// cmodel.c -- model loading #include "cm_local.h" #ifdef BSPC void SetPlaneSignbits (cplane_t *out) { int bits, j; // for fast box on planeside test bits = 0; for (j=0 ; j<3 ; j++) { if (out->normal[j] < 0) { bits |= 1<signbits = bits; } #endif //BSPC // to allow boxes to be treated as brush models, we allocate // some extra indexes along with those needed by the map #define BOX_BRUSHES 1 #define BOX_SIDES 6 #define BOX_LEAFS 2 #define BOX_PLANES 12 #define LL(x) x=LittleLong(x) clipMap_t cm; int c_pointcontents; int c_traces, c_brush_traces, c_patch_traces; byte *cmod_base; #ifndef BSPC cvar_t *cm_noAreas; cvar_t *cm_noCurves; cvar_t *cm_playerCurveClip; #endif cmodel_t box_model; cplane_t *box_planes; cbrush_t *box_brush; void CM_InitBoxHull (void); void CM_FloodAreaConnections (void); /* =============================================================================== MAP LOADING =============================================================================== */ /* ================= CMod_LoadShaders ================= */ void CMod_LoadShaders( lump_t *l ) { dshader_t *in, *out; int i, count; in = (dshader_t *)(cmod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Com_Error (ERR_DROP, "CMod_LoadShaders: funny lump size"); } count = l->filelen / sizeof(*in); if (count < 1) { Com_Error (ERR_DROP, "Map with no shaders"); } cm.shaders = (dshader_t *) Z_Malloc( count * sizeof( *cm.shaders ), TAG_BSP, qfalse); cm.numShaders = count; memcpy( cm.shaders, in, count * sizeof( *cm.shaders ) ); out = cm.shaders; for ( i=0 ; icontentFlags = LittleLong( out->contentFlags ); out->surfaceFlags = LittleLong( out->surfaceFlags ); } } /* ================= CMod_LoadSubmodels ================= */ void CMod_LoadSubmodels( lump_t *l ) { dmodel_t *in; cmodel_t *out; int i, j, count; int *indexes; in = (dmodel_t *)(cmod_base + l->fileofs); if (l->filelen % sizeof(*in)) Com_Error (ERR_DROP, "CMod_LoadSubmodels: funny lump size"); count = l->filelen / sizeof(*in); if (count < 1) { Com_Error (ERR_DROP, "Map with no models"); } if ( count > MAX_SUBMODELS ) { Com_Error( ERR_DROP, "MAX_SUBMODELS (%d) exceeded by %d", MAX_SUBMODELS, count-MAX_SUBMODELS ); } cm.cmodels = (struct cmodel_s *) Z_Malloc( count * sizeof( *cm.cmodels ), TAG_BSP, qtrue ); cm.numSubModels = count; for ( i=0 ; imins[j] = LittleFloat (in->mins[j]) - 1; out->maxs[j] = LittleFloat (in->maxs[j]) + 1; } if ( i == 0 ) { continue; // world model doesn't need other info } // make a "leaf" just to hold the model's brushes and surfaces out->leaf.numLeafBrushes = LittleLong( in->numBrushes ); indexes = (int *) Z_Malloc( out->leaf.numLeafBrushes * 4, TAG_BSP, qfalse); out->leaf.firstLeafBrush = indexes - cm.leafbrushes; for ( j = 0 ; j < out->leaf.numLeafBrushes ; j++ ) { indexes[j] = LittleLong( in->firstBrush ) + j; } out->leaf.numLeafSurfaces = LittleLong( in->numSurfaces ); indexes = (int *) Z_Malloc( out->leaf.numLeafSurfaces * 4, TAG_BSP, qfalse); out->leaf.firstLeafSurface = indexes - cm.leafsurfaces; for ( j = 0 ; j < out->leaf.numLeafSurfaces ; j++ ) { indexes[j] = LittleLong( in->firstSurface ) + j; } } } /* ================= CMod_LoadNodes ================= */ void CMod_LoadNodes( lump_t *l ) { dnode_t *in; int child; cNode_t *out; int i, j, count; in = (dnode_t *)(cmod_base + l->fileofs); if (l->filelen % sizeof(*in)) Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); count = l->filelen / sizeof(*in); if (count < 1) Com_Error (ERR_DROP, "Map has no nodes"); cm.nodes = (cNode_t *) Z_Malloc( count * sizeof( *cm.nodes ), TAG_BSP, qfalse); cm.numNodes = count; out = cm.nodes; for (i=0 ; iplane = cm.planes + LittleLong( in->planeNum ); for (j=0 ; j<2 ; j++) { child = LittleLong (in->children[j]); out->children[j] = child; } } } /* ================= CM_BoundBrush ================= */ void CM_BoundBrush( cbrush_t *b ) { b->bounds[0][0] = -b->sides[0].plane->dist; b->bounds[1][0] = b->sides[1].plane->dist; b->bounds[0][1] = -b->sides[2].plane->dist; b->bounds[1][1] = b->sides[3].plane->dist; b->bounds[0][2] = -b->sides[4].plane->dist; b->bounds[1][2] = b->sides[5].plane->dist; } /* ================= CMod_LoadBrushes ================= */ void CMod_LoadBrushes( lump_t *l ) { dbrush_t *in; cbrush_t *out; int i, count; in = (dbrush_t *)(cmod_base + l->fileofs); if (l->filelen % sizeof(*in)) { Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); } count = l->filelen / sizeof(*in); cm.brushes = (cbrush_t *) Z_Malloc( ( BOX_BRUSHES + count ) * sizeof( *cm.brushes ), TAG_BSP, qfalse); cm.numBrushes = count; out = cm.brushes; for ( i=0 ; isides = cm.brushsides + LittleLong(in->firstSide); out->numsides = LittleLong(in->numSides); out->shaderNum = LittleLong( in->shaderNum ); if ( out->shaderNum < 0 || out->shaderNum >= cm.numShaders ) { Com_Error( ERR_DROP, "CMod_LoadBrushes: bad shaderNum: %i", out->shaderNum ); } out->contents = cm.shaders[out->shaderNum].contentFlags; //TEMP HACK: for water that cuts vis but is not solid!!! if ( cm.shaders[out->shaderNum].surfaceFlags & SURF_SLICK ) { out->contents &= ~CONTENTS_SOLID; } CM_BoundBrush( out ); } } /* ================= CMod_LoadLeafs ================= */ void CMod_LoadLeafs (lump_t *l) { int i; cLeaf_t *out; dleaf_t *in; int count; in = (dleaf_t *)(cmod_base + l->fileofs); if (l->filelen % sizeof(*in)) Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); count = l->filelen / sizeof(*in); if (count < 1) Com_Error (ERR_DROP, "Map with no leafs"); cm.leafs = (cLeaf_t *) Z_Malloc( ( BOX_LEAFS + count ) * sizeof( *cm.leafs ), TAG_BSP, qfalse); cm.numLeafs = count; out = cm.leafs; for ( i=0 ; icluster = LittleLong (in->cluster); out->area = LittleLong (in->area); out->firstLeafBrush = LittleLong (in->firstLeafBrush); out->numLeafBrushes = LittleLong (in->numLeafBrushes); out->firstLeafSurface = LittleLong (in->firstLeafSurface); out->numLeafSurfaces = LittleLong (in->numLeafSurfaces); if (out->cluster >= cm.numClusters) cm.numClusters = out->cluster + 1; if (out->area >= cm.numAreas) cm.numAreas = out->area + 1; } cm.areas = (cArea_t *) Z_Malloc( cm.numAreas * sizeof( *cm.areas ), TAG_BSP, qtrue ); cm.areaPortals = (int *) Z_Malloc( cm.numAreas * cm.numAreas * sizeof( *cm.areaPortals ), TAG_BSP, qtrue ); } /* ================= CMod_LoadPlanes ================= */ void CMod_LoadPlanes (lump_t *l) { int i, j; cplane_t *out; dplane_t *in; int count; int bits; in = (dplane_t *)(cmod_base + l->fileofs); if (l->filelen % sizeof(*in)) Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); count = l->filelen / sizeof(*in); if (count < 1) Com_Error (ERR_DROP, "Map with no planes"); cm.planes = (struct cplane_s *) Z_Malloc( ( BOX_PLANES + count ) * sizeof( *cm.planes ), TAG_BSP, qfalse); cm.numPlanes = count; out = cm.planes; for ( i=0 ; inormal[j] = LittleFloat (in->normal[j]); if (out->normal[j] < 0) bits |= 1<dist = LittleFloat (in->dist); out->type = PlaneTypeForNormal( out->normal ); out->signbits = bits; } } /* ================= CMod_LoadLeafBrushes ================= */ void CMod_LoadLeafBrushes (lump_t *l) { int i; int *out; int *in; int count; in = (int *)(cmod_base + l->fileofs); if (l->filelen % sizeof(*in)) Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); count = l->filelen / sizeof(*in); cm.leafbrushes = (int *) Z_Malloc( ( BOX_BRUSHES + count ) * sizeof( *cm.leafbrushes ), TAG_BSP, qfalse); cm.numLeafBrushes = count; out = cm.leafbrushes; for ( i=0 ; ifileofs); if (l->filelen % sizeof(*in)) Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); count = l->filelen / sizeof(*in); cm.leafsurfaces = (int *) Z_Malloc( count * sizeof( *cm.leafsurfaces ), TAG_BSP, qfalse); cm.numLeafSurfaces = count; out = cm.leafsurfaces; for ( i=0 ; ifileofs); if ( l->filelen % sizeof(*in) ) { Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); } count = l->filelen / sizeof(*in); cm.brushsides = (cbrushside_t *) Z_Malloc( ( BOX_SIDES + count ) * sizeof( *cm.brushsides ), TAG_BSP, qfalse); cm.numBrushSides = count; out = cm.brushsides; for ( i=0 ; iplaneNum ); out->plane = &cm.planes[num]; out->shaderNum = LittleLong( in->shaderNum ); if ( out->shaderNum < 0 || out->shaderNum >= cm.numShaders ) { Com_Error( ERR_DROP, "CMod_LoadBrushSides: bad shaderNum: %i", out->shaderNum ); } out->surfaceFlags = cm.shaders[out->shaderNum].surfaceFlags; } } /* ================= CMod_LoadEntityString ================= */ void CMod_LoadEntityString( lump_t *l ) { cm.entityString = (char *) Z_Malloc( l->filelen, TAG_BSP, qfalse); cm.numEntityChars = l->filelen; memcpy (cm.entityString, cmod_base + l->fileofs, l->filelen); } /* ================= CMod_LoadVisibility ================= */ #define VIS_HEADER 8 void CMod_LoadVisibility( lump_t *l ) { int len; byte *buf; len = l->filelen; if ( !len ) { cm.clusterBytes = ( cm.numClusters + 31 ) & ~31; cm.visibility = (unsigned char *) Z_Malloc( cm.clusterBytes, TAG_BSP, qfalse); memset( cm.visibility, 255, cm.clusterBytes ); return; } buf = cmod_base + l->fileofs; cm.vised = qtrue; cm.visibility = (unsigned char *) Z_Malloc( len, TAG_BSP, qtrue ); cm.numClusters = LittleLong( ((int *)buf)[0] ); cm.clusterBytes = LittleLong( ((int *)buf)[1] ); memcpy (cm.visibility, buf + VIS_HEADER, len - VIS_HEADER ); } //================================================================== /* ================= CMod_LoadPatches ================= */ #define MAX_PATCH_VERTS 1024 void CMod_LoadPatches( lump_t *surfs, lump_t *verts ) { mapVert_t *dv, *dv_p; dsurface_t *in; int count; int i, j; int c; cPatch_t *patch; vec3_t points[MAX_PATCH_VERTS]; int width, height; int shaderNum; in = (dsurface_t *)(cmod_base + surfs->fileofs); if (surfs->filelen % sizeof(*in)) Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); cm.numSurfaces = count = surfs->filelen / sizeof(*in); cm.surfaces = (cPatch_t **) Z_Malloc( cm.numSurfaces * sizeof( cm.surfaces[0] ), TAG_BSP, qtrue ); dv = (mapVert_t *)(cmod_base + verts->fileofs); if (verts->filelen % sizeof(*dv)) Com_Error (ERR_DROP, "MOD_LoadBmodel: funny lump size"); // scan through all the surfaces, but only load patches, // not planar faces for ( i = 0 ; i < count ; i++, in++ ) { if ( LittleLong( in->surfaceType ) != MST_PATCH ) { continue; // ignore other surfaces } // FIXME: check for non-colliding patches cm.surfaces[ i ] = patch = (cPatch_t *) Z_Malloc( sizeof( *patch ), TAG_BSP, qtrue ); // load the full drawverts onto the stack width = LittleLong( in->patchWidth ); height = LittleLong( in->patchHeight ); c = width * height; if ( c > MAX_PATCH_VERTS ) { Com_Error( ERR_DROP, "ParseMesh: MAX_PATCH_VERTS" ); } dv_p = dv + LittleLong( in->firstVert ); for ( j = 0 ; j < c ; j++, dv_p++ ) { points[j][0] = LittleFloat( dv_p->xyz[0] ); points[j][1] = LittleFloat( dv_p->xyz[1] ); points[j][2] = LittleFloat( dv_p->xyz[2] ); } shaderNum = LittleLong( in->shaderNum ); patch->contents = cm.shaders[shaderNum].contentFlags; patch->surfaceFlags = cm.shaders[shaderNum].surfaceFlags; // create the internal facet structure patch->pc = CM_GeneratePatchCollide( width, height, points ); } } //================================================================== #ifdef BSPC /* ================== CM_FreeMap Free any loaded map and all submodels ================== */ void CM_FreeMap(void) { memset( &cm, 0, sizeof( cm ) ); Hunk_ClearHigh(); CM_ClearLevelPatches(); } #endif //BSPC unsigned CM_LumpChecksum(lump_t *lump) { return LittleLong (Com_BlockChecksum (cmod_base + lump->fileofs, lump->filelen)); } unsigned CM_Checksum(dheader_t *header) { unsigned checksums[16]; checksums[0] = CM_LumpChecksum(&header->lumps[LUMP_SHADERS]); checksums[1] = CM_LumpChecksum(&header->lumps[LUMP_LEAFS]); checksums[2] = CM_LumpChecksum(&header->lumps[LUMP_LEAFBRUSHES]); checksums[3] = CM_LumpChecksum(&header->lumps[LUMP_LEAFSURFACES]); checksums[4] = CM_LumpChecksum(&header->lumps[LUMP_PLANES]); checksums[5] = CM_LumpChecksum(&header->lumps[LUMP_BRUSHSIDES]); checksums[6] = CM_LumpChecksum(&header->lumps[LUMP_BRUSHES]); checksums[7] = CM_LumpChecksum(&header->lumps[LUMP_MODELS]); checksums[8] = CM_LumpChecksum(&header->lumps[LUMP_NODES]); checksums[9] = CM_LumpChecksum(&header->lumps[LUMP_SURFACES]); checksums[10] = CM_LumpChecksum(&header->lumps[LUMP_DRAWVERTS]); return LittleLong(Com_BlockChecksum(checksums, 11 * 4)); } /* ================== CM_LoadMap Loads in the map and all submodels ================== */ void *gpvCachedMapDiskImage = NULL; char gsCachedMapDiskImage[MAX_QPATH]; qboolean gbUsingCachedMapDataRightNow = qfalse; // if true, signifies that you can't delete this at the moment!! (used during z_malloc()-fail recovery attempt) // called in response to a "devmapbsp blah" or "devmapall blah" command, do NOT use inside CM_Load unless you pass in qtrue // // new bool return used to see if anything was freed, used during z_malloc failure re-try // qboolean CM_DeleteCachedMap(qboolean bGuaranteedOkToDelete) { qboolean bActuallyFreedSomething = qfalse; if (bGuaranteedOkToDelete || !gbUsingCachedMapDataRightNow) { // dump cached disk image... // if (gpvCachedMapDiskImage) { Z_Free( gpvCachedMapDiskImage ); gpvCachedMapDiskImage = NULL; bActuallyFreedSomething = qtrue; } gsCachedMapDiskImage[0] = '\0'; // force map loader to ignore cached internal BSP structures for next level CM_LoadMap() call... // cm.name[0] = '\0'; } return bActuallyFreedSomething; } static void CM_LoadMap_Actual( const char *name, qboolean clientload, int *checksum ) { const int *buf; int i; dheader_t header; static unsigned last_checksum; if ( !name || !name[0] ) { Com_Error( ERR_DROP, "CM_LoadMap: NULL name" ); } #ifndef BSPC cm_noAreas = Cvar_Get ("cm_noAreas", "0", CVAR_CHEAT); cm_noCurves = Cvar_Get ("cm_noCurves", "0", CVAR_CHEAT); cm_playerCurveClip = Cvar_Get ("cm_playerCurveClip", "1", CVAR_ARCHIVE|CVAR_CHEAT ); #endif Com_DPrintf( "CM_LoadMap( %s, %i )\n", name, clientload ); if ( !strcmp( cm.name, name ) && clientload ) { *checksum = last_checksum; return; } // if there was a cached disk image but the name was empty (ie ERR_DROP happened) or just doesn't match // the current name, then ditch it... // if (gpvCachedMapDiskImage && (gsCachedMapDiskImage[0] == '\0' || strcmp( gsCachedMapDiskImage, name )) ) { Z_Free(gpvCachedMapDiskImage); gpvCachedMapDiskImage = NULL; gsCachedMapDiskImage[0] = '\0'; } // if there's a valid map name, and it's the same as last time (respawn?), and it's the server-load, // then keep the data from last time... // if (name[0] && !strcmp( cm.name, name ) && !clientload ) { // clear some stuff that needs zeroing... // cm.floodvalid = 0; //NO... don't reset this because the brush checkcounts are cached, //so when you load up, brush checkcounts equal the cm.checkcount //and the trace will be skipped (because everything loads and //traces in the same exact order ever time you load the map) cm.checkcount++;// = 0; memset(cm.areas, 0, cm.numAreas * sizeof( *cm.areas )); memset(cm.areaPortals, 0, cm.numAreas * cm.numAreas * sizeof( *cm.areaPortals )); } else { // ... else load map from scratch... // assert(!clientload); // logic check. I'm assuming that a client load doesn't get this far? // free old stuff memset( &cm, 0, sizeof( cm ) ); CM_ClearLevelPatches(); Z_TagFree(TAG_BSP); if ( !name[0] ) { cm.numLeafs = 1; cm.numClusters = 1; cm.numAreas = 1; cm.cmodels = (struct cmodel_s *) Z_Malloc( sizeof( *cm.cmodels ), TAG_BSP, qtrue ); *checksum = 0; return; } // load the file into a buffer that we either discard as usual at the bottom, or if we've got enough memory // then keep it long enough to save the renderer re-loading it, then discard it after that. // fileHandle_t h; const int iBSPLen = FS_FOpenFileRead( name, &h, qfalse ); if(!h) { Com_Error (ERR_DROP, "Couldn't load %s", name); return; } gsCachedMapDiskImage[0] = '\0'; // flag that map isn't valid, until name is filled in gpvCachedMapDiskImage = Z_Malloc( iBSPLen, TAG_BSP_DISKIMAGE, qfalse); FS_Read(gpvCachedMapDiskImage, iBSPLen, h); FS_FCloseFile( h ); buf = (int*) gpvCachedMapDiskImage; // so the rest of the code works as normal // carry on as before... last_checksum = LittleLong (Com_BlockChecksum (buf, iBSPLen)); header = *(dheader_t *)buf; for (i=0 ; i, if we've got enough ram to keep it for the renderer's disk-load... // extern qboolean Sys_LowPhysicalMemory(); if (Sys_LowPhysicalMemory() //|| com_dedicated->integer // no need to check for dedicated in single-player codebase ) { Z_Free( gpvCachedMapDiskImage ); gpvCachedMapDiskImage = NULL; } else { // ... do nothing, and let the renderer free it after it's finished playing with it... // } CM_InitBoxHull (); } *checksum = last_checksum; // do this whether or not the map was cached from last load... // CM_FloodAreaConnections (); // allow this to be cached if it is loaded by the server if ( !clientload ) { Q_strncpyz( cm.name, name, sizeof( cm.name ) ); Q_strncpyz( gsCachedMapDiskImage, name, sizeof(gsCachedMapDiskImage) ); // so the renderer can check it } CM_CleanLeafCache(); } // need a wrapper function around this because of multiple returns, need to ensure bool is correct... // void CM_LoadMap( const char *name, qboolean clientload, int *checksum ) { gbUsingCachedMapDataRightNow = qtrue; // !!!!!!!!!!!!!!!!!! CM_LoadMap_Actual( name, clientload, checksum ); gbUsingCachedMapDataRightNow = qfalse; // !!!!!!!!!!!!!!!!!! } /* ================== CM_ClipHandleToModel ================== */ cmodel_t *CM_ClipHandleToModel( clipHandle_t handle ) { if ( handle < 0 ) { Com_Error( ERR_DROP, "CM_ClipHandleToModel: bad handle %i", handle ); } if ( handle < cm.numSubModels ) { return &cm.cmodels[handle]; } if ( handle == BOX_MODEL_HANDLE ) { return &box_model; } if ( handle < MAX_SUBMODELS ) { Com_Error( ERR_DROP, "CM_ClipHandleToModel: bad handle %i < %i < %i", cm.numSubModels, handle, MAX_SUBMODELS ); } Com_Error( ERR_DROP, "CM_ClipHandleToModel: bad handle %i", handle + MAX_SUBMODELS ); return NULL; } /* ================== CM_InlineModel ================== */ clipHandle_t CM_InlineModel( int index ) { if ( index < 0 || index >= cm.numSubModels ) { Com_Error (ERR_DROP, "CM_InlineModel: bad number (may need to re-BSP map?)"); } return index; } int CM_NumClusters( void ) { return cm.numClusters; } int CM_NumInlineModels( void ) { return cm.numSubModels; } char *CM_EntityString( void ) { return cm.entityString; } int CM_LeafCluster( int leafnum ) { if (leafnum < 0 || leafnum >= cm.numLeafs) { Com_Error (ERR_DROP, "CM_LeafCluster: bad number"); } return cm.leafs[leafnum].cluster; } int CM_LeafArea( int leafnum ) { if ( leafnum < 0 || leafnum >= cm.numLeafs ) { Com_Error (ERR_DROP, "CM_LeafArea: bad number"); } return cm.leafs[leafnum].area; } //======================================================================= /* =================== CM_InitBoxHull Set up the planes and nodes so that the six floats of a bounding box can just be stored out and get a proper clipping hull structure. =================== */ void CM_InitBoxHull (void) { int i; int side; cplane_t *p; cbrushside_t *s; box_planes = &cm.planes[cm.numPlanes]; box_brush = &cm.brushes[cm.numBrushes]; box_brush->numsides = 6; box_brush->sides = cm.brushsides + cm.numBrushSides; box_brush->contents = CONTENTS_BODY; box_model.leaf.numLeafBrushes = 1; // box_model.leaf.firstLeafBrush = cm.numBrushes; box_model.leaf.firstLeafBrush = cm.numLeafBrushes; cm.leafbrushes[cm.numLeafBrushes] = cm.numBrushes; for (i=0 ; i<6 ; i++) { side = i&1; // brush sides s = &cm.brushsides[cm.numBrushSides+i]; s->plane = cm.planes + (cm.numPlanes+i*2+side); s->surfaceFlags = 0; // planes p = &box_planes[i*2]; p->type = i>>1; p->signbits = 0; VectorClear (p->normal); p->normal[i>>1] = 1; p = &box_planes[i*2+1]; p->type = 3 + (i>>1); p->signbits = 0; VectorClear (p->normal); p->normal[i>>1] = -1; SetPlaneSignbits( p ); } } /* =================== CM_HeadnodeForBox To keep everything totally uniform, bounding boxes are turned into small BSP trees instead of being compared directly. =================== */ clipHandle_t CM_TempBoxModel( const vec3_t mins, const vec3_t maxs ) { box_planes[0].dist = maxs[0]; box_planes[1].dist = -maxs[0]; box_planes[2].dist = mins[0]; box_planes[3].dist = -mins[0]; box_planes[4].dist = maxs[1]; box_planes[5].dist = -maxs[1]; box_planes[6].dist = mins[1]; box_planes[7].dist = -mins[1]; box_planes[8].dist = maxs[2]; box_planes[9].dist = -maxs[2]; box_planes[10].dist = mins[2]; box_planes[11].dist = -mins[2]; VectorCopy( mins, box_brush->bounds[0] ); VectorCopy( maxs, box_brush->bounds[1] ); return BOX_MODEL_HANDLE; } /* =================== CM_ModelBounds =================== */ void CM_ModelBounds( clipHandle_t model, vec3_t mins, vec3_t maxs ) { cmodel_t *cmod; cmod = CM_ClipHandleToModel( model ); VectorCopy( cmod->mins, mins ); VectorCopy( cmod->maxs, maxs ); } int CM_ModelContents( clipHandle_t model ) { cmodel_t *cmod; int contents = 0; int i; cmod = CM_ClipHandleToModel( model ); //MCG ADDED - return the contents, too if( cmod->leaf.numLeafBrushes ) // check for brush { int brushNum; for ( i = cmod->leaf.firstLeafBrush; i < cmod->leaf.firstLeafBrush+cmod->leaf.numLeafBrushes; i++ ) { brushNum = cm.leafbrushes[i]; contents |= cm.brushes[brushNum].contents; } } if( cmod->leaf.numLeafSurfaces ) // if not brush, check for patch { int surfaceNum; for ( i = cmod->leaf.firstLeafSurface; i < cmod->leaf.firstLeafSurface+cmod->leaf.numLeafSurfaces; i++ ) { surfaceNum = cm.leafsurfaces[i]; if ( cm.surfaces[surfaceNum] != NULL ) {//HERNH? How could we have a null surf within our cmod->leaf.numLeafSurfaces? contents |= cm.surfaces[surfaceNum]->contents; } } } return contents; } //support for save/load games /* =================== CM_WritePortalState Writes the portal state to a savegame file =================== */ // having to proto this stuff again here is crap, but wtf?... // qboolean SG_Append(unsigned long chid, void *data, int length); int SG_Read(unsigned long chid, void *pvAddress, int iLength, void **ppvAddressPtr = NULL); void CM_WritePortalState () { SG_Append('PRTS', (void *)cm.areaPortals, cm.numAreas * cm.numAreas * sizeof( *cm.areaPortals )); } /* =================== CM_ReadPortalState Reads the portal state from a savegame file and recalculates the area connections =================== */ void CM_ReadPortalState () { SG_Read('PRTS', (void *)cm.areaPortals, cm.numAreas * cm.numAreas * sizeof( *cm.areaPortals )); CM_FloodAreaConnections (); }