Import @armon's depgraph/digraph

This commit is contained in:
Mitchell Hashimoto 2014-05-24 12:47:04 -07:00
parent 046e80361b
commit c0a7e5b98b
13 changed files with 1282 additions and 0 deletions

36
depgraph/dependency.go Normal file
View File

@ -0,0 +1,36 @@
package depgraph
import (
"github.com/hashicorp/terraform/digraph"
)
// Dependency is used to create a directed edge between two nouns.
// One noun may depend on another and provide version constraints
// that cannot be violated
type Dependency struct {
Name string
Meta interface{}
Constraints []Constraint
Source *Noun
Target *Noun
}
// Constraint is used by dependencies to allow arbitrary constraints
// between nouns
type Constraint interface {
Satisfied(head, tail *Noun) (bool, error)
}
// Head returns the source, or dependent noun
func (d *Dependency) Head() digraph.Node {
return d.Source
}
// Tail returns the target, or depended upon noun
func (d *Dependency) Tail() digraph.Node {
return d.Target
}
func (d *Dependency) String() string {
return d.Name
}

155
depgraph/graph.go Normal file
View File

@ -0,0 +1,155 @@
// The depgraph package is used to create and model a dependency graph
// of nouns. Each noun can represent a service, server, application,
// network switch, etc. Nouns can depend on other nouns, and provide
// versioning constraints. Nouns can also have various meta data that
// may be relevant to their construction or configuration.
package depgraph
import (
"fmt"
"github.com/hashicorp/terraform/digraph"
)
// Graph is used to represent the entire dependency graph
type Graph struct {
Name string
Meta interface{}
Nouns []*Noun
Root *Noun
}
// Validate is used to ensure that a few properties of the graph are not violated:
// 1) There must be a single "root", or source on which nothing depends.
// 2) All nouns in the graph must be reachable from the root
// 3) The graph must be cycle free, meaning there are no cicular dependencies
func (g *Graph) Validate() error {
// Convert to node list
nodes := make([]digraph.Node, len(g.Nouns))
for i, n := range g.Nouns {
nodes[i] = n
}
// Create a validate erro
vErr := &ValidateError{}
// Search for all the sources, if we have only 1, it must be the root
if sources := digraph.Sources(nodes); len(sources) != 1 {
vErr.MissingRoot = true
goto CHECK_CYCLES
} else {
g.Root = sources[0].(*Noun)
}
// Check reachability
if unreached := digraph.Unreachable(g.Root, nodes); len(unreached) > 0 {
vErr.Unreachable = make([]*Noun, len(unreached))
for i, u := range unreached {
vErr.Unreachable[i] = u.(*Noun)
}
}
CHECK_CYCLES:
// Check for cycles
if cycles := digraph.StronglyConnectedComponents(nodes, true); len(cycles) > 0 {
vErr.Cycles = make([][]*Noun, len(cycles))
for i, cycle := range cycles {
group := make([]*Noun, len(cycle))
for j, n := range cycle {
group[j] = n.(*Noun)
}
vErr.Cycles[i] = group
}
}
// Return the detailed error
if vErr.MissingRoot || vErr.Unreachable != nil || vErr.Cycles != nil {
return vErr
}
return nil
}
// ValidateError implements the Error interface but provides
// additional information on a validation error
type ValidateError struct {
// If set, then the graph is missing a single root, on which
// there are no depdendencies
MissingRoot bool
// Unreachable are nodes that could not be reached from
// the root noun.
Unreachable []*Noun
// Cycles are groups of strongly connected nodes, which
// form a cycle. This is disallowed.
Cycles [][]*Noun
}
func (v *ValidateError) Error() string {
return "The depedency graph is not valid"
}
// CheckConstraints walks the graph and ensures that all
// user imposed constraints are satisfied.
func (g *Graph) CheckConstraints() error {
// Ensure we have a root
if g.Root == nil {
return fmt.Errorf("Graph must be validated before checking constraint violations")
}
// Create a constraint error
cErr := &ConstraintError{}
// Walk from the root
digraph.DepthFirstWalk(g.Root, func(n digraph.Node) bool {
noun := n.(*Noun)
for _, dep := range noun.Deps {
target := dep.Target
for _, constraint := range dep.Constraints {
ok, err := constraint.Satisfied(noun, target)
if ok {
continue
}
violation := &Violation{
Source: noun,
Target: target,
Dependency: dep,
Constraint: constraint,
Err: err,
}
cErr.Violations = append(cErr.Violations, violation)
}
}
return true
})
if cErr.Violations != nil {
return cErr
}
return nil
}
// ConstraintError is used to return detailed violation
// information from CheckConstraints
type ConstraintError struct {
Violations []*Violation
}
func (c *ConstraintError) Error() string {
return fmt.Sprintf("%d constraint violations", len(c.Violations))
}
// Violation is used to pass along information about
// a constraint violation
type Violation struct {
Source *Noun
Target *Noun
Dependency *Dependency
Constraint Constraint
Err error
}
func (v *Violation) Error() string {
return fmt.Sprintf("Constraint %v between %v and %v violated: %v",
v.Constraint, v.Source, v.Target, v.Err)
}

277
depgraph/graph_test.go Normal file
View File

@ -0,0 +1,277 @@
package depgraph
import (
"fmt"
"strings"
"testing"
)
// ParseNouns is used to parse a string in the format of:
// a -> b ; edge name
// b -> c
// Into a series of nouns and dependencies
func ParseNouns(s string) map[string]*Noun {
lines := strings.Split(s, "\n")
nodes := make(map[string]*Noun)
for _, line := range lines {
var edgeName string
if idx := strings.Index(line, ";"); idx >= 0 {
edgeName = strings.Trim(line[idx+1:], " \t\r\n")
line = line[:idx]
}
parts := strings.SplitN(line, "->", 2)
if len(parts) != 2 {
continue
}
head_name := strings.Trim(parts[0], " \t\r\n")
tail_name := strings.Trim(parts[1], " \t\r\n")
head := nodes[head_name]
if head == nil {
head = &Noun{Name: head_name}
nodes[head_name] = head
}
tail := nodes[tail_name]
if tail == nil {
tail = &Noun{Name: tail_name}
nodes[tail_name] = tail
}
edge := &Dependency{
Name: edgeName,
Source: head,
Target: tail,
}
head.Deps = append(head.Deps, edge)
}
return nodes
}
func NounMapToList(m map[string]*Noun) []*Noun {
list := make([]*Noun, 0, len(m))
for _, n := range m {
list = append(list, n)
}
return list
}
func TestGraph_Validate_NoRoot(t *testing.T) {
nodes := ParseNouns(`a -> b
b -> a`)
list := NounMapToList(nodes)
g := &Graph{Name: "Test", Nouns: list}
err := g.Validate()
if err == nil {
t.Fatalf("expected err")
}
vErr, ok := err.(*ValidateError)
if !ok {
t.Fatalf("expected validate error")
}
if !vErr.MissingRoot {
t.Fatalf("expected missing root")
}
}
func TestGraph_Validate_MultiRoot(t *testing.T) {
nodes := ParseNouns(`a -> b
c -> d`)
list := NounMapToList(nodes)
g := &Graph{Name: "Test", Nouns: list}
err := g.Validate()
if err == nil {
t.Fatalf("expected err")
}
vErr, ok := err.(*ValidateError)
if !ok {
t.Fatalf("expected validate error")
}
if !vErr.MissingRoot {
t.Fatalf("expected missing root")
}
}
func TestGraph_Validate_Unreachable(t *testing.T) {
nodes := ParseNouns(`a -> b
a -> c
b -> d
x -> x`)
list := NounMapToList(nodes)
g := &Graph{Name: "Test", Nouns: list}
err := g.Validate()
if err == nil {
t.Fatalf("expected err")
}
vErr, ok := err.(*ValidateError)
if !ok {
t.Fatalf("expected validate error")
}
if len(vErr.Unreachable) != 1 {
t.Fatalf("expected unreachable")
}
if vErr.Unreachable[0].Name != "x" {
t.Fatalf("bad: %v", vErr.Unreachable[0])
}
}
func TestGraph_Validate_Cycle(t *testing.T) {
nodes := ParseNouns(`a -> b
a -> c
b -> d
d -> b`)
list := NounMapToList(nodes)
g := &Graph{Name: "Test", Nouns: list}
err := g.Validate()
if err == nil {
t.Fatalf("expected err")
}
vErr, ok := err.(*ValidateError)
if !ok {
t.Fatalf("expected validate error")
}
if len(vErr.Cycles) != 1 {
t.Fatalf("expected cycles")
}
cycle := vErr.Cycles[0]
if cycle[0].Name != "d" {
t.Fatalf("bad: %v", cycle)
}
if cycle[1].Name != "b" {
t.Fatalf("bad: %v", cycle)
}
}
func TestGraph_Validate(t *testing.T) {
nodes := ParseNouns(`a -> b
a -> c
b -> d
b -> e
c -> d
c -> e`)
list := NounMapToList(nodes)
g := &Graph{Name: "Test", Nouns: list}
err := g.Validate()
if err != nil {
t.Fatalf("err: %v", err)
}
}
type VersionMeta int
type VersionConstraint struct {
Min int
Max int
}
func (v *VersionConstraint) Satisfied(head, tail *Noun) (bool, error) {
vers := int(tail.Meta.(VersionMeta))
if vers < v.Min {
return false, fmt.Errorf("version %d below minimum %d",
vers, v.Min)
} else if vers > v.Max {
return false, fmt.Errorf("version %d above maximum %d",
vers, v.Max)
}
return true, nil
}
func (v *VersionConstraint) String() string {
return "version"
}
func TestGraph_ConstraintViolation(t *testing.T) {
nodes := ParseNouns(`a -> b
a -> c
b -> d
b -> e
c -> d
c -> e`)
list := NounMapToList(nodes)
// Add a version constraint
vers := &VersionConstraint{1, 3}
// Introduce some constraints
depB := nodes["a"].Deps[0]
depB.Constraints = []Constraint{vers}
depC := nodes["a"].Deps[1]
depC.Constraints = []Constraint{vers}
// Add some versions
nodes["b"].Meta = VersionMeta(0)
nodes["c"].Meta = VersionMeta(4)
g := &Graph{Name: "Test", Nouns: list}
err := g.Validate()
if err != nil {
t.Fatalf("err: %v", err)
}
err = g.CheckConstraints()
if err == nil {
t.Fatalf("Expected err")
}
cErr, ok := err.(*ConstraintError)
if !ok {
t.Fatalf("expected constraint error")
}
if len(cErr.Violations) != 2 {
t.Fatalf("expected 2 violations: %v", cErr)
}
if cErr.Violations[0].Error() != "Constraint version between a and b violated: version 0 below minimum 1" {
t.Fatalf("err: %v", cErr.Violations[0])
}
if cErr.Violations[1].Error() != "Constraint version between a and c violated: version 4 above maximum 3" {
t.Fatalf("err: %v", cErr.Violations[1])
}
}
func TestGraph_Constraint_NoViolation(t *testing.T) {
nodes := ParseNouns(`a -> b
a -> c
b -> d
b -> e
c -> d
c -> e`)
list := NounMapToList(nodes)
// Add a version constraint
vers := &VersionConstraint{1, 3}
// Introduce some constraints
depB := nodes["a"].Deps[0]
depB.Constraints = []Constraint{vers}
depC := nodes["a"].Deps[1]
depC.Constraints = []Constraint{vers}
// Add some versions
nodes["b"].Meta = VersionMeta(2)
nodes["c"].Meta = VersionMeta(3)
g := &Graph{Name: "Test", Nouns: list}
err := g.Validate()
if err != nil {
t.Fatalf("err: %v", err)
}
err = g.CheckConstraints()
if err != nil {
t.Fatalf("err: %v", err)
}
}

27
depgraph/noun.go Normal file
View File

@ -0,0 +1,27 @@
package depgraph
import (
"github.com/hashicorp/terraform/digraph"
)
// Nouns are the key structure of the dependency graph. They can
// be used to represent all objects in the graph. They are linked
// by depedencies.
type Noun struct {
Name string // Opaque name
Meta interface{}
Deps []*Dependency
}
// Edges returns the out-going edges of a Noun
func (n *Noun) Edges() []digraph.Edge {
edges := make([]digraph.Edge, len(n.Deps))
for idx, dep := range n.Deps {
edges[idx] = dep
}
return edges
}
func (n *Noun) String() string {
return n.Name
}

89
digraph/basic.go Normal file
View File

@ -0,0 +1,89 @@
package digraph
import (
"fmt"
"strings"
)
// BasicNode is a digraph Node that has a name and out edges
type BasicNode struct {
Name string
NodeEdges []Edge
}
func (b *BasicNode) Edges() []Edge {
return b.NodeEdges
}
func (b *BasicNode) AddEdge(edge Edge) {
b.NodeEdges = append(b.NodeEdges, edge)
}
func (b *BasicNode) String() string {
if b.Name == "" {
return "Node"
}
return fmt.Sprintf("%v", b.Name)
}
// BasicEdge is a digraph Edge that has a name, head and tail
type BasicEdge struct {
Name string
EdgeHead *BasicNode
EdgeTail *BasicNode
}
func (b *BasicEdge) Head() Node {
return b.EdgeHead
}
// Tail returns the end point of the Edge
func (b *BasicEdge) Tail() Node {
return b.EdgeTail
}
func (b *BasicEdge) String() string {
if b.Name == "" {
return "Edge"
}
return fmt.Sprintf("%v", b.Name)
}
// ParseBasic is used to parse a string in the format of:
// a -> b ; edge name
// b -> c
// Into a series of basic node and basic edges
func ParseBasic(s string) map[string]*BasicNode {
lines := strings.Split(s, "\n")
nodes := make(map[string]*BasicNode)
for _, line := range lines {
var edgeName string
if idx := strings.Index(line, ";"); idx >= 0 {
edgeName = strings.Trim(line[idx+1:], " \t\r\n")
line = line[:idx]
}
parts := strings.SplitN(line, "->", 2)
if len(parts) != 2 {
continue
}
head_name := strings.Trim(parts[0], " \t\r\n")
tail_name := strings.Trim(parts[1], " \t\r\n")
head := nodes[head_name]
if head == nil {
head = &BasicNode{Name: head_name}
nodes[head_name] = head
}
tail := nodes[tail_name]
if tail == nil {
tail = &BasicNode{Name: tail_name}
nodes[tail_name] = tail
}
edge := &BasicEdge{
Name: edgeName,
EdgeHead: head,
EdgeTail: tail,
}
head.AddEdge(edge)
}
return nodes
}

53
digraph/basic_test.go Normal file
View File

@ -0,0 +1,53 @@
package digraph
import (
"fmt"
"testing"
)
func TestParseBasic(t *testing.T) {
spec := `a -> b ; first
b -> c ; second
b -> d ; third
z -> a`
nodes := ParseBasic(spec)
if len(nodes) != 5 {
t.Fatalf("bad: %v", nodes)
}
a := nodes["a"]
if a.Name != "a" {
t.Fatalf("bad: %v", a)
}
aEdges := a.Edges()
if len(aEdges) != 1 {
t.Fatalf("bad: %v", a.Edges())
}
if fmt.Sprintf("%v", aEdges[0]) != "first" {
t.Fatalf("bad: %v", aEdges[0])
}
b := nodes["b"]
if len(b.Edges()) != 2 {
t.Fatalf("bad: %v", b.Edges())
}
c := nodes["c"]
if len(c.Edges()) != 0 {
t.Fatalf("bad: %v", c.Edges())
}
d := nodes["d"]
if len(d.Edges()) != 0 {
t.Fatalf("bad: %v", d.Edges())
}
z := nodes["z"]
zEdges := z.Edges()
if len(zEdges) != 1 {
t.Fatalf("bad: %v", z.Edges())
}
if fmt.Sprintf("%v", zEdges[0]) != "Edge" {
t.Fatalf("bad: %v", zEdges[0])
}
}

34
digraph/digraph.go Normal file
View File

@ -0,0 +1,34 @@
package digraph
// Digraph is used to represent a Directed Graph. This means
// we have a set of nodes, and a set of edges which are directed
// from a source and towards a destination
type Digraph interface {
// Nodes provides all the nodes in the graph
Nodes() []Node
// Sources provides all the source nodes in the graph
Sources() []Node
// Sinks provides all the sink nodes in the graph
Sinks() []Node
// Transpose reverses the edge directions and returns
// a new Digraph
Transpose() Digraph
}
// Node represents a vertex in a Digraph
type Node interface {
// Edges returns the out edges for a given nod
Edges() []Edge
}
// Edge represents a directed edge in a Digraph
type Edge interface {
// Head returns the start point of the Edge
Head() Node
// Tail returns the end point of the Edge
Tail() Node
}

22
digraph/graphviz.go Normal file
View File

@ -0,0 +1,22 @@
package digraph
import (
"fmt"
"io"
)
// GenerateDot is used to emit a GraphViz compatible definition
// for a directed graph. It can be used to dump a .dot file.
func GenerateDot(nodes []Node, w io.Writer) {
w.Write([]byte("digraph {\n"))
defer w.Write([]byte("}\n"))
for _, n := range nodes {
w.Write([]byte(fmt.Sprintf("\t%s;\n", n)))
for _, edge := range n.Edges() {
target := edge.Tail()
line := fmt.Sprintf("\t%s -> %s [label=\"%s\"];\n",
n, target, edge)
w.Write([]byte(line))
}
}
}

57
digraph/graphviz_test.go Normal file
View File

@ -0,0 +1,57 @@
package digraph
import (
"bytes"
"strings"
"testing"
)
func Test_GenerateDot(t *testing.T) {
nodes := ParseBasic(`a -> b ; foo
a -> c
b -> d
b -> e
`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
buf := bytes.NewBuffer(nil)
GenerateDot(nlist, buf)
out := string(buf.Bytes())
if !strings.HasPrefix(out, "digraph {\n") {
t.Fatalf("bad: %v", out)
}
if !strings.HasSuffix(out, "\n}\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\ta;\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\tb;\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\tc;\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\td;\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\te;\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\ta -> b [label=\"foo\"];\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\ta -> c [label=\"Edge\"];\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\tb -> d [label=\"Edge\"];\n") {
t.Fatalf("bad: %v", out)
}
if !strings.Contains(out, "\n\tb -> e [label=\"Edge\"];\n") {
t.Fatalf("bad: %v", out)
}
}

111
digraph/tarjan.go Normal file
View File

@ -0,0 +1,111 @@
package digraph
// sccAcct is used ot pass around accounting information for
// the StronglyConnectedComponents algorithm
type sccAcct struct {
ExcludeSingle bool
NextIndex int
NodeIndex map[Node]int
Stack []Node
SCC [][]Node
}
// visit assigns an index and pushes a node onto the stack
func (s *sccAcct) visit(n Node) int {
idx := s.NextIndex
s.NodeIndex[n] = idx
s.NextIndex++
s.push(n)
return idx
}
// push adds a node to the stack
func (s *sccAcct) push(n Node) {
s.Stack = append(s.Stack, n)
}
// pop removes a node from the stack
func (s *sccAcct) pop() Node {
n := len(s.Stack)
if n == 0 {
return nil
}
node := s.Stack[n-1]
s.Stack = s.Stack[:n-1]
return node
}
// inStack checks if a node is in the stack
func (s *sccAcct) inStack(needle Node) bool {
for _, n := range s.Stack {
if n == needle {
return true
}
}
return false
}
// StronglyConnectedComponents implements Tarjan's algorithm to
// find all the strongly connected components in a graph. This can
// be used to detected any cycles in a graph, as well as which nodes
// partipate in those cycles. excludeSingle is used to exclude strongly
// connected components of size one.
func StronglyConnectedComponents(nodes []Node, excludeSingle bool) [][]Node {
acct := sccAcct{
ExcludeSingle: excludeSingle,
NextIndex: 1,
NodeIndex: make(map[Node]int, len(nodes)),
}
for _, node := range nodes {
// Recurse on any non-visited nodes
if acct.NodeIndex[node] == 0 {
stronglyConnected(&acct, node)
}
}
return acct.SCC
}
func stronglyConnected(acct *sccAcct, node Node) int {
// Initial node visit
index := acct.visit(node)
minIdx := index
for _, edge := range node.Edges() {
target := edge.Tail()
targetIdx := acct.NodeIndex[target]
// Recurse on successor if not yet visited
if targetIdx == 0 {
minIdx = min(minIdx, stronglyConnected(acct, target))
} else if acct.inStack(target) {
// Check if the node is in the stack
minIdx = min(minIdx, targetIdx)
}
}
// Pop the strongly connected components off the stack if
// this is a root node
if index == minIdx {
var scc []Node
for {
n := acct.pop()
scc = append(scc, n)
if n == node {
break
}
}
if !(acct.ExcludeSingle && len(scc) == 1) {
acct.SCC = append(acct.SCC, scc)
}
}
return minIdx
}
func min(a, b int) int {
if a <= b {
return a
}
return b
}

75
digraph/tarjan_test.go Normal file
View File

@ -0,0 +1,75 @@
package digraph
import (
"testing"
)
func TestStronglyConnectedComponents(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
b -> c
c -> b
c -> d
d -> e`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
sccs := StronglyConnectedComponents(nlist, false)
if len(sccs) != 4 {
t.Fatalf("bad: %v", sccs)
}
sccs = StronglyConnectedComponents(nlist, true)
if len(sccs) != 1 {
t.Fatalf("bad: %v", sccs)
}
cycle := sccs[0]
if len(cycle) != 2 {
t.Fatalf("bad: %v", sccs)
}
if cycle[0].(*BasicNode).Name != "c" {
t.Fatalf("bad: %v", cycle)
}
if cycle[1].(*BasicNode).Name != "b" {
t.Fatalf("bad: %v", cycle)
}
}
func TestStronglyConnectedComponents2(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
b -> d
b -> e
c -> f
c -> g
g -> a
`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
sccs := StronglyConnectedComponents(nlist, true)
if len(sccs) != 1 {
t.Fatalf("bad: %v", sccs)
}
cycle := sccs[0]
if len(cycle) != 3 {
t.Fatalf("bad: %v", sccs)
}
if cycle[0].(*BasicNode).Name != "g" {
t.Fatalf("bad: %v", cycle)
}
if cycle[1].(*BasicNode).Name != "c" {
t.Fatalf("bad: %v", cycle)
}
if cycle[2].(*BasicNode).Name != "a" {
t.Fatalf("bad: %v", cycle)
}
}

113
digraph/util.go Normal file
View File

@ -0,0 +1,113 @@
package digraph
// DepthFirstWalk performs a depth-first traversal of the nodes
// that can be reached from the initial input set. The callback is
// invoked for each visited node, and may return false to prevent
// vising any children of the current node
func DepthFirstWalk(node Node, cb func(n Node) bool) {
frontier := []Node{node}
seen := make(map[Node]struct{})
for len(frontier) > 0 {
// Pop the current node
n := len(frontier)
current := frontier[n-1]
frontier = frontier[:n-1]
// Check for potential cycle
if _, ok := seen[current]; ok {
continue
}
seen[current] = struct{}{}
// Visit with the callback
if !cb(current) {
continue
}
// Add any new edges to visit, in reverse order
edges := current.Edges()
for i := len(edges) - 1; i >= 0; i-- {
frontier = append(frontier, edges[i].Tail())
}
}
}
// FilterDegree returns only the nodes with the desired
// degree. This can be used with OutDegree or InDegree
func FilterDegree(degree int, degrees map[Node]int) []Node {
var matching []Node
for n, d := range degrees {
if d == degree {
matching = append(matching, n)
}
}
return matching
}
// InDegree is used to compute the in-degree of nodes
func InDegree(nodes []Node) map[Node]int {
degree := make(map[Node]int, len(nodes))
for _, n := range nodes {
if _, ok := degree[n]; !ok {
degree[n] = 0
}
for _, e := range n.Edges() {
degree[e.Tail()]++
}
}
return degree
}
// OutDegree is used to compute the in-degree of nodes
func OutDegree(nodes []Node) map[Node]int {
degree := make(map[Node]int, len(nodes))
for _, n := range nodes {
degree[n] = len(n.Edges())
}
return degree
}
// Sinks is used to get the nodes with out-degree of 0
func Sinks(nodes []Node) []Node {
return FilterDegree(0, OutDegree(nodes))
}
// Sources is used to get the nodes with in-degree of 0
func Sources(nodes []Node) []Node {
return FilterDegree(0, InDegree(nodes))
}
// Unreachable starts at a given start node, performs
// a DFS from there, and returns the set of unreachable nodes.
func Unreachable(start Node, nodes []Node) []Node {
// DFS from the start ndoe
frontier := []Node{start}
seen := make(map[Node]struct{})
for len(frontier) > 0 {
// Pop the current node
n := len(frontier)
current := frontier[n-1]
frontier = frontier[:n-1]
// Check for potential cycle
if _, ok := seen[current]; ok {
continue
}
seen[current] = struct{}{}
// Add any new edges to visit, in reverse order
edges := current.Edges()
for i := len(edges) - 1; i >= 0; i-- {
frontier = append(frontier, edges[i].Tail())
}
}
// Check for any unseen nodes
var unseen []Node
for _, node := range nodes {
if _, ok := seen[node]; !ok {
unseen = append(unseen, node)
}
}
return unseen
}

233
digraph/util_test.go Normal file
View File

@ -0,0 +1,233 @@
package digraph
import (
"reflect"
"testing"
)
func TestDepthFirstWalk(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
a -> d
b -> e
d -> f
e -> a ; cycle`)
root := nodes["a"]
expected := []string{
"a",
"b",
"e",
"c",
"d",
"f",
}
index := 0
DepthFirstWalk(root, func(n Node) bool {
name := n.(*BasicNode).Name
if expected[index] != name {
t.Fatalf("expected: %v, got %v", expected[index], name)
}
index++
return true
})
}
func TestInDegree(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
a -> d
b -> e
c -> e
d -> f`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
expected := map[string]int{
"a": 0,
"b": 1,
"c": 1,
"d": 1,
"e": 2,
"f": 1,
}
indegree := InDegree(nlist)
for n, d := range indegree {
name := n.(*BasicNode).Name
exp := expected[name]
if exp != d {
t.Fatalf("Expected %d for %s, got %d",
exp, name, d)
}
}
}
func TestOutDegree(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
a -> d
b -> e
c -> e
d -> f`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
expected := map[string]int{
"a": 3,
"b": 1,
"c": 1,
"d": 1,
"e": 0,
"f": 0,
}
outDegree := OutDegree(nlist)
for n, d := range outDegree {
name := n.(*BasicNode).Name
exp := expected[name]
if exp != d {
t.Fatalf("Expected %d for %s, got %d",
exp, name, d)
}
}
}
func TestSinks(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
a -> d
b -> e
c -> e
d -> f`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
sinks := Sinks(nlist)
var haveE, haveF bool
for _, n := range sinks {
name := n.(*BasicNode).Name
switch name {
case "e":
haveE = true
case "f":
haveF = true
}
}
if !haveE || !haveF {
t.Fatalf("missing sink")
}
}
func TestSources(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
a -> d
b -> e
c -> e
d -> f
x -> y`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
sources := Sources(nlist)
if len(sources) != 2 {
t.Fatalf("bad: %v", sources)
}
var haveA, haveX bool
for _, n := range sources {
name := n.(*BasicNode).Name
switch name {
case "a":
haveA = true
case "x":
haveX = true
}
}
if !haveA || !haveX {
t.Fatalf("missing source %v %v", haveA, haveX)
}
}
func TestUnreachable(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
a -> d
b -> e
c -> e
d -> f
f -> a
x -> y
y -> z`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
unreached := Unreachable(nodes["a"], nlist)
if len(unreached) != 3 {
t.Fatalf("bad: %v", unreached)
}
var haveX, haveY, haveZ bool
for _, n := range unreached {
name := n.(*BasicNode).Name
switch name {
case "x":
haveX = true
case "y":
haveY = true
case "z":
haveZ = true
}
}
if !haveX || !haveY || !haveZ {
t.Fatalf("missing %v %v %v", haveX, haveY, haveZ)
}
}
func TestUnreachable2(t *testing.T) {
nodes := ParseBasic(`a -> b
a -> c
a -> d
b -> e
c -> e
d -> f
f -> a
x -> y
y -> z`)
var nlist []Node
for _, n := range nodes {
nlist = append(nlist, n)
}
unreached := Unreachable(nodes["x"], nlist)
if len(unreached) != 6 {
t.Fatalf("bad: %v", unreached)
}
expected := map[string]struct{}{
"a": struct{}{},
"b": struct{}{},
"c": struct{}{},
"d": struct{}{},
"e": struct{}{},
"f": struct{}{},
}
out := map[string]struct{}{}
for _, n := range unreached {
name := n.(*BasicNode).Name
out[name] = struct{}{}
}
if !reflect.DeepEqual(out, expected) {
t.Fatalf("bad: %v %v", out, expected)
}
}