Struct rustc_data_structures::graph::scc::SccsConstruction

pub(in graph::scc) struct SccsConstruction<'c, G: DirectedGraph + WithNumNodes + WithSuccessors, S: Idx> {
    graph: &'c G,
    node_states: IndexVec<G::Node, NodeState<G::Node, S>>,
    node_stack: Vec<G::Node>,
    successors_stack: Vec<S>,
    duplicate_set: FxHashSet<S>,
    scc_data: SccData<S>,
}

Fields

graph: &'c G

node_states: IndexVec<G::Node, NodeState<G::Node, S>>

The state of each node; used during walk to record the stack and after walk to record what cycle each node ended up being in.

node_stack: Vec<G::Node>

The stack of nodes that we are visiting as part of the DFS.

successors_stack: Vec<S>

The stack of successors: as we visit a node, we mark our position in this stack, and when we encounter a successor SCC, we push it on the stack. When we complete an SCC, we can pop everything off the stack that was found along the way.

duplicate_set: FxHashSet<S>

A set used to strip duplicates. As we accumulate successors into the successors_stack, we sometimes get duplicate entries. We use this set to remove those -- we also keep its storage around between successors to amortize memory allocation costs.

scc_data: SccData<S>

Implementations

impl<'c, G, S> SccsConstruction<'c, G, S> where
    G: DirectedGraph + WithNumNodes + WithSuccessors,
    S: Idx, 

pub(in graph::scc) fn construct(
    graph: &'c G
) -> Sccs<G::Node, S>

Identifies SCCs in the graph G and computes the resulting DAG. This uses a variant of Tarjan's algorithm. The high-level summary of the algorithm is that we do a depth-first search. Along the way, we keep a stack of each node whose successors are being visited. We track the depth of each node on this stack (there is no depth if the node is not on the stack). When we find that some node N with depth D can reach some other node N' with lower depth D' (i.e., D' < D), we know that N, N', and all nodes in between them on the stack are part of an SCC.

pub(in graph::scc) fn walk_node(
    &mut self,
    depth: usize,
    node: G::Node
) -> WalkReturn<S>

Visits a node during the DFS. We first examine its current state -- if it is not yet visited (NotVisited), we can push it onto the stack and start walking its successors.

If it is already on the DFS stack it will be in the state BeingVisited. In that case, we have found a cycle and we return the depth from the stack.

Otherwise, we are looking at a node that has already been completely visited. We therefore return WalkReturn::Complete with its associated SCC index.

pub(in graph::scc) fn find_state(
    &mut self,
    r: G::Node
) -> NodeState<G::Node, S>

Fetches the state of the node r. If r is recorded as being in a cycle with some other node r2, then fetches the state of r2 (and updates r to reflect current result). This is basically the "find" part of a standard union-find algorithm (with path compression).

pub(in graph::scc) fn walk_unvisited_node(
    &mut self,
    depth: usize,
    node: G::Node
) -> WalkReturn<S>

Walks a node that has never been visited before.