1
0
mirror of https://github.com/zenorogue/hyperrogue.git synced 2024-11-16 10:14:48 +00:00
hyperrogue/blizzard.cpp
2017-10-30 23:57:16 +01:00

237 lines
6.8 KiB
C++

double randd() { return (rand() % 1000000) / 1000000. + .0000005; }
double cellgfxdist(cell *c, int i) {
return nontruncated ? tessf : (c->type == 6 && (i&1)) ? hexhexdist : crossf;
}
transmatrix cellrelmatrix(cell *c, int i) {
double d = cellgfxdist(c, i);
return ddspin(c, i) * xpush(d) * iddspin(c->mov[i], c->spin(i), euclid ? 0 : S42);
}
hyperpoint randomPointIn(int t) {
while(true) {
hyperpoint h = spin(2*M_PI*(randd()-.5)/t) * tC0(xpush(asinh(randd())));
double d =
nontruncated ? tessf : t == 6 ? hexhexdist : crossf;
if(hdist0(h) < hdist0(xpush(-d) * h))
return spin(2*M_PI/t * (rand() % t)) * h;
}
}
struct snowball {
transmatrix T;
transmatrix global;
snowball *prev;
snowball *next;
double phase;
snowball(int t) { T = rgpushxto0(randomPointIn(t)); phase = randd(); }
};
struct blizzardcell {
cell *c;
int frame;
int tmp;
transmatrix *gm;
char wmap;
int inward, outward, ward;
int qty[MAX_EDGE];
vector<snowball*> inorder, outorder;
int inid, outid;
~blizzardcell() { for(auto i: inorder) delete i; }
};
map<cell*, blizzardcell> blizzardcells;
vector<blizzardcell*> bcells;
int N;
blizzardcell* getbcell(cell *c) {
int i = c->aitmp;
if(i<0 || i >= N) return NULL;
if(bcells[i]->c != c) return NULL;
return bcells[i];
}
void drawBlizzards() {
poly_outline = OUTLINE_NONE;
auto it = blizzardcells.begin();
bcells.clear();
while(it != blizzardcells.end())
if(it->second.frame != frameid || !gmatrix.count(it->first))
it = blizzardcells.erase(it);
else {
it->second.c = it->first;
bcells.push_back(&it->second);
it++;
}
N = size(bcells);
for(int i=0; i<N; i++) {
auto& bc = *bcells[i];
bc.tmp = bc.c->aitmp,
bc.c->aitmp = i;
bc.gm = &gmatrix[bc.c];
bc.wmap = windmap::at(bc.c);
}
for(int i=0; i<N; i++) {
auto& bc = *bcells[i];
cell *c = bc.c;
bc.inward = bc.outward = 0;
for(int i=0; i<c->type; i++) {
int& qty = bc.qty[i];
qty = 0;
cell *c2 = c->mov[i];
if(!c2) continue;
auto bc2 = getbcell(c2);
if(!bc2) continue;
int z = (bc2->wmap - bc.wmap) & 255;
if(z >= windmap::NOWINDBELOW && z < windmap::NOWINDFROM)
bc.outward += qty = z / 8;
z = (-z) & 255;
if(z >= windmap::NOWINDBELOW && z < windmap::NOWINDFROM)
bc.inward += z / 8, qty = -z/8;
}
bc.ward = max(bc.inward, bc.outward);
while(size(bc.inorder) < bc.ward) {
auto sb = new snowball(c->type);
bc.inorder.push_back(sb);
bc.outorder.push_back(sb);
}
for(auto& sb: bc.inorder) sb->prev = sb->next = NULL;
bc.inid = 0;
}
double at = (ticks % 250) / 250.0;
for(int i=0; i<N; i++) {
auto& bc = *bcells[i];
for(auto sb: bc.inorder)
sb->global = (*bc.gm) * sb->T;
}
for(int i=0; i<N; i++) {
auto& bc = *bcells[i];
cell *c = bc.c;
bc.outid = 0;
for(int d=0; d<c->type; d++) for(int k=0; k<bc.qty[d]; k++) {
auto& bc2 = *getbcell(c->mov[d]);
auto& sball = *bc.outorder[bc.outid++];
auto& sball2 = *bc2.inorder[bc2.inid++];
sball.next = &sball2;
sball2.prev = &sball;
hyperpoint t = inverse(sball.global) * tC0(sball2.global);
double at0 = at + sball.phase;
if(at0>1) at0 -= 1;
transmatrix tpartial = sball.global * rspintox(t) * xpush(hdist0(t) * at0);
if(wmascii || wmblack)
queuechr(tpartial, .2, '.', 0xFFFFFF);
else
queuepoly(tpartial, shSnowball, 0xFFFFFF80);
}
}
for(int ii=0; ii<N; ii++) {
auto& bc = *bcells[ii];
if(isNeighbor(bc.c, mouseover)) {
if(againstWind(mouseover, bc.c))
queuepoly(*bc.gm, shHeptaMarker, 0x00C00040);
if(againstWind(bc.c, mouseover))
queuepoly(*bc.gm, shHeptaMarker, 0xC0000040);
}
int B = size(bc.outorder);
if(B<2) continue;
int i = rand() % B;
int j = rand() % (B-1);
if(i==j) j++;
if(1) {
auto& sb1 = *bc.outorder[i];
auto& sb2 = *bc.outorder[j];
double swapcost = 0;
if(sb1.next) swapcost -= hdist(tC0(sb1.global), tC0(sb1.next->global));
if(sb2.next) swapcost -= hdist(tC0(sb2.global), tC0(sb2.next->global));
if(sb1.next) swapcost += hdist(tC0(sb2.global), tC0(sb1.next->global));
if(sb2.next) swapcost += hdist(tC0(sb1.global), tC0(sb2.next->global));
if(swapcost < 0) {
swap(bc.outorder[i], bc.outorder[j]);
swap(sb1.next, sb2.next);
if(sb1.next) sb1.next->prev = &sb1;
if(sb2.next) sb2.next->prev = &sb2;
}
}
if(1) {
auto& sb1 = *bc.inorder[i];
auto& sb2 = *bc.inorder[j];
double swapcost = 0;
if(sb1.prev) swapcost -= hdist(tC0(sb1.global), tC0(sb1.prev->global));
if(sb2.prev) swapcost -= hdist(tC0(sb2.global), tC0(sb2.prev->global));
if(sb1.prev) swapcost += hdist(tC0(sb2.global), tC0(sb1.prev->global));
if(sb2.prev) swapcost += hdist(tC0(sb1.global), tC0(sb2.prev->global));
if(swapcost < 0) {
swap(bc.inorder[i], bc.inorder[j]);
swap(sb1.prev, sb2.prev);
if(sb1.prev) sb1.prev->next = &sb1;
if(sb2.prev) sb2.prev->next = &sb2;
}
}
auto& sbp = *bc.inorder[i];
if(sbp.next && sbp.prev) {
double p1 = sbp.next->phase;
double p2 = sbp.prev->phase;
double d = p2-p1;
if(d<=.5) d+=1;
if(d>=.5) d-=1;
sbp.phase = p1 + d/2;
if(sbp.phase >= 1) sbp.phase -= 1;
if(sbp.phase < 0) sbp.phase += 1;
}
}
for(auto bc: bcells)
bc->c->aitmp = bc->tmp;
}
vector<cell*> arrowtraps;
void drawArrowTraps() {
for(cell *c: arrowtraps) {
auto r = traplimits(c);
try {
transmatrix& t0 = gmatrix.at(r[0]);
transmatrix& t1 = gmatrix.at(r[4]);
queueline(tC0(t0), tC0(t1), 0xFF0000FF, 4, PPR_ITEM);
if((c->wparam & 7) == 3 && !shmup::on) {
// queueline(t0 * randomPointIn(r[0]->type), t1 * randomPointIn(r[1]->type), 0xFFFFFFFF, 4, PPR_ITEM);
int tt = ticks % 401;
for(int u=0; u<2; u++) {
transmatrix& tu = u ? t0 : t1;
transmatrix& tv = u ? t1 : t0;
hyperpoint trel = inverse(tu) * tC0(tv);
transmatrix tpartial = tu * rspintox(trel) * xpush(hdist0(trel) * tt / 401.0);
queuepoly(tpartial * ypush(.05), shTrapArrow, 0xFFFFFFFF);
}
}
}
catch(out_of_range) {}
}
}
auto ccm_blizzard = addHook(clearmemory, 0, [] () {
arrowtraps.clear();
blizzardcells.clear();
bcells.clear();
});