Fix two bugs: yen k=0 underflow, DfsReachable stack overflow

src/directed/dfs.rs

@@ -163,19 +163,21 @@ where
     type Item = N;
 
     fn next(&mut self) -> Option<Self::Item> {
-        let n = self.to_see.pop()?;
-        if self.visited.contains(&n) {
-            return self.next();
-        }
-        self.visited.insert(n.clone());
-        let mut to_insert = Vec::new();
-        for s in (self.successors)(&n) {
-            if !self.visited.contains(&s) {
-                to_insert.push(s.clone());
+        loop {
+            let n = self.to_see.pop()?;
+            if self.visited.contains(&n) {
+                continue;
+            }
+            self.visited.insert(n.clone());
+            let mut to_insert = Vec::new();
+            for s in (self.successors)(&n) {
+                if !self.visited.contains(&s) {
+                    to_insert.push(s.clone());
+                }
             }
+            self.to_see.extend(to_insert.into_iter().rev());
+            return Some(n);
         }
-        self.to_see.extend(to_insert.into_iter().rev());
-        Some(n)
     }
 }
 

src/directed/yen.rs

@@ -108,6 +108,9 @@ where
     IN: IntoIterator<Item = (N, C)>,
     FS: FnMut(&N) -> bool,
 {
+    if k == 0 {
+        return vec![];
+    }
     let Some((n, c)) = dijkstra_internal(start, &mut successors, &mut success) else {
         return vec![];
     };

tests/dfs-reach.rs

@@ -14,3 +14,20 @@ fn issue_511_branches() {
     let it = dfs_reach(0, |&n| [n + 2, n + 5].into_iter().filter(|&x| x <= 10));
     assert_eq!(vec![0, 2, 4, 6, 8, 10, 9, 7, 5], it.collect::<Vec<_>>());
 }
+
+/// Test that `dfs_reach` does not stack overflow when many duplicate nodes
+/// pile up in the `to_see` stack (would previously recurse for each duplicate).
+#[test]
+fn no_stack_overflow_with_duplicates() {
+    // Each node has N successors all pointing to the same next node, creating
+    // many duplicates in to_see. With the old recursive implementation, the
+    // recursion depth could equal the number of duplicates, causing stack overflow.
+    let n = 200_usize;
+    // Node 0 -> [1, 1, 1, ...] (n copies of 1)
+    // Node k -> [k+1, k+1, k+1, ...] (n copies of k+1) for k < n
+    // Node n -> []
+    let result: Vec<usize> =
+        dfs_reach(0usize, |&k| if k < n { vec![k + 1; n] } else { vec![] }).collect();
+    let expected: Vec<usize> = (0..=n).collect();
+    assert_eq!(result, expected);
+}

tests/yen.rs

@@ -177,3 +177,19 @@ fn multiple_equal_cost_paths() {
     assert_eq!(paths[0], (vec!['A', 'B', 'D'], 2));
     assert_eq!(paths[1], (vec!['A', 'C', 'D'], 2));
 }
+
+/// Test that k=0 returns an empty result without panicking.
+#[test]
+fn k_zero() {
+    let result = yen(
+        &'c',
+        |c| match c {
+            'c' => vec![('d', 3)],
+            'd' => vec![],
+            _ => panic!(""),
+        },
+        |c| *c == 'd',
+        0,
+    );
+    assert!(result.is_empty());
+}