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
//! A set of `u32` that iterates in the order `tsl::robin_set<int>` does.
//!
//! WHY THIS EXISTS
//! ---------------
//! `ldpc`'s localized statistics decoder keeps each cluster's bits, checks and candidates in
//! `tsl::robin_set<int>` (Tessil's robin-map, vendored in ldpc), and the order it walks them
//! decides things that change the answer: the order a merged cluster's columns enter the
//! surviving cluster's matrix, and so which columns become pivots. To return `ldpc`'s
//! corrections, LSD (`lsd`) walks its sets in that same order, which this reproduces. It is a
//! pure function of the insertions and removals, because `std::hash<int>` is the identity in
//! both libc++ and libstdc++:
//!
//! - open addressing over a power-of-two array of buckets, a key's ideal bucket `key & mask`;
//! - none allocated until the first insert; then 2 buckets, doubling whenever an insert finds
//! the set holding half as many keys as buckets (max load factor 0.5);
//! - robin-hood insertion: a key probing past a bucket whose occupant sits closer to its own
//! ideal bucket takes that bucket, and the occupant moves on;
//! - removal by backward shift: later keys of the run move back one bucket;
//! - growth reinserts the keys in the old array's bucket order;
//! - `clear` empties the buckets but keeps the array (the minimum load factor is 0).
//!
//! Iteration is bucket order. Robin-map's other rules (rehash after 8192 probes) cannot
//! trigger at the sizes LSD reaches and are left out.
const EMPTY: i16 = -1;
#[derive(Clone, Debug, Default)]
pub(crate) struct RobinSet {
/// (distance from the ideal bucket or `EMPTY`, key).
buckets: Vec<(i16, u32)>,
len: usize,
}
impl RobinSet {
pub(crate) fn new() -> RobinSet {
RobinSet::default()
}
pub(crate) fn len(&self) -> usize {
self.len
}
fn mask(&self) -> usize {
self.buckets.len().saturating_sub(1)
}
fn dist(&self, i: usize) -> i16 {
self.buckets.get(i).map_or(EMPTY, |b| b.0)
}
fn next(&self, i: usize) -> usize {
(i + 1) & self.mask()
}
fn find(&self, key: u32) -> Option<usize> {
let mut i = key as usize & self.mask();
let mut d = 0i16;
while d <= self.dist(i) {
if self.buckets[i].1 == key {
return Some(i);
}
i = self.next(i);
d += 1;
}
None
}
#[cfg(test)]
pub(crate) fn contains(&self, key: u32) -> bool {
self.find(key).is_some()
}
/// Insert `key`; whether it was new.
pub(crate) fn insert(&mut self, key: u32) -> bool {
let mut i = key as usize & self.mask();
let mut d = 0i16;
while d <= self.dist(i) {
if self.buckets[i].1 == key {
return false;
}
i = self.next(i);
d += 1;
}
if self.len >= self.buckets.len() / 2 {
self.rehash(2 * self.buckets.len().max(1));
i = key as usize & self.mask();
d = 0;
while d <= self.dist(i) {
i = self.next(i);
d += 1;
}
}
self.place(i, d, key);
self.len += 1;
true
}
/// Put `key`, `d` from its ideal bucket, at bucket `i` (empty, or holding a key closer to
/// its own), carrying displaced keys on.
fn place(&mut self, mut i: usize, mut d: i16, mut key: u32) {
loop {
let (bd, bk) = self.buckets[i];
if bd == EMPTY {
self.buckets[i] = (d, key);
return;
}
if d > bd {
self.buckets[i] = (d, key);
(d, key) = (bd, bk);
}
i = self.next(i);
d += 1;
}
}
fn rehash(&mut self, count: usize) {
let old = std::mem::replace(&mut self.buckets, vec![(EMPTY, 0); count]);
for (bd, key) in old {
if bd != EMPTY {
// robin-map's insert_value_on_rehash: the same robin-hood walk from distance 0.
let i = key as usize & self.mask();
self.place(i, 0, key);
}
}
}
/// Remove `key`; whether it was there.
pub(crate) fn remove(&mut self, key: u32) -> bool {
let Some(mut prev) = self.find(key) else { return false };
self.buckets[prev].0 = EMPTY;
self.len -= 1;
let mut i = self.next(prev);
while self.buckets[i].0 > 0 {
self.buckets[prev] = (self.buckets[i].0 - 1, self.buckets[i].1);
self.buckets[i].0 = EMPTY;
prev = i;
i = self.next(i);
}
true
}
/// Empty the set, keeping its buckets.
pub(crate) fn clear(&mut self) {
for b in &mut self.buckets {
b.0 = EMPTY;
}
self.len = 0;
}
/// The keys in robin-map's iteration order.
pub(crate) fn iter(&self) -> impl Iterator<Item = u32> + '_ {
self.buckets.iter().filter(|b| b.0 != EMPTY).map(|b| b.1)
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::surface_code::Xorshift;
use std::collections::BTreeSet;
/// The set holds what a BTreeSet holds through any sequence of inserts, removals and
/// clears, and every key sits no further from its ideal bucket than its run allows.
#[test]
fn it_is_a_set() {
let mut rng = Xorshift::new(9);
let (mut a, mut b) = (RobinSet::new(), BTreeSet::new());
for step in 0..20_000 {
let key = (rng.next_u64() % 200) as u32;
match rng.next_u64() % 10 {
0..=5 => assert_eq!(a.insert(key), b.insert(key)),
6..=8 => assert_eq!(a.remove(key), b.remove(&key)),
_ if step % 997 == 0 => {
a.clear();
b.clear();
}
_ => assert_eq!(a.contains(key), b.contains(&key)),
}
assert_eq!(a.len(), b.len());
let mut keys: Vec<u32> = a.iter().collect();
keys.sort_unstable();
assert_eq!(keys, b.iter().copied().collect::<Vec<_>>());
assert!(a.buckets.iter().enumerate().all(|(i, &(d, k))| d == EMPTY || (k as usize + d as usize) & a.mask() == i));
}
}
/// Worked by hand. 5, 1, 9: 5 goes in at 2 buckets (5 & 1 = 1); 1 grows the set to 4
/// buckets (5 → 1, then 1 → 1 is taken, so 1 → 2); 9 grows it to 8 (5 → 5, 1 → 1, then
/// 9 → 1 is taken by 1 at distance 0, so 9 → 2). Order: 1, 9, 5.
#[test]
fn its_order_is_robin_maps() {
let mut s = RobinSet::new();
for k in [5, 1, 9] {
s.insert(k);
}
assert_eq!(s.buckets.len(), 8);
assert_eq!(s.iter().collect::<Vec<_>>(), vec![1, 9, 5]);
// Removing 1 shifts 9 back into its ideal bucket.
s.remove(1);
assert_eq!(s.iter().collect::<Vec<_>>(), vec![9, 5]);
assert_eq!(s.buckets[1], (0, 9));
// A cleared set keeps its 8 buckets: 17 then lands at 17 & 7 = 1, before 3.
s.clear();
s.insert(3);
s.insert(17);
assert_eq!(s.iter().collect::<Vec<_>>(), vec![17, 3]);
}
}