sparrow/engine/
scorer.rs

1use std::collections::HashMap;
2
3/// TF-IDF style file scorer for intelligent context prioritization.
4/// Phase 3 Item 12: scores files by relevance to the task.
5pub struct FileScorer;
6
7impl FileScorer {
8    /// Score a set of files against a task description.
9    /// Returns (file_path, score) sorted by relevance descending.
10    pub fn score_files(
11        task: &str,
12        files: &[crate::memory::FileEntry],
13        symbols: &[crate::memory::SymbolEntry],
14    ) -> Vec<(String, f64)> {
15        let task_words: Vec<String> = tokenize(task);
16        if task_words.is_empty() {
17            return files.iter().map(|f| (f.path.clone(), 0.0)).collect();
18        }
19
20        let mut scores: HashMap<String, f64> = HashMap::new();
21
22        // Score by file path matching
23        for file in files {
24            let path_words = tokenize(&file.path);
25            let score = cosine_similarity_words(&task_words, &path_words);
26            *scores.entry(file.path.clone()).or_insert(0.0) += score * 2.0;
27        }
28
29        // Score by symbol name matching
30        for sym in symbols {
31            let sym_words = tokenize(&sym.name);
32            let score = cosine_similarity_words(&task_words, &sym_words);
33            *scores.entry(sym.file.clone()).or_insert(0.0) += score * 3.0;
34        }
35
36        let mut result: Vec<(String, f64)> = scores.into_iter().collect();
37        result.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap_or(std::cmp::Ordering::Equal));
38        result
39    }
40
41    /// Return top N files, always including explicitly mentioned files
42    pub fn top_files(
43        task: &str,
44        files: &[crate::memory::FileEntry],
45        symbols: &[crate::memory::SymbolEntry],
46        max: usize,
47    ) -> Vec<String> {
48        let mut scored = Self::score_files(task, files, symbols);
49
50        // Always include files explicitly mentioned in the task
51        let task_lower = task.to_lowercase();
52        for file in files {
53            if task_lower.contains(&file.path.to_lowercase()) {
54                if !scored.iter().any(|(p, _)| p == &file.path) {
55                    scored.push((file.path.clone(), 10.0));
56                }
57            }
58        }
59
60        scored.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap_or(std::cmp::Ordering::Equal));
61        scored.truncate(max);
62        scored.into_iter().map(|(p, _)| p).collect()
63    }
64}
65
66fn tokenize(text: &str) -> Vec<String> {
67    text.split(|c: char| !c.is_alphanumeric() && c != '_' && c != '-' && c != '/')
68        .filter(|w| w.len() > 1)
69        .map(|w| w.to_lowercase())
70        .collect()
71}
72
73/// Cosine similarity between two word frequency vectors
74fn cosine_similarity_words(a: &[String], b: &[String]) -> f64 {
75    let mut freq_a: HashMap<&str, f64> = HashMap::new();
76    let mut freq_b: HashMap<&str, f64> = HashMap::new();
77
78    for w in a {
79        *freq_a.entry(w.as_str()).or_insert(0.0) += 1.0;
80    }
81    for w in b {
82        *freq_b.entry(w.as_str()).or_insert(0.0) += 1.0;
83    }
84
85    let all_words: std::collections::HashSet<&str> =
86        freq_a.keys().chain(freq_b.keys()).copied().collect();
87    let mut dot = 0.0;
88    let mut norm_a = 0.0;
89    let mut norm_b = 0.0;
90
91    for w in &all_words {
92        let va = freq_a.get(w).copied().unwrap_or(0.0);
93        let vb = freq_b.get(w).copied().unwrap_or(0.0);
94        dot += va * vb;
95        norm_a += va * va;
96        norm_b += vb * vb;
97    }
98
99    if norm_a == 0.0 || norm_b == 0.0 {
100        0.0
101    } else {
102        dot / (norm_a.sqrt() * norm_b.sqrt())
103    }
104}
105
106// ─── Embedded Python kernel (persistent) ───────────────────────────────────────
107
108pub const PYTHON_KERNEL: &str = r#"
109import sys, json, traceback, io
110namespace = {}
111_stdout = io.StringIO()
112while True:
113    line = sys.stdin.readline()
114    if not line:
115        break
116    try:
117        req = json.loads(line)
118        _stdout = io.StringIO()
119        sys.stdout = _stdout
120        if req.get('mode') == 'eval':
121            result = eval(req['code'], namespace)
122        else:
123            exec(req['code'], namespace)
124            result = None
125        sys.stdout = sys.__stdout__
126        stdout_val = _stdout.getvalue()
127        response = {"id": req['id'], "result": str(result) if result is not None else None, "stdout": stdout_val}
128    except Exception as e:
129        sys.stdout = sys.__stdout__
130        response = {"id": req['id'], "error": traceback.format_exc()}
131    print(json.dumps(response), flush=True)
132"#;
sparrow/engine/scorer.rs

sparrow/engine/
scorer.rs
