minetest/src/content_mapblock.cpp

2160 lines
66 KiB
C++

/*
Minetest
Copyright (C) 2010-2013 celeron55, Perttu Ahola <celeron55@gmail.com>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 2.1 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public License along
with this program; if not, write to the Free Software Foundation, Inc.,
51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include "content_mapblock.h"
#include "util/numeric.h"
#include "util/directiontables.h"
#include "mapblock_mesh.h" // For MapBlock_LightColor() and MeshCollector
#include "settings.h"
#include "nodedef.h"
#include "client/tile.h"
#include "mesh.h"
#include <IMeshManipulator.h>
#include "client.h"
#include "log.h"
#include "noise.h"
// Distance of light extrapolation (for oversized nodes)
// After this distance, it gives up and considers light level constant
#define SMOOTH_LIGHTING_OVERSIZE 1.0
struct LightFrame
{
f32 lightsA[8];
f32 lightsB[8];
u8 light_source;
};
static const v3s16 light_dirs[8] = {
v3s16(-1, -1, -1),
v3s16(-1, -1, 1),
v3s16(-1, 1, -1),
v3s16(-1, 1, 1),
v3s16( 1, -1, -1),
v3s16( 1, -1, 1),
v3s16( 1, 1, -1),
v3s16( 1, 1, 1),
};
// Create a cuboid.
// collector - the MeshCollector for the resulting polygons
// box - the position and size of the box
// tiles - the tiles (materials) to use (for all 6 faces)
// tilecount - number of entries in tiles, 1<=tilecount<=6
// c - colors of the cuboid's six sides
// txc - texture coordinates - this is a list of texture coordinates
// for the opposite corners of each face - therefore, there
// should be (2+2)*6=24 values in the list. Alternatively,
// pass NULL to use the entire texture for each face. The
// order of the faces in the list is up-down-right-left-back-
// front (compatible with ContentFeatures). If you specified
// 0,0,1,1 for each face, that would be the same as
// passing NULL.
// light source - if greater than zero, the box's faces will not be shaded
void makeCuboid(MeshCollector *collector, const aabb3f &box,
TileSpec *tiles, int tilecount, const video::SColor *c,
const f32* txc, const u8 light_source)
{
assert(tilecount >= 1 && tilecount <= 6); // pre-condition
v3f min = box.MinEdge;
v3f max = box.MaxEdge;
if(txc == NULL) {
static const f32 txc_default[24] = {
0,0,1,1,
0,0,1,1,
0,0,1,1,
0,0,1,1,
0,0,1,1,
0,0,1,1
};
txc = txc_default;
}
video::SColor c1 = c[0];
video::SColor c2 = c[1];
video::SColor c3 = c[2];
video::SColor c4 = c[3];
video::SColor c5 = c[4];
video::SColor c6 = c[5];
if (!light_source) {
applyFacesShading(c1, v3f(0, 1, 0));
applyFacesShading(c2, v3f(0, -1, 0));
applyFacesShading(c3, v3f(1, 0, 0));
applyFacesShading(c4, v3f(-1, 0, 0));
applyFacesShading(c5, v3f(0, 0, 1));
applyFacesShading(c6, v3f(0, 0, -1));
}
video::S3DVertex vertices[24] =
{
// up
video::S3DVertex(min.X,max.Y,max.Z, 0,1,0, c1, txc[0],txc[1]),
video::S3DVertex(max.X,max.Y,max.Z, 0,1,0, c1, txc[2],txc[1]),
video::S3DVertex(max.X,max.Y,min.Z, 0,1,0, c1, txc[2],txc[3]),
video::S3DVertex(min.X,max.Y,min.Z, 0,1,0, c1, txc[0],txc[3]),
// down
video::S3DVertex(min.X,min.Y,min.Z, 0,-1,0, c2, txc[4],txc[5]),
video::S3DVertex(max.X,min.Y,min.Z, 0,-1,0, c2, txc[6],txc[5]),
video::S3DVertex(max.X,min.Y,max.Z, 0,-1,0, c2, txc[6],txc[7]),
video::S3DVertex(min.X,min.Y,max.Z, 0,-1,0, c2, txc[4],txc[7]),
// right
video::S3DVertex(max.X,max.Y,min.Z, 1,0,0, c3, txc[ 8],txc[9]),
video::S3DVertex(max.X,max.Y,max.Z, 1,0,0, c3, txc[10],txc[9]),
video::S3DVertex(max.X,min.Y,max.Z, 1,0,0, c3, txc[10],txc[11]),
video::S3DVertex(max.X,min.Y,min.Z, 1,0,0, c3, txc[ 8],txc[11]),
// left
video::S3DVertex(min.X,max.Y,max.Z, -1,0,0, c4, txc[12],txc[13]),
video::S3DVertex(min.X,max.Y,min.Z, -1,0,0, c4, txc[14],txc[13]),
video::S3DVertex(min.X,min.Y,min.Z, -1,0,0, c4, txc[14],txc[15]),
video::S3DVertex(min.X,min.Y,max.Z, -1,0,0, c4, txc[12],txc[15]),
// back
video::S3DVertex(max.X,max.Y,max.Z, 0,0,1, c5, txc[16],txc[17]),
video::S3DVertex(min.X,max.Y,max.Z, 0,0,1, c5, txc[18],txc[17]),
video::S3DVertex(min.X,min.Y,max.Z, 0,0,1, c5, txc[18],txc[19]),
video::S3DVertex(max.X,min.Y,max.Z, 0,0,1, c5, txc[16],txc[19]),
// front
video::S3DVertex(min.X,max.Y,min.Z, 0,0,-1, c6, txc[20],txc[21]),
video::S3DVertex(max.X,max.Y,min.Z, 0,0,-1, c6, txc[22],txc[21]),
video::S3DVertex(max.X,min.Y,min.Z, 0,0,-1, c6, txc[22],txc[23]),
video::S3DVertex(min.X,min.Y,min.Z, 0,0,-1, c6, txc[20],txc[23]),
};
for(int i = 0; i < 6; i++)
{
switch (tiles[MYMIN(i, tilecount-1)].rotation)
{
case 0:
break;
case 1: //R90
for (int x = 0; x < 4; x++)
vertices[i*4+x].TCoords.rotateBy(90,irr::core::vector2df(0, 0));
break;
case 2: //R180
for (int x = 0; x < 4; x++)
vertices[i*4+x].TCoords.rotateBy(180,irr::core::vector2df(0, 0));
break;
case 3: //R270
for (int x = 0; x < 4; x++)
vertices[i*4+x].TCoords.rotateBy(270,irr::core::vector2df(0, 0));
break;
case 4: //FXR90
for (int x = 0; x < 4; x++){
vertices[i*4+x].TCoords.X = 1.0 - vertices[i*4+x].TCoords.X;
vertices[i*4+x].TCoords.rotateBy(90,irr::core::vector2df(0, 0));
}
break;
case 5: //FXR270
for (int x = 0; x < 4; x++){
vertices[i*4+x].TCoords.X = 1.0 - vertices[i*4+x].TCoords.X;
vertices[i*4+x].TCoords.rotateBy(270,irr::core::vector2df(0, 0));
}
break;
case 6: //FYR90
for (int x = 0; x < 4; x++){
vertices[i*4+x].TCoords.Y = 1.0 - vertices[i*4+x].TCoords.Y;
vertices[i*4+x].TCoords.rotateBy(90,irr::core::vector2df(0, 0));
}
break;
case 7: //FYR270
for (int x = 0; x < 4; x++){
vertices[i*4+x].TCoords.Y = 1.0 - vertices[i*4+x].TCoords.Y;
vertices[i*4+x].TCoords.rotateBy(270,irr::core::vector2df(0, 0));
}
break;
case 8: //FX
for (int x = 0; x < 4; x++){
vertices[i*4+x].TCoords.X = 1.0 - vertices[i*4+x].TCoords.X;
}
break;
case 9: //FY
for (int x = 0; x < 4; x++){
vertices[i*4+x].TCoords.Y = 1.0 - vertices[i*4+x].TCoords.Y;
}
break;
default:
break;
}
}
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
for (s32 j = 0; j < 24; j += 4) {
int tileindex = MYMIN(j / 4, tilecount - 1);
collector->append(tiles[tileindex], vertices + j, 4, indices, 6);
}
}
// Create a cuboid.
// collector - the MeshCollector for the resulting polygons
// box - the position and size of the box
// tiles - the tiles (materials) to use (for all 6 faces)
// tilecount - number of entries in tiles, 1<=tilecount<=6
// lights - vertex light levels. The order is the same as in light_dirs
// txc - texture coordinates - this is a list of texture coordinates
// for the opposite corners of each face - therefore, there
// should be (2+2)*6=24 values in the list. Alternatively, pass
// NULL to use the entire texture for each face. The order of
// the faces in the list is up-down-right-left-back-front
// (compatible with ContentFeatures). If you specified 0,0,1,1
// for each face, that would be the same as passing NULL.
// light_source - node light emission
static void makeSmoothLightedCuboid(MeshCollector *collector, const aabb3f &box,
TileSpec *tiles, int tilecount, const u16 *lights , const f32 *txc,
const u8 light_source)
{
assert(tilecount >= 1 && tilecount <= 6); // pre-condition
v3f min = box.MinEdge;
v3f max = box.MaxEdge;
if (txc == NULL) {
static const f32 txc_default[24] = {
0,0,1,1,
0,0,1,1,
0,0,1,1,
0,0,1,1,
0,0,1,1,
0,0,1,1
};
txc = txc_default;
}
static const u8 light_indices[24] = {
3, 7, 6, 2,
0, 4, 5, 1,
6, 7, 5, 4,
3, 2, 0, 1,
7, 3, 1, 5,
2, 6, 4, 0
};
video::S3DVertex vertices[24] = {
// up
video::S3DVertex(min.X, max.Y, max.Z, 0, 1, 0, video::SColor(), txc[0], txc[1]),
video::S3DVertex(max.X, max.Y, max.Z, 0, 1, 0, video::SColor(), txc[2], txc[1]),
video::S3DVertex(max.X, max.Y, min.Z, 0, 1, 0, video::SColor(), txc[2], txc[3]),
video::S3DVertex(min.X, max.Y, min.Z, 0, 1, 0, video::SColor(), txc[0], txc[3]),
// down
video::S3DVertex(min.X, min.Y, min.Z, 0, -1, 0, video::SColor(), txc[4], txc[5]),
video::S3DVertex(max.X, min.Y, min.Z, 0, -1, 0, video::SColor(), txc[6], txc[5]),
video::S3DVertex(max.X, min.Y, max.Z, 0, -1, 0, video::SColor(), txc[6], txc[7]),
video::S3DVertex(min.X, min.Y, max.Z, 0, -1, 0, video::SColor(), txc[4], txc[7]),
// right
video::S3DVertex(max.X, max.Y, min.Z, 1, 0, 0, video::SColor(), txc[ 8], txc[9]),
video::S3DVertex(max.X, max.Y, max.Z, 1, 0, 0, video::SColor(), txc[10], txc[9]),
video::S3DVertex(max.X, min.Y, max.Z, 1, 0, 0, video::SColor(), txc[10], txc[11]),
video::S3DVertex(max.X, min.Y, min.Z, 1, 0, 0, video::SColor(), txc[ 8], txc[11]),
// left
video::S3DVertex(min.X, max.Y, max.Z, -1, 0, 0, video::SColor(), txc[12], txc[13]),
video::S3DVertex(min.X, max.Y, min.Z, -1, 0, 0, video::SColor(), txc[14], txc[13]),
video::S3DVertex(min.X, min.Y, min.Z, -1, 0, 0, video::SColor(), txc[14], txc[15]),
video::S3DVertex(min.X, min.Y, max.Z, -1, 0, 0, video::SColor(), txc[12], txc[15]),
// back
video::S3DVertex(max.X, max.Y, max.Z, 0, 0, 1, video::SColor(), txc[16], txc[17]),
video::S3DVertex(min.X, max.Y, max.Z, 0, 0, 1, video::SColor(), txc[18], txc[17]),
video::S3DVertex(min.X, min.Y, max.Z, 0, 0, 1, video::SColor(), txc[18], txc[19]),
video::S3DVertex(max.X, min.Y, max.Z, 0, 0, 1, video::SColor(), txc[16], txc[19]),
// front
video::S3DVertex(min.X, max.Y, min.Z, 0, 0, -1, video::SColor(), txc[20], txc[21]),
video::S3DVertex(max.X, max.Y, min.Z, 0, 0, -1, video::SColor(), txc[22], txc[21]),
video::S3DVertex(max.X, min.Y, min.Z, 0, 0, -1, video::SColor(), txc[22], txc[23]),
video::S3DVertex(min.X, min.Y, min.Z, 0, 0, -1, video::SColor(), txc[20], txc[23]),
};
for(int i = 0; i < 6; i++) {
switch (tiles[MYMIN(i, tilecount-1)].rotation) {
case 0:
break;
case 1: //R90
for (int x = 0; x < 4; x++)
vertices[i*4+x].TCoords.rotateBy(90,irr::core::vector2df(0, 0));
break;
case 2: //R180
for (int x = 0; x < 4; x++)
vertices[i*4+x].TCoords.rotateBy(180,irr::core::vector2df(0, 0));
break;
case 3: //R270
for (int x = 0; x < 4; x++)
vertices[i*4+x].TCoords.rotateBy(270,irr::core::vector2df(0, 0));
break;
case 4: //FXR90
for (int x = 0; x < 4; x++) {
vertices[i*4+x].TCoords.X = 1.0 - vertices[i*4+x].TCoords.X;
vertices[i*4+x].TCoords.rotateBy(90,irr::core::vector2df(0, 0));
}
break;
case 5: //FXR270
for (int x = 0; x < 4; x++) {
vertices[i*4+x].TCoords.X = 1.0 - vertices[i*4+x].TCoords.X;
vertices[i*4+x].TCoords.rotateBy(270,irr::core::vector2df(0, 0));
}
break;
case 6: //FYR90
for (int x = 0; x < 4; x++) {
vertices[i*4+x].TCoords.Y = 1.0 - vertices[i*4+x].TCoords.Y;
vertices[i*4+x].TCoords.rotateBy(90,irr::core::vector2df(0, 0));
}
break;
case 7: //FYR270
for (int x = 0; x < 4; x++) {
vertices[i*4+x].TCoords.Y = 1.0 - vertices[i*4+x].TCoords.Y;
vertices[i*4+x].TCoords.rotateBy(270,irr::core::vector2df(0, 0));
}
break;
case 8: //FX
for (int x = 0; x < 4; x++) {
vertices[i*4+x].TCoords.X = 1.0 - vertices[i*4+x].TCoords.X;
}
break;
case 9: //FY
for (int x = 0; x < 4; x++) {
vertices[i*4+x].TCoords.Y = 1.0 - vertices[i*4+x].TCoords.Y;
}
break;
default:
break;
}
}
u16 indices[] = {0,1,2,2,3,0};
for (s32 j = 0; j < 24; ++j) {
int tileindex = MYMIN(j / 4, tilecount - 1);
vertices[j].Color = encode_light_and_color(lights[light_indices[j]],
tiles[tileindex].color, light_source);
if (!light_source)
applyFacesShading(vertices[j].Color, vertices[j].Normal);
}
// Add to mesh collector
for (s32 k = 0; k < 6; ++k) {
int tileindex = MYMIN(k, tilecount - 1);
collector->append(tiles[tileindex], vertices + 4 * k, 4, indices, 6);
}
}
// Create a cuboid.
// collector - the MeshCollector for the resulting polygons
// box - the position and size of the box
// tiles - the tiles (materials) to use (for all 6 faces)
// tilecount - number of entries in tiles, 1<=tilecount<=6
// c - color of the cuboid
// txc - texture coordinates - this is a list of texture coordinates
// for the opposite corners of each face - therefore, there
// should be (2+2)*6=24 values in the list. Alternatively,
// pass NULL to use the entire texture for each face. The
// order of the faces in the list is up-down-right-left-back-
// front (compatible with ContentFeatures). If you specified
// 0,0,1,1 for each face, that would be the same as
// passing NULL.
// light source - if greater than zero, the box's faces will not be shaded
void makeCuboid(MeshCollector *collector, const aabb3f &box, TileSpec *tiles,
int tilecount, const video::SColor &c, const f32* txc,
const u8 light_source)
{
video::SColor color[6];
for (u8 i = 0; i < 6; i++)
color[i] = c;
makeCuboid(collector, box, tiles, tilecount, color, txc, light_source);
}
// Gets the base lighting values for a node
// frame - resulting (opaque) data
// p - node position (absolute)
// data - ...
// light_source - node light emission level
static void getSmoothLightFrame(LightFrame *frame, const v3s16 &p, MeshMakeData *data, u8 light_source)
{
for (int k = 0; k < 8; ++k) {
u16 light = getSmoothLight(p, light_dirs[k], data);
frame->lightsA[k] = light & 0xff;
frame->lightsB[k] = light >> 8;
}
frame->light_source = light_source;
}
// Calculates vertex light level
// frame - light values from getSmoothLightFrame()
// vertex_pos - vertex position in the node (coordinates are clamped to [0.0, 1.0] or so)
static u16 blendLight(const LightFrame &frame, const core::vector3df& vertex_pos)
{
f32 x = core::clamp(vertex_pos.X / BS + 0.5, 0.0 - SMOOTH_LIGHTING_OVERSIZE, 1.0 + SMOOTH_LIGHTING_OVERSIZE);
f32 y = core::clamp(vertex_pos.Y / BS + 0.5, 0.0 - SMOOTH_LIGHTING_OVERSIZE, 1.0 + SMOOTH_LIGHTING_OVERSIZE);
f32 z = core::clamp(vertex_pos.Z / BS + 0.5, 0.0 - SMOOTH_LIGHTING_OVERSIZE, 1.0 + SMOOTH_LIGHTING_OVERSIZE);
f32 lightA = 0.0;
f32 lightB = 0.0;
for (int k = 0; k < 8; ++k) {
f32 dx = (k & 4) ? x : 1 - x;
f32 dy = (k & 2) ? y : 1 - y;
f32 dz = (k & 1) ? z : 1 - z;
lightA += dx * dy * dz * frame.lightsA[k];
lightB += dx * dy * dz * frame.lightsB[k];
}
return
core::clamp(core::round32(lightA), 0, 255) |
core::clamp(core::round32(lightB), 0, 255) << 8;
}
// Calculates vertex color to be used in mapblock mesh
// frame - light values from getSmoothLightFrame()
// vertex_pos - vertex position in the node (coordinates are clamped to [0.0, 1.0] or so)
// tile_color - node's tile color
static video::SColor blendLight(const LightFrame &frame,
const core::vector3df& vertex_pos, video::SColor tile_color)
{
u16 light = blendLight(frame, vertex_pos);
return encode_light_and_color(light, tile_color, frame.light_source);
}
static video::SColor blendLight(const LightFrame &frame,
const core::vector3df& vertex_pos, const core::vector3df& vertex_normal,
video::SColor tile_color)
{
video::SColor color = blendLight(frame, vertex_pos, tile_color);
if (!frame.light_source)
applyFacesShading(color, vertex_normal);
return color;
}
static inline void getNeighborConnectingFace(v3s16 p, INodeDefManager *nodedef,
MeshMakeData *data, MapNode n, int v, int *neighbors)
{
MapNode n2 = data->m_vmanip.getNodeNoEx(p);
if (nodedef->nodeboxConnects(n, n2, v))
*neighbors |= v;
}
static void makeAutoLightedCuboid(MeshCollector *collector, MeshMakeData *data,
const v3f &pos, aabb3f box, TileSpec &tile,
/* pre-computed, for non-smooth lighting only */ const video::SColor color,
/* for smooth lighting only */ const LightFrame &frame)
{
f32 dx1 = box.MinEdge.X;
f32 dy1 = box.MinEdge.Y;
f32 dz1 = box.MinEdge.Z;
f32 dx2 = box.MaxEdge.X;
f32 dy2 = box.MaxEdge.Y;
f32 dz2 = box.MaxEdge.Z;
box.MinEdge += pos;
box.MaxEdge += pos;
f32 tx1 = (box.MinEdge.X / BS) + 0.5;
f32 ty1 = (box.MinEdge.Y / BS) + 0.5;
f32 tz1 = (box.MinEdge.Z / BS) + 0.5;
f32 tx2 = (box.MaxEdge.X / BS) + 0.5;
f32 ty2 = (box.MaxEdge.Y / BS) + 0.5;
f32 tz2 = (box.MaxEdge.Z / BS) + 0.5;
f32 txc[24] = {
tx1, 1-tz2, tx2, 1-tz1, // up
tx1, tz1, tx2, tz2, // down
tz1, 1-ty2, tz2, 1-ty1, // right
1-tz2, 1-ty2, 1-tz1, 1-ty1, // left
1-tx2, 1-ty2, 1-tx1, 1-ty1, // back
tx1, 1-ty2, tx2, 1-ty1, // front
};
if (data->m_smooth_lighting) {
u16 lights[8];
for (int j = 0; j < 8; ++j) {
f32 x = (j & 4) ? dx2 : dx1;
f32 y = (j & 2) ? dy2 : dy1;
f32 z = (j & 1) ? dz2 : dz1;
lights[j] = blendLight(frame, core::vector3df(x, y, z));
}
makeSmoothLightedCuboid(collector, box, &tile, 1, lights, txc, frame.light_source);
} else {
makeCuboid(collector, box, &tile, 1, color, txc, frame.light_source);
}
}
static void makeAutoLightedCuboidEx(MeshCollector *collector, MeshMakeData *data,
const v3f &pos, aabb3f box, TileSpec &tile, f32 *txc,
/* pre-computed, for non-smooth lighting only */ const video::SColor color,
/* for smooth lighting only */ const LightFrame &frame)
{
f32 dx1 = box.MinEdge.X;
f32 dy1 = box.MinEdge.Y;
f32 dz1 = box.MinEdge.Z;
f32 dx2 = box.MaxEdge.X;
f32 dy2 = box.MaxEdge.Y;
f32 dz2 = box.MaxEdge.Z;
box.MinEdge += pos;
box.MaxEdge += pos;
if (data->m_smooth_lighting) {
u16 lights[8];
for (int j = 0; j < 8; ++j) {
f32 x = (j & 4) ? dx2 : dx1;
f32 y = (j & 2) ? dy2 : dy1;
f32 z = (j & 1) ? dz2 : dz1;
lights[j] = blendLight(frame, core::vector3df(x, y, z));
}
makeSmoothLightedCuboid(collector, box, &tile, 1, lights, txc, frame.light_source);
} else {
makeCuboid(collector, box, &tile, 1, color, txc, frame.light_source);
}
}
// For use in mapblock_mesh_generate_special
// X,Y,Z of position must be -1,0,1
// This expression is a simplification of
// 3 * 3 * (pos.X + 1) + 3 * (pos.Y + 1) + (pos.Z + 1)
static inline int NeighborToIndex(const v3s16 &pos)
{
return 9 * pos.X + 3 * pos.Y + pos.Z + 13;
}
/*!
* Returns the i-th special tile for a map node.
*/
static TileSpec getSpecialTile(const ContentFeatures &f,
const MapNode &n, u8 i)
{
TileSpec copy = f.special_tiles[i];
if (!copy.has_color)
n.getColor(f, &copy.color);
return copy;
}
/*
TODO: Fix alpha blending for special nodes
Currently only the last element rendered is blended correct
*/
void mapblock_mesh_generate_special(MeshMakeData *data,
MeshCollector &collector)
{
INodeDefManager *nodedef = data->m_client->ndef();
scene::ISceneManager* smgr = data->m_client->getSceneManager();
scene::IMeshManipulator* meshmanip = smgr->getMeshManipulator();
// 0ms
//TimeTaker timer("mapblock_mesh_generate_special()");
/*
Some settings
*/
bool enable_mesh_cache = g_settings->getBool("enable_mesh_cache") &&
!data->m_smooth_lighting; // Mesh cache is not supported with smooth lighting
v3s16 blockpos_nodes = data->m_blockpos*MAP_BLOCKSIZE;
for(s16 z = 0; z < MAP_BLOCKSIZE; z++)
for(s16 y = 0; y < MAP_BLOCKSIZE; y++)
for(s16 x = 0; x < MAP_BLOCKSIZE; x++)
{
v3s16 p(x,y,z);
MapNode n = data->m_vmanip.getNodeNoEx(blockpos_nodes + p);
const ContentFeatures &f = nodedef->get(n);
// Only solidness=0 stuff is drawn here
if(f.solidness != 0)
continue;
if (f.drawtype == NDT_AIRLIKE)
continue;
LightFrame frame;
if (data->m_smooth_lighting)
getSmoothLightFrame(&frame, blockpos_nodes + p, data, f.light_source);
else
frame.light_source = f.light_source;
switch(f.drawtype) {
default:
infostream << "Got " << f.drawtype << std::endl;
FATAL_ERROR("Unknown drawtype");
break;
case NDT_LIQUID:
{
/*
Add water sources to mesh if using new style
*/
TileSpec tile_liquid = getSpecialTile(f, n, 0);
TileSpec tile_liquid_bfculled = getNodeTile(n, p, v3s16(0,0,0), data);
u16 l = getInteriorLight(n, 0, nodedef);
video::SColor c1 = encode_light_and_color(l,
tile_liquid.color, f.light_source);
video::SColor c2 = encode_light_and_color(l,
tile_liquid_bfculled.color, f.light_source);
bool top_is_same_liquid = false;
MapNode ntop = data->m_vmanip.getNodeNoEx(blockpos_nodes + v3s16(x,y+1,z));
content_t c_flowing = nodedef->getId(f.liquid_alternative_flowing);
content_t c_source = nodedef->getId(f.liquid_alternative_source);
if(ntop.getContent() == c_flowing || ntop.getContent() == c_source)
top_is_same_liquid = true;
/*
Generate sides
*/
v3s16 side_dirs[4] = {
v3s16(1,0,0),
v3s16(-1,0,0),
v3s16(0,0,1),
v3s16(0,0,-1),
};
for(u32 i=0; i<4; i++)
{
v3s16 dir = side_dirs[i];
MapNode neighbor = data->m_vmanip.getNodeNoEx(blockpos_nodes + p + dir);
content_t neighbor_content = neighbor.getContent();
const ContentFeatures &n_feat = nodedef->get(neighbor_content);
MapNode n_top = data->m_vmanip.getNodeNoEx(blockpos_nodes + p + dir+ v3s16(0,1,0));
content_t n_top_c = n_top.getContent();
if(neighbor_content == CONTENT_IGNORE)
continue;
/*
If our topside is liquid and neighbor's topside
is liquid, don't draw side face
*/
if(top_is_same_liquid && (n_top_c == c_flowing ||
n_top_c == c_source || n_top_c == CONTENT_IGNORE))
continue;
// Don't draw face if neighbor is blocking the view
if(n_feat.solidness == 2)
continue;
bool neighbor_is_same_liquid = (neighbor_content == c_source
|| neighbor_content == c_flowing);
// Don't draw any faces if neighbor same is liquid and top is
// same liquid
if(neighbor_is_same_liquid && !top_is_same_liquid)
continue;
// Use backface culled material if neighbor doesn't have a
// solidness of 0
const TileSpec *current_tile = &tile_liquid;
video::SColor *c = &c1;
if(n_feat.solidness != 0 || n_feat.visual_solidness != 0) {
current_tile = &tile_liquid_bfculled;
c = &c2;
}
video::S3DVertex vertices[4] =
{
video::S3DVertex(-BS/2,0,BS/2,0,0,0, *c, 0,1),
video::S3DVertex(BS/2,0,BS/2,0,0,0, *c, 1,1),
video::S3DVertex(BS/2,0,BS/2, 0,0,0, *c, 1,0),
video::S3DVertex(-BS/2,0,BS/2, 0,0,0, *c, 0,0),
};
/*
If our topside is liquid, set upper border of face
at upper border of node
*/
if (top_is_same_liquid) {
vertices[2].Pos.Y = 0.5 * BS;
vertices[3].Pos.Y = 0.5 * BS;
} else {
/*
Otherwise upper position of face is liquid level
*/
vertices[2].Pos.Y = 0.5 * BS;
vertices[3].Pos.Y = 0.5 * BS;
}
/*
If neighbor is liquid, lower border of face is liquid level
*/
if (neighbor_is_same_liquid) {
vertices[0].Pos.Y = 0.5 * BS;
vertices[1].Pos.Y = 0.5 * BS;
} else {
/*
If neighbor is not liquid, lower border of face is
lower border of node
*/
vertices[0].Pos.Y = -0.5 * BS;
vertices[1].Pos.Y = -0.5 * BS;
}
for (s32 j = 0; j < 4; j++) {
if(dir == v3s16(0,0,1))
vertices[j].Pos.rotateXZBy(0);
if(dir == v3s16(0,0,-1))
vertices[j].Pos.rotateXZBy(180);
if(dir == v3s16(-1,0,0))
vertices[j].Pos.rotateXZBy(90);
if(dir == v3s16(1,0,-0))
vertices[j].Pos.rotateXZBy(-90);
// Do this to not cause glitches when two liquids are
// side-by-side
/*if(neighbor_is_same_liquid == false){
vertices[j].Pos.X *= 0.98;
vertices[j].Pos.Z *= 0.98;
}*/
if (data->m_smooth_lighting)
vertices[j].Color = blendLight(frame, vertices[j].Pos, current_tile->color);
vertices[j].Pos += intToFloat(p, BS);
}
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
collector.append(*current_tile, vertices, 4, indices, 6);
}
/*
Generate top
*/
if(top_is_same_liquid)
continue;
video::S3DVertex vertices[4] =
{
video::S3DVertex(-BS/2,0,BS/2, 0,0,0, c1, 0,1),
video::S3DVertex(BS/2,0,BS/2, 0,0,0, c1, 1,1),
video::S3DVertex(BS/2,0,-BS/2, 0,0,0, c1, 1,0),
video::S3DVertex(-BS/2,0,-BS/2, 0,0,0, c1, 0,0),
};
for (s32 i = 0; i < 4; i++) {
vertices[i].Pos.Y += 0.5 * BS;
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, tile_liquid.color);
vertices[i].Pos += intToFloat(p, BS);
}
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
collector.append(tile_liquid, vertices, 4, indices, 6);
break;}
case NDT_FLOWINGLIQUID:
{
/*
Add flowing liquid to mesh
*/
TileSpec tile_liquid = getSpecialTile(f, n, 0);
TileSpec tile_liquid_bfculled = getSpecialTile(f, n, 1);
bool top_is_same_liquid = false;
MapNode ntop = data->m_vmanip.getNodeNoEx(blockpos_nodes + v3s16(x,y+1,z));
content_t c_flowing = nodedef->getId(f.liquid_alternative_flowing);
content_t c_source = nodedef->getId(f.liquid_alternative_source);
if(ntop.getContent() == c_flowing || ntop.getContent() == c_source)
top_is_same_liquid = true;
u16 l = 0;
// If this liquid emits light and doesn't contain light, draw
// it at what it emits, for an increased effect
u8 light_source = nodedef->get(n).light_source;
if(light_source != 0){
l = decode_light(light_source);
l = l | (l<<8);
}
// Use the light of the node on top if possible
else if(nodedef->get(ntop).param_type == CPT_LIGHT)
l = getInteriorLight(ntop, 0, nodedef);
// Otherwise use the light of this node (the liquid)
else
l = getInteriorLight(n, 0, nodedef);
video::SColor c1 = encode_light_and_color(l,
tile_liquid.color, f.light_source);
video::SColor c2 = encode_light_and_color(l,
tile_liquid_bfculled.color, f.light_source);
u8 range = rangelim(nodedef->get(c_flowing).liquid_range, 1, 8);
// Neighbor liquid levels (key = relative position)
// Includes current node
struct NeighborData {
f32 level;
content_t content;
u8 flags;
};
NeighborData neighbor_data_matrix[27];
const u8 neighborflag_top_is_same_liquid = 0x01;
v3s16 neighbor_dirs[9] = {
v3s16(0,0,0),
v3s16(0,0,1),
v3s16(0,0,-1),
v3s16(1,0,0),
v3s16(-1,0,0),
v3s16(1,0,1),
v3s16(-1,0,-1),
v3s16(1,0,-1),
v3s16(-1,0,1),
};
for(u32 i=0; i<9; i++)
{
content_t content = CONTENT_AIR;
float level = -0.5 * BS;
u8 flags = 0;
// Check neighbor
v3s16 p2 = p + neighbor_dirs[i];
MapNode n2 = data->m_vmanip.getNodeNoEx(blockpos_nodes + p2);
if(n2.getContent() != CONTENT_IGNORE)
{
content = n2.getContent();
if(n2.getContent() == c_source)
level = 0.5 * BS;
else if(n2.getContent() == c_flowing){
u8 liquid_level = (n2.param2&LIQUID_LEVEL_MASK);
if (liquid_level <= LIQUID_LEVEL_MAX+1-range)
liquid_level = 0;
else
liquid_level -= (LIQUID_LEVEL_MAX+1-range);
level = (-0.5 + ((float)liquid_level + 0.5) / (float)range) * BS;
}
// Check node above neighbor.
// NOTE: This doesn't get executed if neighbor
// doesn't exist
p2.Y += 1;
n2 = data->m_vmanip.getNodeNoEx(blockpos_nodes + p2);
if(n2.getContent() == c_source ||
n2.getContent() == c_flowing)
flags |= neighborflag_top_is_same_liquid;
}
NeighborData &neighbor_data =
neighbor_data_matrix[NeighborToIndex(neighbor_dirs[i])];
neighbor_data.level = level;
neighbor_data.content = content;
neighbor_data.flags = flags;
}
// Corner heights (average between four liquids)
f32 corner_levels[4];
v3s16 halfdirs[4] = {
v3s16(0,0,0),
v3s16(1,0,0),
v3s16(1,0,1),
v3s16(0,0,1),
};
for(u32 i=0; i<4; i++)
{
v3s16 cornerdir = halfdirs[i];
float cornerlevel = 0;
u32 valid_count = 0;
u32 air_count = 0;
for(u32 j=0; j<4; j++)
{
v3s16 neighbordir = cornerdir - halfdirs[j];
NeighborData &neighbor_data =
neighbor_data_matrix[NeighborToIndex(neighbordir)];
content_t content = neighbor_data.content;
// If top is liquid, draw starting from top of node
if (neighbor_data.flags & neighborflag_top_is_same_liquid)
{
cornerlevel = 0.5*BS;
valid_count = 1;
break;
}
// Source is always the same height
else if(content == c_source)
{
cornerlevel = 0.5 * BS;
valid_count = 1;
break;
}
// Flowing liquid has level information
else if(content == c_flowing)
{
cornerlevel += neighbor_data.level;
valid_count++;
}
else if(content == CONTENT_AIR)
{
air_count++;
}
}
if(air_count >= 2)
cornerlevel = -0.5*BS+0.2;
else if(valid_count > 0)
cornerlevel /= valid_count;
corner_levels[i] = cornerlevel;
}
/*
Generate sides
*/
v3s16 side_dirs[4] = {
v3s16(1,0,0),
v3s16(-1,0,0),
v3s16(0,0,1),
v3s16(0,0,-1),
};
s16 side_corners[4][2] = {
{1, 2},
{3, 0},
{2, 3},
{0, 1},
};
for(u32 i=0; i<4; i++)
{
v3s16 dir = side_dirs[i];
NeighborData& neighbor_data =
neighbor_data_matrix[NeighborToIndex(dir)];
/*
If our topside is liquid and neighbor's topside
is liquid, don't draw side face
*/
if (top_is_same_liquid &&
(neighbor_data.flags & neighborflag_top_is_same_liquid))
continue;
content_t neighbor_content = neighbor_data.content;
const ContentFeatures &n_feat = nodedef->get(neighbor_content);
// Don't draw face if neighbor is blocking the view
if(n_feat.solidness == 2)
continue;
bool neighbor_is_same_liquid = (neighbor_content == c_source
|| neighbor_content == c_flowing);
// Don't draw any faces if neighbor same is liquid and top is
// same liquid
if(neighbor_is_same_liquid == true
&& top_is_same_liquid == false)
continue;
// Use backface culled material if neighbor doesn't have a
// solidness of 0
const TileSpec *current_tile = &tile_liquid;
video::SColor *c = &c1;
if(n_feat.solidness != 0 || n_feat.visual_solidness != 0) {
current_tile = &tile_liquid_bfculled;
c = &c2;
}
video::S3DVertex vertices[4] =
{
video::S3DVertex(-BS/2,0,BS/2, 0,0,0, *c, 0,1),
video::S3DVertex(BS/2,0,BS/2, 0,0,0, *c, 1,1),
video::S3DVertex(BS/2,0,BS/2, 0,0,0, *c, 1,0),
video::S3DVertex(-BS/2,0,BS/2, 0,0,0, *c, 0,0),
};
/*
If our topside is liquid, set upper border of face
at upper border of node
*/
if(top_is_same_liquid)
{
vertices[2].Pos.Y = 0.5*BS;
vertices[3].Pos.Y = 0.5*BS;
}
/*
Otherwise upper position of face is corner levels
*/
else
{
vertices[2].Pos.Y = corner_levels[side_corners[i][0]];
vertices[3].Pos.Y = corner_levels[side_corners[i][1]];
}
/*
If neighbor is liquid, lower border of face is corner
liquid levels
*/
if(neighbor_is_same_liquid)
{
vertices[0].Pos.Y = corner_levels[side_corners[i][1]];
vertices[1].Pos.Y = corner_levels[side_corners[i][0]];
}
/*
If neighbor is not liquid, lower border of face is
lower border of node
*/
else
{
vertices[0].Pos.Y = -0.5*BS;
vertices[1].Pos.Y = -0.5*BS;
}
for(s32 j=0; j<4; j++)
{
if(dir == v3s16(0,0,1))
vertices[j].Pos.rotateXZBy(0);
if(dir == v3s16(0,0,-1))
vertices[j].Pos.rotateXZBy(180);
if(dir == v3s16(-1,0,0))
vertices[j].Pos.rotateXZBy(90);
if(dir == v3s16(1,0,-0))
vertices[j].Pos.rotateXZBy(-90);
// Do this to not cause glitches when two liquids are
// side-by-side
/*if(neighbor_is_same_liquid == false){
vertices[j].Pos.X *= 0.98;
vertices[j].Pos.Z *= 0.98;
}*/
if (data->m_smooth_lighting)
vertices[j].Color = blendLight(frame, vertices[j].Pos, current_tile->color);
vertices[j].Pos += intToFloat(p, BS);
}
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
collector.append(*current_tile, vertices, 4, indices, 6);
}
/*
Generate top side, if appropriate
*/
if(top_is_same_liquid == false)
{
video::S3DVertex vertices[4] =
{
video::S3DVertex(-BS/2,0,BS/2, 0,0,0, c1, 0,1),
video::S3DVertex(BS/2,0,BS/2, 0,0,0, c1, 1,1),
video::S3DVertex(BS/2,0,-BS/2, 0,0,0, c1, 1,0),
video::S3DVertex(-BS/2,0,-BS/2, 0,0,0, c1, 0,0),
};
// To get backface culling right, the vertices need to go
// clockwise around the front of the face. And we happened to
// calculate corner levels in exact reverse order.
s32 corner_resolve[4] = {3,2,1,0};
for(s32 i=0; i<4; i++)
{
//vertices[i].Pos.Y += liquid_level;
//vertices[i].Pos.Y += neighbor_levels[v3s16(0,0,0)];
s32 j = corner_resolve[i];
vertices[i].Pos.Y += corner_levels[j];
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, tile_liquid.color);
vertices[i].Pos += intToFloat(p, BS);
}
// Default downwards-flowing texture animation goes from
// -Z towards +Z, thus the direction is +Z.
// Rotate texture to make animation go in flow direction
// Positive if liquid moves towards +Z
f32 dz = (corner_levels[side_corners[3][0]] +
corner_levels[side_corners[3][1]]) -
(corner_levels[side_corners[2][0]] +
corner_levels[side_corners[2][1]]);
// Positive if liquid moves towards +X
f32 dx = (corner_levels[side_corners[1][0]] +
corner_levels[side_corners[1][1]]) -
(corner_levels[side_corners[0][0]] +
corner_levels[side_corners[0][1]]);
f32 tcoord_angle = atan2(dz, dx) * core::RADTODEG ;
v2f tcoord_center(0.5, 0.5);
v2f tcoord_translate(
blockpos_nodes.Z + z,
blockpos_nodes.X + x);
tcoord_translate.rotateBy(tcoord_angle);
tcoord_translate.X -= floor(tcoord_translate.X);
tcoord_translate.Y -= floor(tcoord_translate.Y);
for(s32 i=0; i<4; i++)
{
vertices[i].TCoords.rotateBy(
tcoord_angle,
tcoord_center);
vertices[i].TCoords += tcoord_translate;
}
v2f t = vertices[0].TCoords;
vertices[0].TCoords = vertices[2].TCoords;
vertices[2].TCoords = t;
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
collector.append(tile_liquid, vertices, 4, indices, 6);
}
break;}
case NDT_GLASSLIKE:
{
TileSpec tile = getNodeTile(n, p, v3s16(0,0,0), data);
u16 l = getInteriorLight(n, 1, nodedef);
video::SColor c = encode_light_and_color(l, tile.color,
f.light_source);
for(u32 j=0; j<6; j++)
{
// Check this neighbor
v3s16 dir = g_6dirs[j];
v3s16 n2p = blockpos_nodes + p + dir;
MapNode n2 = data->m_vmanip.getNodeNoEx(n2p);
// Don't make face if neighbor is of same type
if(n2.getContent() == n.getContent())
continue;
video::SColor c2=c;
if(!f.light_source)
applyFacesShading(c2, v3f(dir.X, dir.Y, dir.Z));
// The face at Z+
video::S3DVertex vertices[4] = {
video::S3DVertex(-BS/2,-BS/2,BS/2, dir.X,dir.Y,dir.Z, c2, 1,1),
video::S3DVertex(BS/2,-BS/2,BS/2, dir.X,dir.Y,dir.Z, c2, 0,1),
video::S3DVertex(BS/2,BS/2,BS/2, dir.X,dir.Y,dir.Z, c2, 0,0),
video::S3DVertex(-BS/2,BS/2,BS/2, dir.X,dir.Y,dir.Z, c2, 1,0),
};
// Rotations in the g_6dirs format
if(j == 0) // Z+
for(u16 i=0; i<4; i++)
vertices[i].Pos.rotateXZBy(0);
else if(j == 1) // Y+
for(u16 i=0; i<4; i++)
vertices[i].Pos.rotateYZBy(-90);
else if(j == 2) // X+
for(u16 i=0; i<4; i++)
vertices[i].Pos.rotateXZBy(-90);
else if(j == 3) // Z-
for(u16 i=0; i<4; i++)
vertices[i].Pos.rotateXZBy(180);
else if(j == 4) // Y-
for(u16 i=0; i<4; i++)
vertices[i].Pos.rotateYZBy(90);
else if(j == 5) // X-
for(u16 i=0; i<4; i++)
vertices[i].Pos.rotateXZBy(90);
for (u16 i = 0; i < 4; i++) {
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, vertices[i].Normal, tile.color);
vertices[i].Pos += intToFloat(p, BS);
}
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
collector.append(tile, vertices, 4, indices, 6);
}
break;}
case NDT_GLASSLIKE_FRAMED_OPTIONAL:
// This is always pre-converted to something else
FATAL_ERROR("NDT_GLASSLIKE_FRAMED_OPTIONAL not pre-converted as expected");
break;
case NDT_GLASSLIKE_FRAMED:
{
static const v3s16 dirs[6] = {
v3s16( 0, 1, 0),
v3s16( 0,-1, 0),
v3s16( 1, 0, 0),
v3s16(-1, 0, 0),
v3s16( 0, 0, 1),
v3s16( 0, 0,-1)
};
u16 l = getInteriorLight(n, 1, nodedef);
u8 i;
TileSpec tiles[6];
for (i = 0; i < 6; i++)
tiles[i] = getNodeTile(n, p, dirs[i], data);
video::SColor tile0color = encode_light_and_color(l,
tiles[0].color, f.light_source);
TileSpec glass_tiles[6];
video::SColor glasscolor[6];
if (tiles[1].texture && tiles[2].texture && tiles[3].texture) {
glass_tiles[0] = tiles[2];
glass_tiles[1] = tiles[3];
glass_tiles[2] = tiles[1];
glass_tiles[3] = tiles[1];
glass_tiles[4] = tiles[1];
glass_tiles[5] = tiles[1];
} else {
for (i = 0; i < 6; i++)
glass_tiles[i] = tiles[1];
}
for (i = 0; i < 6; i++)
glasscolor[i] = encode_light_and_color(l, glass_tiles[i].color,
f.light_source);
u8 param2 = n.getParam2();
bool H_merge = ! bool(param2 & 128);
bool V_merge = ! bool(param2 & 64);
param2 = param2 & 63;
v3f pos = intToFloat(p, BS);
static const float a = BS / 2;
static const float g = a - 0.003;
static const float b = .876 * ( BS / 2 );
static const aabb3f frame_edges[12] = {
aabb3f( b, b,-a, a, a, a), // y+
aabb3f(-a, b,-a,-b, a, a), // y+
aabb3f( b,-a,-a, a,-b, a), // y-
aabb3f(-a,-a,-a,-b,-b, a), // y-
aabb3f( b,-a, b, a, a, a), // x+
aabb3f( b,-a,-a, a, a,-b), // x+
aabb3f(-a,-a, b,-b, a, a), // x-
aabb3f(-a,-a,-a,-b, a,-b), // x-
aabb3f(-a, b, b, a, a, a), // z+
aabb3f(-a,-a, b, a,-b, a), // z+
aabb3f(-a,-a,-a, a,-b,-b), // z-
aabb3f(-a, b,-a, a, a,-b) // z-
};
static const aabb3f glass_faces[6] = {
aabb3f(-g, g,-g, g, g, g), // y+
aabb3f(-g,-g,-g, g,-g, g), // y-
aabb3f( g,-g,-g, g, g, g), // x+
aabb3f(-g,-g,-g,-g, g, g), // x-
aabb3f(-g,-g, g, g, g, g), // z+
aabb3f(-g,-g,-g, g, g,-g) // z-
};
// table of node visible faces, 0 = invisible
int visible_faces[6] = {0,0,0,0,0,0};
// table of neighbours, 1 = same type, checked with g_26dirs
int nb[18] = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0};
// g_26dirs to check when only horizontal merge is allowed
int nb_H_dirs[8] = {0,2,3,5,10,11,12,13};
content_t current = n.getContent();
content_t n2c;
MapNode n2;
v3s16 n2p;
// neighbours checks for frames visibility
if (!H_merge && V_merge) {
n2p = blockpos_nodes + p + g_26dirs[1];
n2 = data->m_vmanip.getNodeNoEx(n2p);
n2c = n2.getContent();
if (n2c == current || n2c == CONTENT_IGNORE)
nb[1] = 1;
n2p = blockpos_nodes + p + g_26dirs[4];
n2 = data->m_vmanip.getNodeNoEx(n2p);
n2c = n2.getContent();
if (n2c == current || n2c == CONTENT_IGNORE)
nb[4] = 1;
} else if (H_merge && !V_merge) {
for(i = 0; i < 8; i++) {
n2p = blockpos_nodes + p + g_26dirs[nb_H_dirs[i]];
n2 = data->m_vmanip.getNodeNoEx(n2p);
n2c = n2.getContent();
if (n2c == current || n2c == CONTENT_IGNORE)
nb[nb_H_dirs[i]] = 1;
}
} else if (H_merge && V_merge) {
for(i = 0; i < 18; i++) {
n2p = blockpos_nodes + p + g_26dirs[i];
n2 = data->m_vmanip.getNodeNoEx(n2p);
n2c = n2.getContent();
if (n2c == current || n2c == CONTENT_IGNORE)
nb[i] = 1;
}
}
// faces visibility checks
if (!V_merge) {
visible_faces[0] = 1;
visible_faces[1] = 1;
} else {
for(i = 0; i < 2; i++) {
n2p = blockpos_nodes + p + dirs[i];
n2 = data->m_vmanip.getNodeNoEx(n2p);
n2c = n2.getContent();
if (n2c != current)
visible_faces[i] = 1;
}
}
if (!H_merge) {
visible_faces[2] = 1;
visible_faces[3] = 1;
visible_faces[4] = 1;
visible_faces[5] = 1;
} else {
for(i = 2; i < 6; i++) {
n2p = blockpos_nodes + p + dirs[i];
n2 = data->m_vmanip.getNodeNoEx(n2p);
n2c = n2.getContent();
if (n2c != current)
visible_faces[i] = 1;
}
}
static const u8 nb_triplet[12*3] = {
1,2, 7, 1,5, 6, 4,2,15, 4,5,14,
2,0,11, 2,3,13, 5,0,10, 5,3,12,
0,1, 8, 0,4,16, 3,4,17, 3,1, 9
};
aabb3f box;
for(i = 0; i < 12; i++)
{
int edge_invisible;
if (nb[nb_triplet[i*3+2]])
edge_invisible = nb[nb_triplet[i*3]] & nb[nb_triplet[i*3+1]];
else
edge_invisible = nb[nb_triplet[i*3]] ^ nb[nb_triplet[i*3+1]];
if (edge_invisible)
continue;
box = frame_edges[i];
makeAutoLightedCuboid(&collector, data, pos, box, tiles[0], tile0color, frame);
}
for(i = 0; i < 6; i++)
{
if (!visible_faces[i])
continue;
box = glass_faces[i];
makeAutoLightedCuboid(&collector, data, pos, box, glass_tiles[i], glasscolor[i], frame);
}
if (param2 > 0 && f.special_tiles[0].texture) {
// Interior volume level is in range 0 .. 63,
// convert it to -0.5 .. 0.5
float vlev = (((float)param2 / 63.0 ) * 2.0 - 1.0);
TileSpec tile = getSpecialTile(f, n, 0);
video::SColor special_color = encode_light_and_color(l,
tile.color, f.light_source);
float offset = 0.003;
box = aabb3f(visible_faces[3] ? -b : -a + offset,
visible_faces[1] ? -b : -a + offset,
visible_faces[5] ? -b : -a + offset,
visible_faces[2] ? b : a - offset,
visible_faces[0] ? b * vlev : a * vlev - offset,
visible_faces[4] ? b : a - offset);
makeAutoLightedCuboid(&collector, data, pos, box, tile, special_color, frame);
}
break;}
case NDT_ALLFACES:
{
TileSpec tile_leaves = getNodeTile(n, p,
v3s16(0,0,0), data);
u16 l = getInteriorLight(n, 1, nodedef);
video::SColor c = encode_light_and_color(l,
tile_leaves.color, f.light_source);
v3f pos = intToFloat(p, BS);
aabb3f box(-BS/2,-BS/2,-BS/2,BS/2,BS/2,BS/2);
makeAutoLightedCuboid(&collector, data, pos, box, tile_leaves, c, frame);
break;}
case NDT_ALLFACES_OPTIONAL:
// This is always pre-converted to something else
FATAL_ERROR("NDT_ALLFACES_OPTIONAL not pre-converted");
break;
case NDT_TORCHLIKE:
{
v3s16 dir = n.getWallMountedDir(nodedef);
u8 tileindex = 0;
if(dir == v3s16(0,-1,0)){
tileindex = 0; // floor
} else if(dir == v3s16(0,1,0)){
tileindex = 1; // ceiling
// For backwards compatibility
} else if(dir == v3s16(0,0,0)){
tileindex = 0; // floor
} else {
tileindex = 2; // side
}
TileSpec tile = getNodeTileN(n, p, tileindex, data);
tile.material_flags &= ~MATERIAL_FLAG_BACKFACE_CULLING;
tile.material_flags |= MATERIAL_FLAG_CRACK_OVERLAY;
u16 l = getInteriorLight(n, 1, nodedef);
video::SColor c = encode_light_and_color(l, tile.color,
f.light_source);
float s = BS/2*f.visual_scale;
// Wall at X+ of node
video::S3DVertex vertices[4] =
{
video::S3DVertex(-s,-s,0, 0,0,0, c, 0,1),
video::S3DVertex( s,-s,0, 0,0,0, c, 1,1),
video::S3DVertex( s, s,0, 0,0,0, c, 1,0),
video::S3DVertex(-s, s,0, 0,0,0, c, 0,0),
};
for (s32 i = 0; i < 4; i++)
{
if(dir == v3s16(1,0,0))
vertices[i].Pos.rotateXZBy(0);
if(dir == v3s16(-1,0,0))
vertices[i].Pos.rotateXZBy(180);
if(dir == v3s16(0,0,1))
vertices[i].Pos.rotateXZBy(90);
if(dir == v3s16(0,0,-1))
vertices[i].Pos.rotateXZBy(-90);
if(dir == v3s16(0,-1,0))
vertices[i].Pos.rotateXZBy(45);
if(dir == v3s16(0,1,0))
vertices[i].Pos.rotateXZBy(-45);
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, tile.color);
vertices[i].Pos += intToFloat(p, BS);
}
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
collector.append(tile, vertices, 4, indices, 6);
break;}
case NDT_SIGNLIKE:
{
TileSpec tile = getNodeTileN(n, p, 0, data);
tile.material_flags &= ~MATERIAL_FLAG_BACKFACE_CULLING;
tile.material_flags |= MATERIAL_FLAG_CRACK_OVERLAY;
u16 l = getInteriorLight(n, 0, nodedef);
video::SColor c = encode_light_and_color(l, tile.color,
f.light_source);
float d = (float)BS/16;
float s = BS/2*f.visual_scale;
// Wall at X+ of node
video::S3DVertex vertices[4] =
{
video::S3DVertex(BS/2-d, s, s, 0,0,0, c, 0,0),
video::S3DVertex(BS/2-d, s, -s, 0,0,0, c, 1,0),
video::S3DVertex(BS/2-d, -s, -s, 0,0,0, c, 1,1),
video::S3DVertex(BS/2-d, -s, s, 0,0,0, c, 0,1),
};
v3s16 dir = n.getWallMountedDir(nodedef);
for (s32 i = 0; i < 4; i++)
{
if(dir == v3s16(1,0,0))
vertices[i].Pos.rotateXZBy(0);
if(dir == v3s16(-1,0,0))
vertices[i].Pos.rotateXZBy(180);
if(dir == v3s16(0,0,1))
vertices[i].Pos.rotateXZBy(90);
if(dir == v3s16(0,0,-1))
vertices[i].Pos.rotateXZBy(-90);
if(dir == v3s16(0,-1,0))
vertices[i].Pos.rotateXYBy(-90);
if(dir == v3s16(0,1,0))
vertices[i].Pos.rotateXYBy(90);
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, tile.color);
vertices[i].Pos += intToFloat(p, BS);
}
u16 indices[] = {0,1,2,2,3,0};
// Add to mesh collector
collector.append(tile, vertices, 4, indices, 6);
break;}
case NDT_PLANTLIKE:
{
PseudoRandom rng(x<<8 | z | y<<16);
TileSpec tile = getNodeTileN(n, p, 0, data);
tile.material_flags |= MATERIAL_FLAG_CRACK_OVERLAY;
u16 l = getInteriorLight(n, 1, nodedef);
video::SColor c = encode_light_and_color(l, tile.color,
f.light_source);
float s = BS / 2 * f.visual_scale;
// add sqrt(2) visual scale
if ((f.param_type_2 == CPT2_MESHOPTIONS) && ((n.param2 & 0x10) != 0))
s *= 1.41421;
float random_offset_X = .0;
float random_offset_Z = .0;
if ((f.param_type_2 == CPT2_MESHOPTIONS) && ((n.param2 & 0x8) != 0)) {
random_offset_X = BS * ((rng.next() % 16 / 16.0) * 0.29 - 0.145);
random_offset_Z = BS * ((rng.next() % 16 / 16.0) * 0.29 - 0.145);
}
for (int j = 0; j < 4; j++) {
video::S3DVertex vertices[4] =
{
video::S3DVertex(-s,-BS/2, 0, 0,0,0, c, 0,1),
video::S3DVertex( s,-BS/2, 0, 0,0,0, c, 1,1),
video::S3DVertex( s,-BS/2 + s*2,0, 0,0,0, c, 1,0),
video::S3DVertex(-s,-BS/2 + s*2,0, 0,0,0, c, 0,0),
};
float rotate_degree = 0;
u8 p2mesh = 0;
if (f.param_type_2 == CPT2_DEGROTATE)
rotate_degree = n.param2 * 2;
if (f.param_type_2 != CPT2_MESHOPTIONS) {
if (j == 0) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(46 + rotate_degree);
} else if (j == 1) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(-44 + rotate_degree);
}
} else {
p2mesh = n.param2 & 0x7;
switch (p2mesh) {
case 0:
// x
if (j == 0) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(46);
} else if (j == 1) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(-44);
}
break;
case 1:
// +
if (j == 0) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(91);
} else if (j == 1) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(1);
}
break;
case 2:
// *
if (j == 0) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(121);
} else if (j == 1) {
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(241);
} else { // (j == 2)
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(1);
}
break;
case 3:
// #
switch (j) {
case 0:
for (u16 i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(1);
vertices[i].Pos.Z += BS / 4;
}
break;
case 1:
for (u16 i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(91);
vertices[i].Pos.X += BS / 4;
}
break;
case 2:
for (u16 i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(181);
vertices[i].Pos.Z -= BS / 4;
}
break;
case 3:
for (u16 i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(271);
vertices[i].Pos.X -= BS / 4;
}
break;
}
break;
case 4:
// outward leaning #-like
switch (j) {
case 0:
for (u16 i = 2; i < 4; i++)
vertices[i].Pos.Z -= BS / 2;
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(1);
break;
case 1:
for (u16 i = 2; i < 4; i++)
vertices[i].Pos.Z -= BS / 2;
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(91);
break;
case 2:
for (u16 i = 2; i < 4; i++)
vertices[i].Pos.Z -= BS / 2;
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(181);
break;
case 3:
for (u16 i = 2; i < 4; i++)
vertices[i].Pos.Z -= BS / 2;
for (u16 i = 0; i < 4; i++)
vertices[i].Pos.rotateXZBy(271);
break;
}
break;
}
}
for (int i = 0; i < 4; i++) {
vertices[i].Pos *= f.visual_scale;
vertices[i].Pos.Y += BS/2 * (f.visual_scale - 1);
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, tile.color);
vertices[i].Pos += intToFloat(p, BS);
// move to a random spot to avoid moire
if ((f.param_type_2 == CPT2_MESHOPTIONS) && ((n.param2 & 0x8) != 0)) {
vertices[i].Pos.X += random_offset_X;
vertices[i].Pos.Z += random_offset_Z;
}
// randomly move each face up/down
if ((f.param_type_2 == CPT2_MESHOPTIONS) && ((n.param2 & 0x20) != 0)) {
PseudoRandom yrng(j | x<<16 | z<<8 | y<<24 );
vertices[i].Pos.Y -= BS * ((yrng.next() % 16 / 16.0) * 0.125);
}
}
u16 indices[] = {0, 1, 2, 2, 3, 0};
// Add to mesh collector
collector.append(tile, vertices, 4, indices, 6);
// stop adding faces for meshes with less than 4 faces
if (f.param_type_2 == CPT2_MESHOPTIONS) {
if (((p2mesh == 0) || (p2mesh == 1)) && (j == 1))
break;
else if ((p2mesh == 2) && (j == 2))
break;
} else if (j == 1) {
break;
}
}
break;}
case NDT_FIRELIKE:
{
TileSpec tile = getNodeTileN(n, p, 0, data);
tile.material_flags |= MATERIAL_FLAG_CRACK_OVERLAY;
u16 l = getInteriorLight(n, 1, nodedef);
video::SColor c = encode_light_and_color(l, tile.color,
f.light_source);
float s = BS / 2 * f.visual_scale;
content_t current = n.getContent();
content_t n2c;
MapNode n2;
v3s16 n2p;
static const v3s16 dirs[6] = {
v3s16( 0, 1, 0),
v3s16( 0, -1, 0),
v3s16( 1, 0, 0),
v3s16(-1, 0, 0),
v3s16( 0, 0, 1),
v3s16( 0, 0, -1)
};
int doDraw[6] = {0, 0, 0, 0, 0, 0};
bool drawAllFaces = true;
// Check for adjacent nodes
for (int i = 0; i < 6; i++) {
n2p = blockpos_nodes + p + dirs[i];
n2 = data->m_vmanip.getNodeNoEx(n2p);
n2c = n2.getContent();
if (n2c != CONTENT_IGNORE && n2c != CONTENT_AIR && n2c != current) {
doDraw[i] = 1;
if (drawAllFaces)
drawAllFaces = false;
}
}
for (int j = 0; j < 6; j++) {
video::S3DVertex vertices[4] = {
video::S3DVertex(-s, -BS / 2, 0, 0, 0, 0, c, 0, 1),
video::S3DVertex( s, -BS / 2, 0, 0, 0, 0, c, 1, 1),
video::S3DVertex( s, -BS / 2 + s * 2, 0, 0, 0, 0, c, 1, 0),
video::S3DVertex(-s, -BS / 2 + s * 2, 0, 0, 0, 0, c, 0, 0),
};
// Calculate which faces should be drawn, (top or sides)
if (j == 0 && (drawAllFaces ||
(doDraw[3] == 1 || doDraw[1] == 1))) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(90);
vertices[i].Pos.rotateXYBy(-10);
vertices[i].Pos.X -= 4.0;
}
} else if (j == 1 && (drawAllFaces ||
(doDraw[5] == 1 || doDraw[1] == 1))) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(180);
vertices[i].Pos.rotateYZBy(10);
vertices[i].Pos.Z -= 4.0;
}
} else if (j == 2 && (drawAllFaces ||
(doDraw[2] == 1 || doDraw[1] == 1))) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(270);
vertices[i].Pos.rotateXYBy(10);
vertices[i].Pos.X += 4.0;
}
} else if (j == 3 && (drawAllFaces ||
(doDraw[4] == 1 || doDraw[1] == 1))) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateYZBy(-10);
vertices[i].Pos.Z += 4.0;
}
// Center cross-flames
} else if (j == 4 && (drawAllFaces || doDraw[1] == 1)) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(45);
}
} else if (j == 5 && (drawAllFaces || doDraw[1] == 1)) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateXZBy(-45);
}
// Render flames on bottom of node above
} else if (j == 0 && doDraw[0] == 1 && doDraw[1] == 0) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateYZBy(70);
vertices[i].Pos.rotateXZBy(90);
vertices[i].Pos.Y += 4.84;
vertices[i].Pos.X -= 4.7;
}
} else if (j == 1 && doDraw[0] == 1 && doDraw[1] == 0) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateYZBy(70);
vertices[i].Pos.rotateXZBy(180);
vertices[i].Pos.Y += 4.84;
vertices[i].Pos.Z -= 4.7;
}
} else if (j == 2 && doDraw[0] == 1 && doDraw[1] == 0) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateYZBy(70);
vertices[i].Pos.rotateXZBy(270);
vertices[i].Pos.Y += 4.84;
vertices[i].Pos.X += 4.7;
}
} else if (j == 3 && doDraw[0] == 1 && doDraw[1] == 0) {
for (int i = 0; i < 4; i++) {
vertices[i].Pos.rotateYZBy(70);
vertices[i].Pos.Y += 4.84;
vertices[i].Pos.Z += 4.7;
}
} else {
// Skip faces that aren't adjacent to a node
continue;
}
for (int i = 0; i < 4; i++) {
vertices[i].Pos *= f.visual_scale;
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, tile.color);
vertices[i].Pos += intToFloat(p, BS);
}
u16 indices[] = {0, 1, 2, 2, 3, 0};
// Add to mesh collector
collector.append(tile, vertices, 4, indices, 6);
}
break;}
case NDT_FENCELIKE:
{
TileSpec tile = getNodeTile(n, p, v3s16(0,0,0), data);
TileSpec tile_nocrack = tile;
tile_nocrack.material_flags &= ~MATERIAL_FLAG_CRACK;
// Put wood the right way around in the posts
TileSpec tile_rot = tile;
tile_rot.rotation = 1;
u16 l = getInteriorLight(n, 1, nodedef);
video::SColor c = encode_light_and_color(l, tile.color,
f.light_source);
const f32 post_rad=(f32)BS/8;
const f32 bar_rad=(f32)BS/16;
const f32 bar_len=(f32)(BS/2)-post_rad;
v3f pos = intToFloat(p, BS);
// The post - always present
aabb3f post(-post_rad,-BS/2,-post_rad,post_rad,BS/2,post_rad);
f32 postuv[24]={
6/16.,6/16.,10/16.,10/16.,
6/16.,6/16.,10/16.,10/16.,
0/16.,0,4/16.,1,
4/16.,0,8/16.,1,
8/16.,0,12/16.,1,
12/16.,0,16/16.,1};
makeAutoLightedCuboidEx(&collector, data, pos, post, tile_rot, postuv, c, frame);
// Now a section of fence, +X, if there's a post there
v3s16 p2 = p;
p2.X++;
MapNode n2 = data->m_vmanip.getNodeNoEx(blockpos_nodes + p2);
const ContentFeatures *f2 = &nodedef->get(n2);
if(f2->drawtype == NDT_FENCELIKE)
{
aabb3f bar(-bar_len+BS/2,-bar_rad+BS/4,-bar_rad,
bar_len+BS/2,bar_rad+BS/4,bar_rad);
f32 xrailuv[24]={
0/16.,2/16.,16/16.,4/16.,
0/16.,4/16.,16/16.,6/16.,
6/16.,6/16.,8/16.,8/16.,
10/16.,10/16.,12/16.,12/16.,
0/16.,8/16.,16/16.,10/16.,
0/16.,14/16.,16/16.,16/16.};
makeAutoLightedCuboidEx(&collector, data, pos, bar, tile_nocrack, xrailuv, c, frame);
bar.MinEdge.Y -= BS/2;
bar.MaxEdge.Y -= BS/2;
makeAutoLightedCuboidEx(&collector, data, pos, bar, tile_nocrack, xrailuv, c, frame);
}
// Now a section of fence, +Z, if there's a post there
p2 = p;
p2.Z++;
n2 = data->m_vmanip.getNodeNoEx(blockpos_nodes + p2);
f2 = &nodedef->get(n2);
if(f2->drawtype == NDT_FENCELIKE)
{
aabb3f bar(-bar_rad,-bar_rad+BS/4,-bar_len+BS/2,
bar_rad,bar_rad+BS/4,bar_len+BS/2);
f32 zrailuv[24]={
3/16.,1/16.,5/16.,5/16., // cannot rotate; stretch
4/16.,1/16.,6/16.,5/16., // for wood texture instead
0/16.,9/16.,16/16.,11/16.,
0/16.,6/16.,16/16.,8/16.,
6/16.,6/16.,8/16.,8/16.,
10/16.,10/16.,12/16.,12/16.};
makeAutoLightedCuboidEx(&collector, data, pos, bar, tile_nocrack, zrailuv, c, frame);
bar.MinEdge.Y -= BS/2;
bar.MaxEdge.Y -= BS/2;
makeAutoLightedCuboidEx(&collector, data, pos, bar, tile_nocrack, zrailuv, c, frame);
}
break;}
case NDT_RAILLIKE:
{
bool is_rail_x[6]; /* (-1,-1,0) X (1,-1,0) (-1,0,0) X (1,0,0) (-1,1,0) X (1,1,0) */
bool is_rail_z[6];
content_t thiscontent = n.getContent();
std::string groupname = "connect_to_raillike"; // name of the group that enables connecting to raillike nodes of different kind
int self_group = ((ItemGroupList) nodedef->get(n).groups)[groupname];
u8 index = 0;
for (s8 y0 = -1; y0 <= 1; y0++) {
// Prevent from indexing never used coordinates
for (s8 xz = -1; xz <= 1; xz++) {
if (xz == 0)
continue;
MapNode n_xy = data->m_vmanip.getNodeNoEx(blockpos_nodes + v3s16(x + xz, y + y0, z));
MapNode n_zy = data->m_vmanip.getNodeNoEx(blockpos_nodes + v3s16(x, y + y0, z + xz));
const ContentFeatures &def_xy = nodedef->get(n_xy);
const ContentFeatures &def_zy = nodedef->get(n_zy);
// Check if current node would connect with the rail
is_rail_x[index] = ((def_xy.drawtype == NDT_RAILLIKE
&& ((ItemGroupList) def_xy.groups)[groupname] == self_group)
|| n_xy.getContent() == thiscontent);
is_rail_z[index] = ((def_zy.drawtype == NDT_RAILLIKE
&& ((ItemGroupList) def_zy.groups)[groupname] == self_group)
|| n_zy.getContent() == thiscontent);
index++;
}
}
bool is_rail_x_all[2]; // [0] = negative x, [1] = positive x coordinate from the current node position
bool is_rail_z_all[2];
is_rail_x_all[0] = is_rail_x[0] || is_rail_x[2] || is_rail_x[4];
is_rail_x_all[1] = is_rail_x[1] || is_rail_x[3] || is_rail_x[5];
is_rail_z_all[0] = is_rail_z[0] || is_rail_z[2] || is_rail_z[4];
is_rail_z_all[1] = is_rail_z[1] || is_rail_z[3] || is_rail_z[5];
// reasonable default, flat straight unrotated rail
bool is_straight = true;
int adjacencies = 0;
int angle = 0;
u8 tileindex = 0;
// check for sloped rail
if (is_rail_x[4] || is_rail_x[5] || is_rail_z[4] || is_rail_z[5]) {
adjacencies = 5; // 5 means sloped
is_straight = true; // sloped is always straight
} else {
// is really straight, rails on both sides
is_straight = (is_rail_x_all[0] && is_rail_x_all[1]) || (is_rail_z_all[0] && is_rail_z_all[1]);
adjacencies = is_rail_x_all[0] + is_rail_x_all[1] + is_rail_z_all[0] + is_rail_z_all[1];
}
switch (adjacencies) {
case 1:
if (is_rail_x_all[0] || is_rail_x_all[1])
angle = 90;
break;
case 2:
if (!is_straight)
tileindex = 1; // curved
if (is_rail_x_all[0] && is_rail_x_all[1])
angle = 90;
if (is_rail_z_all[0] && is_rail_z_all[1]) {
if (is_rail_z[4])
angle = 180;
}
else if (is_rail_x_all[0] && is_rail_z_all[0])
angle = 270;
else if (is_rail_x_all[0] && is_rail_z_all[1])
angle = 180;
else if (is_rail_x_all[1] && is_rail_z_all[1])
angle = 90;
break;
case 3:
// here is where the potential to 'switch' a junction is, but not implemented at present
tileindex = 2; // t-junction
if(!is_rail_x_all[1])
angle = 180;
if(!is_rail_z_all[0])
angle = 90;
if(!is_rail_z_all[1])
angle = 270;
break;
case 4:
tileindex = 3; // crossing
break;
case 5: //sloped
if (is_rail_z[4])
angle = 180;
if (is_rail_x[4])
angle = 90;
if (is_rail_x[5])
angle = -90;
break;
default:
break;
}
TileSpec tile = getNodeTileN(n, p, tileindex, data);
tile.material_flags &= ~MATERIAL_FLAG_BACKFACE_CULLING;
tile.material_flags |= MATERIAL_FLAG_CRACK_OVERLAY;
u16 l = getInteriorLight(n, 0, nodedef);
video::SColor c = encode_light_and_color(l, tile.color,
f.light_source);
float d = (float)BS/64;
float s = BS/2;
short g = -1;
if (is_rail_x[4] || is_rail_x[5] || is_rail_z[4] || is_rail_z[5])
g = 1; //Object is at a slope
video::S3DVertex vertices[4] =
{
video::S3DVertex(-s, -s+d, -s, 0, 0, 0, c, 0, 1),
video::S3DVertex( s, -s+d, -s, 0, 0, 0, c, 1, 1),
video::S3DVertex( s, g*s+d, s, 0, 0, 0, c, 1, 0),
video::S3DVertex(-s, g*s+d, s, 0, 0, 0, c, 0, 0),
};
for(s32 i=0; i<4; i++)
{
if(angle != 0)
vertices[i].Pos.rotateXZBy(angle);
if (data->m_smooth_lighting)
vertices[i].Color = blendLight(frame, vertices[i].Pos, tile.color);
vertices[i].Pos += intToFloat(p, BS);
}
u16 indices[] = {0,1,2,2,3,0};
collector.append(tile, vertices, 4, indices, 6);
break;}
case NDT_NODEBOX:
{
static const v3s16 tile_dirs[6] = {
v3s16(0, 1, 0),
v3s16(0, -1, 0),
v3s16(1, 0, 0),
v3s16(-1, 0, 0),
v3s16(0, 0, 1),
v3s16(0, 0, -1)
};
TileSpec tiles[6];
video::SColor colors[6];
for (int j = 0; j < 6; j++) {
// Handles facedir rotation for textures
tiles[j] = getNodeTile(n, p, tile_dirs[j], data);
}
if (!data->m_smooth_lighting) {
u16 l = getInteriorLight(n, 1, nodedef);
for (int j = 0; j < 6; j++)
colors[j] = encode_light_and_color(l, tiles[j].color, f.light_source);
}
v3f pos = intToFloat(p, BS);
int neighbors = 0;
// locate possible neighboring nodes to connect to
if (f.node_box.type == NODEBOX_CONNECTED) {
v3s16 p2 = p;
p2.Y++;
getNeighborConnectingFace(blockpos_nodes + p2, nodedef, data, n, 1, &neighbors);
p2 = p;
p2.Y--;
getNeighborConnectingFace(blockpos_nodes + p2, nodedef, data, n, 2, &neighbors);
p2 = p;
p2.Z--;
getNeighborConnectingFace(blockpos_nodes + p2, nodedef, data, n, 4, &neighbors);
p2 = p;
p2.X--;
getNeighborConnectingFace(blockpos_nodes + p2, nodedef, data, n, 8, &neighbors);
p2 = p;
p2.Z++;
getNeighborConnectingFace(blockpos_nodes + p2, nodedef, data, n, 16, &neighbors);
p2 = p;
p2.X++;
getNeighborConnectingFace(blockpos_nodes + p2, nodedef, data, n, 32, &neighbors);
}
std::vector<aabb3f> boxes;
n.getNodeBoxes(nodedef, &boxes, neighbors);
for (std::vector<aabb3f>::iterator
i = boxes.begin();
i != boxes.end(); ++i) {
aabb3f box = *i;
f32 dx1 = box.MinEdge.X;
f32 dy1 = box.MinEdge.Y;
f32 dz1 = box.MinEdge.Z;
f32 dx2 = box.MaxEdge.X;
f32 dy2 = box.MaxEdge.Y;
f32 dz2 = box.MaxEdge.Z;
box.MinEdge += pos;
box.MaxEdge += pos;
if (box.MinEdge.X > box.MaxEdge.X)
std::swap(box.MinEdge.X, box.MaxEdge.X);
if (box.MinEdge.Y > box.MaxEdge.Y)
std::swap(box.MinEdge.Y, box.MaxEdge.Y);
if (box.MinEdge.Z > box.MaxEdge.Z)
std::swap(box.MinEdge.Z, box.MaxEdge.Z);
//
// Compute texture coords
f32 tx1 = (box.MinEdge.X/BS)+0.5;
f32 ty1 = (box.MinEdge.Y/BS)+0.5;
f32 tz1 = (box.MinEdge.Z/BS)+0.5;
f32 tx2 = (box.MaxEdge.X/BS)+0.5;
f32 ty2 = (box.MaxEdge.Y/BS)+0.5;
f32 tz2 = (box.MaxEdge.Z/BS)+0.5;
f32 txc[24] = {
// up
tx1, 1-tz2, tx2, 1-tz1,
// down
tx1, tz1, tx2, tz2,
// right
tz1, 1-ty2, tz2, 1-ty1,
// left
1-tz2, 1-ty2, 1-tz1, 1-ty1,
// back
1-tx2, 1-ty2, 1-tx1, 1-ty1,
// front
tx1, 1-ty2, tx2, 1-ty1,
};
if (data->m_smooth_lighting) {
u16 lights[8];
for (int j = 0; j < 8; ++j) {
f32 x = (j & 4) ? dx2 : dx1;
f32 y = (j & 2) ? dy2 : dy1;
f32 z = (j & 1) ? dz2 : dz1;
lights[j] = blendLight(frame, core::vector3df(x, y, z));
}
makeSmoothLightedCuboid(&collector, box, tiles, 6, lights, txc, f.light_source);
} else {
makeCuboid(&collector, box, tiles, 6, colors, txc, f.light_source);
}
}
break;}
case NDT_MESH:
{
v3f pos = intToFloat(p, BS);
u16 l = getInteriorLight(n, 1, nodedef);
u8 facedir = 0;
if (f.param_type_2 == CPT2_FACEDIR ||
f.param_type_2 == CPT2_COLORED_FACEDIR) {
facedir = n.getFaceDir(nodedef);
} else if (f.param_type_2 == CPT2_WALLMOUNTED ||
f.param_type_2 == CPT2_COLORED_WALLMOUNTED) {
//convert wallmounted to 6dfacedir.
//when cache enabled, it is already converted
facedir = n.getWallMounted(nodedef);
if (!enable_mesh_cache) {
static const u8 wm_to_6d[6] = {20, 0, 16+1, 12+3, 8, 4+2};
facedir = wm_to_6d[facedir];
}
}
if (!data->m_smooth_lighting && f.mesh_ptr[facedir]) {
// use cached meshes
for (u16 j = 0; j < f.mesh_ptr[0]->getMeshBufferCount(); j++) {
const TileSpec &tile = getNodeTileN(n, p, j, data);
scene::IMeshBuffer *buf = f.mesh_ptr[facedir]->getMeshBuffer(j);
collector.append(tile, (video::S3DVertex *)
buf->getVertices(), buf->getVertexCount(),
buf->getIndices(), buf->getIndexCount(), pos,
encode_light_and_color(l, tile.color, f.light_source),
f.light_source);
}
} else if (f.mesh_ptr[0]) {
// no cache, clone and rotate mesh
scene::IMesh* mesh = cloneMesh(f.mesh_ptr[0]);
rotateMeshBy6dFacedir(mesh, facedir);
recalculateBoundingBox(mesh);
meshmanip->recalculateNormals(mesh, true, false);
for (u16 j = 0; j < mesh->getMeshBufferCount(); j++) {
const TileSpec &tile = getNodeTileN(n, p, j, data);
scene::IMeshBuffer *buf = mesh->getMeshBuffer(j);
video::S3DVertex *vertices = (video::S3DVertex *)buf->getVertices();
u32 vertex_count = buf->getVertexCount();
if (data->m_smooth_lighting) {
for (u16 m = 0; m < vertex_count; ++m) {
video::S3DVertex &vertex = vertices[m];
vertex.Color = blendLight(frame, vertex.Pos, vertex.Normal, tile.color);
vertex.Pos += pos;
}
collector.append(tile, vertices, vertex_count,
buf->getIndices(), buf->getIndexCount());
} else {
collector.append(tile, vertices, vertex_count,
buf->getIndices(), buf->getIndexCount(), pos,
encode_light_and_color(l, tile.color, f.light_source),
f.light_source);
}
}
mesh->drop();
}
break;}
}
}
}