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
/*
* Copyright (c) 2025-2026 Anton Kundenko <singaraiona@gmail.com>
* All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
/* ---------- HNSW Index ----------
*
* Multi-layer proximity graph for approximate nearest neighbor search.
*
* Memory layout per node:
* - Layer 0: up to M_max0 neighbors (default 2*M)
* - Layers 1+: up to M neighbors each
*
* Neighbor lists stored as flat arrays:
* neighbors[node * M_max + i] = neighbor_id (or -1 if unused)
*
* Each layer stores its own neighbor array for all nodes at that layer.
*/
/* Distance metric driving beam search. HNSW requires lower-is-closer;
* we choose the encoding so each metric sorts ascending:
* COSINE → 1 - cos(a, b) range [0, 2]
* L2 → sqrt(sum(sq diff)) range [0, ∞)
* IP → -dot(a, b) range (-∞, ∞) (negated so lower=closer) */
typedef enum ray_hnsw_metric_t;
typedef struct ray_hnsw_layer ray_hnsw_layer_t;
typedef struct ray_hnsw ray_hnsw_t;
/* --- Build / Free / Clone --- */
ray_hnsw_t* ;
void ;
/* Deep-copy an index: duplicates vectors, node levels, and every layer's
* neighbor + node_id arrays. Returns a new fully-owned index with the
* same semantics as the source. Returns NULL on OOM. */
ray_hnsw_t* ;
/* --- Search --- */
/* Returns top-K nearest neighbors as (node_id, distance) pairs.
* out_ids and out_dists must be pre-allocated with k entries.
*
* Return value:
* >= 0 : number of results written (may be < k).
* -1 : allocation failure (OOM) — callers must surface a distinct
* error rather than treat the 0-return as "no matches".
*/
int64_t ;
/* Predicate callback used by the filtered iterative-scan variant below.
* Return true to accept `node_id` into the result set, false to reject.
* Rejected nodes still participate in candidate-graph exploration so
* connectivity through them is preserved — this is the standard
* "iterative scan" shape. */
typedef bool ;
/* Like ray_hnsw_search, but only nodes passing `accept(node_id, ctx)`
* enter the top-K result set. Candidate-queue expansion still traverses
* rejected nodes so their accepted descendants remain reachable.
* Falls back to exhaustive graph exploration for pathologically selective
* filters (bounded by idx->n_nodes).
*
* Return value matches ray_hnsw_search: >= 0 = result count, -1 = OOM. */
int64_t ;
/* --- Accessors --- */
int32_t ;
/* --- Persistence --- */
ray_err_t ;
ray_hnsw_t* ;
ray_hnsw_t* ;
/* RAY_HNSW_H */