Make find dead-symbols linear and add code_only_line tests

find_dead_symbols previously re-scanned every candidate file's lines once
per unique symbol name (O(names × lines)). Replace with a per-file
word → line-number index built once per file and shared across all names,
bounded to tokens that are actual symbol names.

Also add unit coverage for code_only_line's string/comment blanking,
including Rust raw strings with multi-hash delimiters.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

src/ops.rs

@@ -260,8 +260,14 @@ pub fn find_dead_symbols(project: &Path) -> Result<Value> {
     for s in &symbols {
         by_name.entry(s.name.to_lowercase()).or_default().push(s);
     }
+    // Lowercased symbol names, used to bound the per-file word index to tokens
+    // that actually matter (avoids indexing every word in every file).
+    let names: HashSet<&str> = by_name.keys().map(|k| k.as_str()).collect();
 
-    let mut line_cache: HashMap<String, Vec<String>> = HashMap::new();
+    // Per-file word → line-number index, built once per file and shared across
+    // all names. Without this, a file's lines are re-scanned once per symbol
+    // name — O(names × lines) instead of O(lines).
+    let mut word_sites: HashMap<String, HashMap<String, Vec<i64>>> = HashMap::new();
     let mut dead: Vec<Value> = Vec::new();
 
     for (name_lc, syms) in &by_name {
@@ -269,15 +275,12 @@ pub fn find_dead_symbols(project: &Path) -> Result<Value> {
         // contains it (a superset), narrowed by whole-word match.
         let mut sites: Vec<(String, i64)> = Vec::new();
         for rel in search::files_containing(&conn, name_lc, 100_000)? {
-            let lines = match line_cache.entry(rel.clone()) {
+            let index = word_sites
-                std::collections::hash_map::Entry::Occupied(e) => e.get().clone(),
+                .entry(rel.clone())
-                std::collections::hash_map::Entry::Vacant(e) => e
+                .or_insert_with(|| index_words(project, &rel, &names));
-                    .insert(read_all_lines(project, &rel).unwrap_or_default())
+            if let Some(lines) = index.get(name_lc.as_str()) {
-                    .clone(),
+                for &ln in lines {
-            };
+                    sites.push((rel.clone(), ln));
-            for (i, line) in lines.iter().enumerate() {
-                if word_matches_ci(line, name_lc) && !is_import_line(line) {
-                    sites.push((rel.clone(), (i as i64) + 1));
                 }
             }
         }
@@ -302,10 +305,29 @@ pub fn find_dead_symbols(project: &Path) -> Result<Value> {
     Ok(serde_json::json!({"count": dead.len(), "symbols": dead}))
 }
 
-/// Whole-word, case-insensitive match (Python/Rust names may vary in case).
+/// Build a lowercased word → line-number index for one file, keeping only
-fn word_matches_ci(line: &str, word: &str) -> bool {
+/// tokens that are symbol names (from `names`) and skipping import/declaration
-    line.split(|c: char| !c.is_alphanumeric() && c != '_')
+/// lines (which reference, but don't count as call sites).
-        .any(|tok| tok.eq_ignore_ascii_case(word))
+fn index_words(project: &Path, rel: &str, names: &HashSet<&str>) -> HashMap<String, Vec<i64>> {
+    let mut m: HashMap<String, Vec<i64>> = HashMap::new();
+    let Ok(lines) = read_all_lines(project, rel) else {
+        return m;
+    };
+    for (i, line) in lines.iter().enumerate() {
+        if is_import_line(line) {
+            continue;
+        }
+        for tok in line.split(|c: char| !c.is_alphanumeric() && c != '_') {
+            if tok.is_empty() {
+                continue;
+            }
+            let lc = tok.to_lowercase();
+            if names.contains(lc.as_str()) {
+                m.entry(lc).or_default().push((i as i64) + 1);
+            }
+        }
+    }
+    m
 }
 
 /// Decorator keywords that wire a symbol to an external framework. Such symbols
@@ -828,7 +850,7 @@ fn raw_to_nodes(node: &RawNode, base: &str, depth: Option<usize>) -> Vec<TreeNod
 
 #[cfg(test)]
 mod tests {
-    use super::is_entry_point;
+    use super::{code_only_line, is_entry_point};
 
     #[test]
     fn entry_point_is_language_scoped() {
@@ -849,4 +871,30 @@ mod tests {
         assert!(is_entry_point("tests/integration.rs", "rust"));
         assert!(is_entry_point("src/bin/tool.rs", "rust"));
     }
+
+    #[test]
+    fn code_only_line_blanks_strings_and_comments() {
+        // Code references survive.
+        assert!(code_only_line("let x = foo();").contains("foo"));
+        assert!(code_only_line("let s = \"x\"; foo();").contains("foo"));
+
+        // String literals are blanked.
+        assert!(!code_only_line("let x = \"foo\";").contains("foo"));
+        assert!(!code_only_line("let x = 'foo';").contains("foo"));
+        assert!(!code_only_line("let x = `foo`;").contains("foo"));
+        assert!(!code_only_line("let x = \"\"\"foo\"\"\";").contains("foo"));
+
+        // Rust raw strings, including multi-hash delimiters.
+        assert!(!code_only_line("let x = r#\"...foo...\"#;").contains("foo"));
+        assert!(!code_only_line("let x = r##\"...foo...\"##;").contains("foo"));
+        assert!(!code_only_line("let x = br#\"foo\"#;").contains("foo"));
+
+        // Comments (whole-line and trailing) are blanked.
+        assert!(!code_only_line("// foo bar").contains("foo"));
+        assert!(!code_only_line("# foo bar").contains("foo"));
+        assert!(!code_only_line("-- foo bar").contains("foo"));
+        assert!(!code_only_line("/* foo bar */").contains("foo"));
+        assert!(!code_only_line("x = 1; // foo").contains("foo"));
+        assert!(!code_only_line("x = 1; /* foo */ y = 2;").contains("foo"));
+    }
 }