pub fn depth_first_search<G, I, F, C>(graph: G, starts: I, visitor: F) -> Cwhere
    G: IntoNeighbors + Visitable,
    I: IntoIterator<Item = <G as GraphBase>::NodeId>,
    F: FnMut(DfsEvent<<G as GraphBase>::NodeId>) -> C,
    C: ControlFlow,
Expand description

A recursive depth first search.

Starting points are the nodes in the iterator starts (specify just one start vertex x by using Some(x)).

The traversal emits discovery and finish events for each reachable vertex, and edge classification of each reachable edge. visitor is called for each event, see DfsEvent for possible values.

The return value should implement the trait ControlFlow, and can be used to change the control flow of the search.

Control Implements ControlFlow such that Control::Continue resumes the search. Control::Break will stop the visit early, returning the contained value. Control::Prune will stop traversing any additional edges from the current node and proceed immediately to the Finish event.

There are implementations of ControlFlow for (), and Result<C, E> where C: ControlFlow. The implementation for () will continue until finished. For Result, upon encountering an E it will break, otherwise acting the same as C.

*Panics if you attempt to prune a node from its Finish event.

Example returning Control.

Find a path from vertex 0 to 5, and exit the visit as soon as we reach the goal vertex.

use petgraph::prelude::*;
use petgraph::graph::node_index as n;
use petgraph::visit::depth_first_search;
use petgraph::visit::{DfsEvent, Control};

let gr: Graph<(), ()> = Graph::from_edges(&[
    (0, 1), (0, 2), (0, 3),
    (1, 3),
    (2, 3), (2, 4),
    (4, 0), (4, 5),

// record each predecessor, mapping node → node
let mut predecessor = vec![NodeIndex::end(); gr.node_count()];
let start = n(0);
let goal = n(5);
depth_first_search(&gr, Some(start), |event| {
    if let DfsEvent::TreeEdge(u, v) = event {
        predecessor[v.index()] = u;
        if v == goal {
            return Control::Break(v);

let mut next = goal;
let mut path = vec![next];
while next != start {
    let pred = predecessor[next.index()];
    next = pred;
assert_eq!(&path, &[n(0), n(2), n(4), n(5)]);

Example returning a Result.

use petgraph::graph::node_index as n;
use petgraph::prelude::*;
use petgraph::visit::depth_first_search;
use petgraph::visit::{DfsEvent, Time};

let gr: Graph<(), ()> = Graph::from_edges(&[(0, 1), (1, 2), (1, 1), (2, 1)]);
let start = n(0);
let mut back_edges = 0;
let mut discover_time = 0;
// Stop the search, the first time a BackEdge is encountered.
let result = depth_first_search(&gr, Some(start), |event| {
    match event {
        // In the cases where Ok(()) is returned,
        // Result falls back to the implementation of Control on the value ().
        // In the case of (), this is to always return Control::Continue.
        // continuing the search.
        DfsEvent::Discover(_, Time(t)) => {
            discover_time = t;
        DfsEvent::BackEdge(_, _) => {
            back_edges += 1;
            // the implementation of ControlFlow for Result,
            // treats this Err value as Continue::Break
        _ => Ok(()),

// Even though the graph has more than one cycle,
// The number of back_edges visited by the search should always be 1.
assert_eq!(back_edges, 1);
println!("discover time:{:?}", discover_time);
println!("number of backedges encountered: {}", back_edges);
println!("back edge: {:?}", result);