hack on scc

Author: pquerna

pkg/sync/expand/cycle.go

@@ -3,114 +3,68 @@ package expand
 import (
 	"context"
 
+	"github.com/conductorone/baton-sdk/pkg/sync/expand/scc"
 	mapset "github.com/deckarep/golang-set/v2"
 )
 
-const (
-	colorWhite uint8 = iota
-	colorGray
-	colorBlack
-)
-
-// cycleDetector encapsulates coloring state for cycle detection on an
-// EntitlementGraph. Node IDs are dense (1..NextNodeID), so slices are used for
-// O(1) access and zero per-op allocations.
-type cycleDetector struct {
-	g      *EntitlementGraph
-	state  []uint8
-	parent []int
-}
-
-func newCycleDetector(g *EntitlementGraph) *cycleDetector {
-	cd := &cycleDetector{
-		g:      g,
-		state:  make([]uint8, g.NextNodeID+1),
-		parent: make([]int, g.NextNodeID+1),
+// GetFirstCycle given an entitlements graph, return a cycle by node ID if it
+// exists. Returns nil if no cycle exists. If there is a single
+// node pointing to itself, that will count as a cycle.
+func (g *EntitlementGraph) GetFirstCycle(ctx context.Context) []int {
+	if g.HasNoCycles {
+		return nil
 	}
-	for i := range cd.parent {
-		cd.parent[i] = -1
+	comps := g.ComputeCyclicComponents(ctx)
+	if len(comps) == 0 {
+		return nil
 	}
-	return cd
+	return comps[0]
 }
 
-// dfs performs a coloring-based DFS from u, returning the first detected cycle
-// as a slice of node IDs or nil if no cycle is reachable from u.
-func (cd *cycleDetector) dfs(u int) ([]int, bool) {
-	// Self-loop fast path.
-	if nbrs, ok := cd.g.SourcesToDestinations[u]; ok {
-		if _, ok := nbrs[u]; ok {
-			return []int{u}, true
-		}
-	}
-
-	cd.state[u] = colorGray
-	if nbrs, ok := cd.g.SourcesToDestinations[u]; ok {
-		for v := range nbrs {
-			switch cd.state[v] {
-			case colorWhite:
-				cd.parent[v] = u
-				if cyc, ok := cd.dfs(v); ok {
-					return cyc, true
-				}
-			case colorGray:
-				// Back-edge to a node on the current recursion stack.
-				// Reconstruct cycle by walking parents from u back to v (inclusive), then reverse.
-				cycle := make([]int, 0, 8)
-				for x := u; ; x = cd.parent[x] {
-					cycle = append(cycle, x)
-					if x == v || cd.parent[x] == -1 {
-						break
-					}
-				}
-				for i, j := 0, len(cycle)-1; i < j; i, j = i+1, j-1 {
-					cycle[i], cycle[j] = cycle[j], cycle[i]
-				}
-				return cycle, true
-			}
-		}
+// HasCycles returns true if the graph contains any cycle.
+func (g *EntitlementGraph) HasCycles(ctx context.Context) bool {
+	if g.HasNoCycles {
+		return false
 	}
-	cd.state[u] = colorBlack
-	return nil, false
+	return len(g.ComputeCyclicComponents(ctx)) > 0
 }
 
-// FindAny scans all nodes and returns the first detected cycle or nil if none exist.
-func (cd *cycleDetector) FindAny() []int {
-	for nodeID := range cd.g.Nodes {
-		if cd.state[nodeID] != colorWhite {
-			continue
-		}
-		if cyc, ok := cd.dfs(nodeID); ok {
-			return cyc
+func (g *EntitlementGraph) cycleDetectionHelper(
+	nodeID int,
+) ([]int, bool) {
+	reach := g.reachableFrom(nodeID)
+	if len(reach) == 0 {
+		return nil, false
+	}
+	adj := g.toAdjacency(reach)
+	groups := scc.CondenseFWBWGroupsFromAdj(context.Background(), adj, scc.DefaultOptions())
+	for _, comp := range groups {
+		if len(comp) > 1 || (len(comp) == 1 && adj[comp[0]][comp[0]] != 0) {
+			return comp, true
 		}
 	}
-	return nil
+	return nil, false
 }
 
-// FindFrom starts cycle detection from a specific node and returns the first
-// cycle reachable from that node, or nil,false if none.
-func (cd *cycleDetector) FindFrom(start int) ([]int, bool) {
-	return cd.dfs(start)
+func (g *EntitlementGraph) FixCycles(ctx context.Context) error {
+	return g.FixCyclesFromComponents(ctx, g.ComputeCyclicComponents(ctx))
 }
 
-// GetFirstCycle given an entitlements graph, return a cycle by node ID if it
-// exists. Returns nil if no cycle exists. If there is a single
-// node pointing to itself, that will count as a cycle.
-func (g *EntitlementGraph) GetFirstCycle() []int {
+// ComputeCyclicComponents runs SCC once and returns only cyclic components.
+// A component is cyclic if len>1 or a singleton with a self-loop.
+func (g *EntitlementGraph) ComputeCyclicComponents(ctx context.Context) [][]int {
 	if g.HasNoCycles {
 		return nil
 	}
-	cd := newCycleDetector(g)
-	return cd.FindAny()
-}
-
-func (g *EntitlementGraph) cycleDetectionHelper(
-	nodeID int,
-) ([]int, bool) {
-	// Thin wrapper around the coloring-based DFS, starting from a specific node.
-	// The provided visited/currentCycle are ignored here; coloring provides the
-	// necessary state for correctness and performance.
-	cd := newCycleDetector(g)
-	return cd.FindFrom(nodeID)
+	adj := g.toAdjacency(nil)
+	groups := scc.CondenseFWBWGroupsFromAdj(ctx, adj, scc.DefaultOptions())
+	cyclic := make([][]int, 0)
+	for _, comp := range groups {
+		if len(comp) > 1 || (len(comp) == 1 && adj[comp[0]][comp[0]] != 0) {
+			cyclic = append(cyclic, comp)
+		}
+	}
+	return cyclic
 }
 
 // removeNode obliterates a node and all incoming/outgoing edges.
@@ -147,34 +101,33 @@ func (g *EntitlementGraph) removeNode(nodeID int) {
 	delete(g.SourcesToDestinations, nodeID)
 }
 
-// FixCycles if any cycles of nodes exist, merge all nodes in that cycle into a
-// single node and then repeat. Iteration ends when there are no more cycles.
-func (g *EntitlementGraph) FixCycles(ctx context.Context) error {
+// FixCyclesFromComponents merges all provided cyclic components in one pass.
+func (g *EntitlementGraph) FixCyclesFromComponents(ctx context.Context, cyclic [][]int) error {
 	if g.HasNoCycles {
 		return nil
 	}
-	for {
+	if len(cyclic) == 0 {
+		g.HasNoCycles = true
+		return nil
+	}
+	for _, comp := range cyclic {
 		select {
 		case <-ctx.Done():
 			return ctx.Err()
 		default:
 		}
-		cycle := g.GetFirstCycle()
-		if cycle == nil {
-			g.HasNoCycles = true
-			return nil
-		}
-
-		if err := g.fixCycle(cycle); err != nil {
+		if err := g.fixCycle(comp); err != nil {
 			return err
 		}
 	}
+	g.HasNoCycles = true
+	return nil
 }
 
 // fixCycle takes a list of Node IDs that form a cycle and merges them into a
 // single, new node.
 func (g *EntitlementGraph) fixCycle(nodeIDs []int) error {
-	entitlementIDs := mapset.NewSet[string]()
+	entitlementIDs := mapset.NewThreadUnsafeSet[string]()
 	outgoingEdgesToResourceTypeIDs := map[int]mapset.Set[string]{}
 	incomingEdgesToResourceTypeIDs := map[int]mapset.Set[string]{}
 	for _, nodeID := range nodeIDs {
@@ -190,7 +143,7 @@ func (g *EntitlementGraph) fixCycle(nodeIDs []int) error {
 					if edge, ok := g.Edges[edgeID]; ok {
 						resourceTypeIDs, ok := incomingEdgesToResourceTypeIDs[sourceNodeID]
 						if !ok {
-							resourceTypeIDs = mapset.NewSet[string]()
+							resourceTypeIDs = mapset.NewThreadUnsafeSet[string]()
 						}
 						for _, resourceTypeID := range edge.ResourceTypeIDs {
 							resourceTypeIDs.Add(resourceTypeID)
@@ -206,7 +159,7 @@ func (g *EntitlementGraph) fixCycle(nodeIDs []int) error {
 					if edge, ok := g.Edges[edgeID]; ok {
 						resourceTypeIDs, ok := outgoingEdgesToResourceTypeIDs[destinationNodeID]
 						if !ok {
-							resourceTypeIDs = mapset.NewSet[string]()
+							resourceTypeIDs = mapset.NewThreadUnsafeSet[string]()
 						}
 						for _, resourceTypeID := range edge.ResourceTypeIDs {
 							resourceTypeIDs.Add(resourceTypeID)

pkg/sync/expand/cycle_benchmark_test.go

@@ -151,14 +151,16 @@ func BenchmarkCycleDetectionHelper(b *testing.B) {
 }
 
 func BenchmarkGetFirstCycle(b *testing.B) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
 	sizes := []int{100, 1000}
 
 	for _, n := range sizes {
 		b.Run(fmt.Sprintf("ring-%d", n), func(b *testing.B) {
 			g := buildRing(b, n)
 			b.ResetTimer()
 			for i := 0; i < b.N; i++ {
-				_ = g.GetFirstCycle()
+				_ = g.GetFirstCycle(ctx)
 			}
 		})
 	}
@@ -168,7 +170,7 @@ func BenchmarkGetFirstCycle(b *testing.B) {
 			g := buildChain(b, n)
 			b.ResetTimer()
 			for i := 0; i < b.N; i++ {
-				_ = g.GetFirstCycle()
+				_ = g.GetFirstCycle(ctx)
 			}
 		})
 	}

pkg/sync/expand/graph.go

@@ -309,3 +309,64 @@ func (g *EntitlementGraph) DeleteEdge(ctx context.Context, srcEntitlementID stri
 	}
 	return nil
 }
+
+// toAdjacency builds an adjacency map for SCC. If nodesSubset is non-nil, only
+// include those nodes (and edges between them). Always include all nodes in the
+// subset as keys, even if they have zero outgoing edges.
+func (g *EntitlementGraph) toAdjacency(nodesSubset map[int]struct{}) map[int]map[int]int {
+	adj := make(map[int]map[int]int, len(g.Nodes))
+	include := func(id int) bool {
+		if nodesSubset == nil {
+			return true
+		}
+		_, ok := nodesSubset[id]
+		return ok
+	}
+
+	// Ensure keys for all included nodes.
+	for id := range g.Nodes {
+		if include(id) {
+			adj[id] = make(map[int]int)
+		}
+	}
+
+	// Add edges where both endpoints are included.
+	for src, dsts := range g.SourcesToDestinations {
+		if !include(src) {
+			continue
+		}
+		row := adj[src]
+		for dst := range dsts {
+			if include(dst) {
+				row[dst] = 1
+			}
+		}
+	}
+	return adj
+}
+
+// reachableFrom computes the set of node IDs reachable from start over
+// SourcesToDestinations using an iterative BFS.
+func (g *EntitlementGraph) reachableFrom(start int) map[int]struct{} {
+	if _, ok := g.Nodes[start]; !ok {
+		return nil
+	}
+	visited := make(map[int]struct{}, 16)
+	queue := make([]int, 0, 16)
+	queue = append(queue, start)
+	visited[start] = struct{}{}
+	for len(queue) > 0 {
+		u := queue[0]
+		queue = queue[1:]
+		if nbrs, ok := g.SourcesToDestinations[u]; ok {
+			for v := range nbrs {
+				if _, seen := visited[v]; seen {
+					continue
+				}
+				visited[v] = struct{}{}
+				queue = append(queue, v)
+			}
+		}
+	}
+	return visited
+}

pkg/sync/expand/graph_test.go

@@ -160,7 +160,7 @@ func TestGetFirstCycle(t *testing.T) {
 	for _, testCase := range testCases {
 		t.Run(testCase.message, func(t *testing.T) {
 			graph := parseExpression(t, ctx, testCase.expression)
-			cycle := graph.GetFirstCycle()
+			cycle := graph.GetFirstCycle(ctx)
 			if testCase.expectedCycleSize == 0 {
 				require.Nil(t, cycle)
 			} else {
@@ -189,7 +189,7 @@ func TestHandleCycle(t *testing.T) {
 
 			graph := parseExpression(t, ctx, testCase.expression)
 
-			cycle := graph.GetFirstCycle()
+			cycle := graph.GetFirstCycle(ctx)
 			expectedCycles := createNodeIDList(testCase.expectedCycles)
 			require.NotNil(t, cycle)
 			found := false
@@ -205,7 +205,7 @@ func TestHandleCycle(t *testing.T) {
 			require.NoError(t, err, graph.Str())
 			err = graph.Validate()
 			require.NoError(t, err)
-			cycle = graph.GetFirstCycle()
+			cycle = graph.GetFirstCycle(ctx)
 			require.Nil(t, cycle)
 		})
 	}
@@ -230,7 +230,7 @@ func TestHandleComplexCycle(t *testing.T) {
 	require.Equal(t, 0, len(graph.Edges))
 	require.Equal(t, 3, len(graph.GetEntitlements()))
 
-	cycle := graph.GetFirstCycle()
+	cycle := graph.GetFirstCycle(ctx)
 	require.Nil(t, cycle)
 }
 
@@ -257,7 +257,7 @@ func TestHandleCliqueCycle(t *testing.T) {
 		require.Equal(t, 0, len(graph.Edges))
 		require.Equal(t, 3, len(graph.GetEntitlements()))
 
-		cycle := graph.GetFirstCycle()
+		cycle := graph.GetFirstCycle(ctx)
 		require.Nil(t, cycle)
 	}
 }