omnidiff 0.2.0

Fast, robust, syntax-aware code diffing using tree-sitter ASTs
Documentation 
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

/*  This file is part of the OmniDiff code diffing tool.
 *
 *  Copyright (C) 2026 Marko Ivankovic
 *
 *  This program is free software: you can redistribute it and/or modify
 *  it under the terms of the GNU Affero General Public License as published
 *  by the Free Software Foundation, either version 3 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 Affero General Public License for more details.
 *
 *  You should have received a copy of the GNU Affero General Public License
 *  along with this program. If not, see <https://www.gnu.org/licenses/>.
 */

//! A *similarity*-preserving sketch, as opposed to the four *equality* hashes in
//! [`crate::code::hash`], which say nothing once two subtrees differ in one token.
//!
//! A bottom-k MinHash sketch of the set of *leaf* hashes in a node's subtree, built bottom-up with
//! the equality hashes and compared in O(k) whatever the subtree size. Leaves, not every
//! descendant: one changed token flips the full hash of every ancestor, so a one-token edit would
//! cost O(depth) elements instead of one.
//!
//! [`SimilaritySketch::jaccard`] is exact when both subtrees have fewer than [`SKETCH_WIDTH`]
//! distinct leaves and an estimate above. Rank and gate with it; never conclude equality from it.

/// Number of retained bottom-k values: small per node, and exact for most real subtrees.
pub const SKETCH_WIDTH: usize = 16;

/// Fixed, never per-process: a random seed would make similarity, and so matching, differ between
/// runs on identical input.
const SKETCH_SEED: u64 = 0x9E37_79B9_7F4A_7C15;

/// SplitMix64, the MinHash permutation: Merkle hashes of similar nodes correlate in their low
/// bits, and the k smallest images must be an unbiased sample.
fn mix(value: u64) -> u64 {
    let mut z = value.wrapping_add(SKETCH_SEED);
    z = (z ^ (z >> 30)).wrapping_mul(0xBF58_476D_1CE4_E5B9);
    z = (z ^ (z >> 27)).wrapping_mul(0x94D0_49BB_1331_11EB);
    z ^ (z >> 31)
}

/// The k smallest *distinct* `mix`ed leaf hashes in a node's subtree, ascending.
///
/// Distinct, so a sketch is a function of the leaf-hash *set*, not of merge order.
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct SimilaritySketch {
    values: [u64; SKETCH_WIDTH],
    /// How many of `values` are populated. `< SKETCH_WIDTH` means the sketch is the complete set.
    len: u8,
}

impl SimilaritySketch {
    /// The sketch of a leaf: the single element `{mix(full_hash)}`.
    pub fn leaf(full_hash: u64) -> Self {
        let mut values = [0u64; SKETCH_WIDTH];
        values[0] = mix(full_hash);
        Self { values, len: 1 }
    }

    /// The sketch of an internal node: the bottom-k of the union of its children's sketches.
    ///
    /// Equal to the bottom-k of the union of the children's sets (the bottom-k merge property),
    /// which is why one bottom-up pass sketches the whole tree.
    pub fn merge(children: impl IntoIterator<Item = Self>) -> Self {
        let mut pool: Vec<u64> = Vec::new();
        for child in children {
            pool.extend_from_slice(child.as_slice());
        }
        pool.sort_unstable();
        pool.dedup();
        pool.truncate(SKETCH_WIDTH);

        let mut values = [0u64; SKETCH_WIDTH];
        values[..pool.len()].copy_from_slice(&pool);
        Self {
            values,
            len: pool.len() as u8,
        }
    }

    fn as_slice(&self) -> &[u64] {
        &self.values[..self.len as usize]
    }

    /// True when this sketch holds the subtree's complete set of distinct leaf hashes, so
    /// [`jaccard`](Self::jaccard) against another complete sketch is exact rather than estimated.
    #[cfg(test)]
    pub fn is_exact(&self) -> bool {
        (self.len as usize) < SKETCH_WIDTH
    }

    /// Estimated Jaccard similarity of the two subtrees' leaf-hash sets, in `[0.0, 1.0]`.
    ///
    /// Standard bottom-k estimator: the fraction of `U`, the k smallest values of the union, present
    /// in both. The divisor is `|U|`, not `k`, or small subtrees would under-report similarity.
    pub fn jaccard(&self, other: &Self) -> f32 {
        let (a, b) = (self.as_slice(), other.as_slice());
        if a.is_empty() || b.is_empty() {
            return 0.0;
        }

        let (mut i, mut j) = (0usize, 0usize);
        let (mut union_size, mut shared) = (0usize, 0usize);
        while union_size < SKETCH_WIDTH && (i < a.len() || j < b.len()) {
            match (a.get(i), b.get(j)) {
                (Some(&x), Some(&y)) if x == y => {
                    shared += 1;
                    i += 1;
                    j += 1;
                }
                (Some(&x), Some(&y)) if x < y => i += 1,
                (Some(_), Some(_)) => j += 1,
                (Some(_), None) => i += 1,
                (None, Some(_)) => j += 1,
                (None, None) => unreachable!("loop condition guarantees one side has values left"),
            }
            union_size += 1;
        }

        shared as f32 / union_size as f32
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn sketch_of(leaf_hashes: &[u64]) -> SimilaritySketch {
        SimilaritySketch::merge(leaf_hashes.iter().map(|&h| SimilaritySketch::leaf(h)))
    }

    #[test]
    fn identical_sets_score_one_and_disjoint_sets_score_zero() {
        let a = sketch_of(&[1, 2, 3, 4]);
        assert_eq!(a.jaccard(&sketch_of(&[1, 2, 3, 4])), 1.0);
        assert_eq!(a.jaccard(&sketch_of(&[90, 91, 92, 93])), 0.0);
    }

    #[test]
    fn small_sets_are_exact_not_estimated() {
        // Dividing by SKETCH_WIDTH instead of `|U|` would report 4/16.
        let a = sketch_of(&[1, 2, 3, 4]);
        let b = sketch_of(&[1, 2, 3, 4, 5]);
        assert!(a.is_exact() && b.is_exact());
        assert_eq!(a.jaccard(&b), 0.8);
    }

    #[test]
    fn one_changed_leaf_out_of_many_stays_near_one() {
        let shared: Vec<u64> = (0..40).collect();
        let mut changed = shared.clone();
        changed[7] = 1_000;
        let similarity = sketch_of(&shared).jaccard(&sketch_of(&changed));
        assert!(
            similarity > 0.8,
            "one differing leaf in 40 scored {similarity}"
        );
    }

    #[test]
    fn merge_is_order_independent() {
        let forward = sketch_of(&[5, 9, 1, 7, 3]);
        let backward = sketch_of(&[3, 7, 1, 9, 5]);
        assert_eq!(forward, backward);
    }

    #[test]
    fn merging_is_associative_over_intermediate_nodes() {
        let flat = sketch_of(&(0..60).collect::<Vec<_>>());
        let nested = SimilaritySketch::merge([
            sketch_of(&(0..20).collect::<Vec<_>>()),
            SimilaritySketch::merge([
                sketch_of(&(20..40).collect::<Vec<_>>()),
                sketch_of(&(40..60).collect::<Vec<_>>()),
            ]),
        ]);
        assert_eq!(flat, nested);
    }

    #[test]
    fn saturated_sketches_estimate_large_set_similarity() {
        let a: Vec<u64> = (0..1_000).collect();
        let b: Vec<u64> = (500..1_500).collect(); // true Jaccard = 500/1500 = 0.333...
        let estimate = sketch_of(&a).jaccard(&sketch_of(&b));
        assert!(
            (0.1..=0.6).contains(&estimate),
            "estimate {estimate} is nowhere near the true 0.333"
        );
    }
}