lean_ctx/core/graph_analysis/
mod.rs

1//! Static dependency-graph analyses (God-Nodes, import cycles).
2//!
3//! These run on the *real* directed dependency edges — `import` and `reexport` —
4//! and deliberately exclude the co-location heuristics (`sibling`, `cochange`)
5//! and the ambiguous `module` edges. That way the results reflect genuine code
6//! dependencies (graphify-style) instead of directory layout coincidences.
7
8mod centrality;
9mod cycles;
10mod god_nodes;
11mod surprising;
12
13pub use centrality::{BridgeCentrality, compute_bridge_centrality, compute_bridge_nodes};
14pub use cycles::{ImportCycle, find_import_cycles};
15pub use god_nodes::{GodNode, compute_god_nodes};
16pub use surprising::{SurprisingConnection, find_surprising_connections};
17
18use crate::core::graph_provider::EdgeInfo;
19
20/// Edge kinds that represent a genuine *directed* code dependency
21/// (`from` depends on `to`).
22pub(crate) const DEP_EDGE_KINDS: [&str; 2] = ["import", "reexport"];
23
24/// True when an edge kind is a genuine directed dependency.
25pub fn is_dependency_kind(kind: &str) -> bool {
26    DEP_EDGE_KINDS.contains(&kind)
27}
28
29/// Directed dependency edges as `(from, to)` pairs, with self-loops removed.
30/// Borrows from `edges` to avoid allocating new strings.
31pub fn dependency_edges(edges: &[EdgeInfo]) -> Vec<(&str, &str)> {
32    edges
33        .iter()
34        .filter(|e| is_dependency_kind(&e.kind))
35        .filter(|e| e.from != e.to)
36        .map(|e| (e.from.as_str(), e.to.as_str()))
37        .collect()
38}
39
40/// Confidence that an edge represents a *real* relationship, in `0.0..=1.0`.
41///
42/// Explicit code references (`import`/`reexport`) are certain; structural
43/// `module` grouping is fairly reliable; the co-location/co-change heuristics
44/// are weaker and scale with their observed `weight` (e.g. how often two files
45/// changed together). The dashboard styles edges by this score so heuristic
46/// links read as faint/dashed while real dependencies stay solid (#273).
47pub fn edge_confidence(kind: &str, weight: f64) -> f64 {
48    match kind {
49        "import" => 1.0,
50        "reexport" => 0.95,
51        "module" => 0.6,
52        // More co-changes ⇒ more confidence, with diminishing returns.
53        "cochange" => (0.30 + weight.max(0.0).ln_1p() * 0.18).clamp(0.30, 0.85),
54        "sibling" => 0.25,
55        _ => 0.5,
56    }
57}
58
59#[cfg(test)]
60mod confidence_tests {
61    use super::edge_confidence;
62
63    #[test]
64    fn explicit_refs_rank_above_heuristics() {
65        let import = edge_confidence("import", 0.0);
66        let reexport = edge_confidence("reexport", 0.0);
67        let module = edge_confidence("module", 0.0);
68        let sibling = edge_confidence("sibling", 0.0);
69        assert!(import >= reexport);
70        assert!(reexport > module);
71        assert!(module > sibling);
72        assert!((0.0..=1.0).contains(&sibling));
73    }
74
75    #[test]
76    fn cochange_scales_with_weight_and_is_bounded() {
77        let low = edge_confidence("cochange", 1.0);
78        let high = edge_confidence("cochange", 50.0);
79        assert!(high > low, "more co-changes should raise confidence");
80        assert!((0.30..=0.85).contains(&low));
81        assert!((0.30..=0.85).contains(&high));
82    }
83
84    #[test]
85    fn unknown_kind_is_neutral() {
86        assert_eq!(edge_confidence("mystery", 0.0), 0.5);
87    }
88}
lean_ctx/core/graph_analysis/mod.rs

lean_ctx/core/graph_analysis/
mod.rs
