⚡ Bolt: Optimize transitive_callers with set difference

.jules/bolt.md

@@ -0,0 +1,3 @@
+## 2026-06-07 - Python loop overhead in graph traversal
+**Learning:** In highly connected graphs within a static analysis engine, looping over sets in Python (`for mod in callers: if mod not in closure: ...`) can introduce significant bytecode execution overhead.
+**Action:** Replace Python loops checking membership with fast C-level set operations (e.g., `callers - closure`) and use list-based stacks instead of intermediate set constructions for graph frontiers. This avoids O(N) Python iteration in hot paths.

src/wardline/scanner/taint/reverse_edge_index.py

@@ -66,15 +66,15 @@ def callers_of(self, callee_module: str) -> frozenset[str]:
     def transitive_callers(self, seeds: frozenset[str]) -> frozenset[str]:
         """``seeds`` plus every transitively-reverse-reachable module."""
         closure: set[str] = set(seeds)
-        frontier: set[str] = set(seeds)
+        frontier: list[str] = list(seeds)
+        get_callers = self._reverse.get
+
         while frontier:
-            next_frontier: set[str] = set()
-            for mod in frontier:
-                if mod not in self._reverse:
-                    continue
-                for caller_mod in self._reverse[mod]:
-                    if caller_mod not in closure:
-                        closure.add(caller_mod)
-                        next_frontier.add(caller_mod)
-            frontier = next_frontier
+            mod = frontier.pop()
+            callers = get_callers(mod)
+            if callers:
+                new_callers = callers - closure
+                if new_callers:
+                    closure.update(new_callers)
+                    frontier.extend(new_callers)
         return frozenset(closure)