SRB2/src/w_wad.c
2023-07-31 15:32:22 +02:00

2656 lines
68 KiB
C

// SONIC ROBO BLAST 2
//-----------------------------------------------------------------------------
// Copyright (C) 1993-1996 by id Software, Inc.
// Copyright (C) 1998-2000 by DooM Legacy Team.
// Copyright (C) 1999-2023 by Sonic Team Junior.
//
// This program is free software distributed under the
// terms of the GNU General Public License, version 2.
// See the 'LICENSE' file for more details.
//-----------------------------------------------------------------------------
/// \file w_wad.c
/// \brief Handles WAD file header, directory, lump I/O
#ifdef HAVE_ZLIB
#ifndef _MSC_VER
#ifndef _LARGEFILE64_SOURCE
#define _LARGEFILE64_SOURCE
#endif
#endif
#ifndef _LFS64_LARGEFILE
#define _LFS64_LARGEFILE
#endif
#ifndef _FILE_OFFSET_BITS
#define _FILE_OFFSET_BITS 0
#endif
#include <zlib.h>
#endif
#ifdef __GNUC__
#include <unistd.h>
#endif
#define ZWAD
#ifdef ZWAD
#include <errno.h>
#include "lzf.h"
#endif
#include "doomdef.h"
#include "doomstat.h"
#include "doomtype.h"
#include "w_wad.h"
#include "z_zone.h"
#include "fastcmp.h"
#include "filesrch.h"
#include "d_main.h"
#include "d_netfil.h"
#include "d_clisrv.h"
#include "dehacked.h"
#include "r_defs.h"
#include "r_data.h"
#include "r_textures.h"
#include "r_patch.h"
#include "r_picformats.h"
#include "i_time.h"
#include "i_system.h"
#include "i_video.h" // rendermode
#include "md5.h"
#include "lua_script.h"
#ifdef SCANTHINGS
#include "p_setup.h" // P_ScanThings
#endif
#include "m_misc.h" // M_MapNumber
#include "g_game.h" // G_SetGameModified
#ifdef HWRENDER
#include "hardware/hw_main.h"
#include "hardware/hw_glob.h"
#endif
#ifdef _DEBUG
#include "console.h"
#endif
#ifndef O_BINARY
#define O_BINARY 0
#endif
typedef struct
{
const char *name;
size_t len;
} lumpchecklist_t;
// Must be a power of two
#define LUMPNUMCACHESIZE 64
typedef struct lumpnum_cache_s
{
char lumpname[32];
lumpnum_t lumpnum;
} lumpnum_cache_t;
static lumpnum_cache_t lumpnumcache[LUMPNUMCACHESIZE];
static UINT16 lumpnumcacheindex = 0;
//===========================================================================
// GLOBALS
//===========================================================================
UINT16 numwadfiles; // number of active wadfiles
wadfile_t **wadfiles; // 0 to numwadfiles-1 are valid
// W_Shutdown
// Closes all of the WAD files before quitting
// If not done on a Mac then open wad files
// can prevent removable media they are on from
// being ejected
void W_Shutdown(void)
{
while (numwadfiles--)
{
wadfile_t *wad = wadfiles[numwadfiles];
if (wad->handle)
fclose(wad->handle);
Z_Free(wad->filename);
if (wad->path)
Z_Free(wad->path);
while (wad->numlumps--)
{
if (wad->lumpinfo[wad->numlumps].diskpath)
Z_Free(wad->lumpinfo[wad->numlumps].diskpath);
Z_Free(wad->lumpinfo[wad->numlumps].longname);
Z_Free(wad->lumpinfo[wad->numlumps].fullname);
}
Z_Free(wad->lumpinfo);
Z_Free(wad);
}
Z_Free(wadfiles);
}
//===========================================================================
// LUMP BASED ROUTINES
//===========================================================================
// W_AddFile
// All files are optional, but at least one file must be
// found (PWAD, if all required lumps are present).
// Files with a .wad extension are wadlink files
// with multiple lumps.
// Other files are single lumps with the base filename
// for the lump name.
static char filenamebuf[MAX_WADPATH];
// W_OpenWadFile
// Helper function for opening the WAD file.
// Returns the FILE * handle for the file, or NULL if not found or could not be opened
// If "useerrors" is true then print errors in the console, else just don't bother
// "filename" may be modified to have the correct path the actual file is located in, if necessary
FILE *W_OpenWadFile(const char **filename, boolean useerrors)
{
FILE *handle;
// Officially, strncpy should not have overlapping buffers, since W_VerifyNMUSlumps is called after this, and it
// changes filename to point at filenamebuf, it would technically be doing that. I doubt any issue will occur since
// they point to the same location, but it's better to be safe and this is a simple change.
if (filenamebuf != *filename)
{
strncpy(filenamebuf, *filename, MAX_WADPATH);
filenamebuf[MAX_WADPATH - 1] = '\0';
*filename = filenamebuf;
}
// open wad file
if ((handle = fopen(*filename, "rb")) == NULL)
{
// If we failed to load the file with the path as specified by
// the user, strip the directories and search for the file.
nameonly(filenamebuf);
// If findfile finds the file, the full path will be returned
// in filenamebuf == *filename.
if (findfile(filenamebuf, NULL, true))
{
if ((handle = fopen(*filename, "rb")) == NULL)
{
if (useerrors)
CONS_Alert(CONS_ERROR, M_GetText("Can't open %s\n"), *filename);
return NULL;
}
}
else
{
if (useerrors)
CONS_Alert(CONS_ERROR, M_GetText("File %s not found.\n"), *filename);
return NULL;
}
}
return handle;
}
// Look for all DEHACKED and Lua scripts inside a PK3 archive.
static void W_LoadDehackedLumpsPK3(UINT16 wadnum, boolean mainfile)
{
UINT16 posStart, posEnd;
posStart = W_CheckNumForFullNamePK3("Init.lua", wadnum, 0);
if (posStart != INT16_MAX)
{
LUA_LoadLump(wadnum, posStart, true);
}
else
{
posStart = W_CheckNumForFolderStartPK3("Lua/", wadnum, 0);
if (posStart != INT16_MAX)
{
posEnd = W_CheckNumForFolderEndPK3("Lua/", wadnum, posStart);
for (; posStart < posEnd; posStart++)
LUA_LoadLump(wadnum, posStart, true);
}
}
posStart = W_CheckNumForFolderStartPK3("SOC/", wadnum, 0);
if (posStart != INT16_MAX)
{
posEnd = W_CheckNumForFolderEndPK3("SOC/", wadnum, posStart);
for(; posStart < posEnd; posStart++)
{
lumpinfo_t *lump_p = &wadfiles[wadnum]->lumpinfo[posStart];
size_t length = strlen(wadfiles[wadnum]->filename) + 1 + strlen(lump_p->fullname); // length of file name, '|', and lump name
char *name = malloc(length + 1);
sprintf(name, "%s|%s", wadfiles[wadnum]->filename, lump_p->fullname);
name[length] = '\0';
CONS_Printf(M_GetText("Loading SOC from %s\n"), name);
DEH_LoadDehackedLumpPwad(wadnum, posStart, mainfile);
free(name);
}
}
}
// search for all DEHACKED lump in all wads and load it
static void W_LoadDehackedLumps(UINT16 wadnum, boolean mainfile)
{
UINT16 lump;
// Find Lua scripts before SOCs to allow new A_Actions in SOC editing.
{
lumpinfo_t *lump_p = wadfiles[wadnum]->lumpinfo;
for (lump = 0; lump < wadfiles[wadnum]->numlumps; lump++, lump_p++)
if (memcmp(lump_p->name,"LUA_",4)==0)
LUA_LoadLump(wadnum, lump, true);
}
{
lumpinfo_t *lump_p = wadfiles[wadnum]->lumpinfo;
for (lump = 0; lump < wadfiles[wadnum]->numlumps; lump++, lump_p++)
if (memcmp(lump_p->name,"SOC_",4)==0) // Check for generic SOC lump
{ // shameless copy+paste of code from LUA_LoadLump
size_t length = strlen(wadfiles[wadnum]->filename) + 1 + strlen(lump_p->fullname); // length of file name, '|', and lump name
char *name = malloc(length + 1);
sprintf(name, "%s|%s", wadfiles[wadnum]->filename, lump_p->fullname);
name[length] = '\0';
CONS_Printf(M_GetText("Loading SOC from %s\n"), name);
DEH_LoadDehackedLumpPwad(wadnum, lump, mainfile);
free(name);
}
else if (memcmp(lump_p->name,"MAINCFG",8)==0) // Check for MAINCFG
{
CONS_Printf(M_GetText("Loading main config from %s\n"), wadfiles[wadnum]->filename);
DEH_LoadDehackedLumpPwad(wadnum, lump, mainfile);
}
else if (memcmp(lump_p->name,"OBJCTCFG",8)==0) // Check for OBJCTCFG
{
CONS_Printf(M_GetText("Loading object config from %s\n"), wadfiles[wadnum]->filename);
DEH_LoadDehackedLumpPwad(wadnum, lump, mainfile);
}
}
#ifdef SCANTHINGS
// Scan maps for emblems 'n shit
{
lumpinfo_t *lump_p = wadfiles[wadnum]->lumpinfo;
for (lump = 0; lump < wadfiles[wadnum]->numlumps; lump++, lump_p++)
{
const char *name = lump_p->name;
if (name[0] == 'M' && name[1] == 'A' && name[2] == 'P' && name[5]=='\0')
{
INT16 mapnum = (INT16)M_MapNumber(name[3], name[4]);
P_ScanThings(mapnum, wadnum, lump + ML_THINGS);
}
}
}
#endif
}
/** Compute MD5 message digest for bytes read from STREAM of this filname.
*
* The resulting message digest number will be written into the 16 bytes
* beginning at RESBLOCK.
*
* \param filename path of file
* \param resblock resulting MD5 checksum
* \return 0 if MD5 checksum was made, and is at resblock, 1 if error was found
*/
static INT32 W_MakeFileMD5(const char *filename, void *resblock)
{
#ifdef NOMD5
(void)filename;
memset(resblock, 0x00, 16);
#else
FILE *fhandle;
if ((fhandle = fopen(filename, "rb")) != NULL)
{
tic_t t = I_GetTime();
CONS_Debug(DBG_SETUP, "Making MD5 for %s\n",filename);
if (md5_stream(fhandle, resblock) == 1)
{
fclose(fhandle);
return 1;
}
CONS_Debug(DBG_SETUP, "MD5 calc for %s took %f seconds\n",
filename, (float)(I_GetTime() - t)/NEWTICRATE);
fclose(fhandle);
return 0;
}
#endif
return 1;
}
// Invalidates the cache of lump numbers. Call this whenever a wad is added.
static void W_InvalidateLumpnumCache(void)
{
memset(lumpnumcache, 0, sizeof (lumpnumcache));
}
/** Detect a file type.
* \todo Actually detect the wad/pkzip headers and whatnot, instead of just checking the extensions.
*/
static restype_t ResourceFileDetect (const char* filename)
{
if (!stricmp(&filename[strlen(filename) - 4], ".pk3"))
return RET_PK3;
if (!stricmp(&filename[strlen(filename) - 4], ".soc"))
return RET_SOC;
if (!stricmp(&filename[strlen(filename) - 4], ".lua"))
return RET_LUA;
return RET_WAD;
}
/** Create a 1-lump lumpinfo_t for standalone files.
*/
static lumpinfo_t* ResGetLumpsStandalone (FILE* handle, UINT16* numlumps, const char* lumpname)
{
lumpinfo_t* lumpinfo = Z_Calloc(sizeof (*lumpinfo), PU_STATIC, NULL);
lumpinfo->position = 0;
fseek(handle, 0, SEEK_END);
lumpinfo->size = ftell(handle);
fseek(handle, 0, SEEK_SET);
strcpy(lumpinfo->name, lumpname);
lumpinfo->hash = quickncasehash(lumpname, 8);
// Allocate the lump's long name.
lumpinfo->longname = Z_Malloc(9 * sizeof(char), PU_STATIC, NULL);
strcpy(lumpinfo->longname, lumpname);
lumpinfo->longname[8] = '\0';
// Allocate the lump's full name.
lumpinfo->fullname = Z_Malloc(9 * sizeof(char), PU_STATIC, NULL);
strcpy(lumpinfo->fullname, lumpname);
lumpinfo->fullname[8] = '\0';
*numlumps = 1;
return lumpinfo;
}
/** Create a lumpinfo_t array for a WAD file.
*/
static lumpinfo_t* ResGetLumpsWad (FILE* handle, UINT16* nlmp, const char* filename)
{
UINT16 numlumps = *nlmp;
lumpinfo_t* lumpinfo;
size_t i;
INT32 compressed = 0;
wadinfo_t header;
lumpinfo_t *lump_p;
filelump_t *fileinfo;
void *fileinfov;
// read the header
if (fread(&header, 1, sizeof header, handle) < sizeof header)
{
CONS_Alert(CONS_ERROR, M_GetText("Can't read wad header because %s\n"), M_FileError(handle));
return NULL;
}
if (memcmp(header.identification, "ZWAD", 4) == 0)
compressed = 1;
else if (memcmp(header.identification, "IWAD", 4) != 0
&& memcmp(header.identification, "PWAD", 4) != 0
&& memcmp(header.identification, "SDLL", 4) != 0)
{
CONS_Alert(CONS_ERROR, M_GetText("Invalid WAD header\n"));
return NULL;
}
header.numlumps = LONG(header.numlumps);
header.infotableofs = LONG(header.infotableofs);
// read wad file directory
i = header.numlumps * sizeof (*fileinfo);
fileinfov = fileinfo = malloc(i);
if (fseek(handle, header.infotableofs, SEEK_SET) == -1
|| fread(fileinfo, 1, i, handle) < i)
{
CONS_Alert(CONS_ERROR, M_GetText("Corrupt wadfile directory (%s)\n"), M_FileError(handle));
free(fileinfov);
return NULL;
}
numlumps = header.numlumps;
// fill in lumpinfo for this wad
lump_p = lumpinfo = Z_Malloc(numlumps * sizeof (*lumpinfo), PU_STATIC, NULL);
for (i = 0; i < numlumps; i++, lump_p++, fileinfo++)
{
lump_p->position = LONG(fileinfo->filepos);
lump_p->size = lump_p->disksize = LONG(fileinfo->size);
lump_p->diskpath = NULL;
if (compressed) // wad is compressed, lump might be
{
UINT32 realsize = 0;
if (fseek(handle, lump_p->position, SEEK_SET)
== -1 || fread(&realsize, 1, sizeof realsize,
handle) < sizeof realsize)
{
I_Error("corrupt compressed file: %s; maybe %s", /// \todo Avoid the bailout?
filename, M_FileError(handle));
}
realsize = LONG(realsize);
if (realsize != 0)
{
lump_p->size = realsize;
lump_p->compression = CM_LZF;
}
else
{
lump_p->size -= 4;
lump_p->compression = CM_NOCOMPRESSION;
}
lump_p->position += 4;
lump_p->disksize -= 4;
}
else
lump_p->compression = CM_NOCOMPRESSION;
memset(lump_p->name, 0x00, 9);
strncpy(lump_p->name, fileinfo->name, 8);
lump_p->hash = quickncasehash(lump_p->name, 8);
// Allocate the lump's long name.
lump_p->longname = Z_Malloc(9 * sizeof(char), PU_STATIC, NULL);
strncpy(lump_p->longname, fileinfo->name, 8);
lump_p->longname[8] = '\0';
// Allocate the lump's full name.
lump_p->fullname = Z_Malloc(9 * sizeof(char), PU_STATIC, NULL);
strncpy(lump_p->fullname, fileinfo->name, 8);
lump_p->fullname[8] = '\0';
}
free(fileinfov);
*nlmp = numlumps;
return lumpinfo;
}
/** Optimized pattern search in a file.
*/
static boolean ResFindSignature (FILE* handle, char endPat[], UINT32 startpos)
{
char *s;
int c;
fseek(handle, startpos, SEEK_SET);
s = endPat;
while((c = fgetc(handle)) != EOF)
{
if (*s != c && s > endPat) // No match?
s = endPat; // We "reset" the counter by sending the s pointer back to the start of the array.
if (*s == c)
{
s++;
if (*s == 0x00) // The array pointer has reached the key char which marks the end. It means we have matched the signature.
{
return true;
}
}
}
return false;
}
#if defined(_MSC_VER)
#pragma pack(1)
#endif
typedef struct zend_s
{
char signature[4];
UINT16 diskpos;
UINT16 cdirdisk;
UINT16 diskentries;
UINT16 entries;
UINT32 cdirsize;
UINT32 cdiroffset;
UINT16 commentlen;
} ATTRPACK zend_t;
typedef struct zentry_s
{
char signature[4];
UINT16 version;
UINT16 versionneeded;
UINT16 flags;
UINT16 compression;
UINT16 modtime;
UINT16 moddate;
UINT32 CRC32;
UINT32 compsize;
UINT32 size;
UINT16 namelen;
UINT16 xtralen;
UINT16 commlen;
UINT16 diskstart;
UINT16 attrint;
UINT32 attrext;
UINT32 offset;
} ATTRPACK zentry_t;
typedef struct zlentry_s
{
char signature[4];
UINT16 versionneeded;
UINT16 flags;
UINT16 compression;
UINT16 modtime;
UINT16 moddate;
UINT32 CRC32;
UINT32 compsize;
UINT32 size;
UINT16 namelen;
UINT16 xtralen;
} ATTRPACK zlentry_t;
#if defined(_MSC_VER)
#pragma pack()
#endif
/** Create a lumpinfo_t array for a PKZip file.
*/
static lumpinfo_t* ResGetLumpsZip (FILE* handle, UINT16* nlmp)
{
zend_t zend;
zentry_t zentry;
zlentry_t zlentry;
UINT16 numlumps = *nlmp;
lumpinfo_t* lumpinfo;
lumpinfo_t *lump_p;
size_t i;
char pat_central[] = {0x50, 0x4b, 0x01, 0x02, 0x00};
char pat_end[] = {0x50, 0x4b, 0x05, 0x06, 0x00};
// Look for central directory end signature near end of file.
// Contains entry number (number of lumps), and central directory start offset.
fseek(handle, 0, SEEK_END);
if (!ResFindSignature(handle, pat_end, max(0, ftell(handle) - (22 + 65536))))
{
CONS_Alert(CONS_ERROR, "Missing central directory\n");
return NULL;
}
fseek(handle, -4, SEEK_CUR);
if (fread(&zend, 1, sizeof zend, handle) < sizeof zend)
{
CONS_Alert(CONS_ERROR, "Corrupt central directory (%s)\n", M_FileError(handle));
return NULL;
}
numlumps = zend.entries;
lump_p = lumpinfo = Z_Malloc(numlumps * sizeof (*lumpinfo), PU_STATIC, NULL);
fseek(handle, zend.cdiroffset, SEEK_SET);
for (i = 0; i < numlumps; i++, lump_p++)
{
char* fullname;
char* trimname;
char* dotpos;
if (fread(&zentry, 1, sizeof(zentry_t), handle) < sizeof(zentry_t))
{
CONS_Alert(CONS_ERROR, "Failed to read central directory (%s)\n", M_FileError(handle));
Z_Free(lumpinfo);
return NULL;
}
if (memcmp(zentry.signature, pat_central, 4))
{
CONS_Alert(CONS_ERROR, "Central directory is corrupt\n");
Z_Free(lumpinfo);
return NULL;
}
lump_p->position = zentry.offset; // NOT ACCURATE YET: we still need to read the local entry to find our true position
lump_p->disksize = zentry.compsize;
lump_p->diskpath = NULL;
lump_p->size = zentry.size;
fullname = malloc(zentry.namelen + 1);
if (fgets(fullname, zentry.namelen + 1, handle) != fullname)
{
CONS_Alert(CONS_ERROR, "Unable to read lumpname (%s)\n", M_FileError(handle));
Z_Free(lumpinfo);
free(fullname);
return NULL;
}
// Strip away file address and extension for the 8char name.
if ((trimname = strrchr(fullname, '/')) != 0)
trimname++;
else
trimname = fullname; // Care taken for root files.
if ((dotpos = strrchr(trimname, '.')) == 0)
dotpos = fullname + strlen(fullname); // Watch for files without extension.
memset(lump_p->name, '\0', 9); // Making sure they're initialized to 0. Is it necessary?
strncpy(lump_p->name, trimname, min(8, dotpos - trimname));
lump_p->hash = quickncasehash(lump_p->name, 8);
lump_p->longname = Z_Calloc(dotpos - trimname + 1, PU_STATIC, NULL);
strlcpy(lump_p->longname, trimname, dotpos - trimname + 1);
lump_p->fullname = Z_Calloc(zentry.namelen + 1, PU_STATIC, NULL);
strncpy(lump_p->fullname, fullname, zentry.namelen);
switch(zentry.compression)
{
case 0:
lump_p->compression = CM_NOCOMPRESSION;
break;
#ifdef HAVE_ZLIB
case 8:
lump_p->compression = CM_DEFLATE;
break;
#endif
case 14:
lump_p->compression = CM_LZF;
break;
default:
CONS_Alert(CONS_WARNING, "%s: Unsupported compression method\n", fullname);
lump_p->compression = CM_UNSUPPORTED;
break;
}
free(fullname);
// skip and ignore comments/extra fields
if (fseek(handle, zentry.xtralen + zentry.commlen, SEEK_CUR) != 0)
{
CONS_Alert(CONS_ERROR, "Central directory is corrupt\n");
Z_Free(lumpinfo);
return NULL;
}
}
// Adjust lump position values properly
for (i = 0, lump_p = lumpinfo; i < numlumps; i++, lump_p++)
{
// skip and ignore comments/extra fields
if ((fseek(handle, lump_p->position, SEEK_SET) != 0) || (fread(&zlentry, 1, sizeof(zlentry_t), handle) < sizeof(zlentry_t)))
{
CONS_Alert(CONS_ERROR, "Local headers for lump %s are corrupt\n", lump_p->fullname);
Z_Free(lumpinfo);
return NULL;
}
lump_p->position += sizeof(zlentry_t) + zlentry.namelen + zlentry.xtralen;
}
*nlmp = numlumps;
return lumpinfo;
}
static INT32 CheckPathsNotEqual(const char *path1, const char *path2)
{
INT32 stat = samepaths(path1, path2);
if (stat == 1)
return 0;
else if (stat < 0)
return -1;
return 1;
}
// Returns 1 if the path is valid, 0 if not, and -1 if there was an error.
INT32 W_IsPathToFolderValid(const char *path)
{
INT32 stat;
// Remove path delimiters.
const char *p = path + (strlen(path) - 1);
while (*p == '\\' || *p == '/' || *p == ':')
{
p--;
if (p < path)
return 0;
}
// Check if the path is a directory.
stat = pathisdirectory(path);
if (stat == 0)
return 0;
else if (stat < 0)
{
// The path doesn't exist, so it can't be a directory.
if (direrror == ENOENT)
return 0;
return -1;
}
// Don't add your home, you sodding tic tac.
stat = CheckPathsNotEqual(path, srb2home);
if (stat != 1)
return stat;
// Do the same checks for SRB2's path, and the current directory.
stat = CheckPathsNotEqual(path, srb2path);
if (stat != 1)
return stat;
stat = CheckPathsNotEqual(path, ".");
if (stat != 1)
return stat;
return 1;
}
// Checks if the combination of the first path and the second path are valid.
// If they are, the concatenated path is returned.
static char *CheckConcatFolderPath(const char *startpath, const char *path)
{
if (concatpaths(path, startpath) == 1)
{
char *fn;
if (startpath)
{
size_t len = strlen(startpath) + strlen(path) + strlen(PATHSEP) + 1;
fn = ZZ_Alloc(len);
snprintf(fn, len, "%s" PATHSEP "%s", startpath, path);
}
else
fn = Z_StrDup(path);
return fn;
}
return NULL;
}
// Looks for the first valid full path for a folder.
// Returns NULL if the folder doesn't exist, or it isn't valid.
char *W_GetFullFolderPath(const char *path)
{
// Check the path by itself first.
char *fn = CheckConcatFolderPath(NULL, path);
if (fn)
return fn;
#define checkpath(startpath) \
fn = CheckConcatFolderPath(startpath, path); \
if (fn) \
return fn
checkpath(srb2home); // Then, look in srb2home.
checkpath(srb2path); // Now, look in srb2path.
checkpath("."); // Finally, look in the current directory.
#undef checkpath
return NULL;
}
// Loads files from a folder into a lumpinfo structure.
static lumpinfo_t *ResGetLumpsFolder(const char *path, UINT16 *nlmp, UINT16 *nfolders)
{
return getdirectoryfiles(path, nlmp, nfolders);
}
static UINT16 W_InitFileError (const char *filename, boolean exitworthy)
{
if (exitworthy)
{
#ifdef _DEBUG
CONS_Error(va("%s was not found or not valid.\nCheck the log for more details.\n", filename));
#else
I_Error("%s was not found or not valid.\nCheck the log for more details.\n", filename);
#endif
}
else
CONS_Printf(M_GetText("Errors occurred while loading %s; not added.\n"), filename);
return INT16_MAX;
}
static void W_ReadFileShaders(wadfile_t *wadfile)
{
#ifdef HWRENDER
if (rendermode == render_opengl && (vid.glstate == VID_GL_LIBRARY_LOADED))
{
HWR_LoadCustomShadersFromFile(numwadfiles - 1, W_FileHasFolders(wadfile));
HWR_CompileShaders();
}
#else
(void)wadfile;
#endif
}
// Allocate a wadfile, setup the lumpinfo (directory) and
// lumpcache, add the wadfile to the current active wadfiles
//
// now returns index into wadfiles[], you can get wadfile_t *
// with:
// wadfiles[<return value>]
//
// return -1 in case of problem
//
// Can now load dehacked files (.soc)
//
UINT16 W_InitFile(const char *filename, boolean mainfile, boolean startup)
{
FILE *handle;
lumpinfo_t *lumpinfo = NULL;
wadfile_t *wadfile;
restype_t type;
UINT16 numlumps = 0;
#ifndef NOMD5
size_t i;
#endif
UINT8 md5sum[16];
int important;
if (!(refreshdirmenu & REFRESHDIR_ADDFILE))
refreshdirmenu = REFRESHDIR_NORMAL|REFRESHDIR_ADDFILE; // clean out cons_alerts that happened earlier
if (refreshdirname)
Z_Free(refreshdirname);
if (dirmenu)
{
refreshdirname = Z_StrDup(filename);
nameonly(refreshdirname);
}
else
refreshdirname = NULL;
//CONS_Debug(DBG_SETUP, "Loading %s\n", filename);
// Check if the game reached the limit of active wadfiles.
if (numwadfiles >= MAX_WADFILES)
{
CONS_Alert(CONS_ERROR, M_GetText("Maximum wad files reached\n"));
refreshdirmenu |= REFRESHDIR_MAX;
return W_InitFileError(filename, startup);
}
// open wad file
if ((handle = W_OpenWadFile(&filename, true)) == NULL)
return W_InitFileError(filename, startup);
important = W_VerifyNMUSlumps(filename, startup);
if (important == -1)
{
fclose(handle);
return INT16_MAX;
}
important = !important;
#ifndef NOMD5
//
// w-waiiiit!
// Let's not add a wad file if the MD5 matches
// an MD5 of an already added WAD file!
//
W_MakeFileMD5(filename, md5sum);
for (i = 0; i < numwadfiles; i++)
{
if (wadfiles[i]->type == RET_FOLDER)
continue;
if (!memcmp(wadfiles[i]->md5sum, md5sum, 16))
{
CONS_Alert(CONS_ERROR, M_GetText("%s is already loaded\n"), filename);
if (handle)
fclose(handle);
return W_InitFileError(filename, false);
}
}
#endif
switch(type = ResourceFileDetect(filename))
{
case RET_SOC:
lumpinfo = ResGetLumpsStandalone(handle, &numlumps, "OBJCTCFG");
break;
case RET_LUA:
lumpinfo = ResGetLumpsStandalone(handle, &numlumps, "LUA_INIT");
break;
case RET_PK3:
lumpinfo = ResGetLumpsZip(handle, &numlumps);
break;
case RET_WAD:
lumpinfo = ResGetLumpsWad(handle, &numlumps, filename);
break;
default:
CONS_Alert(CONS_ERROR, "Unsupported file format\n");
}
if (lumpinfo == NULL)
{
fclose(handle);
return W_InitFileError(filename, startup);
}
if (important && !mainfile)
{
//G_SetGameModified(true);
modifiedgame = true; // avoid savemoddata being set to false
}
//
// link wad file to search files
//
wadfile = Z_Malloc(sizeof (*wadfile), PU_STATIC, NULL);
wadfile->filename = Z_StrDup(filename);
wadfile->path = NULL;
wadfile->type = type;
wadfile->handle = handle;
wadfile->numlumps = numlumps;
wadfile->foldercount = 0;
wadfile->lumpinfo = lumpinfo;
wadfile->important = important;
fseek(handle, 0, SEEK_END);
wadfile->filesize = (unsigned)ftell(handle);
wadfile->type = type;
// already generated, just copy it over
M_Memcpy(&wadfile->md5sum, &md5sum, 16);
//
// set up caching
//
Z_Calloc(numlumps * sizeof (*wadfile->lumpcache), PU_STATIC, &wadfile->lumpcache);
Z_Calloc(numlumps * sizeof (*wadfile->patchcache), PU_STATIC, &wadfile->patchcache);
//
// add the wadfile
//
CONS_Printf(M_GetText("Added file %s (%u lumps)\n"), filename, numlumps);
wadfiles = Z_Realloc(wadfiles, sizeof(wadfile_t *) * (numwadfiles + 1), PU_STATIC, NULL);
wadfiles[numwadfiles] = wadfile;
numwadfiles++; // must come BEFORE W_LoadDehackedLumps, so any addfile called by COM_BufInsertText called by Lua doesn't overwrite what we just loaded
// Read shaders from file
W_ReadFileShaders(wadfile);
// TODO: HACK ALERT - Load Lua & SOC stuff right here. I feel like this should be out of this place, but... Let's stick with this for now.
switch (wadfile->type)
{
case RET_WAD:
W_LoadDehackedLumps(numwadfiles - 1, mainfile);
break;
case RET_PK3:
W_LoadDehackedLumpsPK3(numwadfiles - 1, mainfile);
break;
case RET_SOC:
CONS_Printf(M_GetText("Loading SOC from %s\n"), wadfile->filename);
DEH_LoadDehackedLumpPwad(numwadfiles - 1, 0, mainfile);
break;
case RET_LUA:
LUA_LoadLump(numwadfiles - 1, 0, true);
break;
default:
break;
}
W_InvalidateLumpnumCache();
return wadfile->numlumps;
}
//
// Loads a folder as a WAD.
//
UINT16 W_InitFolder(const char *path, boolean mainfile, boolean startup)
{
lumpinfo_t *lumpinfo = NULL;
wadfile_t *wadfile;
UINT16 numlumps = 0;
UINT16 foldercount;
size_t i;
char *fn, *fullpath;
const char *p;
int important;
INT32 stat;
if (!(refreshdirmenu & REFRESHDIR_ADDFILE))
refreshdirmenu = REFRESHDIR_NORMAL|REFRESHDIR_ADDFILE; // clean out cons_alerts that happened earlier
if (refreshdirname)
Z_Free(refreshdirname);
if (dirmenu)
refreshdirname = Z_StrDup(path);
else
refreshdirname = NULL;
if (numwadfiles >= MAX_WADFILES)
{
CONS_Alert(CONS_ERROR, M_GetText("Maximum wad files reached\n"));
refreshdirmenu |= REFRESHDIR_MAX;
return W_InitFileError(path, startup);
}
important = 1; /// \todo Implement a W_VerifyFolder.
// Remove path delimiters.
p = path + (strlen(path) - 1);
while (*p == '\\' || *p == '/' || *p == ':')
{
p--;
if (p < path)
{
CONS_Alert(CONS_ERROR, M_GetText("Path %s is invalid\n"), path);
return W_InitFileError(path, startup);
}
}
p++;
// Allocate the new path name.
i = (p - path) + 1;
fn = ZZ_Alloc(i);
strlcpy(fn, path, i);
// Don't add an empty path.
if (M_IsStringEmpty(fn))
{
CONS_Alert(CONS_ERROR, M_GetText("Folder name is empty\n"));
Z_Free(fn);
if (startup)
return W_InitFileError("A folder", true);
else
return W_InitFileError("a folder", false);
}
// Check if the path is valid.
stat = W_IsPathToFolderValid(fn);
if (stat != 1)
{
if (stat == 0)
CONS_Alert(CONS_ERROR, M_GetText("Path %s is invalid\n"), fn);
else if (stat < 0)
{
#ifndef AVOID_ERRNO
CONS_Alert(CONS_ERROR, M_GetText("Could not stat %s: %s\n"), fn, strerror(direrror));
#else
CONS_Alert(CONS_ERROR, M_GetText("Could not stat %s\n"), fn);
#endif
}
Z_Free(fn);
return W_InitFileError(path, startup);
}
// Get the full path for this folder.
fullpath = W_GetFullFolderPath(fn);
if (fullpath == NULL)
{
CONS_Alert(CONS_ERROR, M_GetText("Path %s is invalid\n"), fn);
Z_Free(fn);
return W_InitFileError(path, startup);
}
// Check if the folder is already added.
for (i = 0; i < numwadfiles; i++)
{
if (wadfiles[i]->type != RET_FOLDER)
continue;
if (samepaths(wadfiles[i]->path, fullpath) > 0)
{
CONS_Alert(CONS_ERROR, M_GetText("%s is already loaded\n"), path);
Z_Free(fn);
Z_Free(fullpath);
return W_InitFileError(path, false);
}
}
lumpinfo = ResGetLumpsFolder(fullpath, &numlumps, &foldercount);
if (lumpinfo == NULL)
{
if (!numlumps)
CONS_Alert(CONS_ERROR, M_GetText("Folder %s is empty\n"), path);
else if (numlumps == UINT16_MAX)
CONS_Alert(CONS_ERROR, M_GetText("Folder %s contains too many files\n"), path);
else
CONS_Alert(CONS_ERROR, M_GetText("Unknown error enumerating files from folder %s\n"), path);
Z_Free(fn);
Z_Free(fullpath);
return W_InitFileError(path, startup);
}
if (important && !mainfile)
G_SetGameModified(true);
wadfile = Z_Malloc(sizeof (*wadfile), PU_STATIC, NULL);
wadfile->filename = fn;
wadfile->path = fullpath;
wadfile->type = RET_FOLDER;
wadfile->handle = NULL;
wadfile->numlumps = numlumps;
wadfile->foldercount = foldercount;
wadfile->lumpinfo = lumpinfo;
wadfile->important = important;
// Irrelevant.
wadfile->filesize = 0;
memset(wadfile->md5sum, 0x00, 16);
Z_Calloc(numlumps * sizeof (*wadfile->lumpcache), PU_STATIC, &wadfile->lumpcache);
Z_Calloc(numlumps * sizeof (*wadfile->patchcache), PU_STATIC, &wadfile->patchcache);
CONS_Printf(M_GetText("Added folder %s (%u files, %u folders)\n"), fn, numlumps, foldercount);
wadfiles = Z_Realloc(wadfiles, sizeof(wadfile_t *) * (numwadfiles + 1), PU_STATIC, NULL);
wadfiles[numwadfiles] = wadfile;
numwadfiles++;
W_ReadFileShaders(wadfile);
W_LoadDehackedLumpsPK3(numwadfiles - 1, mainfile);
W_InvalidateLumpnumCache();
return wadfile->numlumps;
}
/** Tries to load a series of files.
* All files are wads unless they have an extension of ".soc" or ".lua".
*
* Each file is optional, but at least one file must be found or an error will
* result. Lump names can appear multiple times. The name searcher looks
* backwards, so a later file overrides all earlier ones.
*
* \param filenames A null-terminated list of files to use.
*/
void W_InitMultipleFiles(addfilelist_t *list)
{
size_t i = 0;
for (; i < list->numfiles; i++)
{
const char *fn = list->files[i];
char pathsep = fn[strlen(fn) - 1];
boolean mainfile = (numwadfiles < mainwads);
//CONS_Debug(DBG_SETUP, "Loading %s\n", fn);
if (pathsep == '\\' || pathsep == '/')
W_InitFolder(fn, mainfile, true);
else
W_InitFile(fn, mainfile, true);
}
}
/** Make sure a lump number is valid.
* Compiles away to nothing if PARANOIA is not defined.
*/
static boolean TestValidLump(UINT16 wad, UINT16 lump)
{
I_Assert(wad < numwadfiles);
if (!wadfiles[wad]) // make sure the wad file exists
return false;
I_Assert(lump < wadfiles[wad]->numlumps);
if (lump >= wadfiles[wad]->numlumps) // make sure the lump exists
return false;
return true;
}
const char *W_CheckNameForNumPwad(UINT16 wad, UINT16 lump)
{
if (lump >= wadfiles[wad]->numlumps || !TestValidLump(wad, 0))
return NULL;
return wadfiles[wad]->lumpinfo[lump].name;
}
const char *W_CheckNameForNum(lumpnum_t lumpnum)
{
return W_CheckNameForNumPwad(WADFILENUM(lumpnum),LUMPNUM(lumpnum));
}
//
// Same as the original, but checks in one pwad only.
// wadid is a wad number
// (Used for sprites loading)
//
// 'startlump' is the lump number to start the search
//
UINT16 W_CheckNumForNamePwad(const char *name, UINT16 wad, UINT16 startlump)
{
UINT16 i;
static char uname[8 + 1];
UINT32 hash;
if (!TestValidLump(wad,0))
return INT16_MAX;
strlcpy(uname, name, sizeof uname);
strupr(uname);
hash = quickncasehash(uname, 8);
//
// scan forward
// start at 'startlump', useful parameter when there are multiple
// resources with the same name
//
if (startlump < wadfiles[wad]->numlumps)
{
lumpinfo_t *lump_p = wadfiles[wad]->lumpinfo + startlump;
for (i = startlump; i < wadfiles[wad]->numlumps; i++, lump_p++)
if (lump_p->hash == hash && !strncmp(lump_p->name, uname, sizeof(uname) - 1))
return i;
}
// not found.
return INT16_MAX;
}
//
// Like W_CheckNumForNamePwad, but can find entries with long names
//
// Should be the only version, but that's not possible until we fix
// all the instances of non null-terminated strings in the codebase...
//
UINT16 W_CheckNumForLongNamePwad(const char *name, UINT16 wad, UINT16 startlump)
{
UINT16 i;
static char uname[256 + 1];
if (!TestValidLump(wad,0))
return INT16_MAX;
strlcpy(uname, name, sizeof uname);
strupr(uname);
//
// scan forward
// start at 'startlump', useful parameter when there are multiple
// resources with the same name
//
if (startlump < wadfiles[wad]->numlumps)
{
lumpinfo_t *lump_p = wadfiles[wad]->lumpinfo + startlump;
for (i = startlump; i < wadfiles[wad]->numlumps; i++, lump_p++)
if (!strcmp(lump_p->longname, uname))
return i;
}
// not found.
return INT16_MAX;
}
UINT16
W_CheckNumForMarkerStartPwad (const char *name, UINT16 wad, UINT16 startlump)
{
UINT16 marker;
marker = W_CheckNumForNamePwad(name, wad, startlump);
if (marker != INT16_MAX)
marker++; // Do not count the first marker
return marker;
}
// Look for the first lump from a folder.
UINT16 W_CheckNumForFolderStartPK3(const char *name, UINT16 wad, UINT16 startlump)
{
size_t name_length;
INT32 i;
lumpinfo_t *lump_p = wadfiles[wad]->lumpinfo + startlump;
name_length = strlen(name);
for (i = startlump; i < wadfiles[wad]->numlumps; i++, lump_p++)
{
if (strnicmp(name, lump_p->fullname, name_length) == 0)
{
/* SLADE is special and puts a single directory entry. Skip that. */
if (strlen(lump_p->fullname) == name_length)
i++;
break;
}
}
return i;
}
// In a PK3 type of resource file, it looks for the next lumpinfo entry that doesn't share the specified pathfile.
// Useful for finding folder ends.
// Returns the position of the lumpinfo entry.
UINT16 W_CheckNumForFolderEndPK3(const char *name, UINT16 wad, UINT16 startlump)
{
INT32 i;
lumpinfo_t *lump_p = wadfiles[wad]->lumpinfo + startlump;
for (i = startlump; i < wadfiles[wad]->numlumps; i++, lump_p++)
{
if (strnicmp(name, lump_p->fullname, strlen(name)))
break;
}
return i;
}
// In a PK3 type of resource file, it looks for an entry with the specified full name.
// Returns lump position in PK3's lumpinfo, or INT16_MAX if not found.
UINT16 W_CheckNumForFullNamePK3(const char *name, UINT16 wad, UINT16 startlump)
{
INT32 i;
lumpinfo_t *lump_p = wadfiles[wad]->lumpinfo + startlump;
for (i = startlump; i < wadfiles[wad]->numlumps; i++, lump_p++)
{
if (!strnicmp(name, lump_p->fullname, strlen(name)))
{
return i;
}
}
// Not found at all?
return INT16_MAX;
}
//
// W_CheckNumForName
// Returns LUMPERROR if name not found.
//
lumpnum_t W_CheckNumForName(const char *name)
{
INT32 i;
lumpnum_t check = INT16_MAX;
if (!*name) // some doofus gave us an empty string?
return LUMPERROR;
// Check the lumpnumcache first. Loop backwards so that we check
// most recent entries first
for (i = lumpnumcacheindex + LUMPNUMCACHESIZE; i > lumpnumcacheindex; i--)
{
if (!lumpnumcache[i & (LUMPNUMCACHESIZE - 1)].lumpname[8]
&& strncmp(lumpnumcache[i & (LUMPNUMCACHESIZE - 1)].lumpname, name, 8) == 0)
{
lumpnumcacheindex = i & (LUMPNUMCACHESIZE - 1);
return lumpnumcache[lumpnumcacheindex].lumpnum;
}
}
// scan wad files backwards so patch lump files take precedence
for (i = numwadfiles - 1; i >= 0; i--)
{
check = W_CheckNumForNamePwad(name,(UINT16)i,0);
if (check != INT16_MAX)
break; //found it
}
if (check == INT16_MAX) return LUMPERROR;
else
{
// Update the cache.
lumpnumcacheindex = (lumpnumcacheindex + 1) & (LUMPNUMCACHESIZE - 1);
memset(lumpnumcache[lumpnumcacheindex].lumpname, '\0', 32);
strncpy(lumpnumcache[lumpnumcacheindex].lumpname, name, 8);
lumpnumcache[lumpnumcacheindex].lumpnum = (i<<16)+check;
return lumpnumcache[lumpnumcacheindex].lumpnum;
}
}
//
// Like W_CheckNumForName, but can find entries with long names
//
// Should be the only version, but that's not possible until we fix
// all the instances of non null-terminated strings in the codebase...
//
lumpnum_t W_CheckNumForLongName(const char *name)
{
INT32 i;
lumpnum_t check = INT16_MAX;
if (!*name) // some doofus gave us an empty string?
return LUMPERROR;
// Check the lumpnumcache first. Loop backwards so that we check
// most recent entries first
for (i = lumpnumcacheindex + LUMPNUMCACHESIZE; i > lumpnumcacheindex; i--)
{
if (strcmp(lumpnumcache[i & (LUMPNUMCACHESIZE - 1)].lumpname, name) == 0)
{
lumpnumcacheindex = i & (LUMPNUMCACHESIZE - 1);
return lumpnumcache[lumpnumcacheindex].lumpnum;
}
}
// scan wad files backwards so patch lump files take precedence
for (i = numwadfiles - 1; i >= 0; i--)
{
check = W_CheckNumForLongNamePwad(name,(UINT16)i,0);
if (check != INT16_MAX)
break; //found it
}
if (check == INT16_MAX) return LUMPERROR;
else
{
if (strlen(name) < 32)
{
// Update the cache.
lumpnumcacheindex = (lumpnumcacheindex + 1) & (LUMPNUMCACHESIZE - 1);
memset(lumpnumcache[lumpnumcacheindex].lumpname, '\0', 32);
strlcpy(lumpnumcache[lumpnumcacheindex].lumpname, name, 32);
lumpnumcache[lumpnumcacheindex].lumpnum = (i << 16) + check;
}
return (i << 16) + check;
}
}
// Look for valid map data through all added files in descendant order.
// Get a map marker for WADs, and a standalone WAD file lump inside PK3s.
// TODO: Make it search through cache first, maybe...?
lumpnum_t W_CheckNumForMap(const char *name)
{
UINT32 hash = quickncasehash(name, 8);
UINT16 lumpNum, end;
UINT32 i;
lumpinfo_t *p;
for (i = numwadfiles - 1; i < numwadfiles; i--)
{
if (wadfiles[i]->type == RET_WAD)
{
for (lumpNum = 0; lumpNum < wadfiles[i]->numlumps; lumpNum++)
{
p = wadfiles[i]->lumpinfo + lumpNum;
if (p->hash == hash && !strncmp(name, p->name, 8))
return (i<<16) + lumpNum;
}
}
else if (W_FileHasFolders(wadfiles[i]))
{
lumpNum = W_CheckNumForFolderStartPK3("maps/", i, 0);
if (lumpNum != INT16_MAX)
end = W_CheckNumForFolderEndPK3("maps/", i, lumpNum);
else
continue;
// Now look for the specified map.
for (; lumpNum < end; lumpNum++)
{
p = wadfiles[i]->lumpinfo + lumpNum;
if (p->hash == hash && !strnicmp(name, p->name, 8))
{
const char *extension = strrchr(p->fullname, '.');
if (!(extension && stricmp(extension, ".wad")))
return (i<<16) + lumpNum;
}
}
}
}
return LUMPERROR;
}
//
// W_GetNumForName
//
// Calls W_CheckNumForName, but bombs out if not found.
//
lumpnum_t W_GetNumForName(const char *name)
{
lumpnum_t i;
i = W_CheckNumForName(name);
if (i == LUMPERROR)
I_Error("W_GetNumForName: %s not found!\n", name);
return i;
}
//
// Like W_GetNumForName, but can find entries with long names
//
// Should be the only version, but that's not possible until we fix
// all the instances of non null-terminated strings in the codebase...
//
lumpnum_t W_GetNumForLongName(const char *name)
{
lumpnum_t i;
i = W_CheckNumForLongName(name);
if (i == LUMPERROR)
I_Error("W_GetNumForLongName: %s not found!\n", name);
return i;
}
//
// W_CheckNumForNameInBlock
// Checks only in blocks from blockstart lump to blockend lump
//
lumpnum_t W_CheckNumForNameInBlock(const char *name, const char *blockstart, const char *blockend)
{
INT32 i;
lumpnum_t bsid, beid;
lumpnum_t check = INT16_MAX;
// scan wad files backwards so patch lump files take precedence
for (i = numwadfiles - 1; i >= 0; i--)
{
if (wadfiles[i]->type == RET_WAD)
{
bsid = W_CheckNumForNamePwad(blockstart, (UINT16)i, 0);
if (bsid == INT16_MAX)
continue; // Start block doesn't exist?
beid = W_CheckNumForNamePwad(blockend, (UINT16)i, 0);
if (beid == INT16_MAX)
continue; // End block doesn't exist?
check = W_CheckNumForNamePwad(name, (UINT16)i, bsid);
if (check < beid)
return (i<<16)+check; // found it, in our constraints
}
}
return LUMPERROR;
}
// Used by Lua. Case sensitive lump checking, quickly...
#include "fastcmp.h"
UINT8 W_LumpExists(const char *name)
{
INT32 i,j;
for (i = numwadfiles - 1; i >= 0; i--)
{
lumpinfo_t *lump_p = wadfiles[i]->lumpinfo;
for (j = 0; j < wadfiles[i]->numlumps; ++j, ++lump_p)
if (fastcmp(lump_p->longname, name))
return true;
}
return false;
}
size_t W_LumpLengthPwad(UINT16 wad, UINT16 lump)
{
lumpinfo_t *l;
if (!TestValidLump(wad, lump))
return 0;
l = wadfiles[wad]->lumpinfo + lump;
// Open the external file for this lump, if the WAD is a folder.
if (wadfiles[wad]->type == RET_FOLDER)
{
// pathisdirectory calls stat, so if anything wrong has happened,
// this is the time to be aware of it.
INT32 stat = pathisdirectory(l->diskpath);
if (stat < 0)
{
#ifndef AVOID_ERRNO
if (direrror == ENOENT)
I_Error("W_LumpLengthPwad: file %s doesn't exist", l->diskpath);
else
I_Error("W_LumpLengthPwad: could not stat %s: %s", l->diskpath, strerror(direrror));
#else
I_Error("W_LumpLengthPwad: could not access %s", l->diskpath);
#endif
}
else if (stat == 1) // Path is a folder.
return 0;
else
{
FILE *handle = fopen(l->diskpath, "rb");
if (handle == NULL)
I_Error("W_LumpLengthPwad: could not open file %s", l->diskpath);
fseek(handle, 0, SEEK_END);
l->size = l->disksize = ftell(handle);
fclose(handle);
}
}
return l->size;
}
/** Returns the buffer size needed to load the given lump.
*
* \param lump Lump number to look at.
* \return Buffer size needed, in bytes.
*/
size_t W_LumpLength(lumpnum_t lumpnum)
{
return W_LumpLengthPwad(WADFILENUM(lumpnum),LUMPNUM(lumpnum));
}
//
// W_IsLumpWad
// Is the lump a WAD? (presumably not in a WAD)
//
boolean W_IsLumpWad(lumpnum_t lumpnum)
{
if (W_FileHasFolders(wadfiles[WADFILENUM(lumpnum)]))
{
const char *lumpfullName = (wadfiles[WADFILENUM(lumpnum)]->lumpinfo + LUMPNUM(lumpnum))->fullname;
if (strlen(lumpfullName) < 4)
return false; // can't possibly be a WAD can it?
return !strnicmp(lumpfullName + strlen(lumpfullName) - 4, ".wad", 4);
}
return false; // WADs should never be inside WADs as far as SRB2 is concerned
}
//
// W_IsLumpFolder
// Is the lump a folder? (not in a WAD obviously)
//
boolean W_IsLumpFolder(UINT16 wad, UINT16 lump)
{
if (W_FileHasFolders(wadfiles[wad]))
{
const char *name = wadfiles[wad]->lumpinfo[lump].fullname;
return (name[strlen(name)-1] == '/'); // folders end in '/'
}
return false; // WADs don't have folders
}
#ifdef HAVE_ZLIB
/* report a zlib or i/o error */
void zerr(int ret)
{
CONS_Printf("zpipe: ");
switch (ret) {
case Z_ERRNO:
if (ferror(stdin))
CONS_Printf("error reading stdin\n");
if (ferror(stdout))
CONS_Printf("error writing stdout\n");
break;
case Z_STREAM_ERROR:
CONS_Printf("invalid compression level\n");
break;
case Z_DATA_ERROR:
CONS_Printf("invalid or incomplete deflate data\n");
break;
case Z_MEM_ERROR:
CONS_Printf("out of memory\n");
break;
case Z_VERSION_ERROR:
CONS_Printf("zlib version mismatch!\n");
}
}
#endif
/** Reads bytes from the head of a lump.
* Note: If the lump is compressed, the whole thing has to be read anyway.
*
* \param wad Wad number to read from.
* \param lump Lump number to read from.
* \param dest Buffer in memory to serve as destination.
* \param size Number of bytes to read.
* \param offest Number of bytes to offset.
* \return Number of bytes read (should equal size).
* \sa W_ReadLump, W_RawReadLumpHeader
*/
size_t W_ReadLumpHeaderPwad(UINT16 wad, UINT16 lump, void *dest, size_t size, size_t offset)
{
size_t lumpsize, bytesread;
lumpinfo_t *l;
FILE *handle = NULL;
if (!TestValidLump(wad, lump))
return 0;
l = wadfiles[wad]->lumpinfo + lump;
// Open the external file for this lump, if the WAD is a folder.
if (wadfiles[wad]->type == RET_FOLDER)
{
// pathisdirectory calls stat, so if anything wrong has happened,
// this is the time to be aware of it.
INT32 stat = pathisdirectory(l->diskpath);
if (stat < 0)
{
#ifndef AVOID_ERRNO
if (direrror == ENOENT)
I_Error("W_ReadLumpHeaderPwad: file %s doesn't exist", l->diskpath);
else
I_Error("W_ReadLumpHeaderPwad: could not stat %s: %s", l->diskpath, strerror(direrror));
#else
I_Error("W_ReadLumpHeaderPwad: could not access %s", l->diskpath);
#endif
}
else if (stat == 1) // Path is a folder.
return 0;
else
{
handle = fopen(l->diskpath, "rb");
if (handle == NULL)
I_Error("W_ReadLumpHeaderPwad: could not open file %s", l->diskpath);
// Find length of file
fseek(handle, 0, SEEK_END);
l->size = l->disksize = ftell(handle);
}
}
lumpsize = wadfiles[wad]->lumpinfo[lump].size;
// empty resource (usually markers like S_START, F_END ..)
if (!lumpsize || lumpsize<offset)
{
if (wadfiles[wad]->type == RET_FOLDER)
fclose(handle);
return 0;
}
// zero size means read all the lump
if (!size || size+offset > lumpsize)
size = lumpsize - offset;
// Let's get the raw lump data.
// We setup the desired file handle to read the lump data.
if (wadfiles[wad]->type != RET_FOLDER)
handle = wadfiles[wad]->handle;
fseek(handle, (long)(l->position + offset), SEEK_SET);
// But let's not copy it yet. We support different compression formats on lumps, so we need to take that into account.
switch(wadfiles[wad]->lumpinfo[lump].compression)
{
case CM_NOCOMPRESSION: // If it's uncompressed, we directly write the data into our destination, and return the bytes read.
bytesread = fread(dest, 1, size, handle);
if (wadfiles[wad]->type == RET_FOLDER)
fclose(handle);
#ifdef NO_PNG_LUMPS
if (Picture_IsLumpPNG((UINT8 *)dest, bytesread))
Picture_ThrowPNGError(l->fullname, wadfiles[wad]->filename);
#endif
return bytesread;
case CM_LZF: // Is it LZF compressed? Used by ZWADs.
{
#ifdef ZWAD
char *rawData; // The lump's raw data.
char *decData; // Lump's decompressed real data.
size_t retval; // Helper var, lzf_decompress returns 0 when an error occurs.
rawData = Z_Malloc(l->disksize, PU_STATIC, NULL);
decData = Z_Malloc(l->size, PU_STATIC, NULL);
if (fread(rawData, 1, l->disksize, handle) < l->disksize)
I_Error("wad %d, lump %d: cannot read compressed data", wad, lump);
retval = lzf_decompress(rawData, l->disksize, decData, l->size);
#ifndef AVOID_ERRNO
if (retval == 0) // If this was returned, check if errno was set
{
// errno is a global var set by the lzf functions when something goes wrong.
if (errno == E2BIG)
I_Error("wad %d, lump %d: compressed data too big (bigger than %s)", wad, lump, sizeu1(l->size));
else if (errno == EINVAL)
I_Error("wad %d, lump %d: invalid compressed data", wad, lump);
}
// Otherwise, fall back on below error (if zero was actually the correct size then ???)
#endif
if (retval != l->size)
{
I_Error("wad %d, lump %d: decompressed to wrong number of bytes (expected %s, got %s)", wad, lump, sizeu1(l->size), sizeu2(retval));
}
if (!decData) // Did we get no data at all?
return 0;
M_Memcpy(dest, decData + offset, size);
Z_Free(rawData);
Z_Free(decData);
#ifdef NO_PNG_LUMPS
if (Picture_IsLumpPNG((UINT8 *)dest, size))
Picture_ThrowPNGError(l->fullname, wadfiles[wad]->filename);
#endif
return size;
#else
//I_Error("ZWAD files not supported on this platform.");
return 0;
#endif
}
#ifdef HAVE_ZLIB
case CM_DEFLATE: // Is it compressed via DEFLATE? Very common in ZIPs/PK3s, also what most doom-related editors support.
{
UINT8 *rawData; // The lump's raw data.
UINT8 *decData; // Lump's decompressed real data.
int zErr; // Helper var.
z_stream strm;
unsigned long rawSize = l->disksize;
unsigned long decSize = l->size;
rawData = Z_Malloc(rawSize, PU_STATIC, NULL);
decData = Z_Malloc(decSize, PU_STATIC, NULL);
if (fread(rawData, 1, rawSize, handle) < rawSize)
I_Error("wad %d, lump %d: cannot read compressed data", wad, lump);
strm.zalloc = Z_NULL;
strm.zfree = Z_NULL;
strm.opaque = Z_NULL;
strm.total_in = strm.avail_in = rawSize;
strm.total_out = strm.avail_out = decSize;
strm.next_in = rawData;
strm.next_out = decData;
zErr = inflateInit2(&strm, -15);
if (zErr == Z_OK)
{
zErr = inflate(&strm, Z_FINISH);
if (zErr == Z_STREAM_END)
{
M_Memcpy(dest, decData, size);
}
else
{
size = 0;
zerr(zErr);
}
(void)inflateEnd(&strm);
}
else
{
size = 0;
zerr(zErr);
}
Z_Free(rawData);
Z_Free(decData);
#ifdef NO_PNG_LUMPS
if (Picture_IsLumpPNG((UINT8 *)dest, size))
Picture_ThrowPNGError(l->fullname, wadfiles[wad]->filename);
#endif
return size;
}
#endif
default:
I_Error("wad %d, lump %d: unsupported compression type!", wad, lump);
}
return 0;
}
size_t W_ReadLumpHeader(lumpnum_t lumpnum, void *dest, size_t size, size_t offset)
{
return W_ReadLumpHeaderPwad(WADFILENUM(lumpnum), LUMPNUM(lumpnum), dest, size, offset);
}
/** Reads a lump into memory.
*
* \param lump Lump number to read from.
* \param dest Buffer in memory to serve as destination. Size must be >=
* W_LumpLength().
* \sa W_ReadLumpHeader
*/
void W_ReadLump(lumpnum_t lumpnum, void *dest)
{
W_ReadLumpHeaderPwad(WADFILENUM(lumpnum),LUMPNUM(lumpnum),dest,0,0);
}
void W_ReadLumpPwad(UINT16 wad, UINT16 lump, void *dest)
{
W_ReadLumpHeaderPwad(wad, lump, dest, 0, 0);
}
// ==========================================================================
// W_CacheLumpNum
// ==========================================================================
void *W_CacheLumpNumPwad(UINT16 wad, UINT16 lump, INT32 tag)
{
lumpcache_t *lumpcache;
if (!TestValidLump(wad,lump))
return NULL;
lumpcache = wadfiles[wad]->lumpcache;
if (!lumpcache[lump])
{
void *ptr = Z_Malloc(W_LumpLengthPwad(wad, lump), tag, &lumpcache[lump]);
W_ReadLumpHeaderPwad(wad, lump, ptr, 0, 0); // read the lump in full
}
else
Z_ChangeTag(lumpcache[lump], tag);
return lumpcache[lump];
}
void *W_CacheLumpNum(lumpnum_t lumpnum, INT32 tag)
{
return W_CacheLumpNumPwad(WADFILENUM(lumpnum),LUMPNUM(lumpnum),tag);
}
//
// W_CacheLumpNumForce
//
// Forces the lump to be loaded, even if it already is!
//
void *W_CacheLumpNumForce(lumpnum_t lumpnum, INT32 tag)
{
UINT16 wad, lump;
void *ptr;
wad = WADFILENUM(lumpnum);
lump = LUMPNUM(lumpnum);
if (!TestValidLump(wad,lump))
return NULL;
ptr = Z_Malloc(W_LumpLengthPwad(wad, lump), tag, NULL);
W_ReadLumpHeaderPwad(wad, lump, ptr, 0, 0); // read the lump in full
return ptr;
}
//
// W_IsLumpCached
//
// If a lump is already cached return true, otherwise
// return false.
//
// no outside code uses the PWAD form, for now
static boolean W_IsLumpCachedPWAD(UINT16 wad, UINT16 lump, void *ptr)
{
void *lcache;
if (!TestValidLump(wad, lump))
return false;
lcache = wadfiles[wad]->lumpcache[lump];
if (ptr)
{
if (ptr == lcache)
return true;
}
else if (lcache)
return true;
return false;
}
boolean W_IsLumpCached(lumpnum_t lumpnum, void *ptr)
{
return W_IsLumpCachedPWAD(WADFILENUM(lumpnum),LUMPNUM(lumpnum), ptr);
}
//
// W_IsPatchCached
//
// If a patch is already cached return true, otherwise
// return false.
//
// no outside code uses the PWAD form, for now
static boolean W_IsPatchCachedPWAD(UINT16 wad, UINT16 lump, void *ptr)
{
void *lcache;
if (!TestValidLump(wad, lump))
return false;
lcache = wadfiles[wad]->patchcache[lump];
if (ptr)
{
if (ptr == lcache)
return true;
}
else if (lcache)
return true;
return false;
}
boolean W_IsPatchCached(lumpnum_t lumpnum, void *ptr)
{
return W_IsPatchCachedPWAD(WADFILENUM(lumpnum),LUMPNUM(lumpnum), ptr);
}
// ==========================================================================
// W_CacheLumpName
// ==========================================================================
void *W_CacheLumpName(const char *name, INT32 tag)
{
return W_CacheLumpNum(W_GetNumForName(name), tag);
}
// ==========================================================================
// CACHING OF GRAPHIC PATCH RESOURCES
// ==========================================================================
// Graphic 'patches' are loaded, and if necessary, converted into the format
// the most useful for the current rendermode. For software renderer, the
// graphic patches are kept as is. For the hardware renderer, graphic patches
// are 'unpacked', and are kept into the cache in that unpacked format, and
// the heap memory cache then acts as a 'level 2' cache just after the
// graphics card memory.
//
// Cache a patch into heap memory, convert the patch format as necessary
//
void *W_CacheSoftwarePatchNumPwad(UINT16 wad, UINT16 lump, INT32 tag)
{
lumpcache_t *lumpcache = NULL;
if (!TestValidLump(wad, lump))
return NULL;
lumpcache = wadfiles[wad]->patchcache;
if (!lumpcache[lump])
{
size_t len = W_LumpLengthPwad(wad, lump);
void *ptr, *dest, *lumpdata = Z_Malloc(len, PU_STATIC, NULL);
// read the lump in full
W_ReadLumpHeaderPwad(wad, lump, lumpdata, 0, 0);
ptr = lumpdata;
#ifndef NO_PNG_LUMPS
if (Picture_IsLumpPNG((UINT8 *)lumpdata, len))
ptr = Picture_PNGConvert((UINT8 *)lumpdata, PICFMT_DOOMPATCH, NULL, NULL, NULL, NULL, len, &len, 0);
#endif
dest = Z_Calloc(sizeof(patch_t), tag, &lumpcache[lump]);
Patch_Create(ptr, len, dest);
Z_Free(ptr);
}
else
Z_ChangeTag(lumpcache[lump], tag);
return lumpcache[lump];
}
void *W_CacheSoftwarePatchNum(lumpnum_t lumpnum, INT32 tag)
{
return W_CacheSoftwarePatchNumPwad(WADFILENUM(lumpnum),LUMPNUM(lumpnum),tag);
}
void *W_CachePatchNumPwad(UINT16 wad, UINT16 lump, INT32 tag)
{
patch_t *patch;
if (!TestValidLump(wad, lump))
return NULL;
patch = W_CacheSoftwarePatchNumPwad(wad, lump, tag);
#ifdef HWRENDER
// Software-only compile cache the data without conversion
if (rendermode == render_soft || rendermode == render_none)
#endif
return (void *)patch;
#ifdef HWRENDER
Patch_CreateGL(patch);
return (void *)patch;
#endif
}
void *W_CachePatchNum(lumpnum_t lumpnum, INT32 tag)
{
return W_CachePatchNumPwad(WADFILENUM(lumpnum),LUMPNUM(lumpnum),tag);
}
void W_UnlockCachedPatch(void *patch)
{
if (!patch)
return;
// The hardware code does its own memory management, as its patches
// have different lifetimes from software's.
#ifdef HWRENDER
if (rendermode == render_opengl)
HWR_UnlockCachedPatch((GLPatch_t *)((patch_t *)patch)->hardware);
else
#endif
Z_Unlock(patch);
}
void *W_CachePatchName(const char *name, INT32 tag)
{
lumpnum_t num;
num = W_CheckNumForName(name);
if (num == LUMPERROR)
return W_CachePatchNum(W_GetNumForName("MISSING"), tag);
return W_CachePatchNum(num, tag);
}
void *W_CachePatchLongName(const char *name, INT32 tag)
{
lumpnum_t num;
num = W_CheckNumForLongName(name);
if (num == LUMPERROR)
return W_CachePatchNum(W_GetNumForLongName("MISSING"), tag);
return W_CachePatchNum(num, tag);
}
#ifndef NOMD5
#define MD5_LEN 16
/**
* Prints an MD5 string into a human-readable textual format.
*
* \param md5 The md5 in binary form -- MD5_LEN (16) bytes.
* \param buf Where to print the textual form. Needs 2*MD5_LEN+1 (33) bytes.
* \author Graue <graue@oceanbase.org>
*/
#define MD5_FORMAT \
"%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x%02x"
static void PrintMD5String(const UINT8 *md5, char *buf)
{
snprintf(buf, 2*MD5_LEN+1, MD5_FORMAT,
md5[0], md5[1], md5[2], md5[3],
md5[4], md5[5], md5[6], md5[7],
md5[8], md5[9], md5[10], md5[11],
md5[12], md5[13], md5[14], md5[15]);
}
#endif
/** Verifies a file's MD5 is as it should be.
* For releases, used as cheat prevention -- if the MD5 doesn't match, a
* fatal error is thrown. In debug mode, an MD5 mismatch only triggers a
* warning.
*
* \param wadfilenum Number of the loaded wad file to check.
* \param matchmd5 The MD5 sum this wad should have, expressed as a
* textual string.
* \author Graue <graue@oceanbase.org>
*/
void W_VerifyFileMD5(UINT16 wadfilenum, const char *matchmd5)
{
#ifdef NOMD5
(void)wadfilenum;
(void)matchmd5;
#else
UINT8 realmd5[MD5_LEN];
INT32 ix;
I_Assert(strlen(matchmd5) == 2*MD5_LEN);
I_Assert(wadfilenum < numwadfiles);
// Convert an md5 string like "7d355827fa8f981482246d6c95f9bd48"
// into a real md5.
for (ix = 0; ix < 2*MD5_LEN; ix++)
{
INT32 n, c = matchmd5[ix];
if (isdigit(c))
n = c - '0';
else
{
I_Assert(isxdigit(c));
if (isupper(c)) n = c - 'A' + 10;
else n = c - 'a' + 10;
}
if (ix & 1) realmd5[ix>>1] = (UINT8)(realmd5[ix>>1]+n);
else realmd5[ix>>1] = (UINT8)(n<<4);
}
if (memcmp(realmd5, wadfiles[wadfilenum]->md5sum, 16))
{
char actualmd5text[2*MD5_LEN+1];
PrintMD5String(wadfiles[wadfilenum]->md5sum, actualmd5text);
#ifdef _DEBUG
CONS_Printf
#else
I_Error
#endif
(M_GetText("File is old, is corrupt or has been modified:\n%s\nFound MD5: %s\nWanted MD5: %s\n"), wadfiles[wadfilenum]->filename, actualmd5text, matchmd5);
}
#endif
}
// Verify versions for different archive
// formats. checklist assumed to be valid.
static int
W_VerifyName (const char *name, lumpchecklist_t *checklist, boolean status)
{
size_t j;
for (j = 0; checklist[j].len && checklist[j].name; ++j)
{
if (( strncasecmp(name, checklist[j].name,
checklist[j].len) != false ) == status)
{
return true;
}
}
return false;
}
static int
W_VerifyWAD (FILE *fp, lumpchecklist_t *checklist, boolean status)
{
size_t i;
// assume wad file
wadinfo_t header;
filelump_t lumpinfo;
// read the header
if (fread(&header, 1, sizeof header, fp) == sizeof header
&& header.numlumps < INT16_MAX
&& strncmp(header.identification, "ZWAD", 4)
&& strncmp(header.identification, "IWAD", 4)
&& strncmp(header.identification, "PWAD", 4)
&& strncmp(header.identification, "SDLL", 4))
{
return true;
}
header.numlumps = LONG(header.numlumps);
header.infotableofs = LONG(header.infotableofs);
// let seek to the lumpinfo list
if (fseek(fp, header.infotableofs, SEEK_SET) == -1)
return true;
for (i = 0; i < header.numlumps; i++)
{
// fill in lumpinfo for this wad file directory
if (fread(&lumpinfo, sizeof (lumpinfo), 1 , fp) != 1)
return true;
lumpinfo.filepos = LONG(lumpinfo.filepos);
lumpinfo.size = LONG(lumpinfo.size);
if (lumpinfo.size == 0)
continue;
if (! W_VerifyName(lumpinfo.name, checklist, status))
return false;
}
return true;
}
// List of blacklisted folders to use when checking the PK3
static lumpchecklist_t folderblacklist[] =
{
{"Lua/", 4},
{"SOC/", 4},
{"Sprites/", 8},
{"Textures/", 9},
{"Patches/", 8},
{"Flats/", 6},
{"Fades/", 6},
{NULL, 0},
};
static int
W_VerifyPK3 (FILE *fp, lumpchecklist_t *checklist, boolean status)
{
int verified = true;
zend_t zend;
zentry_t zentry;
zlentry_t zlentry;
long file_size;/* size of zip file */
long data_size;/* size of data inside zip file */
long old_position;
UINT16 numlumps;
size_t i;
char pat_central[] = {0x50, 0x4b, 0x01, 0x02, 0x00};
char pat_end[] = {0x50, 0x4b, 0x05, 0x06, 0x00};
char lumpname[9];
// Haha the ResGetLumpsZip function doesn't
// check for file errors, so neither will I.
// Central directory bullshit
fseek(fp, 0, SEEK_END);
file_size = ftell(fp);
if (!ResFindSignature(fp, pat_end, max(0, ftell(fp) - (22 + 65536))))
return true;
fseek(fp, -4, SEEK_CUR);
if (fread(&zend, 1, sizeof zend, fp) < sizeof zend)
return true;
data_size = sizeof zend;
numlumps = zend.entries;
fseek(fp, zend.cdiroffset, SEEK_SET);
for (i = 0; i < numlumps; i++)
{
char* fullname;
char* trimname;
char* dotpos;
if (fread(&zentry, 1, sizeof(zentry_t), fp) < sizeof(zentry_t))
return true;
if (memcmp(zentry.signature, pat_central, 4))
return true;
if (verified == true)
{
fullname = malloc(zentry.namelen + 1);
if (fgets(fullname, zentry.namelen + 1, fp) != fullname)
return true;
// Strip away file address and extension for the 8char name.
if ((trimname = strrchr(fullname, '/')) != 0)
trimname++;
else
trimname = fullname; // Care taken for root files.
if (*trimname) // Ignore directories, well kinda
{
if ((dotpos = strrchr(trimname, '.')) == 0)
dotpos = fullname + strlen(fullname); // Watch for files without extension.
memset(lumpname, '\0', 9); // Making sure they're initialized to 0. Is it necessary?
strncpy(lumpname, trimname, min(8, dotpos - trimname));
if (! W_VerifyName(lumpname, checklist, status))
verified = false;
// Check for directories next, if it's blacklisted it will return false
else if (W_VerifyName(fullname, folderblacklist, status))
verified = false;
}
free(fullname);
// skip and ignore comments/extra fields
if (fseek(fp, zentry.xtralen + zentry.commlen, SEEK_CUR) != 0)
return true;
}
else
{
if (fseek(fp, zentry.namelen + zentry.xtralen + zentry.commlen, SEEK_CUR) != 0)
return true;
}
data_size +=
sizeof zentry + zentry.namelen + zentry.xtralen + zentry.commlen;
old_position = ftell(fp);
if (fseek(fp, zentry.offset, SEEK_SET) != 0)
return true;
if (fread(&zlentry, 1, sizeof(zlentry_t), fp) < sizeof (zlentry_t))
return true;
data_size +=
sizeof zlentry + zlentry.namelen + zlentry.xtralen + zlentry.compsize;
fseek(fp, old_position, SEEK_SET);
}
if (data_size < file_size)
{
const char * error = "ZIP file has holes (%ld extra bytes)\n";
CONS_Alert(CONS_ERROR, error, (file_size - data_size));
return -1;
}
else if (data_size > file_size)
{
const char * error = "Reported size of ZIP file contents exceeds file size (%ld extra bytes)\n";
CONS_Alert(CONS_ERROR, error, (data_size - file_size));
return -1;
}
else
{
return verified;
}
}
// Note: This never opens lumps themselves and therefore doesn't have to
// deal with compressed lumps.
static int W_VerifyFile(const char *filename, lumpchecklist_t *checklist,
boolean status)
{
FILE *handle;
int goodfile = false;
if (!checklist)
I_Error("No checklist for %s\n", filename);
// open wad file
if ((handle = W_OpenWadFile(&filename, false)) == NULL)
return -1;
if (stricmp(&filename[strlen(filename) - 4], ".pk3") == 0)
goodfile = W_VerifyPK3(handle, checklist, status);
else
{
// detect wad file by the absence of the other supported extensions
if (stricmp(&filename[strlen(filename) - 4], ".soc")
&& stricmp(&filename[strlen(filename) - 4], ".lua"))
{
goodfile = W_VerifyWAD(handle, checklist, status);
}
}
fclose(handle);
return goodfile;
}
/** Checks a wad for lumps other than music and sound.
* Used during game load to verify music.dta is a good file and during a
* netgame join (on the server side) to see if a wad is important enough to
* be sent.
*
* \param filename Filename of the wad to check.
* \param exit_on_error Whether to exit upon file error.
* \return 1 if file contains only music/sound lumps, 0 if it contains other
* stuff (maps, sprites, dehacked lumps, and so on). -1 if there no
* file exists with that filename
* \author Alam Arias
*/
int W_VerifyNMUSlumps(const char *filename, boolean exit_on_error)
{
// MIDI, MOD/S3M/IT/XM/OGG/MP3/WAV, WAVE SFX
// ENDOOM text and palette lumps
lumpchecklist_t NMUSlist[] =
{
{"D_", 2}, // MIDI music
{"O_", 2}, // Digital music
{"DS", 2}, // Sound effects
{"ENDOOM", 6}, // ENDOOM text lump
{"PLAYPAL", 7}, // Palette changes
{"PAL", 3}, // Palette changes
{"COLORMAP", 8}, // Colormap changes
{"CLM", 3}, // Colormap changes
{"TRANS", 5}, // Translucency map changes
{"CONSBACK", 8}, // Console Background graphic
{"SAVE", 4}, // Save Select graphics here and below
{"BLACXLVL", 8},
{"GAMEDONE", 8},
{"CONT", 4}, // Continue icons on saves (probably not used anymore)
{"STNONEX", 7}, // "X" graphic
{"ULTIMATE", 8}, // Ultimate no-save
{"SLCT", 4}, // Level select "cursor"
{"LSSTATIC", 8}, // Level select static
{"BLANKLV", 7}, // "?" level images
{"CRFNT", 5}, // Sonic 1 font changes
{"NTFNT", 5}, // Character Select font changes
{"NTFNO", 5}, // Character Select font (outline)
{"LTFNT", 5}, // Level title font changes
{"TTL", 3}, // Act number changes
{"STCFN", 5}, // Console font changes
{"TNYFN", 5}, // Tiny console font changes
{"STLIVE", 6}, // Life graphics, background and the "X" that shows under skin's HUDNAME
{"CROSHAI", 7}, // First person crosshairs
{"INTERSC", 7}, // Default intermission backgrounds (co-op)
{"SPECTILE", 8}, // Special stage intermission background
{"STT", 3}, // Acceptable HUD changes (Score Time Rings)
{"YB_", 3}, // Intermission graphics, goes with the above
{"RESULT", 6}, // Used in intermission for competitive modes, above too :3
{"RACE", 4}, // Race mode graphics, 321go
{"SRB2BACK", 8}, // MP intermission background
{"M_", 2}, // Menu stuff
{"LT", 2}, // Titlecard changes
{"HOMING", 6}, // Emerald hunt radar
{"HOMITM", 6}, // Emblem radar
{"CHARFG", 6}, // Character select menu
{"CHARBG", 6},
{"RECATK", 6}, // Record Attack menu
{"RECCLOCK", 8},
{"NTSATK", 6}, // NiGHTS Mode menu
{"NTSSONC", 7},
{"SLID", 4}, // Continue
{"CONT", 4},
{"MINICAPS", 8}, // NiGHTS graphics here and below
{"BLUESTAT", 8}, // Sphere status
{"BYELSTAT", 8},
{"ORNGSTAT", 8},
{"REDSTAT", 7},
{"YELSTAT", 7},
{"NBRACKET", 8},
{"NGHTLINK", 8},
{"NGT", 3}, // Link numbers
{"NARROW", 6},
{"NREDAR", 6},
{"NSS", 3},
{"NBON", 4},
{"NRNG", 4},
{"NHUD", 4},
{"CAPS", 4},
{"DRILL", 5},
{"GRADE", 5},
{"MINUS5", 6},
{"NGRTIMER", 8}, // NiGHTS Mode timer
{"MUSICDEF", 8}, // Song definitions (thanks kart)
{"SHADERS", 7}, // OpenGL shader definitions
{"SH_", 3}, // GLSL shader
{NULL, 0},
};
int status = W_VerifyFile(filename, NMUSlist, false);
if (status == -1)
W_InitFileError(filename, exit_on_error);
return status;
}
/** \brief Generates a virtual resource used for level data loading.
*
* \param lumpnum_t reference
* \return Virtual resource
*
*/
virtres_t* vres_GetMap(lumpnum_t lumpnum)
{
UINT32 i;
virtres_t* vres = NULL;
virtlump_t* vlumps = NULL;
size_t numlumps = 0;
if (W_IsLumpWad(lumpnum))
{
// Remember that we're assuming that the WAD will have a specific set of lumps in a specific order.
UINT8 *wadData = W_CacheLumpNum(lumpnum, PU_LEVEL);
filelump_t *fileinfo = (filelump_t *)(wadData + ((wadinfo_t *)wadData)->infotableofs);
numlumps = ((wadinfo_t *)wadData)->numlumps;
vlumps = Z_Malloc(sizeof(virtlump_t)*numlumps, PU_LEVEL, NULL);
// Build the lumps.
for (i = 0; i < numlumps; i++)
{
vlumps[i].size = (size_t)(((filelump_t *)(fileinfo + i))->size);
// Play it safe with the name in this case.
memcpy(vlumps[i].name, (fileinfo + i)->name, 8);
vlumps[i].name[8] = '\0';
vlumps[i].data = Z_Malloc(vlumps[i].size, PU_LEVEL, NULL); // This is memory inefficient, sorry about that.
memcpy(vlumps[i].data, wadData + (fileinfo + i)->filepos, vlumps[i].size);
}
Z_Free(wadData);
}
else
{
// Count number of lumps until the end of resource OR up until next "MAPXX" lump.
lumpnum_t lumppos = lumpnum + 1;
for (i = LUMPNUM(lumppos); i < wadfiles[WADFILENUM(lumpnum)]->numlumps; i++, lumppos++, numlumps++)
if (memcmp(W_CheckNameForNum(lumppos), "MAP", 3) == 0)
break;
numlumps++;
vlumps = Z_Malloc(sizeof(virtlump_t)*numlumps, PU_LEVEL, NULL);
for (i = 0; i < numlumps; i++, lumpnum++)
{
vlumps[i].size = W_LumpLength(lumpnum);
memcpy(vlumps[i].name, W_CheckNameForNum(lumpnum), 8);
vlumps[i].name[8] = '\0';
vlumps[i].data = W_CacheLumpNum(lumpnum, PU_LEVEL);
}
}
vres = Z_Malloc(sizeof(virtres_t), PU_LEVEL, NULL);
vres->vlumps = vlumps;
vres->numlumps = numlumps;
return vres;
}
/** \brief Frees zone memory for a given virtual resource.
*
* \param Virtual resource
*/
void vres_Free(virtres_t* vres)
{
while (vres->numlumps--)
Z_Free(vres->vlumps[vres->numlumps].data);
Z_Free(vres->vlumps);
Z_Free(vres);
}
/** (Debug) Prints lumps from a virtual resource into console.
*/
/*
static void vres_Diag(const virtres_t* vres)
{
UINT32 i;
for (i = 0; i < vres->numlumps; i++)
CONS_Printf("%s\n", vres->vlumps[i].name);
}
*/
/** \brief Finds a lump in a given virtual resource.
*
* \param Virtual resource
* \param Lump name to look for
* \return Virtual lump if found, NULL otherwise
*
*/
virtlump_t* vres_Find(const virtres_t* vres, const char* name)
{
UINT32 i;
for (i = 0; i < vres->numlumps; i++)
if (fastcmp(name, vres->vlumps[i].name))
return &vres->vlumps[i];
return NULL;
}