cmd/digraph/digraph.go - tools - Git at Google

 // Copyright 2019 The Go Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.

 /*
 The digraph command performs queries over unlabelled directed graphs
 represented in text form.  It is intended to integrate nicely with
 typical UNIX command pipelines.

 Usage:

 	your-application | digraph [command]

 The support commands are:

 	nodes
 		the set of all nodes
 	degree
 		the in-degree and out-degree of each node
 	transpose
 		the reverse of the input edges
 	preds <node> ...
 		the set of immediate predecessors of the specified nodes
 	succs <node> ...
 		the set of immediate successors of the specified nodes
 	forward <node> ...
 		the set of nodes transitively reachable from the specified nodes
 	reverse <node> ...
 		the set of nodes that transitively reach the specified nodes
 	somepath <node> <node>
 		the list of nodes on some arbitrary path from the first node to the second
 	allpaths <node> <node>
 		the set of nodes on all paths from the first node to the second
 	sccs
 		all strongly connected components (one per line)
 	scc <node>
 		the set of nodes nodes strongly connected to the specified one
 	focus <node>
 		the subgraph containing all directed paths that pass through the specified node

 Input format:

 Each line contains zero or more words. Words are separated by unquoted
 whitespace; words may contain Go-style double-quoted portions, allowing spaces
 and other characters to be expressed.

 Each word declares a node, and if there are more than one, an edge from the
 first to each subsequent one. The graph is provided on the standard input.

 For instance, the following (acyclic) graph specifies a partial order among the
 subtasks of getting dressed:

 	$ cat clothes.txt
 	socks shoes
 	"boxer shorts" pants
 	pants belt shoes
 	shirt tie sweater
 	sweater jacket
 	hat

 The line "shirt tie sweater" indicates the two edges shirt -> tie and
 shirt -> sweater, not shirt -> tie -> sweater.

 Example usage:

 Using digraph with existing Go tools:

 	$ go mod graph | digraph nodes # Operate on the Go module graph.
 	$ go list -m all | digraph nodes # Operate on the Go package graph.

 Show the transitive closure of imports of the digraph tool itself:
 	$ go list -f '{{.ImportPath}} {{join .Imports " "}}' ... | digraph forward golang.org/x/tools/cmd/digraph

 Show which clothes (see above) must be donned before a jacket:
 	$ digraph reverse jacket

 */
 package main // import "golang.org/x/tools/cmd/digraph"

 // TODO(adonovan):
 // - support input files other than stdin
 // - support alternative formats (AT&T GraphViz, CSV, etc),
 //   a comment syntax, etc.
 // - allow queries to nest, like Blaze query language.

 import (
 	"bufio"
 	"bytes"
 	"errors"
 	"flag"
 	"fmt"
 	"io"
 	"os"
 	"sort"
 	"strconv"
 	"strings"
 	"unicode"
 	"unicode/utf8"
 )

 func usage() {
 	fmt.Fprintf(os.Stderr, `Usage: your-application | digraph [command]

 The support commands are:
 	nodes
 		the set of all nodes
 	degree
 		the in-degree and out-degree of each node
 	transpose
 		the reverse of the input edges
 	preds <node> ...
 		the set of immediate predecessors of the specified nodes
 	succs <node> ...
 		the set of immediate successors of the specified nodes
 	forward <node> ...
 		the set of nodes transitively reachable from the specified nodes
 	reverse <node> ...
 		the set of nodes that transitively reach the specified nodes
 	somepath <node> <node>
 		the list of nodes on some arbitrary path from the first node to the second
 	allpaths <node> <node>
 		the set of nodes on all paths from the first node to the second
 	sccs
 		all strongly connected components (one per line)
 	scc <node>
 		the set of nodes nodes strongly connected to the specified one
 	focus <node>
 		the subgraph containing all directed paths that pass through the specified node
 `)
 	os.Exit(2)
 }

 func main() {
 	flag.Usage = usage
 	flag.Parse()

 	args := flag.Args()
 	if len(args) == 0 {
 		usage()
 	}

 	if err := digraph(args[0], args[1:]); err != nil {
 		fmt.Fprintf(os.Stderr, "digraph: %s\n", err)
 		os.Exit(1)
 	}
 }

 type nodelist []string

 func (l nodelist) println(sep string) {
 	for i, node := range l {
 		if i > 0 {
 			fmt.Fprint(stdout, sep)
 		}
 		fmt.Fprint(stdout, node)
 	}
 	fmt.Fprintln(stdout)
 }

 type nodeset map[string]bool // TODO(deklerk): change bool to struct to reduce memory footprint

 func (s nodeset) sort() nodelist {
 	nodes := make(nodelist, len(s))
 	var i int
 	for node := range s {
 		nodes[i] = node
 		i++
 	}
 	sort.Strings(nodes)
 	return nodes
 }

 func (s nodeset) addAll(x nodeset) {
 	for node := range x {
 		s[node] = true
 	}
 }

 // A graph maps nodes to the non-nil set of their immediate successors.
 type graph map[string]nodeset

 func (g graph) addNode(node string) nodeset {
 	edges := g[node]
 	if edges == nil {
 		edges = make(nodeset)
 		g[node] = edges
 	}
 	return edges
 }

 func (g graph) addEdges(from string, to ...string) {
 	edges := g.addNode(from)
 	for _, to := range to {
 		g.addNode(to)
 		edges[to] = true
 	}
 }

 func (g graph) reachableFrom(roots nodeset) nodeset {
 	seen := make(nodeset)
 	var visit func(node string)
 	visit = func(node string) {
 		if !seen[node] {
 			seen[node] = true
 			for e := range g[node] {
 				visit(e)
 			}
 		}
 	}
 	for root := range roots {
 		visit(root)
 	}
 	return seen
 }

 func (g graph) transpose() graph {
 	rev := make(graph)
 	for node, edges := range g {
 		rev.addNode(node)
 		for succ := range edges {
 			rev.addEdges(succ, node)
 		}
 	}
 	return rev
 }

 func (g graph) sccs() []nodeset {
 	// Kosaraju's algorithm---Tarjan is overkill here.

 	// Forward pass.
 	S := make(nodelist, 0, len(g)) // postorder stack
 	seen := make(nodeset)
 	var visit func(node string)
 	visit = func(node string) {
 		if !seen[node] {
 			seen[node] = true
 			for e := range g[node] {
 				visit(e)
 			}
 			S = append(S, node)
 		}
 	}
 	for node := range g {
 		visit(node)
 	}

 	// Reverse pass.
 	rev := g.transpose()
 	var scc nodeset
 	seen = make(nodeset)
 	var rvisit func(node string)
 	rvisit = func(node string) {
 		if !seen[node] {
 			seen[node] = true
 			scc[node] = true
 			for e := range rev[node] {
 				rvisit(e)
 			}
 		}
 	}
 	var sccs []nodeset
 	for len(S) > 0 {
 		top := S[len(S)-1]
 		S = S[:len(S)-1] // pop
 		if !seen[top] {
 			scc = make(nodeset)
 			rvisit(top)
 			sccs = append(sccs, scc)
 		}
 	}
 	return sccs
 }

 func (g graph) allpaths(from, to string) error {
 	// Mark all nodes to "to".
 	seen := make(nodeset) // value of seen[x] indicates whether x is on some path to "to"
 	var visit func(node string) bool
 	visit = func(node string) bool {
 		reachesTo, ok := seen[node]
 		if !ok {
 			reachesTo = node == to
 			seen[node] = reachesTo
 			for e := range g[node] {
 				if visit(e) {
 					reachesTo = true
 				}
 			}
 			if reachesTo && node != to {
 				seen[node] = true
 			}
 		}
 		return reachesTo
 	}
 	visit(from)

 	// For each marked node, collect its marked successors.
 	var edges []string
 	for n := range seen {
 		for succ := range g[n] {
 			if seen[succ] {
 				edges = append(edges, n+" "+succ)
 			}
 		}
 	}

 	// Sort (so that this method is deterministic) and print edges.
 	sort.Strings(edges)
 	for _, e := range edges {
 		fmt.Fprintln(stdout, e)
 	}

 	return nil
 }

 func (g graph) somepath(from, to string) error {
 	type edge struct{ from, to string }
 	seen := make(nodeset)
 	var dfs func(path []edge, from string) bool
 	dfs = func(path []edge, from string) bool {
 		if !seen[from] {
 			seen[from] = true
 			if from == to {
 				// fmt.Println(path, len(path), cap(path))
 				// Print and unwind.
 				for _, e := range path {
 					fmt.Fprintln(stdout, e.from+" "+e.to)
 				}
 				return true
 			}
 			for e := range g[from] {
 				if dfs(append(path, edge{from: from, to: e}), e) {
 					return true
 				}
 			}
 		}
 		return false
 	}
 	maxEdgesInGraph := len(g) * (len(g) - 1)
 	if !dfs(make([]edge, 0, maxEdgesInGraph), from) {
 		return fmt.Errorf("no path from %q to %q", from, to)
 	}
 	return nil
 }

 func parse(rd io.Reader) (graph, error) {
 	g := make(graph)

 	var linenum int
 	in := bufio.NewScanner(rd)
 	for in.Scan() {
 		linenum++
 		// Split into words, honoring double-quotes per Go spec.
 		words, err := split(in.Text())
 		if err != nil {
 			return nil, fmt.Errorf("at line %d: %v", linenum, err)
 		}
 		if len(words) > 0 {
 			g.addEdges(words[0], words[1:]...)
 		}
 	}
 	if err := in.Err(); err != nil {
 		return nil, err
 	}
 	return g, nil
 }

 // Overridable for testing purposes.
 var stdin io.Reader = os.Stdin
 var stdout io.Writer = os.Stdout

 func digraph(cmd string, args []string) error {
 	// Parse the input graph.
 	g, err := parse(stdin)
 	if err != nil {
 		return err
 	}

 	// Parse the command line.
 	switch cmd {
 	case "nodes":
 		if len(args) != 0 {
 			return fmt.Errorf("usage: digraph nodes")
 		}
 		nodes := make(nodeset)
 		for node := range g {
 			nodes[node] = true
 		}
 		nodes.sort().println("\n")

 	case "degree":
 		if len(args) != 0 {
 			return fmt.Errorf("usage: digraph degree")
 		}
 		nodes := make(nodeset)
 		for node := range g {
 			nodes[node] = true
 		}
 		rev := g.transpose()
 		for _, node := range nodes.sort() {
 			fmt.Fprintf(stdout, "%d\t%d\t%s\n", len(rev[node]), len(g[node]), node)
 		}

 	case "transpose":
 		if len(args) != 0 {
 			return fmt.Errorf("usage: digraph transpose")
 		}
 		var revEdges []string
 		for node, succs := range g.transpose() {
 			for succ := range succs {
 				revEdges = append(revEdges, fmt.Sprintf("%s %s", node, succ))
 			}
 		}
 		sort.Strings(revEdges) // make output deterministic
 		for _, e := range revEdges {
 			fmt.Fprintln(stdout, e)
 		}

 	case "succs", "preds":
 		if len(args) == 0 {
 			return fmt.Errorf("usage: digraph %s <node> ...", cmd)
 		}
 		g := g
 		if cmd == "preds" {
 			g = g.transpose()
 		}
 		result := make(nodeset)
 		for _, root := range args {
 			edges := g[root]
 			if edges == nil {
 				return fmt.Errorf("no such node %q", root)
 			}
 			result.addAll(edges)
 		}
 		result.sort().println("\n")

 	case "forward", "reverse":
 		if len(args) == 0 {
 			return fmt.Errorf("usage: digraph %s <node> ...", cmd)
 		}
 		roots := make(nodeset)
 		for _, root := range args {
 			if g[root] == nil {
 				return fmt.Errorf("no such node %q", root)
 			}
 			roots[root] = true
 		}
 		g := g
 		if cmd == "reverse" {
 			g = g.transpose()
 		}
 		g.reachableFrom(roots).sort().println("\n")

 	case "somepath":
 		if len(args) != 2 {
 			return fmt.Errorf("usage: digraph somepath <from> <to>")
 		}
 		from, to := args[0], args[1]
 		if g[from] == nil {
 			return fmt.Errorf("no such 'from' node %q", from)
 		}
 		if g[to] == nil {
 			return fmt.Errorf("no such 'to' node %q", to)
 		}
 		if err := g.somepath(from, to); err != nil {
 			return err
 		}

 	case "allpaths":
 		if len(args) != 2 {
 			return fmt.Errorf("usage: digraph allpaths <from> <to>")
 		}
 		from, to := args[0], args[1]
 		if g[from] == nil {
 			return fmt.Errorf("no such 'from' node %q", from)
 		}
 		if g[to] == nil {
 			return fmt.Errorf("no such 'to' node %q", to)
 		}
 		if err := g.allpaths(from, to); err != nil {
 			return err
 		}

 	case "sccs":
 		if len(args) != 0 {
 			return fmt.Errorf("usage: digraph sccs")
 		}
 		for _, scc := range g.sccs() {
 			scc.sort().println(" ")
 		}

 	case "scc":
 		if len(args) != 1 {
 			return fmt.Errorf("usage: digraph scc <node>")
 		}
 		node := args[0]
 		if g[node] == nil {
 			return fmt.Errorf("no such node %q", node)
 		}
 		for _, scc := range g.sccs() {
 			if scc[node] {
 				scc.sort().println("\n")
 				break
 			}
 		}

 	case "focus":
 		if len(args) != 1 {
 			return fmt.Errorf("usage: digraph focus <node>")
 		}
 		node := args[0]
 		if g[node] == nil {
 			return fmt.Errorf("no such node %q", node)
 		}

 		edges := make(map[string]struct{})
 		for from := range g.reachableFrom(nodeset{node: true}) {
 			for to := range g[from] {
 				edges[fmt.Sprintf("%s %s", from, to)] = struct{}{}
 			}
 		}

 		gtrans := g.transpose()
 		for from := range gtrans.reachableFrom(nodeset{node: true}) {
 			for to := range gtrans[from] {
 				edges[fmt.Sprintf("%s %s", to, from)] = struct{}{}
 			}
 		}

 		edgesSorted := make([]string, 0, len(edges))
 		for e := range edges {
 			edgesSorted = append(edgesSorted, e)
 		}
 		sort.Strings(edgesSorted)
 		fmt.Fprintln(stdout, strings.Join(edgesSorted, "\n"))

 	default:
 		return fmt.Errorf("no such command %q", cmd)
 	}

 	return nil
 }

 // -- Utilities --------------------------------------------------------

 // split splits a line into words, which are generally separated by
 // spaces, but Go-style double-quoted string literals are also supported.
 // (This approximates the behaviour of the Bourne shell.)
 //
 //   `one "two three"` -> ["one" "two three"]
 //   `a"\n"b` -> ["a\nb"]
 //
 func split(line string) ([]string, error) {
 	var (
 		words   []string
 		inWord  bool
 		current bytes.Buffer
 	)

 	for len(line) > 0 {
 		r, size := utf8.DecodeRuneInString(line)
 		if unicode.IsSpace(r) {
 			if inWord {
 				words = append(words, current.String())
 				current.Reset()
 				inWord = false
 			}
 		} else if r == '"' {
 			var ok bool
 			size, ok = quotedLength(line)
 			if !ok {
 				return nil, errors.New("invalid quotation")
 			}
 			s, err := strconv.Unquote(line[:size])
 			if err != nil {
 				return nil, err
 			}
 			current.WriteString(s)
 			inWord = true
 		} else {
 			current.WriteRune(r)
 			inWord = true
 		}
 		line = line[size:]
 	}
 	if inWord {
 		words = append(words, current.String())
 	}
 	return words, nil
 }

 // quotedLength returns the length in bytes of the prefix of input that
 // contain a possibly-valid double-quoted Go string literal.
 //
 // On success, n is at least two (""); input[:n] may be passed to
 // strconv.Unquote to interpret its value, and input[n:] contains the
 // rest of the input.
 //
 // On failure, quotedLength returns false, and the entire input can be
 // passed to strconv.Unquote if an informative error message is desired.
 //
 // quotedLength does not and need not detect all errors, such as
 // invalid hex or octal escape sequences, since it assumes
 // strconv.Unquote will be applied to the prefix.  It guarantees only
 // that if there is a prefix of input containing a valid string literal,
 // its length is returned.
 //
 // TODO(adonovan): move this into a strconv-like utility package.
 //
 func quotedLength(input string) (n int, ok bool) {
 	var offset int

 	// next returns the rune at offset, or -1 on EOF.
 	// offset advances to just after that rune.
 	next := func() rune {
 		if offset < len(input) {
 			r, size := utf8.DecodeRuneInString(input[offset:])
 			offset += size
 			return r
 		}
 		return -1
 	}

 	if next() != '"' {
 		return // error: not a quotation
 	}

 	for {
 		r := next()
 		if r == '\n' || r < 0 {
 			return // error: string literal not terminated
 		}
 		if r == '"' {
 			return offset, true // success
 		}
 		if r == '\\' {
 			var skip int
 			switch next() {
 			case 'a', 'b', 'f', 'n', 'r', 't', 'v', '\\', '"':
 				skip = 0
 			case '0', '1', '2', '3', '4', '5', '6', '7':
 				skip = 2
 			case 'x':
 				skip = 2
 			case 'u':
 				skip = 4
 			case 'U':
 				skip = 8
 			default:
 				return // error: invalid escape
 			}

 			for i := 0; i < skip; i++ {
 				next()
 			}
 		}
 	}
 }
	// Copyright 2019 The Go Authors. All rights reserved.
	// Use of this source code is governed by a BSD-style
	// license that can be found in the LICENSE file.

	/*
	The digraph command performs queries over unlabelled directed graphs
	represented in text form. It is intended to integrate nicely with
	typical UNIX command pipelines.

	Usage:

	your-application \| digraph [command]

	The support commands are:

	nodes
	the set of all nodes
	degree
	the in-degree and out-degree of each node
	transpose
	the reverse of the input edges
	preds <node> ...
	the set of immediate predecessors of the specified nodes
	succs <node> ...
	the set of immediate successors of the specified nodes
	forward <node> ...
	the set of nodes transitively reachable from the specified nodes
	reverse <node> ...
	the set of nodes that transitively reach the specified nodes
	somepath <node> <node>
	the list of nodes on some arbitrary path from the first node to the second
	allpaths <node> <node>
	the set of nodes on all paths from the first node to the second
	sccs
	all strongly connected components (one per line)
	scc <node>
	the set of nodes nodes strongly connected to the specified one
	focus <node>
	the subgraph containing all directed paths that pass through the specified node

	Input format:

	Each line contains zero or more words. Words are separated by unquoted
	whitespace; words may contain Go-style double-quoted portions, allowing spaces
	and other characters to be expressed.

	Each word declares a node, and if there are more than one, an edge from the
	first to each subsequent one. The graph is provided on the standard input.

	For instance, the following (acyclic) graph specifies a partial order among the
	subtasks of getting dressed:

	$ cat clothes.txt
	socks shoes
	"boxer shorts" pants
	pants belt shoes
	shirt tie sweater
	sweater jacket
	hat

	The line "shirt tie sweater" indicates the two edges shirt -> tie and
	shirt -> sweater, not shirt -> tie -> sweater.

	Example usage:

	Using digraph with existing Go tools:

	$ go mod graph \| digraph nodes # Operate on the Go module graph.
	$ go list -m all \| digraph nodes # Operate on the Go package graph.

	Show the transitive closure of imports of the digraph tool itself:
	$ go list -f '{{.ImportPath}} {{join .Imports " "}}' ... \| digraph forward golang.org/x/tools/cmd/digraph

	Show which clothes (see above) must be donned before a jacket:
	$ digraph reverse jacket

	*/
	package main // import "golang.org/x/tools/cmd/digraph"

	// TODO(adonovan):
	// - support input files other than stdin
	// - support alternative formats (AT&T GraphViz, CSV, etc),
	// a comment syntax, etc.
	// - allow queries to nest, like Blaze query language.

	import (
	"bufio"
	"bytes"
	"errors"
	"flag"
	"fmt"
	"io"
	"os"
	"sort"
	"strconv"
	"strings"
	"unicode"
	"unicode/utf8"
	)

	func usage() {
	fmt.Fprintf(os.Stderr, `Usage: your-application \| digraph [command]

	The support commands are:
	nodes
	the set of all nodes
	degree
	the in-degree and out-degree of each node
	transpose
	the reverse of the input edges
	preds <node> ...
	the set of immediate predecessors of the specified nodes
	succs <node> ...
	the set of immediate successors of the specified nodes
	forward <node> ...
	the set of nodes transitively reachable from the specified nodes
	reverse <node> ...
	the set of nodes that transitively reach the specified nodes
	somepath <node> <node>
	the list of nodes on some arbitrary path from the first node to the second
	allpaths <node> <node>
	the set of nodes on all paths from the first node to the second
	sccs
	all strongly connected components (one per line)
	scc <node>
	the set of nodes nodes strongly connected to the specified one
	focus <node>
	the subgraph containing all directed paths that pass through the specified node
	`)
	os.Exit(2)
	}

	func main() {
	flag.Usage = usage
	flag.Parse()

	args := flag.Args()
	if len(args) == 0 {
	usage()
	}

	if err := digraph(args[0], args[1:]); err != nil {
	fmt.Fprintf(os.Stderr, "digraph: %s\n", err)
	os.Exit(1)
	}
	}

	type nodelist []string

	func (l nodelist) println(sep string) {
	for i, node := range l {
	if i > 0 {
	fmt.Fprint(stdout, sep)
	}
	fmt.Fprint(stdout, node)
	}
	fmt.Fprintln(stdout)
	}

	type nodeset map[string]bool // TODO(deklerk): change bool to struct to reduce memory footprint

	func (s nodeset) sort() nodelist {
	nodes := make(nodelist, len(s))
	var i int
	for node := range s {
	nodes[i] = node
	i++
	}
	sort.Strings(nodes)
	return nodes
	}

	func (s nodeset) addAll(x nodeset) {
	for node := range x {
	s[node] = true
	}
	}

	// A graph maps nodes to the non-nil set of their immediate successors.
	type graph map[string]nodeset

	func (g graph) addNode(node string) nodeset {
	edges := g[node]
	if edges == nil {
	edges = make(nodeset)
	g[node] = edges
	}
	return edges
	}

	func (g graph) addEdges(from string, to ...string) {
	edges := g.addNode(from)
	for _, to := range to {
	g.addNode(to)
	edges[to] = true
	}
	}

	func (g graph) reachableFrom(roots nodeset) nodeset {
	seen := make(nodeset)
	var visit func(node string)
	visit = func(node string) {
	if !seen[node] {
	seen[node] = true
	for e := range g[node] {
	visit(e)
	}
	}
	}
	for root := range roots {
	visit(root)
	}
	return seen
	}

	func (g graph) transpose() graph {
	rev := make(graph)
	for node, edges := range g {
	rev.addNode(node)
	for succ := range edges {
	rev.addEdges(succ, node)
	}
	}
	return rev
	}

	func (g graph) sccs() []nodeset {
	// Kosaraju's algorithm---Tarjan is overkill here.

	// Forward pass.
	S := make(nodelist, 0, len(g)) // postorder stack
	seen := make(nodeset)
	var visit func(node string)
	visit = func(node string) {
	if !seen[node] {
	seen[node] = true
	for e := range g[node] {
	visit(e)
	}
	S = append(S, node)
	}
	}
	for node := range g {
	visit(node)
	}

	// Reverse pass.
	rev := g.transpose()
	var scc nodeset
	seen = make(nodeset)
	var rvisit func(node string)
	rvisit = func(node string) {
	if !seen[node] {
	seen[node] = true
	scc[node] = true
	for e := range rev[node] {
	rvisit(e)
	}
	}
	}
	var sccs []nodeset
	for len(S) > 0 {
	top := S[len(S)-1]
	S = S[:len(S)-1] // pop
	if !seen[top] {
	scc = make(nodeset)
	rvisit(top)
	sccs = append(sccs, scc)
	}
	}
	return sccs
	}

	func (g graph) allpaths(from, to string) error {
	// Mark all nodes to "to".
	seen := make(nodeset) // value of seen[x] indicates whether x is on some path to "to"
	var visit func(node string) bool
	visit = func(node string) bool {
	reachesTo, ok := seen[node]
	if !ok {
	reachesTo = node == to
	seen[node] = reachesTo
	for e := range g[node] {
	if visit(e) {
	reachesTo = true
	}
	}
	if reachesTo && node != to {
	seen[node] = true
	}
	}
	return reachesTo
	}
	visit(from)

	// For each marked node, collect its marked successors.
	var edges []string
	for n := range seen {
	for succ := range g[n] {
	if seen[succ] {
	edges = append(edges, n+" "+succ)
	}
	}
	}

	// Sort (so that this method is deterministic) and print edges.
	sort.Strings(edges)
	for _, e := range edges {
	fmt.Fprintln(stdout, e)
	}

	return nil
	}

	func (g graph) somepath(from, to string) error {
	type edge struct{ from, to string }
	seen := make(nodeset)
	var dfs func(path []edge, from string) bool
	dfs = func(path []edge, from string) bool {
	if !seen[from] {
	seen[from] = true
	if from == to {
	// fmt.Println(path, len(path), cap(path))
	// Print and unwind.
	for _, e := range path {
	fmt.Fprintln(stdout, e.from+" "+e.to)
	}
	return true
	}
	for e := range g[from] {
	if dfs(append(path, edge{from: from, to: e}), e) {
	return true
	}
	}
	}
	return false
	}
	maxEdgesInGraph := len(g) * (len(g) - 1)
	if !dfs(make([]edge, 0, maxEdgesInGraph), from) {
	return fmt.Errorf("no path from %q to %q", from, to)
	}
	return nil
	}

	func parse(rd io.Reader) (graph, error) {
	g := make(graph)

	var linenum int
	in := bufio.NewScanner(rd)
	for in.Scan() {
	linenum++
	// Split into words, honoring double-quotes per Go spec.
	words, err := split(in.Text())
	if err != nil {
	return nil, fmt.Errorf("at line %d: %v", linenum, err)
	}
	if len(words) > 0 {
	g.addEdges(words[0], words[1:]...)
	}
	}
	if err := in.Err(); err != nil {
	return nil, err
	}
	return g, nil
	}

	// Overridable for testing purposes.
	var stdin io.Reader = os.Stdin
	var stdout io.Writer = os.Stdout

	func digraph(cmd string, args []string) error {
	// Parse the input graph.
	g, err := parse(stdin)
	if err != nil {
	return err
	}

	// Parse the command line.
	switch cmd {
	case "nodes":
	if len(args) != 0 {
	return fmt.Errorf("usage: digraph nodes")
	}
	nodes := make(nodeset)
	for node := range g {
	nodes[node] = true
	}
	nodes.sort().println("\n")

	case "degree":
	if len(args) != 0 {
	return fmt.Errorf("usage: digraph degree")
	}
	nodes := make(nodeset)
	for node := range g {
	nodes[node] = true
	}
	rev := g.transpose()
	for _, node := range nodes.sort() {
	fmt.Fprintf(stdout, "%d\t%d\t%s\n", len(rev[node]), len(g[node]), node)
	}

	case "transpose":
	if len(args) != 0 {
	return fmt.Errorf("usage: digraph transpose")
	}
	var revEdges []string
	for node, succs := range g.transpose() {
	for succ := range succs {
	revEdges = append(revEdges, fmt.Sprintf("%s %s", node, succ))
	}
	}
	sort.Strings(revEdges) // make output deterministic
	for _, e := range revEdges {
	fmt.Fprintln(stdout, e)
	}

	case "succs", "preds":
	if len(args) == 0 {
	return fmt.Errorf("usage: digraph %s <node> ...", cmd)
	}
	g := g
	if cmd == "preds" {
	g = g.transpose()
	}
	result := make(nodeset)
	for _, root := range args {
	edges := g[root]
	if edges == nil {
	return fmt.Errorf("no such node %q", root)
	}
	result.addAll(edges)
	}
	result.sort().println("\n")

	case "forward", "reverse":
	if len(args) == 0 {
	return fmt.Errorf("usage: digraph %s <node> ...", cmd)
	}
	roots := make(nodeset)
	for _, root := range args {
	if g[root] == nil {
	return fmt.Errorf("no such node %q", root)
	}
	roots[root] = true
	}
	g := g
	if cmd == "reverse" {
	g = g.transpose()
	}
	g.reachableFrom(roots).sort().println("\n")

	case "somepath":
	if len(args) != 2 {
	return fmt.Errorf("usage: digraph somepath <from> <to>")
	}
	from, to := args[0], args[1]
	if g[from] == nil {
	return fmt.Errorf("no such 'from' node %q", from)
	}
	if g[to] == nil {
	return fmt.Errorf("no such 'to' node %q", to)
	}
	if err := g.somepath(from, to); err != nil {
	return err
	}

	case "allpaths":
	if len(args) != 2 {
	return fmt.Errorf("usage: digraph allpaths <from> <to>")
	}
	from, to := args[0], args[1]
	if g[from] == nil {
	return fmt.Errorf("no such 'from' node %q", from)
	}
	if g[to] == nil {
	return fmt.Errorf("no such 'to' node %q", to)
	}
	if err := g.allpaths(from, to); err != nil {
	return err
	}

	case "sccs":
	if len(args) != 0 {
	return fmt.Errorf("usage: digraph sccs")
	}
	for _, scc := range g.sccs() {
	scc.sort().println(" ")
	}

	case "scc":
	if len(args) != 1 {
	return fmt.Errorf("usage: digraph scc <node>")
	}
	node := args[0]
	if g[node] == nil {
	return fmt.Errorf("no such node %q", node)
	}
	for _, scc := range g.sccs() {
	if scc[node] {
	scc.sort().println("\n")
	break
	}
	}

	case "focus":
	if len(args) != 1 {
	return fmt.Errorf("usage: digraph focus <node>")
	}
	node := args[0]
	if g[node] == nil {
	return fmt.Errorf("no such node %q", node)
	}

	edges := make(map[string]struct{})
	for from := range g.reachableFrom(nodeset{node: true}) {
	for to := range g[from] {
	edges[fmt.Sprintf("%s %s", from, to)] = struct{}{}
	}
	}

	gtrans := g.transpose()
	for from := range gtrans.reachableFrom(nodeset{node: true}) {
	for to := range gtrans[from] {
	edges[fmt.Sprintf("%s %s", to, from)] = struct{}{}
	}
	}

	edgesSorted := make([]string, 0, len(edges))
	for e := range edges {
	edgesSorted = append(edgesSorted, e)
	}
	sort.Strings(edgesSorted)
	fmt.Fprintln(stdout, strings.Join(edgesSorted, "\n"))

	default:
	return fmt.Errorf("no such command %q", cmd)
	}

	return nil
	}

	// -- Utilities --------------------------------------------------------

	// split splits a line into words, which are generally separated by
	// spaces, but Go-style double-quoted string literals are also supported.
	// (This approximates the behaviour of the Bourne shell.)
	//
	// `one "two three"` -> ["one" "two three"]
	// `a"\n"b` -> ["a\nb"]
	//
	func split(line string) ([]string, error) {
	var (
	words []string
	inWord bool
	current bytes.Buffer
	)

	for len(line) > 0 {
	r, size := utf8.DecodeRuneInString(line)
	if unicode.IsSpace(r) {
	if inWord {
	words = append(words, current.String())
	current.Reset()
	inWord = false
	}
	} else if r == '"' {
	var ok bool
	size, ok = quotedLength(line)
	if !ok {
	return nil, errors.New("invalid quotation")
	}
	s, err := strconv.Unquote(line[:size])
	if err != nil {
	return nil, err
	}
	current.WriteString(s)
	inWord = true
	} else {
	current.WriteRune(r)
	inWord = true
	}
	line = line[size:]
	}
	if inWord {
	words = append(words, current.String())
	}
	return words, nil
	}

	// quotedLength returns the length in bytes of the prefix of input that
	// contain a possibly-valid double-quoted Go string literal.
	//
	// On success, n is at least two (""); input[:n] may be passed to
	// strconv.Unquote to interpret its value, and input[n:] contains the
	// rest of the input.
	//
	// On failure, quotedLength returns false, and the entire input can be
	// passed to strconv.Unquote if an informative error message is desired.
	//
	// quotedLength does not and need not detect all errors, such as
	// invalid hex or octal escape sequences, since it assumes
	// strconv.Unquote will be applied to the prefix. It guarantees only
	// that if there is a prefix of input containing a valid string literal,
	// its length is returned.
	//
	// TODO(adonovan): move this into a strconv-like utility package.
	//
	func quotedLength(input string) (n int, ok bool) {
	var offset int

	// next returns the rune at offset, or -1 on EOF.
	// offset advances to just after that rune.
	next := func() rune {
	if offset < len(input) {
	r, size := utf8.DecodeRuneInString(input[offset:])
	offset += size
	return r
	}
	return -1
	}

	if next() != '"' {
	return // error: not a quotation
	}

	for {
	r := next()
	if r == '\n' \|\| r < 0 {
	return // error: string literal not terminated
	}
	if r == '"' {
	return offset, true // success
	}
	if r == '\\' {
	var skip int
	switch next() {
	case 'a', 'b', 'f', 'n', 'r', 't', 'v', '\\', '"':
	skip = 0
	case '0', '1', '2', '3', '4', '5', '6', '7':
	skip = 2
	case 'x':
	skip = 2
	case 'u':
	skip = 4
	case 'U':
	skip = 8
	default:
	return // error: invalid escape
	}

	for i := 0; i < skip; i++ {
	next()
	}
	}
	}
	}