
|
//
|
|
//==========================================================================
|
|
|
|
const char *FWadCollection::GetWadName (int wadnum) const
|
|
{
|
|
const char *name, *slash;
|
|
|
|
if ((DWORD)wadnum >= NumWads)
|
|
{
|
|
return NULL;
|
|
}
|
|
|
|
name = Wads[wadnum]->Name;
|
|
slash = strrchr (name, '/');
|
|
return slash != NULL ? slash+1 : name;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_GetWadFullName
|
|
//
|
|
// Returns the name of the given wad, including any path
|
|
//
|
|
//==========================================================================
|
|
|
|
const char *FWadCollection::GetWadFullName (int wadnum) const
|
|
{
|
|
if ((unsigned int)wadnum >= NumWads)
|
|
{
|
|
return NULL;
|
|
}
|
|
|
|
return Wads[wadnum]->Name;
|
|
}
|
|
|
|
|
|
//==========================================================================
|
|
//
|
|
// IsUncompressedFile
|
|
//
|
|
// Returns true when the lump is available as an uncompressed portion of
|
|
// a file. The music player can play such lumps by streaming but anything
|
|
// else has to be loaded into memory first.
|
|
//
|
|
//==========================================================================
|
|
|
|
bool FWadCollection::IsUncompressedFile(int lump) const
|
|
{
|
|
/*
|
|
if ((unsigned)lump >= (unsigned)NumLumps)
|
|
{
|
|
I_Error ("IsUncompressedFile: %u >= NumLumps",lump);
|
|
}
|
|
|
|
LumpRecord * l = &LumpInfo[lump];
|
|
|
|
if (l->flags & LUMPF_COMPRESSED) return false;
|
|
else if (Wads[l->wadnum]->MemoryData!=NULL) return false;
|
|
*/
|
|
return true;
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// W_SkinHack
|
|
//
|
|
// Tests a wad file to see if it contains an S_SKIN marker. If it does,
|
|
// every lump in the wad is moved into a new namespace. Because skins are
|
|
// only supposed to replace player sprites, sounds, or faces, this should
|
|
// not be a problem. Yes, there are skins that replace more than that, but
|
|
// they are such a pain, and breaking them like this was done on purpose.
|
|
// This also renames any S_SKINxx lumps to just S_SKIN.
|
|
//
|
|
//==========================================================================
|
|
|
|
void FWadCollection::SkinHack (int baselump)
|
|
{
|
|
bool skinned = false;
|
|
bool hasmap = false;
|
|
size_t i;
|
|
|
|
for (i = baselump; i < NumLumps; i++)
|
|
{
|
|
if (LumpInfo[i].name[0] == 'S' &&
|
|
LumpInfo[i].name[1] == '_' &&
|
|
LumpInfo[i].name[2] == 'S' &&
|
|
LumpInfo[i].name[3] == 'K' &&
|
|
LumpInfo[i].name[4] == 'I' &&
|
|
LumpInfo[i].name[5] == 'N')
|
|
{ // Wad has at least one skin.
|
|
LumpInfo[i].name[6] = LumpInfo[i].name[7] = 0;
|
|
if (!skinned)
|
|
{
|
|
skinned = true;
|
|
size_t j;
|
|
|
|
for (j = baselump; j < NumLumps; j++)
|
|
{
|
|
// Using the baselump as the namespace is safe, because
|
|
// zdoom.wad guarantees the first possible baselump
|
|
// passed to this function is a largish number.
|
|
LumpInfo[j].namespc = baselump;
|
|
}
|
|
}
|
|
}
|
|
if (LumpInfo[i].name[0] == 'M' &&
|
|
LumpInfo[i].name[1] == 'A' &&
|
|
LumpInfo[i].name[2] == 'P')
|
|
{
|
|
hasmap = true;
|
|
}
|
|
}
|
|
if (skinned && hasmap)
|
|
{
|
|
Printf (TEXTCOLOR_BLUE
|
|
"The maps in %s will not be loaded because it has a skin.\n"
|
|
TEXTCOLOR_BLUE
|
|
"You should remove the skin from the wad to play these maps.\n",
|
|
Wads[LumpInfo[baselump].wadnum]->Name);
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// BloodCrypt
|
|
//
|
|
//==========================================================================
|
|
|
|
static void BloodCrypt (void *data, int key, int len)
|
|
{
|
|
int p = (BYTE)key, i;
|
|
|
|
for (i = 0; i < len; ++i)
|
|
{
|
|
((BYTE *)data)[i] ^= (unsigned char)(p+(i>>1));
|
|
}
|
|
}
|
|
|
|
//==========================================================================
|
|
//
|
|
// PrintLastError
|
|
//
|
|
//==========================================================================
|
|
|
|
#ifdef _WIN32
|
|
#define WIN32_LEAN_AND_MEAN
|
|
#include <windows.h>
|
|
|
|
static void PrintLastError ()
|
|
{
|
|
char *lpMsgBuf;
|
|
FormatMessageA(
|
|
FORMAT_MESSAGE_ALLOCATE_BUFFER |
|
|
FORMAT_MESSAGE_FROM_SYSTEM |
|
|
FORMAT_MESSAGE_IGNORE_INSERTS,
|
|
NULL,
|
|
GetLastError(),
|
|
MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), // Default language
|
|
(LPTSTR)&lpMsgBuf,
|
|
0,
|
|
NULL
|
|
);
|
|
Printf (TEXTCOLOR_RED " %s\n", lpMsgBuf);
|
|
// Free the buffer.
|
|
LocalFree( lpMsgBuf );
|
|
}
|
|
#else
|
|
static void PrintLastError ()
|
|
{
|
|
Printf (TEXTCOLOR_RED " %s\n", strerror(errno));
|
|
}
|
|
#endif
|
|
|
|
// WadFileRecord ------------------------------------------------------------
|
|
|
|
FWadCollection::WadFileRecord::WadFileRecord (FILE *file)
|
|
: FileReader(file), Name(NULL), FirstLump(0), LastLump(0)
|
|
{
|
|
}
|
|
|
|
FWadCollection::WadFileRecord::~WadFileRecord ()
|
|
{
|
|
if (Name != NULL)
|
|
{
|
|
delete[] Name;
|
|
}
|
|
}
|
|
|
|
// FWadLump -----------------------------------------------------------------
|
|
|
|
FWadLump::FWadLump ()
|
|
: FileReader ()
|
|
{
|
|
}
|
|
|
|
FWadLump::FWadLump (const FWadLump ©)
|
|
{
|
|
File = copy.File;
|
|
}
|
|
|
|
FWadLump::FWadLump (const FileReader &other, long length)
|
|
: FileReader (other, length)
|
|
{
|
|
}
|
|
|
|
FWadLump::FWadLump (FILE *file, long length)
|
|
: FileReader (file, length)
|
|
{
|
|
}
|
|
|
|
// FMemLump -----------------------------------------------------------------
|
|
|
|
FMemLump::FMemLump ()
|
|
: Block (NULL)
|
|
{
|
|
}
|
|
|
|
#ifdef __GNUC__
|
|
FMemLump::FMemLump (const FMemLump ©)
|
|
#else
|
|
FMemLump::FMemLump (FMemLump ©)
|
|
#endif
|
|
{
|
|
Block = copy.Block;
|
|
const_cast<FMemLump *>(©)->Block = NULL;
|
|
}
|
|
|
|
#ifdef __GNUC__
|
|
FMemLump &FMemLump::operator = (const FMemLump ©)
|
|
#else
|
|
FMemLump &FMemLump::operator = (FMemLump ©)
|
|
#endif
|
|
{
|
|
if (Block != NULL)
|
|
{
|
|
delete[] Block;
|
|
}
|
|
Block = copy.Block;
|
|
const_cast<FMemLump *>(©)->Block = NULL;
|
|
return *this;
|
|
}
|
|
|
|
FMemLump::FMemLump (BYTE *data)
|
|
: Block (data)
|
|
{
|
|
}
|
|
|
|
FMemLump::~FMemLump ()
|
|
{
|
|
if (Block != NULL)
|
|
{
|
|
delete[] Block;
|
|
}
|
|
}
|