1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
//! A uniform grid over the XY ground plane, for nearest-thing-under-a-point
//! queries.
//!
//! Both hot lookups on a city map are the same shape: given `(x, y)`, find the
//! nearest of thousands of things that each occupy a small patch of ground.
//! Scanning all of them is O(map) per query, and a consumer runs these every
//! tick for every body, so the cost scales with map size times fleet size.
//!
//! The grid buckets items by their XY bounding box, then answers a query by
//! walking cells outward in rings and stopping as soon as the best candidate
//! found is closer than anything the next ring could hold. Items are bucketed
//! by bounding box, which is a superset of their geometry, so nothing that
//! could win is ever skipped.
/// An axis-aligned XY footprint.
#[derive(Debug, Clone, Copy)]
pub(crate) struct Aabb {
pub min_x: f32,
pub max_x: f32,
pub min_y: f32,
pub max_y: f32,
}
impl Aabb {
/// The box around a set of XY points, or `None` if there are none.
pub fn around(points: impl IntoIterator<Item = (f32, f32)>) -> Option<Self> {
points.into_iter().fold(None, |acc: Option<Self>, (x, y)| {
Some(match acc {
None => Self {
min_x: x,
max_x: x,
min_y: y,
max_y: y,
},
Some(b) => Self {
min_x: b.min_x.min(x),
max_x: b.max_x.max(x),
min_y: b.min_y.min(y),
max_y: b.max_y.max(y),
},
})
})
}
/// This box grown by `r` on every side.
pub fn padded(self, r: f32) -> Self {
Self {
min_x: self.min_x - r,
max_x: self.max_x + r,
min_y: self.min_y - r,
max_y: self.max_y + r,
}
}
fn merge(self, other: Self) -> Self {
Self {
min_x: self.min_x.min(other.min_x),
max_x: self.max_x.max(other.max_x),
min_y: self.min_y.min(other.min_y),
max_y: self.max_y.max(other.max_y),
}
}
/// Squared distance from `(x, y)` to this box; zero inside it. A lower
/// bound on the distance to whatever the box contains, so a candidate
/// whose box is already further than the best hit can be rejected without
/// touching its geometry.
pub fn dist2(&self, x: f32, y: f32) -> f32 {
let dx = (self.min_x - x).max(0.0).max(x - self.max_x);
let dy = (self.min_y - y).max(0.0).max(y - self.max_y);
dx * dx + dy * dy
}
}
/// Item indices bucketed into square cells.
#[derive(Debug, Clone, Default)]
pub(crate) struct Grid {
min_x: f32,
min_y: f32,
/// The far corner of the items' extent. `cols * cell` can round to just
/// short of it.
max_x: f32,
max_y: f32,
cell: f32,
cols: usize,
rows: usize,
/// Row-major `rows * cols` buckets of indices into the caller's item list.
cells: Vec<Vec<u32>>,
}
/// Cap on a grid side, so a map with one enormous item can't allocate a
/// pathological number of cells.
const MAX_DIM: usize = 1024;
/// Items per cell to aim for. Low enough that a ring holds few candidates,
/// high enough that the ring walk isn't all bookkeeping.
const ITEMS_PER_CELL: f32 = 2.0;
impl Grid {
/// Bucket `bounds` (one box per item, indexed as given). An empty list
/// yields an empty grid, which answers every query with `None`.
pub fn build(bounds: &[Aabb]) -> Self {
let Some(extent) = bounds.iter().copied().reduce(Aabb::merge) else {
return Self::default();
};
let (w, h) = (extent.max_x - extent.min_x, extent.max_y - extent.min_y);
let target = (bounds.len() as f32 / ITEMS_PER_CELL).max(1.0);
// Area per cell, floored so a perfectly straight road (zero extent on
// one axis) still produces a usable cell size.
let area = w.max(1e-3) * h.max(1e-3);
let cell = (area / target)
.sqrt()
.max(w.max(h) / MAX_DIM as f32)
.max(1e-3);
let cols = ((w / cell).ceil() as usize).clamp(1, MAX_DIM);
let rows = ((h / cell).ceil() as usize).clamp(1, MAX_DIM);
// Re-derive the cell size from the clamped dimensions, so the grid is
// guaranteed to span the whole extent.
let cell = (w / cols as f32).max(h / rows as f32).max(1e-3);
let mut grid = Self {
min_x: extent.min_x,
min_y: extent.min_y,
max_x: extent.max_x,
max_y: extent.max_y,
cell,
cols,
rows,
cells: vec![Vec::new(); cols * rows],
};
for (i, b) in bounds.iter().enumerate() {
let (lo_x, hi_x) = (grid.col(b.min_x), grid.col(b.max_x));
let (lo_y, hi_y) = (grid.row(b.min_y), grid.row(b.max_y));
for iy in lo_y..=hi_y {
for ix in lo_x..=hi_x {
grid.cells[iy * cols + ix].push(i as u32);
}
}
}
grid
}
fn col(&self, x: f32) -> usize {
Self::bucket(x - self.min_x, self.cell, self.cols)
}
fn row(&self, y: f32) -> usize {
Self::bucket(y - self.min_y, self.cell, self.rows)
}
fn bucket(offset: f32, cell: f32, count: usize) -> usize {
// A NaN query coordinate lands in cell 0 rather than panicking: it can
// only fail to find anything, which is what a NaN deserves.
let i = (offset / cell).floor();
if i.is_nan() || i < 0.0 {
0
} else {
(i as usize).min(count - 1)
}
}
/// The items whose footprint could contain `(x, y)`. Everything that does
/// contain it is here; some of what is here does not. Empty when the point
/// falls outside the grid's extent, or the grid holds nothing.
pub fn at(&self, x: f32, y: f32) -> &[u32] {
if self.cells.is_empty() || !self.covers(x, y) {
return &[];
}
&self.cells[self.row(y) * self.cols + self.col(x)]
}
/// Whether `(x, y)` is inside the grid's extent. Outside it, `col`/`row`
/// clamp to an edge cell, which is right for a nearest-item walk and wrong
/// for a containment lookup.
fn covers(&self, x: f32, y: f32) -> bool {
(self.min_x..=self.max_x).contains(&x) && (self.min_y..=self.max_y).contains(&y)
}
/// The nearest item to `(x, y)`, by whatever `consider` measures. Exact
/// distance ties go to the lowest item index, so the answer matches a
/// linear scan of the item list and does not depend on the grid's layout.
///
/// `consider` is called with an item index and the best squared distance
/// found so far, and returns that item's squared distance and result, or
/// `None` to reject it. Passing the running best lets a caller cheaply
/// discard an item that cannot win before evaluating its geometry. That
/// is also what keeps a large item appearing in many cells from being
/// re-measured in each one. It is an upper bound to beat, not to match:
/// an item exactly at that distance must still be offered, or the tie
/// break cannot see it.
///
/// `None` for a point that isn't finite. An item whose distance is NaN,
/// such as one with a NaN vertex, never wins.
pub fn nearest<T>(
&self,
x: f32,
y: f32,
mut consider: impl FnMut(u32, f32) -> Option<(f32, T)>,
) -> Option<T> {
if self.cells.is_empty() || !x.is_finite() || !y.is_finite() {
return None;
}
let (cx, cy) = (self.col(x) as isize, self.row(y) as isize);
let mut best: Option<(f32, u32, T)> = None;
for r in 0..=self.cols.max(self.rows) as isize {
// Cells in ring `r` or beyond sit at least `(r - 1)` whole cells
// away from anywhere inside the query's own cell. Once the best
// hit is nearer than that, no further ring can improve on it.
if let Some((d2, _, _)) = &best {
let floor = (r - 1).max(0) as f32 * self.cell;
if *d2 <= floor * floor {
break;
}
}
let mut in_bounds = false;
for (ix, iy) in self.ring(cx, cy, r) {
in_bounds = true;
for &item in &self.cells[iy * self.cols + ix] {
let ceiling = best.as_ref().map_or(f32::INFINITY, |(d2, _, _)| *d2);
let Some((d2, value)) = consider(item, ceiling).filter(|(d2, _)| !d2.is_nan())
else {
continue;
};
let wins = best
.as_ref()
.is_none_or(|(b, bi, _)| (d2, item) < (*b, *bi));
if wins {
best = Some((d2, item, value));
}
}
}
if !in_bounds {
break; // the ring has outgrown the grid; so will every later one
}
}
best.map(|(_, _, value)| value)
}
/// The in-bounds cells at Chebyshev distance `r` from `(cx, cy)`.
fn ring(&self, cx: isize, cy: isize, r: isize) -> impl Iterator<Item = (usize, usize)> + '_ {
let (cols, rows) = (self.cols as isize, self.rows as isize);
(cy - r..=cy + r)
.filter(move |iy| (0..rows).contains(iy))
.flat_map(move |iy| {
// Interior rows contribute only their two end columns; the top
// and bottom rows of the ring contribute all of theirs.
let edge = iy == cy - r || iy == cy + r;
let step = if edge { 1 } else { (2 * r).max(1) };
(cx - r..=cx + r)
.step_by(step as usize)
.filter(move |ix| (0..cols).contains(ix))
.map(move |ix| (ix as usize, iy as usize))
})
}
}
#[cfg(test)]
mod tests {
use super::*;
/// One point per item, as a degenerate box.
fn points(ps: &[(f32, f32)]) -> Vec<Aabb> {
ps.iter()
.map(|&(x, y)| Aabb::around([(x, y)]).unwrap())
.collect()
}
/// Brute-force nearest, as the reference answer.
fn brute(ps: &[(f32, f32)], x: f32, y: f32) -> Option<usize> {
ps.iter()
.enumerate()
.min_by(|(_, a), (_, b)| {
let d = |p: &(f32, f32)| (p.0 - x).powi(2) + (p.1 - y).powi(2);
d(a).total_cmp(&d(b))
})
.map(|(i, _)| i)
}
fn nearest(grid: &Grid, ps: &[(f32, f32)], x: f32, y: f32) -> Option<usize> {
grid.nearest(x, y, |i, _| {
let p = ps[i as usize];
Some(((p.0 - x).powi(2) + (p.1 - y).powi(2), i as usize))
})
}
#[test]
fn an_empty_grid_finds_nothing() {
let grid = Grid::build(&[]);
assert_eq!(nearest(&grid, &[], 0.0, 0.0), None);
}
#[test]
fn agrees_with_brute_force_on_a_scattered_cloud() {
// A deterministic pseudo-random spread, queried from inside, outside,
// and far past the grid. The ring walk must never stop early on a
// candidate a later ring would have beaten.
let ps: Vec<(f32, f32)> = (0..400)
.map(|i| {
let a = i as f32 * 2.399_963; // golden-angle spiral
(
a.cos() * (i as f32).sqrt() * 9.0,
a.sin() * (i as f32).sqrt() * 9.0,
)
})
.collect();
let grid = Grid::build(&points(&ps));
for k in 0..200 {
let t = k as f32 * 0.31;
for (x, y) in [
(t.cos() * 150.0, t.sin() * 150.0),
(t * 2.0 - 100.0, t * -1.5 + 60.0),
(5000.0, -5000.0),
] {
assert_eq!(
nearest(&grid, &ps, x, y),
brute(&ps, x, y),
"disagreed at ({x}, {y})"
);
}
}
}
#[test]
fn a_collinear_cloud_still_indexes() {
// Zero extent on one axis would divide by zero in the cell sizing.
let ps: Vec<(f32, f32)> = (0..50).map(|i| (i as f32 * 3.0, 7.0)).collect();
let grid = Grid::build(&points(&ps));
for k in 0..50 {
let x = k as f32 * 3.1 - 10.0;
assert_eq!(nearest(&grid, &ps, x, 7.0), brute(&ps, x, 7.0));
assert_eq!(nearest(&grid, &ps, x, 500.0), brute(&ps, x, 500.0));
}
}
#[test]
fn coincident_items_and_a_single_item_both_resolve() {
let ps = vec![(1.0, 1.0); 8];
let grid = Grid::build(&points(&ps));
assert!(nearest(&grid, &ps, 9.0, 9.0).is_some());
let one = vec![(4.0, -2.0)];
let grid = Grid::build(&points(&one));
assert_eq!(nearest(&grid, &one, 100.0, 100.0), Some(0));
}
#[test]
fn a_non_finite_query_never_panics() {
let ps = vec![(0.0, 0.0), (10.0, 10.0)];
let grid = Grid::build(&points(&ps));
for bad in [f32::NAN, f32::INFINITY, f32::NEG_INFINITY] {
let _ = nearest(&grid, &ps, bad, 0.0);
let _ = nearest(&grid, &ps, 0.0, bad);
}
}
#[test]
fn a_box_spanning_the_map_is_still_found_from_anywhere() {
// A single item whose footprint covers everything: it lands in every
// cell, and the dist2 prune has to stop it being re-measured forever.
let wide = Aabb {
min_x: -100.0,
max_x: 100.0,
min_y: -100.0,
max_y: 100.0,
};
let mut bounds = points(&[(90.0, 90.0), (-90.0, -90.0)]);
bounds.push(wide);
let grid = Grid::build(&bounds);
let hit = grid.nearest(0.0, 0.0, |i, best| {
let d2 = bounds[i as usize].dist2(0.0, 0.0);
(d2 <= best).then_some((d2, i))
});
assert_eq!(hit, Some(2), "the enclosing box is at distance zero");
}
#[test]
fn dist2_is_zero_inside_and_grows_outside() {
let b = Aabb {
min_x: 0.0,
max_x: 2.0,
min_y: 0.0,
max_y: 2.0,
};
assert_eq!(b.dist2(1.0, 1.0), 0.0);
assert_eq!(b.dist2(2.0, 0.0), 0.0);
assert_eq!(b.dist2(5.0, 1.0), 9.0);
assert_eq!(b.dist2(-3.0, -4.0), 25.0);
}
}