Merge pull request #1010 from hashicorp/f-ast-graph

core: formalize internals to do graph transformation steps and walking over an AST tree for execution

Makefile

@@ -40,6 +40,14 @@ updatedeps:
 		| sort -u \
 		| xargs go get -f -u -v
 
+cover:
+	@go tool cover 2>/dev/null; if [ $$? -eq 3 ]; then \
+		go get -u golang.org/x/tools/cmd/cover; \
+	fi
+	go test $(TEST) -coverprofile=coverage.out
+	go tool cover -html=coverage.out
+	rm coverage.out
+
 # vet runs the Go source code static analysis tool `vet` to find
 # any common errors.
 vet:

command/graph.go

@@ -58,11 +58,7 @@ func (c *GraphCommand) Run(args []string) int {
 		return 1
 	}
 
-	opts := &terraform.GraphDotOpts{
-		ModuleDepth: moduleDepth,
-	}
-
-	c.Ui.Output(terraform.GraphDot(g, opts))
+	c.Ui.Output(terraform.GraphDot(g, nil))
 
 	return 0
 }

command/graph_test.go

@@ -26,7 +26,7 @@ func TestGraph(t *testing.T) {
 	}
 
 	output := ui.OutputWriter.String()
-	if !strings.Contains(output, "digraph {") {
+	if !strings.Contains(output, "provider.test") {
 		t.Fatalf("doesn't look like digraph: %s", output)
 	}
 }
@@ -73,7 +73,7 @@ func TestGraph_noArgs(t *testing.T) {
 	}
 
 	output := ui.OutputWriter.String()
-	if !strings.Contains(output, "digraph {") {
+	if !strings.Contains(output, "provider.test") {
 		t.Fatalf("doesn't look like digraph: %s", output)
 	}
 }
@@ -99,7 +99,7 @@ func TestGraph_plan(t *testing.T) {
 	}
 
 	output := ui.OutputWriter.String()
-	if !strings.Contains(output, "digraph {") {
+	if !strings.Contains(output, "provider.test") {
 		t.Fatalf("doesn't look like digraph: %s", output)
 	}
 }

config/raw_config.go

@@ -92,6 +92,51 @@ func (r *RawConfig) Interpolate(vs map[string]ast.Variable) error {
 	})
 }
 
+// Merge merges another RawConfig into this one (overriding any conflicting
+// values in this config) and returns a new config. The original config
+// is not modified.
+func (r *RawConfig) Merge(other *RawConfig) *RawConfig {
+	// Merge the raw configurations
+	raw := make(map[string]interface{})
+	for k, v := range r.Raw {
+		raw[k] = v
+	}
+	for k, v := range other.Raw {
+		raw[k] = v
+	}
+
+	// Create the result
+	result, err := NewRawConfig(raw)
+	if err != nil {
+		panic(err)
+	}
+
+	// Merge the interpolated results
+	result.config = make(map[string]interface{})
+	for k, v := range r.config {
+		result.config[k] = v
+	}
+	for k, v := range other.config {
+		result.config[k] = v
+	}
+
+	// Build the unknown keys
+	unknownKeys := make(map[string]struct{})
+	for _, k := range r.unknownKeys {
+		unknownKeys[k] = struct{}{}
+	}
+	for _, k := range other.unknownKeys {
+		unknownKeys[k] = struct{}{}
+	}
+
+	result.unknownKeys = make([]string, 0, len(unknownKeys))
+	for k, _ := range unknownKeys {
+		result.unknownKeys = append(result.unknownKeys, k)
+	}
+
+	return result
+}
+
 func (r *RawConfig) init() error {
 	r.config = r.Raw
 	r.Interpolations = nil

config/raw_config_test.go

@@ -114,6 +114,87 @@ func TestRawConfig_double(t *testing.T) {
 	}
 }
 
+func TestRawConfig_merge(t *testing.T) {
+	raw1 := map[string]interface{}{
+		"foo": "${var.foo}",
+		"bar": "${var.bar}",
+	}
+
+	rc1, err := NewRawConfig(raw1)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	{
+		vars := map[string]ast.Variable{
+			"var.foo": ast.Variable{
+				Value: "foovalue",
+				Type:  ast.TypeString,
+			},
+			"var.bar": ast.Variable{
+				Value: "nope",
+				Type:  ast.TypeString,
+			},
+		}
+		if err := rc1.Interpolate(vars); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	raw2 := map[string]interface{}{
+		"bar": "${var.bar}",
+		"baz": "${var.baz}",
+	}
+
+	rc2, err := NewRawConfig(raw2)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	{
+		vars := map[string]ast.Variable{
+			"var.bar": ast.Variable{
+				Value: "barvalue",
+				Type:  ast.TypeString,
+			},
+			"var.baz": ast.Variable{
+				Value: UnknownVariableValue,
+				Type:  ast.TypeString,
+			},
+		}
+		if err := rc2.Interpolate(vars); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	// Merge the two
+	rc3 := rc1.Merge(rc2)
+
+	// Raw should be merged
+	raw3 := map[string]interface{}{
+		"foo": "${var.foo}",
+		"bar": "${var.bar}",
+		"baz": "${var.baz}",
+	}
+	if !reflect.DeepEqual(rc3.Raw, raw3) {
+		t.Fatalf("bad: %#v", rc3.Raw)
+	}
+
+	actual := rc3.Config()
+	expected := map[string]interface{}{
+		"foo": "foovalue",
+		"bar": "barvalue",
+	}
+	if !reflect.DeepEqual(actual, expected) {
+		t.Fatalf("bad: %#v", actual)
+	}
+
+	expectedKeys := []string{"baz"}
+	if !reflect.DeepEqual(rc3.UnknownKeys(), expectedKeys) {
+		t.Fatalf("bad: %#v", rc3.UnknownKeys())
+	}
+}
+
 func TestRawConfig_syntax(t *testing.T) {
 	raw := map[string]interface{}{
 		"foo": "${var",

dag/dag.go

@@ -0,0 +1,163 @@
+package dag
+
+import (
+	"fmt"
+	"strings"
+	"sync"
+
+	"github.com/hashicorp/go-multierror"
+)
+
+// AcyclicGraph is a specialization of Graph that cannot have cycles. With
+// this property, we get the property of sane graph traversal.
+type AcyclicGraph struct {
+	Graph
+}
+
+// WalkFunc is the callback used for walking the graph.
+type WalkFunc func(Vertex) error
+
+// Root returns the root of the DAG, or an error.
+//
+// Complexity: O(V)
+func (g *AcyclicGraph) Root() (Vertex, error) {
+	roots := make([]Vertex, 0, 1)
+	for _, v := range g.Vertices() {
+		if g.UpEdges(v).Len() == 0 {
+			roots = append(roots, v)
+		}
+	}
+
+	if len(roots) > 1 {
+		// TODO(mitchellh): make this error message a lot better
+		return nil, fmt.Errorf("multiple roots: %#v", roots)
+	}
+
+	if len(roots) == 0 {
+		return nil, fmt.Errorf("no roots found")
+	}
+
+	return roots[0], nil
+}
+
+// Validate validates the DAG. A DAG is valid if it has a single root
+// with no cycles.
+func (g *AcyclicGraph) Validate() error {
+	if _, err := g.Root(); err != nil {
+		return err
+	}
+
+	// Look for cycles of more than 1 component
+	var err error
+	var cycles [][]Vertex
+	for _, cycle := range StronglyConnected(&g.Graph) {
+		if len(cycle) > 1 {
+			cycles = append(cycles, cycle)
+		}
+	}
+	if len(cycles) > 0 {
+		for _, cycle := range cycles {
+			cycleStr := make([]string, len(cycle))
+			for j, vertex := range cycle {
+				cycleStr[j] = VertexName(vertex)
+			}
+
+			err = multierror.Append(err, fmt.Errorf(
+				"Cycle: %s", strings.Join(cycleStr, ", ")))
+		}
+	}
+
+	// Look for cycles to self
+	for _, e := range g.Edges() {
+		if e.Source() == e.Target() {
+			err = multierror.Append(err, fmt.Errorf(
+				"Self reference: %s", VertexName(e.Source())))
+		}
+	}
+
+	return err
+}
+
+// Walk walks the graph, calling your callback as each node is visited.
+// This will walk nodes in parallel if it can. Because the walk is done
+// in parallel, the error returned will be a multierror.
+func (g *AcyclicGraph) Walk(cb WalkFunc) error {
+	// Cache the vertices since we use it multiple times
+	vertices := g.Vertices()
+
+	// Build the waitgroup that signals when we're done
+	var wg sync.WaitGroup
+	wg.Add(len(vertices))
+	doneCh := make(chan struct{})
+	go func() {
+		defer close(doneCh)
+		wg.Wait()
+	}()
+
+	// The map of channels to watch to wait for vertices to finish
+	vertMap := make(map[Vertex]chan struct{})
+	for _, v := range vertices {
+		vertMap[v] = make(chan struct{})
+	}
+
+	// The map of whether a vertex errored or not during the walk
+	var errLock sync.Mutex
+	var errs error
+	errMap := make(map[Vertex]bool)
+	for _, v := range vertices {
+		// Build our list of dependencies and the list of channels to
+		// wait on until we start executing for this vertex.
+		depsRaw := g.DownEdges(v).List()
+		deps := make([]Vertex, len(depsRaw))
+		depChs := make([]<-chan struct{}, len(deps))
+		for i, raw := range depsRaw {
+			deps[i] = raw.(Vertex)
+			depChs[i] = vertMap[deps[i]]
+		}
+
+		// Get our channel so that we can close it when we're done
+		ourCh := vertMap[v]
+
+		// Start the goroutine to wait for our dependencies
+		readyCh := make(chan bool)
+		go func(deps []Vertex, chs []<-chan struct{}, readyCh chan<- bool) {
+			// First wait for all the dependencies
+			for _, ch := range chs {
+				<-ch
+			}
+
+			// Then, check the map to see if any of our dependencies failed
+			errLock.Lock()
+			defer errLock.Unlock()
+			for _, dep := range deps {
+				if errMap[dep] {
+					readyCh <- false
+					return
+				}
+			}
+
+			readyCh <- true
+		}(deps, depChs, readyCh)
+
+		// Start the goroutine that executes
+		go func(v Vertex, doneCh chan<- struct{}, readyCh <-chan bool) {
+			defer close(doneCh)
+			defer wg.Done()
+
+			var err error
+			if ready := <-readyCh; ready {
+				err = cb(v)
+			}
+
+			errLock.Lock()
+			defer errLock.Unlock()
+			if err != nil {
+				errMap[v] = true
+				errs = multierror.Append(errs, err)
+			}
+		}(v, ourCh, readyCh)
+	}
+
+	<-doneCh
+	return errs
+}

dag/dag_test.go

@@ -0,0 +1,158 @@
+package dag
+
+import (
+	"fmt"
+	"reflect"
+	"sync"
+	"testing"
+)
+
+func TestAcyclicGraphRoot(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(3, 2))
+	g.Connect(BasicEdge(3, 1))
+
+	if root, err := g.Root(); err != nil {
+		t.Fatalf("err: %s", err)
+	} else if root != 3 {
+		t.Fatalf("bad: %#v", root)
+	}
+}
+
+func TestAcyclicGraphRoot_cycle(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(1, 2))
+	g.Connect(BasicEdge(2, 3))
+	g.Connect(BasicEdge(3, 1))
+
+	if _, err := g.Root(); err == nil {
+		t.Fatal("should error")
+	}
+}
+
+func TestAcyclicGraphRoot_multiple(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(3, 2))
+
+	if _, err := g.Root(); err == nil {
+		t.Fatal("should error")
+	}
+}
+
+func TestAcyclicGraphValidate(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(3, 2))
+	g.Connect(BasicEdge(3, 1))
+
+	if err := g.Validate(); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+}
+
+func TestAcyclicGraphValidate_cycle(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(3, 2))
+	g.Connect(BasicEdge(3, 1))
+	g.Connect(BasicEdge(1, 2))
+	g.Connect(BasicEdge(2, 1))
+
+	if err := g.Validate(); err == nil {
+		t.Fatal("should error")
+	}
+}
+
+func TestAcyclicGraphValidate_cycleSelf(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Connect(BasicEdge(1, 1))
+
+	if err := g.Validate(); err == nil {
+		t.Fatal("should error")
+	}
+}
+
+func TestAcyclicGraphWalk(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(3, 2))
+	g.Connect(BasicEdge(3, 1))
+
+	var visits []Vertex
+	var lock sync.Mutex
+	err := g.Walk(func(v Vertex) error {
+		lock.Lock()
+		defer lock.Unlock()
+		visits = append(visits, v)
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	expected := [][]Vertex{
+		{1, 2, 3},
+		{2, 1, 3},
+	}
+	for _, e := range expected {
+		if reflect.DeepEqual(visits, e) {
+			return
+		}
+	}
+
+	t.Fatalf("bad: %#v", visits)
+}
+
+func TestAcyclicGraphWalk_error(t *testing.T) {
+	var g AcyclicGraph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(3, 2))
+	g.Connect(BasicEdge(3, 1))
+
+	var visits []Vertex
+	var lock sync.Mutex
+	err := g.Walk(func(v Vertex) error {
+		lock.Lock()
+		defer lock.Unlock()
+
+		if v == 2 {
+			return fmt.Errorf("error")
+		}
+
+		visits = append(visits, v)
+		return nil
+	})
+	if err == nil {
+		t.Fatal("should error")
+	}
+
+	expected := [][]Vertex{
+		{1},
+	}
+	for _, e := range expected {
+		if reflect.DeepEqual(visits, e) {
+			return
+		}
+	}
+
+	t.Fatalf("bad: %#v", visits)
+}

dag/edge.go

@@ -0,0 +1,37 @@
+package dag
+
+import (
+	"fmt"
+)
+
+// Edge represents an edge in the graph, with a source and target vertex.
+type Edge interface {
+	Source() Vertex
+	Target() Vertex
+
+	Hashable
+}
+
+// BasicEdge returns an Edge implementation that simply tracks the source
+// and target given as-is.
+func BasicEdge(source, target Vertex) Edge {
+	return &basicEdge{S: source, T: target}
+}
+
+// basicEdge is a basic implementation of Edge that has the source and
+// target vertex.
+type basicEdge struct {
+	S, T Vertex
+}
+
+func (e *basicEdge) Hashcode() interface{} {
+	return fmt.Sprintf("%p-%p", e.S, e.T)
+}
+
+func (e *basicEdge) Source() Vertex {
+	return e.S
+}
+
+func (e *basicEdge) Target() Vertex {
+	return e.T
+}

dag/edge_test.go

@@ -0,0 +1,26 @@
+package dag
+
+import (
+	"testing"
+)
+
+func TestBasicEdgeHashcode(t *testing.T) {
+	e1 := BasicEdge(1, 2)
+	e2 := BasicEdge(1, 2)
+	if e1.Hashcode() != e2.Hashcode() {
+		t.Fatalf("bad")
+	}
+}
+
+func TestBasicEdgeHashcode_pointer(t *testing.T) {
+	type test struct {
+		Value string
+	}
+
+	v1, v2 := &test{"foo"}, &test{"bar"}
+	e1 := BasicEdge(v1, v2)
+	e2 := BasicEdge(v1, v2)
+	if e1.Hashcode() != e2.Hashcode() {
+		t.Fatalf("bad")
+	}
+}

dag/graph.go

@@ -0,0 +1,219 @@
+package dag
+
+import (
+	"bytes"
+	"fmt"
+	"sort"
+	"sync"
+)
+
+// Graph is used to represent a dependency graph.
+type Graph struct {
+	vertices  *Set
+	edges     *Set
+	downEdges map[Vertex]*Set
+	upEdges   map[Vertex]*Set
+	once      sync.Once
+}
+
+// Vertex of the graph.
+type Vertex interface{}
+
+// NamedVertex is an optional interface that can be implemented by Vertex
+// to give it a human-friendly name that is used for outputting the graph.
+type NamedVertex interface {
+	Vertex
+	Name() string
+}
+
+// Vertices returns the list of all the vertices in the graph.
+func (g *Graph) Vertices() []Vertex {
+	list := g.vertices.List()
+	result := make([]Vertex, len(list))
+	for i, v := range list {
+		result[i] = v.(Vertex)
+	}
+
+	return result
+}
+
+// Edges returns the list of all the edges in the graph.
+func (g *Graph) Edges() []Edge {
+	list := g.edges.List()
+	result := make([]Edge, len(list))
+	for i, v := range list {
+		result[i] = v.(Edge)
+	}
+
+	return result
+}
+
+// Add adds a vertex to the graph. This is safe to call multiple time with
+// the same Vertex.
+func (g *Graph) Add(v Vertex) Vertex {
+	g.once.Do(g.init)
+	g.vertices.Add(v)
+	return v
+}
+
+// Remove removes a vertex from the graph. This will also remove any
+// edges with this vertex as a source or target.
+func (g *Graph) Remove(v Vertex) Vertex {
+	// Delete the vertex itself
+	g.vertices.Delete(v)
+
+	// Delete the edges to non-existent things
+	for _, target := range g.DownEdges(v).List() {
+		g.RemoveEdge(BasicEdge(v, target))
+	}
+	for _, source := range g.UpEdges(v).List() {
+		g.RemoveEdge(BasicEdge(source, v))
+	}
+
+	return nil
+}
+
+// Replace replaces the original Vertex with replacement. If the original
+// does not exist within the graph, then false is returned. Otherwise, true
+// is returned.
+func (g *Graph) Replace(original, replacement Vertex) bool {
+	// If we don't have the original, we can't do anything
+	if !g.vertices.Include(original) {
+		return false
+	}
+
+	// Add our new vertex, then copy all the edges
+	g.Add(replacement)
+	for _, target := range g.DownEdges(original).List() {
+		g.Connect(BasicEdge(replacement, target))
+	}
+	for _, source := range g.UpEdges(original).List() {
+		g.Connect(BasicEdge(source, replacement))
+	}
+
+	// Remove our old vertex, which will also remove all the edges
+	g.Remove(original)
+
+	return true
+}
+
+// RemoveEdge removes an edge from the graph.
+func (g *Graph) RemoveEdge(edge Edge) {
+	g.once.Do(g.init)
+
+	// Delete the edge from the set
+	g.edges.Delete(edge)
+
+	// Delete the up/down edges
+	if s, ok := g.downEdges[edge.Source()]; ok {
+		s.Delete(edge.Target())
+	}
+	if s, ok := g.upEdges[edge.Target()]; ok {
+		s.Delete(edge.Source())
+	}
+}
+
+// DownEdges returns the outward edges from the source Vertex v.
+func (g *Graph) DownEdges(v Vertex) *Set {
+	g.once.Do(g.init)
+	return g.downEdges[v]
+}
+
+// UpEdges returns the inward edges to the destination Vertex v.
+func (g *Graph) UpEdges(v Vertex) *Set {
+	g.once.Do(g.init)
+	return g.upEdges[v]
+}
+
+// Connect adds an edge with the given source and target. This is safe to
+// call multiple times with the same value. Note that the same value is
+// verified through pointer equality of the vertices, not through the
+// value of the edge itself.
+func (g *Graph) Connect(edge Edge) {
+	g.once.Do(g.init)
+
+	source := edge.Source()
+	target := edge.Target()
+
+	// Do we have this already? If so, don't add it again.
+	if s, ok := g.downEdges[source]; ok && s.Include(target) {
+		return
+	}
+
+	// Add the edge to the set
+	g.edges.Add(edge)
+
+	// Add the down edge
+	s, ok := g.downEdges[source]
+	if !ok {
+		s = new(Set)
+		g.downEdges[source] = s
+	}
+	s.Add(target)
+
+	// Add the up edge
+	s, ok = g.upEdges[target]
+	if !ok {
+		s = new(Set)
+		g.upEdges[target] = s
+	}
+	s.Add(source)
+}
+
+// String outputs some human-friendly output for the graph structure.
+func (g *Graph) String() string {
+	var buf bytes.Buffer
+
+	// Build the list of node names and a mapping so that we can more
+	// easily alphabetize the output to remain deterministic.
+	vertices := g.Vertices()
+	names := make([]string, 0, len(vertices))
+	mapping := make(map[string]Vertex, len(vertices))
+	for _, v := range vertices {
+		name := VertexName(v)
+		names = append(names, name)
+		mapping[name] = v
+	}
+	sort.Strings(names)
+
+	// Write each node in order...
+	for _, name := range names {
+		v := mapping[name]
+		targets := g.downEdges[v]
+
+		buf.WriteString(fmt.Sprintf("%s\n", name))
+
+		// Alphabetize dependencies
+		deps := make([]string, 0, targets.Len())
+		for _, target := range targets.List() {
+			deps = append(deps, VertexName(target))
+		}
+		sort.Strings(deps)
+
+		// Write dependencies
+		for _, d := range deps {
+			buf.WriteString(fmt.Sprintf("  %s\n", d))
+		}
+	}
+
+	return buf.String()
+}
+
+func (g *Graph) init() {
+	g.vertices = new(Set)
+	g.edges = new(Set)
+	g.downEdges = make(map[Vertex]*Set)
+	g.upEdges = make(map[Vertex]*Set)
+}
+
+// VertexName returns the name of a vertex.
+func VertexName(raw Vertex) string {
+	switch v := raw.(type) {
+	case NamedVertex:
+		return v.Name()
+	case fmt.Stringer:
+		return fmt.Sprintf("%s", v)
+	default:
+		return fmt.Sprintf("%v", v)
+	}
+}

dag/graph_test.go

@@ -0,0 +1,90 @@
+package dag
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestGraph_empty(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphEmptyStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func TestGraph_basic(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(1, 3))
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphBasicStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func TestGraph_remove(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(1, 3))
+	g.Remove(3)
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphRemoveStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func TestGraph_replace(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Add(3)
+	g.Connect(BasicEdge(1, 2))
+	g.Connect(BasicEdge(2, 3))
+	g.Replace(2, 42)
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphReplaceStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+const testGraphBasicStr = `
+1
+  3
+2
+3
+`
+
+const testGraphEmptyStr = `
+1
+2
+3
+`
+
+const testGraphRemoveStr = `
+1
+2
+`
+
+const testGraphReplaceStr = `
+1
+  42
+3
+42
+  3
+`

dag/set.go

@@ -0,0 +1,72 @@
+package dag
+
+import (
+	"sync"
+)
+
+// Set is a set data structure.
+type Set struct {
+	m    map[interface{}]interface{}
+	once sync.Once
+}
+
+// Hashable is the interface used by set to get the hash code of a value.
+// If this isn't given, then the value of the item being added to the set
+// itself is used as the comparison value.
+type Hashable interface {
+	Hashcode() interface{}
+}
+
+// Add adds an item to the set
+func (s *Set) Add(v interface{}) {
+	s.once.Do(s.init)
+	s.m[s.code(v)] = v
+}
+
+// Delete removes an item from the set.
+func (s *Set) Delete(v interface{}) {
+	s.once.Do(s.init)
+	delete(s.m, s.code(v))
+}
+
+// Include returns true/false of whether a value is in the set.
+func (s *Set) Include(v interface{}) bool {
+	s.once.Do(s.init)
+	_, ok := s.m[s.code(v)]
+	return ok
+}
+
+// Len is the number of items in the set.
+func (s *Set) Len() int {
+	if s == nil {
+		return 0
+	}
+
+	return len(s.m)
+}
+
+// List returns the list of set elements.
+func (s *Set) List() []interface{} {
+	if s == nil {
+		return nil
+	}
+
+	r := make([]interface{}, 0, len(s.m))
+	for _, v := range s.m {
+		r = append(r, v)
+	}
+
+	return r
+}
+
+func (s *Set) code(v interface{}) interface{} {
+	if h, ok := v.(Hashable); ok {
+		return h.Hashcode()
+	}
+
+	return v
+}
+
+func (s *Set) init() {
+	s.m = make(map[interface{}]interface{})
+}

dag/tarjan.go

@@ -0,0 +1,107 @@
+package dag
+
+// StronglyConnected returns the list of strongly connected components
+// within the Graph g. This information is primarily used by this package
+// for cycle detection, but strongly connected components have widespread
+// use.
+func StronglyConnected(g *Graph) [][]Vertex {
+	vs := g.Vertices()
+	acct := sccAcct{
+		NextIndex:   1,
+		VertexIndex: make(map[Vertex]int, len(vs)),
+	}
+	for _, v := range vs {
+		// Recurse on any non-visited nodes
+		if acct.VertexIndex[v] == 0 {
+			stronglyConnected(&acct, g, v)
+		}
+	}
+	return acct.SCC
+}
+
+func stronglyConnected(acct *sccAcct, g *Graph, v Vertex) int {
+	// Initial vertex visit
+	index := acct.visit(v)
+	minIdx := index
+
+	for _, raw := range g.DownEdges(v).List() {
+		target := raw.(Vertex)
+		targetIdx := acct.VertexIndex[target]
+
+		// Recurse on successor if not yet visited
+		if targetIdx == 0 {
+			minIdx = min(minIdx, stronglyConnected(acct, g, target))
+		} else if acct.inStack(target) {
+			// Check if the vertex is in the stack
+			minIdx = min(minIdx, targetIdx)
+		}
+	}
+
+	// Pop the strongly connected components off the stack if
+	// this is a root vertex
+	if index == minIdx {
+		var scc []Vertex
+		for {
+			v2 := acct.pop()
+			scc = append(scc, v2)
+			if v2 == v {
+				break
+			}
+		}
+
+		acct.SCC = append(acct.SCC, scc)
+	}
+
+	return minIdx
+}
+
+func min(a, b int) int {
+	if a <= b {
+		return a
+	}
+	return b
+}
+
+// sccAcct is used ot pass around accounting information for
+// the StronglyConnectedComponents algorithm
+type sccAcct struct {
+	NextIndex   int
+	VertexIndex map[Vertex]int
+	Stack       []Vertex
+	SCC         [][]Vertex
+}
+
+// visit assigns an index and pushes a vertex onto the stack
+func (s *sccAcct) visit(v Vertex) int {
+	idx := s.NextIndex
+	s.VertexIndex[v] = idx
+	s.NextIndex++
+	s.push(v)
+	return idx
+}
+
+// push adds a vertex to the stack
+func (s *sccAcct) push(n Vertex) {
+	s.Stack = append(s.Stack, n)
+}
+
+// pop removes a vertex from the stack
+func (s *sccAcct) pop() Vertex {
+	n := len(s.Stack)
+	if n == 0 {
+		return nil
+	}
+	vertex := s.Stack[n-1]
+	s.Stack = s.Stack[:n-1]
+	return vertex
+}
+
+// inStack checks if a vertex is in the stack
+func (s *sccAcct) inStack(needle Vertex) bool {
+	for _, n := range s.Stack {
+		if n == needle {
+			return true
+		}
+	}
+	return false
+}

dag/tarjan_test.go

@@ -0,0 +1,86 @@
+package dag
+
+import (
+	"sort"
+	"strings"
+	"testing"
+)
+
+func TestGraphStronglyConnected(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Connect(BasicEdge(1, 2))
+	g.Connect(BasicEdge(2, 1))
+
+	actual := strings.TrimSpace(testSCCStr(StronglyConnected(&g)))
+	expected := strings.TrimSpace(testGraphStronglyConnectedStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func TestGraphStronglyConnected_two(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Connect(BasicEdge(1, 2))
+	g.Connect(BasicEdge(2, 1))
+	g.Add(3)
+
+	actual := strings.TrimSpace(testSCCStr(StronglyConnected(&g)))
+	expected := strings.TrimSpace(testGraphStronglyConnectedTwoStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func TestGraphStronglyConnected_three(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Connect(BasicEdge(1, 2))
+	g.Connect(BasicEdge(2, 1))
+	g.Add(3)
+	g.Add(4)
+	g.Add(5)
+	g.Add(6)
+	g.Connect(BasicEdge(4, 5))
+	g.Connect(BasicEdge(5, 6))
+	g.Connect(BasicEdge(6, 4))
+
+	actual := strings.TrimSpace(testSCCStr(StronglyConnected(&g)))
+	expected := strings.TrimSpace(testGraphStronglyConnectedThreeStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func testSCCStr(list [][]Vertex) string {
+	var lines []string
+	for _, vs := range list {
+		result := make([]string, len(vs))
+		for i, v := range vs {
+			result[i] = VertexName(v)
+		}
+
+		sort.Strings(result)
+		lines = append(lines, strings.Join(result, ","))
+	}
+
+	sort.Strings(lines)
+	return strings.Join(lines, "\n")
+}
+
+const testGraphStronglyConnectedStr = `1,2`
+
+const testGraphStronglyConnectedTwoStr = `
+1,2
+3
+`
+
+const testGraphStronglyConnectedThreeStr = `
+1,2
+3
+4,5,6
+`

terraform/diff.go

@@ -333,6 +333,10 @@ func (d *InstanceDiff) Empty() bool {
 	return !d.Destroy && len(d.Attributes) == 0
 }
 
+func (d *InstanceDiff) GoString() string {
+	return fmt.Sprintf("*%#v", *d)
+}
+
 // RequiresNew returns true if the diff requires the creation of a new
 // resource (implying the destruction of the old).
 func (d *InstanceDiff) RequiresNew() bool {

terraform/eval.go

@@ -0,0 +1,59 @@
+package terraform
+
+import (
+	"log"
+	"strings"
+)
+
+// EvalNode is the interface that must be implemented by graph nodes to
+// evaluate/execute.
+type EvalNode interface {
+	// Eval evaluates this node with the given context. The second parameter
+	// are the argument values. These will match in order and 1-1 with the
+	// results of the Args() return value.
+	Eval(EvalContext) (interface{}, error)
+}
+
+// GraphNodeEvalable is the interface that graph nodes must implement
+// to enable valuation.
+type GraphNodeEvalable interface {
+	EvalTree() EvalNode
+}
+
+// EvalEarlyExitError is a special error return value that can be returned
+// by eval nodes that does an early exit.
+type EvalEarlyExitError struct{}
+
+func (EvalEarlyExitError) Error() string { return "early exit" }
+
+// Eval evaluates the given EvalNode with the given context, properly
+// evaluating all args in the correct order.
+func Eval(n EvalNode, ctx EvalContext) (interface{}, error) {
+	// Call the lower level eval which doesn't understand early exit,
+	// and if we early exit, it isn't an error.
+	result, err := EvalRaw(n, ctx)
+	if err != nil {
+		if _, ok := err.(EvalEarlyExitError); ok {
+			return nil, nil
+		}
+	}
+
+	return result, err
+}
+
+// EvalRaw is like Eval except that it returns all errors, even if they
+// signal something normal such as EvalEarlyExitError.
+func EvalRaw(n EvalNode, ctx EvalContext) (interface{}, error) {
+	path := "unknown"
+	if ctx != nil {
+		path = strings.Join(ctx.Path(), ".")
+	}
+
+	log.Printf("[DEBUG] %s: eval: %T", path, n)
+	output, err := n.Eval(ctx)
+	if err != nil {
+		log.Printf("[ERROR] %s: eval: %T, err: %s", path, n, err)
+	}
+
+	return output, err
+}

terraform/eval_apply.go

@@ -0,0 +1,297 @@
+package terraform
+
+import (
+	"fmt"
+	"log"
+	"strconv"
+
+	"github.com/hashicorp/go-multierror"
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalApply is an EvalNode implementation that writes the diff to
+// the full diff.
+type EvalApply struct {
+	Info      *InstanceInfo
+	State     **InstanceState
+	Diff      **InstanceDiff
+	Provider  *ResourceProvider
+	Output    **InstanceState
+	CreateNew *bool
+	Error     *error
+}
+
+// TODO: test
+func (n *EvalApply) Eval(ctx EvalContext) (interface{}, error) {
+	diff := *n.Diff
+	provider := *n.Provider
+	state := *n.State
+
+	// If we have no diff, we have nothing to do!
+	if diff.Empty() {
+		log.Printf(
+			"[DEBUG] apply: %s: diff is empty, doing nothing.", n.Info.Id)
+		return nil, nil
+	}
+
+	// Remove any output values from the diff
+	for k, ad := range diff.Attributes {
+		if ad.Type == DiffAttrOutput {
+			delete(diff.Attributes, k)
+		}
+	}
+
+	// If the state is nil, make it non-nil
+	if state == nil {
+		state = new(InstanceState)
+	}
+	state.init()
+
+	// Flag if we're creating a new instance
+	if n.CreateNew != nil {
+		*n.CreateNew = (state.ID == "" && !diff.Destroy) || diff.RequiresNew()
+	}
+
+	{
+		// Call pre-apply hook
+		err := ctx.Hook(func(h Hook) (HookAction, error) {
+			return h.PreApply(n.Info, state, diff)
+		})
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	// With the completed diff, apply!
+	log.Printf("[DEBUG] apply: %s: executing Apply", n.Info.Id)
+	state, err := provider.Apply(n.Info, state, diff)
+	if state == nil {
+		state = new(InstanceState)
+	}
+	state.init()
+
+	// Force the "id" attribute to be our ID
+	if state.ID != "" {
+		state.Attributes["id"] = state.ID
+	}
+
+	// If the value is the unknown variable value, then it is an error.
+	// In this case we record the error and remove it from the state
+	for ak, av := range state.Attributes {
+		if av == config.UnknownVariableValue {
+			err = multierror.Append(err, fmt.Errorf(
+				"Attribute with unknown value: %s", ak))
+			delete(state.Attributes, ak)
+		}
+	}
+
+	// Write the final state
+	if n.Output != nil {
+		*n.Output = state
+	}
+
+	// If there are no errors, then we append it to our output error
+	// if we have one, otherwise we just output it.
+	if err != nil {
+		if n.Error != nil {
+			*n.Error = multierror.Append(*n.Error, err)
+		} else {
+			return nil, err
+		}
+	}
+
+	return nil, nil
+}
+
+// EvalApplyPost is an EvalNode implementation that does the post-Apply work
+type EvalApplyPost struct {
+	Info  *InstanceInfo
+	State **InstanceState
+	Error *error
+}
+
+// TODO: test
+func (n *EvalApplyPost) Eval(ctx EvalContext) (interface{}, error) {
+	state := *n.State
+
+	{
+		// Call post-apply hook
+		err := ctx.Hook(func(h Hook) (HookAction, error) {
+			return h.PostApply(n.Info, state, *n.Error)
+		})
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	return nil, *n.Error
+}
+
+// EvalApplyProvisioners is an EvalNode implementation that executes
+// the provisioners for a resource.
+//
+// TODO(mitchellh): This should probably be split up into a more fine-grained
+// ApplyProvisioner (single) that is looped over.
+type EvalApplyProvisioners struct {
+	Info           *InstanceInfo
+	State          **InstanceState
+	Resource       *config.Resource
+	InterpResource *Resource
+	CreateNew      *bool
+	Tainted        *bool
+	Error          *error
+}
+
+// TODO: test
+func (n *EvalApplyProvisioners) Eval(ctx EvalContext) (interface{}, error) {
+	state := *n.State
+
+	if !*n.CreateNew {
+		// If we're not creating a new resource, then don't run provisioners
+		return nil, nil
+	}
+
+	if len(n.Resource.Provisioners) == 0 {
+		// We have no provisioners, so don't do anything
+		return nil, nil
+	}
+
+	if n.Error != nil && *n.Error != nil {
+		// We're already errored creating, so mark as tainted and continue
+		if n.Tainted != nil {
+			*n.Tainted = true
+		}
+
+		// We're already tainted, so just return out
+		return nil, nil
+	}
+
+	{
+		// Call pre hook
+		err := ctx.Hook(func(h Hook) (HookAction, error) {
+			return h.PreProvisionResource(n.Info, state)
+		})
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	// If there are no errors, then we append it to our output error
+	// if we have one, otherwise we just output it.
+	err := n.apply(ctx)
+	if n.Tainted != nil {
+		*n.Tainted = err != nil
+	}
+	if err != nil {
+		if n.Error != nil {
+			*n.Error = multierror.Append(*n.Error, err)
+		} else {
+			return nil, err
+		}
+	}
+
+	{
+		// Call post hook
+		err := ctx.Hook(func(h Hook) (HookAction, error) {
+			return h.PostProvisionResource(n.Info, state)
+		})
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	return nil, nil
+}
+
+func (n *EvalApplyProvisioners) apply(ctx EvalContext) error {
+	state := *n.State
+
+	// Store the original connection info, restore later
+	origConnInfo := state.Ephemeral.ConnInfo
+	defer func() {
+		state.Ephemeral.ConnInfo = origConnInfo
+	}()
+
+	for _, prov := range n.Resource.Provisioners {
+		// Get the provisioner
+		provisioner := ctx.Provisioner(prov.Type)
+
+		// Interpolate the provisioner config
+		provConfig, err := ctx.Interpolate(prov.RawConfig, n.InterpResource)
+		if err != nil {
+			return err
+		}
+
+		// Interpolate the conn info, since it may contain variables
+		connInfo, err := ctx.Interpolate(prov.ConnInfo, n.InterpResource)
+		if err != nil {
+			return err
+		}
+
+		// Merge the connection information
+		overlay := make(map[string]string)
+		if origConnInfo != nil {
+			for k, v := range origConnInfo {
+				overlay[k] = v
+			}
+		}
+		for k, v := range connInfo.Config {
+			switch vt := v.(type) {
+			case string:
+				overlay[k] = vt
+			case int64:
+				overlay[k] = strconv.FormatInt(vt, 10)
+			case int32:
+				overlay[k] = strconv.FormatInt(int64(vt), 10)
+			case int:
+				overlay[k] = strconv.FormatInt(int64(vt), 10)
+			case float32:
+				overlay[k] = strconv.FormatFloat(float64(vt), 'f', 3, 32)
+			case float64:
+				overlay[k] = strconv.FormatFloat(vt, 'f', 3, 64)
+			case bool:
+				overlay[k] = strconv.FormatBool(vt)
+			default:
+				overlay[k] = fmt.Sprintf("%v", vt)
+			}
+		}
+		state.Ephemeral.ConnInfo = overlay
+
+		{
+			// Call pre hook
+			err := ctx.Hook(func(h Hook) (HookAction, error) {
+				return h.PreProvision(n.Info, prov.Type)
+			})
+			if err != nil {
+				return err
+			}
+		}
+
+		// The output function
+		outputFn := func(msg string) {
+			ctx.Hook(func(h Hook) (HookAction, error) {
+				h.ProvisionOutput(n.Info, prov.Type, msg)
+				return HookActionContinue, nil
+			})
+		}
+
+		// Invoke the Provisioner
+		output := CallbackUIOutput{OutputFn: outputFn}
+		if err := provisioner.Apply(&output, state, provConfig); err != nil {
+			return err
+		}
+
+		{
+			// Call post hook
+			err := ctx.Hook(func(h Hook) (HookAction, error) {
+				return h.PostProvision(n.Info, prov.Type)
+			})
+			if err != nil {
+				return err
+			}
+		}
+	}
+
+	return nil
+
+}

terraform/eval_context.go

@@ -0,0 +1,224 @@
+package terraform
+
+import (
+	"sync"
+
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalContext is the interface that is given to eval nodes to execute.
+type EvalContext interface {
+	// Path is the current module path.
+	Path() []string
+
+	// Hook is used to call hook methods. The callback is called for each
+	// hook and should return the hook action to take and the error.
+	Hook(func(Hook) (HookAction, error)) error
+
+	// Input is the UIInput object for interacting with the UI.
+	Input() UIInput
+
+	// InitProvider initializes the provider with the given name and
+	// returns the implementation of the resource provider or an error.
+	//
+	// It is an error to initialize the same provider more than once.
+	InitProvider(string) (ResourceProvider, error)
+
+	// Provider gets the provider instance with the given name (already
+	// initialized) or returns nil if the provider isn't initialized.
+	Provider(string) ResourceProvider
+
+	// ConfigureProvider configures the provider with the given
+	// configuration. This is a separate context call because this call
+	// is used to store the provider configuration for inheritance lookups
+	// with ParentProviderConfig().
+	ConfigureProvider(string, *ResourceConfig) error
+	ParentProviderConfig(string) *ResourceConfig
+
+	// ProviderInput and SetProviderInput are used to configure providers
+	// from user input.
+	ProviderInput(string) map[string]interface{}
+	SetProviderInput(string, map[string]interface{})
+
+	// InitProvisioner initializes the provisioner with the given name and
+	// returns the implementation of the resource provisioner or an error.
+	//
+	// It is an error to initialize the same provisioner more than once.
+	InitProvisioner(string) (ResourceProvisioner, error)
+
+	// Provisioner gets the provisioner instance with the given name (already
+	// initialized) or returns nil if the provisioner isn't initialized.
+	Provisioner(string) ResourceProvisioner
+
+	// Interpolate takes the given raw configuration and completes
+	// the interpolations, returning the processed ResourceConfig.
+	//
+	// The resource argument is optional. If given, it is the resource
+	// that is currently being acted upon.
+	Interpolate(*config.RawConfig, *Resource) (*ResourceConfig, error)
+
+	// SetVariables sets the variables for interpolation. These variables
+	// should not have a "var." prefix. For example: "var.foo" should be
+	// "foo" as the key.
+	SetVariables(map[string]string)
+
+	// Diff returns the global diff as well as the lock that should
+	// be used to modify that diff.
+	Diff() (*Diff, *sync.RWMutex)
+
+	// State returns the global state as well as the lock that should
+	// be used to modify that state.
+	State() (*State, *sync.RWMutex)
+}
+
+// MockEvalContext is a mock version of EvalContext that can be used
+// for tests.
+type MockEvalContext struct {
+	HookCalled bool
+	HookError  error
+
+	InputCalled bool
+	InputInput  UIInput
+
+	InitProviderCalled   bool
+	InitProviderName     string
+	InitProviderProvider ResourceProvider
+	InitProviderError    error
+
+	ProviderCalled   bool
+	ProviderName     string
+	ProviderProvider ResourceProvider
+
+	ProviderInputCalled bool
+	ProviderInputName   string
+	ProviderInputConfig map[string]interface{}
+
+	SetProviderInputCalled bool
+	SetProviderInputName   string
+	SetProviderInputConfig map[string]interface{}
+
+	ConfigureProviderCalled bool
+	ConfigureProviderName   string
+	ConfigureProviderConfig *ResourceConfig
+	ConfigureProviderError  error
+
+	ParentProviderConfigCalled bool
+	ParentProviderConfigName   string
+	ParentProviderConfigConfig *ResourceConfig
+
+	InitProvisionerCalled      bool
+	InitProvisionerName        string
+	InitProvisionerProvisioner ResourceProvisioner
+	InitProvisionerError       error
+
+	ProvisionerCalled      bool
+	ProvisionerName        string
+	ProvisionerProvisioner ResourceProvisioner
+
+	InterpolateCalled       bool
+	InterpolateConfig       *config.RawConfig
+	InterpolateResource     *Resource
+	InterpolateConfigResult *ResourceConfig
+	InterpolateError        error
+
+	PathCalled bool
+	PathPath   []string
+
+	SetVariablesCalled    bool
+	SetVariablesVariables map[string]string
+
+	DiffCalled bool
+	DiffDiff   *Diff
+	DiffLock   *sync.RWMutex
+
+	StateCalled bool
+	StateState  *State
+	StateLock   *sync.RWMutex
+}
+
+func (c *MockEvalContext) Hook(fn func(Hook) (HookAction, error)) error {
+	c.HookCalled = true
+	return c.HookError
+}
+
+func (c *MockEvalContext) Input() UIInput {
+	c.InputCalled = true
+	return c.InputInput
+}
+
+func (c *MockEvalContext) InitProvider(n string) (ResourceProvider, error) {
+	c.InitProviderCalled = true
+	c.InitProviderName = n
+	return c.InitProviderProvider, c.InitProviderError
+}
+
+func (c *MockEvalContext) Provider(n string) ResourceProvider {
+	c.ProviderCalled = true
+	c.ProviderName = n
+	return c.ProviderProvider
+}
+
+func (c *MockEvalContext) ConfigureProvider(n string, cfg *ResourceConfig) error {
+	c.ConfigureProviderCalled = true
+	c.ConfigureProviderName = n
+	c.ConfigureProviderConfig = cfg
+	return c.ConfigureProviderError
+}
+
+func (c *MockEvalContext) ParentProviderConfig(n string) *ResourceConfig {
+	c.ParentProviderConfigCalled = true
+	c.ParentProviderConfigName = n
+	return c.ParentProviderConfigConfig
+}
+
+func (c *MockEvalContext) ProviderInput(n string) map[string]interface{} {
+	c.ProviderInputCalled = true
+	c.ProviderInputName = n
+	return c.ProviderInputConfig
+}
+
+func (c *MockEvalContext) SetProviderInput(n string, cfg map[string]interface{}) {
+	c.SetProviderInputCalled = true
+	c.SetProviderInputName = n
+	c.SetProviderInputConfig = cfg
+}
+
+func (c *MockEvalContext) InitProvisioner(n string) (ResourceProvisioner, error) {
+	c.InitProvisionerCalled = true
+	c.InitProvisionerName = n
+	return c.InitProvisionerProvisioner, c.InitProvisionerError
+}
+
+func (c *MockEvalContext) Provisioner(n string) ResourceProvisioner {
+	c.ProvisionerCalled = true
+	c.ProvisionerName = n
+	return c.ProvisionerProvisioner
+}
+
+func (c *MockEvalContext) Interpolate(
+	config *config.RawConfig, resource *Resource) (*ResourceConfig, error) {
+	c.InterpolateCalled = true
+	c.InterpolateConfig = config
+	c.InterpolateResource = resource
+	return c.InterpolateConfigResult, c.InterpolateError
+}
+
+func (c *MockEvalContext) Path() []string {
+	c.PathCalled = true
+	return c.PathPath
+}
+
+func (c *MockEvalContext) SetVariables(vs map[string]string) {
+	c.SetVariablesCalled = true
+	c.SetVariablesVariables = vs
+}
+
+func (c *MockEvalContext) Diff() (*Diff, *sync.RWMutex) {
+	c.DiffCalled = true
+	return c.DiffDiff, c.DiffLock
+}
+
+func (c *MockEvalContext) State() (*State, *sync.RWMutex) {
+	c.StateCalled = true
+	return c.StateState, c.StateLock
+}

terraform/eval_context_builtin.go

@@ -0,0 +1,230 @@
+package terraform
+
+import (
+	"fmt"
+	"log"
+	"sync"
+
+	"github.com/hashicorp/terraform/config"
+)
+
+// BuiltinEvalContext is an EvalContext implementation that is used by
+// Terraform by default.
+type BuiltinEvalContext struct {
+	PathValue           []string
+	Interpolater        *Interpolater
+	Hooks               []Hook
+	InputValue          UIInput
+	Providers           map[string]ResourceProviderFactory
+	ProviderCache       map[string]ResourceProvider
+	ProviderConfigCache map[string]*ResourceConfig
+	ProviderInputConfig map[string]map[string]interface{}
+	ProviderLock        *sync.Mutex
+	Provisioners        map[string]ResourceProvisionerFactory
+	ProvisionerCache    map[string]ResourceProvisioner
+	ProvisionerLock     *sync.Mutex
+	DiffValue           *Diff
+	DiffLock            *sync.RWMutex
+	StateValue          *State
+	StateLock           *sync.RWMutex
+
+	once sync.Once
+}
+
+func (ctx *BuiltinEvalContext) Hook(fn func(Hook) (HookAction, error)) error {
+	for _, h := range ctx.Hooks {
+		action, err := fn(h)
+		if err != nil {
+			return err
+		}
+
+		switch action {
+		case HookActionContinue:
+			continue
+		case HookActionHalt:
+			// Return an early exit error to trigger an early exit
+			log.Printf("[WARN] Early exit triggered by hook: %T", h)
+			return EvalEarlyExitError{}
+		}
+	}
+
+	return nil
+}
+
+func (ctx *BuiltinEvalContext) Input() UIInput {
+	return ctx.InputValue
+}
+
+func (ctx *BuiltinEvalContext) InitProvider(n string) (ResourceProvider, error) {
+	ctx.once.Do(ctx.init)
+
+	// If we already initialized, it is an error
+	if p := ctx.Provider(n); p != nil {
+		return nil, fmt.Errorf("Provider '%s' already initialized", n)
+	}
+
+	// Warning: make sure to acquire these locks AFTER the call to Provider
+	// above, since it also acquires locks.
+	ctx.ProviderLock.Lock()
+	defer ctx.ProviderLock.Unlock()
+
+	f, ok := ctx.Providers[n]
+	if !ok {
+		return nil, fmt.Errorf("Provider '%s' not found", n)
+	}
+
+	p, err := f()
+	if err != nil {
+		return nil, err
+	}
+
+	providerPath := make([]string, len(ctx.Path())+1)
+	copy(providerPath, ctx.Path())
+	providerPath[len(providerPath)-1] = n
+
+	ctx.ProviderCache[PathCacheKey(providerPath)] = p
+	return p, nil
+}
+
+func (ctx *BuiltinEvalContext) Provider(n string) ResourceProvider {
+	ctx.once.Do(ctx.init)
+
+	ctx.ProviderLock.Lock()
+	defer ctx.ProviderLock.Unlock()
+
+	providerPath := make([]string, len(ctx.Path())+1)
+	copy(providerPath, ctx.Path())
+	providerPath[len(providerPath)-1] = n
+
+	return ctx.ProviderCache[PathCacheKey(providerPath)]
+}
+
+func (ctx *BuiltinEvalContext) ConfigureProvider(
+	n string, cfg *ResourceConfig) error {
+	p := ctx.Provider(n)
+	if p == nil {
+		return fmt.Errorf("Provider '%s' not initialized", n)
+	}
+
+	// Save the configuration
+	ctx.ProviderLock.Lock()
+	ctx.ProviderConfigCache[PathCacheKey(ctx.Path())] = cfg
+	ctx.ProviderLock.Unlock()
+
+	return p.Configure(cfg)
+}
+
+func (ctx *BuiltinEvalContext) ProviderInput(n string) map[string]interface{} {
+	ctx.ProviderLock.Lock()
+	defer ctx.ProviderLock.Unlock()
+
+	return ctx.ProviderInputConfig[n]
+}
+
+func (ctx *BuiltinEvalContext) SetProviderInput(n string, c map[string]interface{}) {
+	ctx.ProviderLock.Lock()
+	defer ctx.ProviderLock.Unlock()
+
+	ctx.ProviderInputConfig[n] = c
+}
+
+func (ctx *BuiltinEvalContext) ParentProviderConfig(n string) *ResourceConfig {
+	ctx.ProviderLock.Lock()
+	defer ctx.ProviderLock.Unlock()
+
+	path := ctx.Path()
+	for i := len(path) - 1; i >= 1; i-- {
+		k := PathCacheKey(path[:i])
+		if v, ok := ctx.ProviderConfigCache[k]; ok {
+			return v
+		}
+	}
+
+	return nil
+}
+
+func (ctx *BuiltinEvalContext) InitProvisioner(
+	n string) (ResourceProvisioner, error) {
+	ctx.once.Do(ctx.init)
+
+	// If we already initialized, it is an error
+	if p := ctx.Provisioner(n); p != nil {
+		return nil, fmt.Errorf("Provisioner '%s' already initialized", n)
+	}
+
+	// Warning: make sure to acquire these locks AFTER the call to Provisioner
+	// above, since it also acquires locks.
+	ctx.ProvisionerLock.Lock()
+	defer ctx.ProvisionerLock.Unlock()
+
+	f, ok := ctx.Provisioners[n]
+	if !ok {
+		return nil, fmt.Errorf("Provisioner '%s' not found", n)
+	}
+
+	p, err := f()
+	if err != nil {
+		return nil, err
+	}
+
+	ctx.ProvisionerCache[PathCacheKey(ctx.Path())] = p
+	return p, nil
+}
+
+func (ctx *BuiltinEvalContext) Provisioner(n string) ResourceProvisioner {
+	ctx.once.Do(ctx.init)
+
+	ctx.ProvisionerLock.Lock()
+	defer ctx.ProvisionerLock.Unlock()
+
+	return ctx.ProvisionerCache[PathCacheKey(ctx.Path())]
+}
+
+func (ctx *BuiltinEvalContext) Interpolate(
+	cfg *config.RawConfig, r *Resource) (*ResourceConfig, error) {
+	if cfg != nil {
+		scope := &InterpolationScope{
+			Path:     ctx.Path(),
+			Resource: r,
+		}
+		vs, err := ctx.Interpolater.Values(scope, cfg.Variables)
+		if err != nil {
+			return nil, err
+		}
+
+		// Do the interpolation
+		if err := cfg.Interpolate(vs); err != nil {
+			return nil, err
+		}
+	}
+
+	result := NewResourceConfig(cfg)
+	result.interpolateForce()
+	return result, nil
+}
+
+func (ctx *BuiltinEvalContext) Path() []string {
+	return ctx.PathValue
+}
+
+func (ctx *BuiltinEvalContext) SetVariables(vs map[string]string) {
+	for k, v := range vs {
+		ctx.Interpolater.Variables[k] = v
+	}
+}
+
+func (ctx *BuiltinEvalContext) Diff() (*Diff, *sync.RWMutex) {
+	return ctx.DiffValue, ctx.DiffLock
+}
+
+func (ctx *BuiltinEvalContext) State() (*State, *sync.RWMutex) {
+	return ctx.StateValue, ctx.StateLock
+}
+
+func (ctx *BuiltinEvalContext) init() {
+	// We nil-check the things below because they're meant to be configured,
+	// and we just default them to non-nil.
+	if ctx.Providers == nil {
+		ctx.Providers = make(map[string]ResourceProviderFactory)
+	}
+}

terraform/eval_count.go

@@ -0,0 +1,50 @@
+package terraform
+
+import (
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalCountFixZeroOneBoundary is an EvalNode that fixes up the state
+// when there is a resource count with zero/one boundary, i.e. fixing
+// a resource named "aws_instance.foo" to "aws_instance.foo.0" and vice-versa.
+type EvalCountFixZeroOneBoundary struct {
+	Resource *config.Resource
+}
+
+// TODO: test
+func (n *EvalCountFixZeroOneBoundary) Eval(ctx EvalContext) (interface{}, error) {
+	// Get the count, important for knowing whether we're supposed to
+	// be adding the zero, or trimming it.
+	count, err := n.Resource.Count()
+	if err != nil {
+		return nil, err
+	}
+
+	// Figure what to look for and what to replace it with
+	hunt := n.Resource.Id()
+	replace := hunt + ".0"
+	if count < 2 {
+		hunt, replace = replace, hunt
+	}
+
+	state, lock := ctx.State()
+
+	// Get a lock so we can access this instance and potentially make
+	// changes to it.
+	lock.Lock()
+	defer lock.Unlock()
+
+	// Look for the module state. If we don't have one, then it doesn't matter.
+	mod := state.ModuleByPath(ctx.Path())
+	if mod == nil {
+		return nil, nil
+	}
+
+	// Look for the resource state. If we don't have one, then it is okay.
+	if rs, ok := mod.Resources[hunt]; ok {
+		mod.Resources[replace] = rs
+		delete(mod.Resources, hunt)
+	}
+
+	return nil, nil
+}

terraform/eval_diff.go

@@ -0,0 +1,299 @@
+package terraform
+
+import (
+	"fmt"
+	"log"
+)
+
+// EvalCompareDiff is an EvalNode implementation that compares two diffs
+// and errors if the diffs are not equal.
+type EvalCompareDiff struct {
+	Info     *InstanceInfo
+	One, Two **InstanceDiff
+}
+
+// TODO: test
+func (n *EvalCompareDiff) Eval(ctx EvalContext) (interface{}, error) {
+	one, two := *n.One, *n.Two
+
+	// If either are nil, let them be empty
+	if one == nil {
+		one = new(InstanceDiff)
+		one.init()
+	}
+	if two == nil {
+		two = new(InstanceDiff)
+		two.init()
+	}
+	oneId := one.Attributes["id"]
+	twoId := two.Attributes["id"]
+	delete(one.Attributes, "id")
+	delete(two.Attributes, "id")
+	defer func() {
+		if oneId != nil {
+			one.Attributes["id"] = oneId
+		}
+		if twoId != nil {
+			two.Attributes["id"] = twoId
+		}
+	}()
+
+	if !one.Same(two) {
+		log.Printf("[ERROR] %s: diff's didn't match", n.Info.Id)
+		log.Printf("[ERROR] %s: diff one: %#v", n.Info.Id, one)
+		log.Printf("[ERROR] %s: diff two: %#v", n.Info.Id, two)
+		return nil, fmt.Errorf(
+			"%s: diffs didn't match during apply. This is a bug with "+
+				"Terraform and should be reported.", n.Info.Id)
+	}
+
+	return nil, nil
+}
+
+// EvalDiff is an EvalNode implementation that does a refresh for
+// a resource.
+type EvalDiff struct {
+	Info        *InstanceInfo
+	Config      **ResourceConfig
+	Provider    *ResourceProvider
+	State       **InstanceState
+	Output      **InstanceDiff
+	OutputState **InstanceState
+}
+
+// TODO: test
+func (n *EvalDiff) Eval(ctx EvalContext) (interface{}, error) {
+	state := *n.State
+	config := *n.Config
+	provider := *n.Provider
+
+	// Call pre-diff hook
+	err := ctx.Hook(func(h Hook) (HookAction, error) {
+		return h.PreDiff(n.Info, state)
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	// The state for the diff must never be nil
+	diffState := state
+	if diffState == nil {
+		diffState = new(InstanceState)
+	}
+	diffState.init()
+
+	// Diff!
+	diff, err := provider.Diff(n.Info, diffState, config)
+	if err != nil {
+		return nil, err
+	}
+	if diff == nil {
+		diff = new(InstanceDiff)
+	}
+
+	// Require a destroy if there is no ID and it requires new.
+	if diff.RequiresNew() && state != nil && state.ID != "" {
+		diff.Destroy = true
+	}
+
+	// If we're creating a new resource, compute its ID
+	if diff.RequiresNew() || state == nil || state.ID == "" {
+		var oldID string
+		if state != nil {
+			oldID = state.Attributes["id"]
+		}
+
+		// Add diff to compute new ID
+		diff.init()
+		diff.Attributes["id"] = &ResourceAttrDiff{
+			Old:         oldID,
+			NewComputed: true,
+			RequiresNew: true,
+			Type:        DiffAttrOutput,
+		}
+	}
+
+	// Call post-refresh hook
+	err = ctx.Hook(func(h Hook) (HookAction, error) {
+		return h.PostDiff(n.Info, diff)
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	// Update our output
+	*n.Output = diff
+
+	// Update the state if we care
+	if n.OutputState != nil {
+		*n.OutputState = state
+
+		// Merge our state so that the state is updated with our plan
+		if !diff.Empty() && n.OutputState != nil {
+			*n.OutputState = state.MergeDiff(diff)
+		}
+	}
+
+	return nil, nil
+}
+
+// EvalDiffDestroy is an EvalNode implementation that returns a plain
+// destroy diff.
+type EvalDiffDestroy struct {
+	Info   *InstanceInfo
+	State  **InstanceState
+	Output **InstanceDiff
+}
+
+// TODO: test
+func (n *EvalDiffDestroy) Eval(ctx EvalContext) (interface{}, error) {
+	state := *n.State
+
+	// If there is no state or we don't have an ID, we're already destroyed
+	if state == nil || state.ID == "" {
+		return nil, nil
+	}
+
+	// Call pre-diff hook
+	err := ctx.Hook(func(h Hook) (HookAction, error) {
+		return h.PreDiff(n.Info, state)
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	// The diff
+	diff := &InstanceDiff{Destroy: true}
+
+	// Call post-diff hook
+	err = ctx.Hook(func(h Hook) (HookAction, error) {
+		return h.PostDiff(n.Info, diff)
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	// Update our output
+	*n.Output = diff
+
+	return nil, nil
+}
+
+// EvalDiffDestroyModule is an EvalNode implementation that writes the diff to
+// the full diff.
+type EvalDiffDestroyModule struct {
+	Path []string
+}
+
+// TODO: test
+func (n *EvalDiffDestroyModule) Eval(ctx EvalContext) (interface{}, error) {
+	diff, lock := ctx.Diff()
+
+	// Acquire the lock so that we can do this safely concurrently
+	lock.Lock()
+	defer lock.Unlock()
+
+	// Write the diff
+	modDiff := diff.ModuleByPath(n.Path)
+	if modDiff == nil {
+		modDiff = diff.AddModule(n.Path)
+	}
+	modDiff.Destroy = true
+
+	return nil, nil
+}
+
+// EvalDiffTainted is an EvalNode implementation that writes the diff to
+// the full diff.
+type EvalDiffTainted struct {
+	Name string
+	Diff **InstanceDiff
+}
+
+// TODO: test
+func (n *EvalDiffTainted) Eval(ctx EvalContext) (interface{}, error) {
+	state, lock := ctx.State()
+
+	// Get a read lock so we can access this instance
+	lock.RLock()
+	defer lock.RUnlock()
+
+	// Look for the module state. If we don't have one, then it doesn't matter.
+	mod := state.ModuleByPath(ctx.Path())
+	if mod == nil {
+		return nil, nil
+	}
+
+	// Look for the resource state. If we don't have one, then it is okay.
+	rs := mod.Resources[n.Name]
+	if rs == nil {
+		return nil, nil
+	}
+
+	// If we have tainted, then mark it on the diff
+	if len(rs.Tainted) > 0 {
+		(*n.Diff).DestroyTainted = true
+	}
+
+	return nil, nil
+}
+
+// EvalReadDiff is an EvalNode implementation that writes the diff to
+// the full diff.
+type EvalReadDiff struct {
+	Name string
+	Diff **InstanceDiff
+}
+
+func (n *EvalReadDiff) Eval(ctx EvalContext) (interface{}, error) {
+	diff, lock := ctx.Diff()
+
+	// Acquire the lock so that we can do this safely concurrently
+	lock.Lock()
+	defer lock.Unlock()
+
+	// Write the diff
+	modDiff := diff.ModuleByPath(ctx.Path())
+	if modDiff == nil {
+		return nil, nil
+	}
+
+	*n.Diff = modDiff.Resources[n.Name]
+
+	return nil, nil
+}
+
+// EvalWriteDiff is an EvalNode implementation that writes the diff to
+// the full diff.
+type EvalWriteDiff struct {
+	Name string
+	Diff **InstanceDiff
+}
+
+// TODO: test
+func (n *EvalWriteDiff) Eval(ctx EvalContext) (interface{}, error) {
+	diff, lock := ctx.Diff()
+
+	// The diff to write, if its empty it should write nil
+	diffVal := *n.Diff
+	if diffVal.Empty() {
+		diffVal = nil
+	}
+
+	// Acquire the lock so that we can do this safely concurrently
+	lock.Lock()
+	defer lock.Unlock()
+
+	// Write the diff
+	modDiff := diff.ModuleByPath(ctx.Path())
+	if modDiff == nil {
+		modDiff = diff.AddModule(ctx.Path())
+	}
+	if diffVal != nil {
+		modDiff.Resources[n.Name] = diffVal
+	} else {
+		delete(modDiff.Resources, n.Name)
+	}
+
+	return nil, nil
+}

terraform/eval_filter.go

@@ -0,0 +1,25 @@
+package terraform
+
+// EvalNodeFilterFunc is the callback used to replace a node with
+// another to node. To not do the replacement, just return the input node.
+type EvalNodeFilterFunc func(EvalNode) EvalNode
+
+// EvalNodeFilterable is an interface that can be implemented by
+// EvalNodes to allow filtering of sub-elements. Note that this isn't
+// a common thing to implement and you probably don't need it.
+type EvalNodeFilterable interface {
+	EvalNode
+	Filter(EvalNodeFilterFunc)
+}
+
+// EvalFilter runs the filter on the given node and returns the
+// final filtered value. This should be called rather than checking
+// the EvalNode directly since this will properly handle EvalNodeFilterables.
+func EvalFilter(node EvalNode, fn EvalNodeFilterFunc) EvalNode {
+	if f, ok := node.(EvalNodeFilterable); ok {
+		f.Filter(fn)
+		return node
+	}
+
+	return fn(node)
+}

terraform/eval_filter_operation.go

@@ -0,0 +1,49 @@
+package terraform
+
+// EvalNodeOpFilterable is an interface that EvalNodes can implement
+// to be filterable by the operation that is being run on Terraform.
+type EvalNodeOpFilterable interface {
+	IncludeInOp(walkOperation) bool
+}
+
+// EvalNodeFilterOp returns a filter function that filters nodes that
+// include themselves in specific operations.
+func EvalNodeFilterOp(op walkOperation) EvalNodeFilterFunc {
+	return func(n EvalNode) EvalNode {
+		include := true
+		if of, ok := n.(EvalNodeOpFilterable); ok {
+			include = of.IncludeInOp(op)
+		}
+		if include {
+			return n
+		}
+
+		return EvalNoop{}
+	}
+}
+
+// EvalOpFilter is an EvalNode implementation that is a proxy to
+// another node but filters based on the operation.
+type EvalOpFilter struct {
+	// Ops is the list of operations to include this node in.
+	Ops []walkOperation
+
+	// Node is the node to execute
+	Node EvalNode
+}
+
+// TODO: test
+func (n *EvalOpFilter) Eval(ctx EvalContext) (interface{}, error) {
+	return EvalRaw(n.Node, ctx)
+}
+
+// EvalNodeOpFilterable impl.
+func (n *EvalOpFilter) IncludeInOp(op walkOperation) bool {
+	for _, v := range n.Ops {
+		if v == op {
+			return true
+		}
+	}
+
+	return false
+}

terraform/eval_if.go

@@ -0,0 +1,21 @@
+package terraform
+
+// EvalIf is an EvalNode that is a conditional.
+type EvalIf struct {
+	If   func(EvalContext) (bool, error)
+	Node EvalNode
+}
+
+// TODO: test
+func (n *EvalIf) Eval(ctx EvalContext) (interface{}, error) {
+	yes, err := n.If(ctx)
+	if err != nil {
+		return nil, err
+	}
+
+	if yes {
+		return EvalRaw(n.Node, ctx)
+	}
+
+	return nil, nil
+}

terraform/eval_interpolate.go

@@ -0,0 +1,26 @@
+package terraform
+
+import (
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalInterpolate is an EvalNode implementation that takes a raw
+// configuration and interpolates it.
+type EvalInterpolate struct {
+	Config   *config.RawConfig
+	Resource *Resource
+	Output   **ResourceConfig
+}
+
+func (n *EvalInterpolate) Eval(ctx EvalContext) (interface{}, error) {
+	rc, err := ctx.Interpolate(n.Config, n.Resource)
+	if err != nil {
+		return nil, err
+	}
+
+	if n.Output != nil {
+		*n.Output = rc
+	}
+
+	return nil, nil
+}

terraform/eval_interpolate_test.go

@@ -0,0 +1,37 @@
+package terraform
+
+import (
+	"reflect"
+	"testing"
+
+	"github.com/hashicorp/terraform/config"
+)
+
+func TestEvalInterpolate_impl(t *testing.T) {
+	var _ EvalNode = new(EvalInterpolate)
+}
+
+func TestEvalInterpolate(t *testing.T) {
+	config, err := config.NewRawConfig(map[string]interface{}{})
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	var actual *ResourceConfig
+	n := &EvalInterpolate{Config: config, Output: &actual}
+	result := testResourceConfig(t, map[string]interface{}{})
+	ctx := &MockEvalContext{InterpolateConfigResult: result}
+	if _, err := n.Eval(ctx); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+	if actual != result {
+		t.Fatalf("bad: %#v", actual)
+	}
+
+	if !ctx.InterpolateCalled {
+		t.Fatal("should be called")
+	}
+	if !reflect.DeepEqual(ctx.InterpolateConfig, config) {
+		t.Fatalf("bad: %#v", ctx.InterpolateConfig)
+	}
+}

terraform/eval_noop.go

@@ -0,0 +1,8 @@
+package terraform
+
+// EvalNoop is an EvalNode that does nothing.
+type EvalNoop struct{}
+
+func (EvalNoop) Eval(EvalContext) (interface{}, error) {
+	return nil, nil
+}

terraform/eval_output.go

@@ -0,0 +1,63 @@
+package terraform
+
+import (
+	"fmt"
+
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalWriteOutput is an EvalNode implementation that writes the output
+// for the given name to the current state.
+type EvalWriteOutput struct {
+	Name  string
+	Value *config.RawConfig
+}
+
+// TODO: test
+func (n *EvalWriteOutput) Eval(ctx EvalContext) (interface{}, error) {
+	cfg, err := ctx.Interpolate(n.Value, nil)
+	if err != nil {
+		// Ignore it
+	}
+
+	state, lock := ctx.State()
+	if state == nil {
+		return nil, fmt.Errorf("cannot write state to nil state")
+	}
+
+	// Get a write lock so we can access this instance
+	lock.Lock()
+	defer lock.Unlock()
+
+	// Look for the module state. If we don't have one, create it.
+	mod := state.ModuleByPath(ctx.Path())
+	if mod == nil {
+		mod = state.AddModule(ctx.Path())
+	}
+
+	// Get the value from the config
+	var valueRaw interface{} = config.UnknownVariableValue
+	if cfg != nil {
+		var ok bool
+		valueRaw, ok = cfg.Get("value")
+		if !ok {
+			valueRaw = ""
+		}
+		if cfg.IsComputed("value") {
+			valueRaw = config.UnknownVariableValue
+		}
+	}
+
+	// If it is a list of values, get the first one
+	if list, ok := valueRaw.([]interface{}); ok {
+		valueRaw = list[0]
+	}
+	if _, ok := valueRaw.(string); !ok {
+		return nil, fmt.Errorf("output %s is not a string", n.Name)
+	}
+
+	// Write the output
+	mod.Outputs[n.Name] = valueRaw.(string)
+
+	return nil, nil
+}

terraform/eval_provider.go

@@ -0,0 +1,108 @@
+package terraform
+
+import (
+	"fmt"
+
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalConfigProvider is an EvalNode implementation that configures
+// a provider that is already initialized and retrieved.
+type EvalConfigProvider struct {
+	Provider string
+	Config   **ResourceConfig
+}
+
+func (n *EvalConfigProvider) Eval(ctx EvalContext) (interface{}, error) {
+	cfg := *n.Config
+
+	// If we have a configuration set, then use that
+	if input := ctx.ProviderInput(n.Provider); input != nil {
+		rc, err := config.NewRawConfig(input)
+		if err != nil {
+			return nil, err
+		}
+
+		merged := cfg.raw.Merge(rc)
+		cfg = NewResourceConfig(merged)
+	}
+
+	// Get the parent configuration if there is one
+	if parent := ctx.ParentProviderConfig(n.Provider); parent != nil {
+		merged := cfg.raw.Merge(parent.raw)
+		cfg = NewResourceConfig(merged)
+	}
+
+	return nil, ctx.ConfigureProvider(n.Provider, cfg)
+}
+
+// EvalInitProvider is an EvalNode implementation that initializes a provider
+// and returns nothing. The provider can be retrieved again with the
+// EvalGetProvider node.
+type EvalInitProvider struct {
+	Name string
+}
+
+func (n *EvalInitProvider) Eval(ctx EvalContext) (interface{}, error) {
+	return ctx.InitProvider(n.Name)
+}
+
+// EvalGetProvider is an EvalNode implementation that retrieves an already
+// initialized provider instance for the given name.
+type EvalGetProvider struct {
+	Name   string
+	Output *ResourceProvider
+}
+
+func (n *EvalGetProvider) Eval(ctx EvalContext) (interface{}, error) {
+	result := ctx.Provider(n.Name)
+	if result == nil {
+		return nil, fmt.Errorf("provider %s not initialized", n.Name)
+	}
+
+	if n.Output != nil {
+		*n.Output = result
+	}
+
+	return nil, nil
+}
+
+// EvalInputProvider is an EvalNode implementation that asks for input
+// for the given provider configurations.
+type EvalInputProvider struct {
+	Name     string
+	Provider *ResourceProvider
+	Config   *config.RawConfig
+}
+
+func (n *EvalInputProvider) Eval(ctx EvalContext) (interface{}, error) {
+	// If we already configured this provider, then don't do this again
+	if v := ctx.ProviderInput(n.Name); v != nil {
+		return nil, nil
+	}
+
+	rc := NewResourceConfig(n.Config)
+	rc.Config = make(map[string]interface{})
+
+	// Wrap the input into a namespace
+	input := &PrefixUIInput{
+		IdPrefix:    fmt.Sprintf("provider.%s", n.Name),
+		QueryPrefix: fmt.Sprintf("provider.%s.", n.Name),
+		UIInput:     ctx.Input(),
+	}
+
+	// Go through each provider and capture the input necessary
+	// to satisfy it.
+	config, err := (*n.Provider).Input(input, rc)
+	if err != nil {
+		return nil, fmt.Errorf(
+			"Error configuring %s: %s", n.Name, err)
+	}
+
+	if config != nil && len(config.Config) > 0 {
+		// Set the configuration
+		ctx.SetProviderInput(n.Name, config.Config)
+	}
+
+	return nil, nil
+}

terraform/eval_provider_test.go

@@ -0,0 +1,72 @@
+package terraform
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestEvalConfigProvider_impl(t *testing.T) {
+	var _ EvalNode = new(EvalConfigProvider)
+}
+
+func TestEvalConfigProvider(t *testing.T) {
+	config := testResourceConfig(t, map[string]interface{}{})
+	provider := &MockResourceProvider{}
+	n := &EvalConfigProvider{Config: &config}
+
+	ctx := &MockEvalContext{ProviderProvider: provider}
+	if _, err := n.Eval(ctx); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	if !ctx.ConfigureProviderCalled {
+		t.Fatal("should be called")
+	}
+	if !reflect.DeepEqual(ctx.ConfigureProviderConfig, config) {
+		t.Fatalf("bad: %#v", ctx.ConfigureProviderConfig)
+	}
+}
+
+func TestEvalInitProvider_impl(t *testing.T) {
+	var _ EvalNode = new(EvalInitProvider)
+}
+
+func TestEvalInitProvider(t *testing.T) {
+	n := &EvalInitProvider{Name: "foo"}
+	provider := &MockResourceProvider{}
+	ctx := &MockEvalContext{InitProviderProvider: provider}
+	if _, err := n.Eval(ctx); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	if !ctx.InitProviderCalled {
+		t.Fatal("should be called")
+	}
+	if ctx.InitProviderName != "foo" {
+		t.Fatalf("bad: %#v", ctx.InitProviderName)
+	}
+}
+
+func TestEvalGetProvider_impl(t *testing.T) {
+	var _ EvalNode = new(EvalGetProvider)
+}
+
+func TestEvalGetProvider(t *testing.T) {
+	var actual ResourceProvider
+	n := &EvalGetProvider{Name: "foo", Output: &actual}
+	provider := &MockResourceProvider{}
+	ctx := &MockEvalContext{ProviderProvider: provider}
+	if _, err := n.Eval(ctx); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+	if actual != provider {
+		t.Fatalf("bad: %#v", actual)
+	}
+
+	if !ctx.ProviderCalled {
+		t.Fatal("should be called")
+	}
+	if ctx.ProviderName != "foo" {
+		t.Fatalf("bad: %#v", ctx.ProviderName)
+	}
+}

terraform/eval_provisioner.go

@@ -0,0 +1,36 @@
+package terraform
+
+import (
+	"fmt"
+)
+
+// EvalInitProvisioner is an EvalNode implementation that initializes a provisioner
+// and returns nothing. The provisioner can be retrieved again with the
+// EvalGetProvisioner node.
+type EvalInitProvisioner struct {
+	Name string
+}
+
+func (n *EvalInitProvisioner) Eval(ctx EvalContext) (interface{}, error) {
+	return ctx.InitProvisioner(n.Name)
+}
+
+// EvalGetProvisioner is an EvalNode implementation that retrieves an already
+// initialized provisioner instance for the given name.
+type EvalGetProvisioner struct {
+	Name   string
+	Output *ResourceProvisioner
+}
+
+func (n *EvalGetProvisioner) Eval(ctx EvalContext) (interface{}, error) {
+	result := ctx.Provisioner(n.Name)
+	if result == nil {
+		return nil, fmt.Errorf("provisioner %s not initialized", n.Name)
+	}
+
+	if n.Output != nil {
+		*n.Output = result
+	}
+
+	return result, nil
+}

terraform/eval_provisioner_test.go

@@ -0,0 +1,49 @@
+package terraform
+
+import (
+	"testing"
+)
+
+func TestEvalInitProvisioner_impl(t *testing.T) {
+	var _ EvalNode = new(EvalInitProvisioner)
+}
+
+func TestEvalInitProvisioner(t *testing.T) {
+	n := &EvalInitProvisioner{Name: "foo"}
+	provisioner := &MockResourceProvisioner{}
+	ctx := &MockEvalContext{InitProvisionerProvisioner: provisioner}
+	if _, err := n.Eval(ctx); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	if !ctx.InitProvisionerCalled {
+		t.Fatal("should be called")
+	}
+	if ctx.InitProvisionerName != "foo" {
+		t.Fatalf("bad: %#v", ctx.InitProvisionerName)
+	}
+}
+
+func TestEvalGetProvisioner_impl(t *testing.T) {
+	var _ EvalNode = new(EvalGetProvisioner)
+}
+
+func TestEvalGetProvisioner(t *testing.T) {
+	var actual ResourceProvisioner
+	n := &EvalGetProvisioner{Name: "foo", Output: &actual}
+	provisioner := &MockResourceProvisioner{}
+	ctx := &MockEvalContext{ProvisionerProvisioner: provisioner}
+	if _, err := n.Eval(ctx); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+	if actual != provisioner {
+		t.Fatalf("bad: %#v", actual)
+	}
+
+	if !ctx.ProvisionerCalled {
+		t.Fatal("should be called")
+	}
+	if ctx.ProvisionerName != "foo" {
+		t.Fatalf("bad: %#v", ctx.ProvisionerName)
+	}
+}

terraform/eval_refresh.go

@@ -0,0 +1,54 @@
+package terraform
+
+import (
+	"log"
+)
+
+// EvalRefresh is an EvalNode implementation that does a refresh for
+// a resource.
+type EvalRefresh struct {
+	Provider *ResourceProvider
+	State    **InstanceState
+	Info     *InstanceInfo
+	Output   **InstanceState
+}
+
+// TODO: test
+func (n *EvalRefresh) Eval(ctx EvalContext) (interface{}, error) {
+	provider := *n.Provider
+	state := *n.State
+
+	// If we have no state, we don't do any refreshing
+	if state == nil {
+		log.Printf("[DEBUG] refresh: %s: no state, not refreshing", n.Info.Id)
+		return nil, nil
+	}
+
+	// Call pre-refresh hook
+	err := ctx.Hook(func(h Hook) (HookAction, error) {
+		return h.PreRefresh(n.Info, state)
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	// Refresh!
+	state, err = provider.Refresh(n.Info, state)
+	if err != nil {
+		return nil, err
+	}
+
+	// Call post-refresh hook
+	err = ctx.Hook(func(h Hook) (HookAction, error) {
+		return h.PostRefresh(n.Info, state)
+	})
+	if err != nil {
+		return nil, err
+	}
+
+	if n.Output != nil {
+		*n.Output = state
+	}
+
+	return nil, nil
+}

terraform/eval_resource.go

@@ -0,0 +1,13 @@
+package terraform
+
+// EvalInstanceInfo is an EvalNode implementation that fills in the
+// InstanceInfo as much as it can.
+type EvalInstanceInfo struct {
+	Info *InstanceInfo
+}
+
+// TODO: test
+func (n *EvalInstanceInfo) Eval(ctx EvalContext) (interface{}, error) {
+	n.Info.ModulePath = ctx.Path()
+	return nil, nil
+}

terraform/eval_sequence.go

@@ -0,0 +1,23 @@
+package terraform
+
+// EvalSequence is an EvalNode that evaluates in sequence.
+type EvalSequence struct {
+	Nodes []EvalNode
+}
+
+func (n *EvalSequence) Eval(ctx EvalContext) (interface{}, error) {
+	for _, n := range n.Nodes {
+		if _, err := EvalRaw(n, ctx); err != nil {
+			return nil, err
+		}
+	}
+
+	return nil, nil
+}
+
+// EvalNodeFilterable impl.
+func (n *EvalSequence) Filter(fn EvalNodeFilterFunc) {
+	for i, node := range n.Nodes {
+		n.Nodes[i] = fn(node)
+	}
+}

terraform/eval_sequence_test.go

@@ -0,0 +1,9 @@
+package terraform
+
+import (
+	"testing"
+)
+
+func TestEvalSequence_impl(t *testing.T) {
+	var _ EvalNodeFilterable = new(EvalSequence)
+}

terraform/eval_state.go

@@ -0,0 +1,196 @@
+package terraform
+
+import (
+	"fmt"
+)
+
+// EvalReadState is an EvalNode implementation that reads the
+// InstanceState for a specific resource out of the state.
+type EvalReadState struct {
+	Name         string
+	Tainted      bool
+	TaintedIndex int
+	Output       **InstanceState
+}
+
+// TODO: test
+func (n *EvalReadState) Eval(ctx EvalContext) (interface{}, error) {
+	state, lock := ctx.State()
+
+	// Get a read lock so we can access this instance
+	lock.RLock()
+	defer lock.RUnlock()
+
+	// Look for the module state. If we don't have one, then it doesn't matter.
+	mod := state.ModuleByPath(ctx.Path())
+	if mod == nil {
+		return nil, nil
+	}
+
+	// Look for the resource state. If we don't have one, then it is okay.
+	rs := mod.Resources[n.Name]
+	if rs == nil {
+		return nil, nil
+	}
+
+	var result *InstanceState
+	if !n.Tainted {
+		// Return the primary
+		result = rs.Primary
+	} else {
+		// Get the index. If it is negative, then we get the last one
+		idx := n.TaintedIndex
+		if idx < 0 {
+			idx = len(rs.Tainted) - 1
+		}
+
+		if idx < len(rs.Tainted) {
+			// Return the proper tainted resource
+			result = rs.Tainted[n.TaintedIndex]
+		}
+	}
+
+	// Write the result to the output pointer
+	if n.Output != nil {
+		*n.Output = result
+	}
+
+	return result, nil
+}
+
+// EvalWriteState is an EvalNode implementation that reads the
+// InstanceState for a specific resource out of the state.
+type EvalWriteState struct {
+	Name                string
+	ResourceType        string
+	Dependencies        []string
+	State               **InstanceState
+	Tainted             *bool
+	TaintedIndex        int
+	TaintedClearPrimary bool
+}
+
+// TODO: test
+func (n *EvalWriteState) Eval(ctx EvalContext) (interface{}, error) {
+	state, lock := ctx.State()
+	if state == nil {
+		return nil, fmt.Errorf("cannot write state to nil state")
+	}
+
+	// Get a write lock so we can access this instance
+	lock.Lock()
+	defer lock.Unlock()
+
+	// Look for the module state. If we don't have one, create it.
+	mod := state.ModuleByPath(ctx.Path())
+	if mod == nil {
+		mod = state.AddModule(ctx.Path())
+	}
+
+	// Look for the resource state.
+	rs := mod.Resources[n.Name]
+	if rs == nil {
+		rs = &ResourceState{}
+		rs.init()
+		mod.Resources[n.Name] = rs
+	}
+	rs.Type = n.ResourceType
+	rs.Dependencies = n.Dependencies
+
+	if n.Tainted != nil && *n.Tainted {
+		if n.TaintedIndex != -1 {
+			rs.Tainted[n.TaintedIndex] = *n.State
+		} else {
+			rs.Tainted = append(rs.Tainted, *n.State)
+		}
+
+		if n.TaintedClearPrimary {
+			rs.Primary = nil
+		}
+	} else {
+		// Set the primary state
+		rs.Primary = *n.State
+	}
+	println(fmt.Sprintf("%#v", rs))
+
+	return nil, nil
+}
+
+// EvalDeposeState is an EvalNode implementation that takes the primary
+// out of a state and makes it tainted. This is done at the beggining of
+// create-before-destroy calls so that the create can create while preserving
+// the old state of the to-be-destroyed resource.
+type EvalDeposeState struct {
+	Name string
+}
+
+// TODO: test
+func (n *EvalDeposeState) Eval(ctx EvalContext) (interface{}, error) {
+	state, lock := ctx.State()
+
+	// Get a read lock so we can access this instance
+	lock.RLock()
+	defer lock.RUnlock()
+
+	// Look for the module state. If we don't have one, then it doesn't matter.
+	mod := state.ModuleByPath(ctx.Path())
+	if mod == nil {
+		return nil, nil
+	}
+
+	// Look for the resource state. If we don't have one, then it is okay.
+	rs := mod.Resources[n.Name]
+	if rs == nil {
+		return nil, nil
+	}
+
+	// If we don't have a primary, we have nothing to depose
+	if rs.Primary == nil {
+		return nil, nil
+	}
+
+	// Depose to the tainted
+	rs.Tainted = append(rs.Tainted, rs.Primary)
+	rs.Primary = nil
+
+	return nil, nil
+}
+
+// EvalUndeposeState is an EvalNode implementation that reads the
+// InstanceState for a specific resource out of the state.
+type EvalUndeposeState struct {
+	Name string
+}
+
+// TODO: test
+func (n *EvalUndeposeState) Eval(ctx EvalContext) (interface{}, error) {
+	state, lock := ctx.State()
+
+	// Get a read lock so we can access this instance
+	lock.RLock()
+	defer lock.RUnlock()
+
+	// Look for the module state. If we don't have one, then it doesn't matter.
+	mod := state.ModuleByPath(ctx.Path())
+	if mod == nil {
+		return nil, nil
+	}
+
+	// Look for the resource state. If we don't have one, then it is okay.
+	rs := mod.Resources[n.Name]
+	if rs == nil {
+		return nil, nil
+	}
+
+	// If we don't have any tainted, then we don't have anything to do
+	if len(rs.Tainted) == 0 {
+		return nil, nil
+	}
+
+	// Undepose to the tainted
+	idx := len(rs.Tainted) - 1
+	rs.Primary = rs.Tainted[idx]
+	rs.Tainted[idx] = nil
+
+	return nil, nil
+}

terraform/eval_test.go

@@ -0,0 +1,40 @@
+package terraform
+
+import (
+	"testing"
+)
+
+func TestMockEvalContext_impl(t *testing.T) {
+	var _ EvalContext = new(MockEvalContext)
+}
+
+func TestEval(t *testing.T) {
+	var result int
+	n := &testEvalAdd{
+		Items:  []int{10, 32},
+		Result: &result,
+	}
+
+	if _, err := Eval(n, nil); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	if result != 42 {
+		t.Fatalf("bad: %#v", result)
+	}
+}
+
+type testEvalAdd struct {
+	Items  []int
+	Result *int
+}
+
+func (n *testEvalAdd) Eval(ctx EvalContext) (interface{}, error) {
+	result := 0
+	for _, item := range n.Items {
+		result += item
+	}
+
+	*n.Result = result
+	return nil, nil
+}

terraform/eval_validate.go

@@ -0,0 +1,141 @@
+package terraform
+
+import (
+	"fmt"
+
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalValidateError is the error structure returned if there were
+// validation errors.
+type EvalValidateError struct {
+	Warnings []string
+	Errors   []error
+}
+
+func (e *EvalValidateError) Error() string {
+	return fmt.Sprintf("Warnings: %s. Errors: %s", e.Warnings, e.Errors)
+}
+
+// EvalValidateCount is an EvalNode implementation that validates
+// the count of a resource.
+type EvalValidateCount struct {
+	Resource *config.Resource
+}
+
+// TODO: test
+func (n *EvalValidateCount) Eval(ctx EvalContext) (interface{}, error) {
+	var count int
+	var errs []error
+	var err error
+	if _, err := ctx.Interpolate(n.Resource.RawCount, nil); err != nil {
+		errs = append(errs, fmt.Errorf(
+			"Failed to interpolate count: %s", err))
+		goto RETURN
+	}
+
+	count, err = n.Resource.Count()
+	if err != nil {
+		// If we can't get the count during validation, then
+		// just replace it with the number 1.
+		c := n.Resource.RawCount.Config()
+		c[n.Resource.RawCount.Key] = "1"
+		count = 1
+	}
+
+	if count < 0 {
+		errs = append(errs, fmt.Errorf(
+			"Count is less than zero: %d", count))
+	}
+
+RETURN:
+	return nil, &EvalValidateError{
+		Errors: errs,
+	}
+}
+
+// EvalValidateProvider is an EvalNode implementation that validates
+// the configuration of a resource.
+type EvalValidateProvider struct {
+	ProviderName string
+	Provider     *ResourceProvider
+	Config       **ResourceConfig
+}
+
+func (n *EvalValidateProvider) Eval(ctx EvalContext) (interface{}, error) {
+	provider := *n.Provider
+	config := *n.Config
+
+	// Get the parent configuration if there is one
+	if parent := ctx.ParentProviderConfig(n.ProviderName); parent != nil {
+		merged := parent.raw.Merge(config.raw)
+		config = NewResourceConfig(merged)
+	}
+
+	warns, errs := provider.Validate(config)
+	if len(warns) == 0 && len(errs) == 0 {
+		return nil, nil
+	}
+
+	return nil, &EvalValidateError{
+		Warnings: warns,
+		Errors:   errs,
+	}
+}
+
+// EvalValidateProvisioner is an EvalNode implementation that validates
+// the configuration of a resource.
+type EvalValidateProvisioner struct {
+	Provisioner *ResourceProvisioner
+	Config      **ResourceConfig
+}
+
+func (n *EvalValidateProvisioner) Eval(ctx EvalContext) (interface{}, error) {
+	provisioner := *n.Provisioner
+	config := *n.Config
+	warns, errs := provisioner.Validate(config)
+	if len(warns) == 0 && len(errs) == 0 {
+		return nil, nil
+	}
+
+	return nil, &EvalValidateError{
+		Warnings: warns,
+		Errors:   errs,
+	}
+}
+
+// EvalValidateResource is an EvalNode implementation that validates
+// the configuration of a resource.
+type EvalValidateResource struct {
+	Provider     *ResourceProvider
+	Config       **ResourceConfig
+	ResourceName string
+	ResourceType string
+}
+
+func (n *EvalValidateResource) Eval(ctx EvalContext) (interface{}, error) {
+	// TODO: test
+
+	provider := *n.Provider
+	cfg := *n.Config
+	warns, errs := provider.ValidateResource(n.ResourceType, cfg)
+
+	// If the resouce name doesn't match the name regular
+	// expression, show a warning.
+	if !config.NameRegexp.Match([]byte(n.ResourceName)) {
+		warns = append(warns, fmt.Sprintf(
+			"%s: resource name can only contain letters, numbers, "+
+				"dashes, and underscores.\n"+
+				"This will be an error in Terraform 0.4",
+			n.ResourceName))
+	}
+
+	if len(warns) == 0 && len(errs) == 0 {
+		return nil, nil
+	}
+
+	return nil, &EvalValidateError{
+		Warnings: warns,
+		Errors:   errs,
+	}
+}

terraform/eval_variable.go

@@ -0,0 +1,46 @@
+package terraform
+
+import (
+	"github.com/hashicorp/terraform/config"
+)
+
+// EvalSetVariables is an EvalNode implementation that sets the variables
+// explicitly for interpolation later.
+type EvalSetVariables struct {
+	Variables map[string]string
+}
+
+// TODO: test
+func (n *EvalSetVariables) Eval(ctx EvalContext) (interface{}, error) {
+	ctx.SetVariables(n.Variables)
+	return nil, nil
+}
+
+// EvalVariableBlock is an EvalNode implementation that evaluates the
+// given configuration, and uses the final values as a way to set the
+// mapping.
+type EvalVariableBlock struct {
+	Config    **ResourceConfig
+	Variables map[string]string
+}
+
+// TODO: test
+func (n *EvalVariableBlock) Eval(ctx EvalContext) (interface{}, error) {
+	// Clear out the existing mapping
+	for k, _ := range n.Variables {
+		delete(n.Variables, k)
+	}
+
+	// Get our configuration
+	rc := *n.Config
+	for k, v := range rc.Config {
+		n.Variables[k] = v.(string)
+	}
+	for k, _ := range rc.Raw {
+		if _, ok := n.Variables[k]; !ok {
+			n.Variables[k] = config.UnknownVariableValue
+		}
+	}
+
+	return nil, nil
+}

terraform/evaltree_provider.go

@@ -0,0 +1,61 @@
+package terraform
+
+import (
+	"github.com/hashicorp/terraform/config"
+)
+
+// ProviderEvalTree returns the evaluation tree for initializing and
+// configuring providers.
+func ProviderEvalTree(n string, config *config.RawConfig) EvalNode {
+	var provider ResourceProvider
+	var resourceConfig *ResourceConfig
+
+	seq := make([]EvalNode, 0, 5)
+	seq = append(seq, &EvalInitProvider{Name: n})
+
+	// Input stuff
+	seq = append(seq, &EvalOpFilter{
+		Ops: []walkOperation{walkInput},
+		Node: &EvalSequence{
+			Nodes: []EvalNode{
+				&EvalGetProvider{
+					Name:   n,
+					Output: &provider,
+				},
+				&EvalInputProvider{
+					Name:     n,
+					Provider: &provider,
+					Config:   config,
+				},
+			},
+		},
+	})
+
+	// Apply stuff
+	seq = append(seq, &EvalOpFilter{
+		Ops: []walkOperation{walkValidate, walkRefresh, walkPlan, walkApply},
+		Node: &EvalSequence{
+			Nodes: []EvalNode{
+				&EvalGetProvider{
+					Name:   n,
+					Output: &provider,
+				},
+				&EvalInterpolate{
+					Config: config,
+					Output: &resourceConfig,
+				},
+				&EvalValidateProvider{
+					ProviderName: n,
+					Provider:     &provider,
+					Config:       &resourceConfig,
+				},
+				&EvalConfigProvider{
+					Provider: n,
+					Config:   &resourceConfig,
+				},
+			},
+		},
+	})
+
+	return &EvalSequence{Nodes: seq}
+}

terraform/graph_builder.go

@@ -0,0 +1,113 @@
+package terraform
+
+import (
+	"log"
+
+	"github.com/hashicorp/terraform/config/module"
+)
+
+// GraphBuilder is an interface that can be implemented and used with
+// Terraform to build the graph that Terraform walks.
+type GraphBuilder interface {
+	// Build builds the graph for the given module path. It is up to
+	// the interface implementation whether this build should expand
+	// the graph or not.
+	Build(path []string) (*Graph, error)
+}
+
+// BasicGraphBuilder is a GraphBuilder that builds a graph out of a
+// series of transforms and validates the graph is a valid structure.
+type BasicGraphBuilder struct {
+	Steps []GraphTransformer
+}
+
+func (b *BasicGraphBuilder) Build(path []string) (*Graph, error) {
+	g := &Graph{Path: path}
+	for _, step := range b.Steps {
+		if err := step.Transform(g); err != nil {
+			return g, err
+		}
+
+		log.Printf("[TRACE] Graph after step %T:\n\n%s", step, g.String())
+	}
+
+	// Validate the graph structure
+	if err := g.Validate(); err != nil {
+		log.Printf("[ERROR] Graph validation failed. Graph:\n\n%s", g.String())
+		return nil, err
+	}
+
+	return g, nil
+}
+
+// BuiltinGraphBuilder is responsible for building the complete graph that
+// Terraform uses for execution. It is an opinionated builder that defines
+// the step order required to build a complete graph as is used and expected
+// by Terraform.
+//
+// If you require a custom graph, you'll have to build it up manually
+// on your own by building a new GraphBuilder implementation.
+type BuiltinGraphBuilder struct {
+	// Root is the root module of the graph to build.
+	Root *module.Tree
+
+	// Diff is the diff. The proper module diffs will be looked up.
+	Diff *Diff
+
+	// State is the global state. The proper module states will be looked
+	// up by graph path.
+	State *State
+
+	// Providers is the list of providers supported.
+	Providers []string
+
+	// Provisioners is the list of provisioners supported.
+	Provisioners []string
+}
+
+// Build builds the graph according to the steps returned by Steps.
+func (b *BuiltinGraphBuilder) Build(path []string) (*Graph, error) {
+	basic := &BasicGraphBuilder{
+		Steps: b.Steps(),
+	}
+
+	return basic.Build(path)
+}
+
+// Steps returns the ordered list of GraphTransformers that must be executed
+// to build a complete graph.
+func (b *BuiltinGraphBuilder) Steps() []GraphTransformer {
+	return []GraphTransformer{
+		// Create all our resources from the configuration and state
+		&ConfigTransformer{Module: b.Root},
+		&OrphanTransformer{State: b.State, Module: b.Root},
+
+		// Provider-related transformations
+		&MissingProviderTransformer{Providers: b.Providers},
+		&ProviderTransformer{},
+		&PruneProviderTransformer{},
+
+		// Provisioner-related transformations
+		&MissingProvisionerTransformer{Provisioners: b.Provisioners},
+		&ProvisionerTransformer{},
+		&PruneProvisionerTransformer{},
+
+		// Run our vertex-level transforms
+		&VertexTransformer{
+			Transforms: []GraphVertexTransformer{
+				// Expand any statically expanded nodes, such as module graphs
+				&ExpandTransform{
+					Builder: b,
+				},
+			},
+		},
+
+		// Create the destruction nodes
+		&DestroyTransformer{},
+		&CreateBeforeDestroyTransformer{},
+		&PruneDestroyTransformer{Diff: b.Diff, State: b.State},
+
+		// Make sure we create one root
+		&RootTransformer{},
+	}
+}

terraform/graph_builder_test.go

@@ -0,0 +1,156 @@
+package terraform
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/hashicorp/terraform/dag"
+)
+
+func TestBasicGraphBuilder_impl(t *testing.T) {
+	var _ GraphBuilder = new(BasicGraphBuilder)
+}
+
+func TestBasicGraphBuilder(t *testing.T) {
+	b := &BasicGraphBuilder{
+		Steps: []GraphTransformer{
+			&testBasicGraphBuilderTransform{1},
+		},
+	}
+
+	g, err := b.Build(RootModulePath)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	if !reflect.DeepEqual(g.Path, RootModulePath) {
+		t.Fatalf("bad: %#v", g.Path)
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testBasicGraphBuilderStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func TestBasicGraphBuilder_validate(t *testing.T) {
+	b := &BasicGraphBuilder{
+		Steps: []GraphTransformer{
+			&testBasicGraphBuilderTransform{1},
+			&testBasicGraphBuilderTransform{2},
+		},
+	}
+
+	_, err := b.Build(RootModulePath)
+	if err == nil {
+		t.Fatal("should error")
+	}
+}
+
+func TestBuiltinGraphBuilder_impl(t *testing.T) {
+	var _ GraphBuilder = new(BuiltinGraphBuilder)
+}
+
+// This test is not meant to test all the transforms but rather just
+// to verify we get some basic sane graph out. Special tests to ensure
+// specific ordering of steps should be added in other tests.
+func TestBuiltinGraphBuilder(t *testing.T) {
+	b := &BuiltinGraphBuilder{
+		Root: testModule(t, "graph-builder-basic"),
+	}
+
+	g, err := b.Build(RootModulePath)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testBuiltinGraphBuilderBasicStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+// This tests a cycle we got when a CBD resource depends on a non-CBD
+// resource. This cycle shouldn't happen in the general case anymore.
+func TestBuiltinGraphBuilder_cbdDepNonCbd(t *testing.T) {
+	b := &BuiltinGraphBuilder{
+		Root: testModule(t, "graph-builder-cbd-non-cbd"),
+	}
+
+	_, err := b.Build(RootModulePath)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+}
+
+/*
+TODO: This exposes a really bad bug we need to fix after we merge
+the f-ast-branch. This bug still exists in master.
+
+// This test tests that the graph builder properly expands modules.
+func TestBuiltinGraphBuilder_modules(t *testing.T) {
+	b := &BuiltinGraphBuilder{
+		Root: testModule(t, "graph-builder-modules"),
+	}
+
+	g, err := b.Build(RootModulePath)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testBuiltinGraphBuilderModuleStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+*/
+
+type testBasicGraphBuilderTransform struct {
+	V dag.Vertex
+}
+
+func (t *testBasicGraphBuilderTransform) Transform(g *Graph) error {
+	g.Add(t.V)
+	return nil
+}
+
+const testBasicGraphBuilderStr = `
+1
+`
+
+const testBuiltinGraphBuilderBasicStr = `
+aws_instance.db
+  aws_instance.db (destroy tainted)
+  provider.aws
+aws_instance.db (destroy tainted)
+  aws_instance.web (destroy tainted)
+  provider.aws
+aws_instance.web
+  aws_instance.db
+  aws_instance.web (destroy tainted)
+  provider.aws
+aws_instance.web (destroy tainted)
+  provider.aws
+provider.aws
+`
+
+const testBuiltinGraphBuilderModuleStr = `
+aws_instance.web
+  aws_instance.web (destroy)
+aws_instance.web (destroy)
+  aws_security_group.firewall
+  module.consul (expanded)
+  provider.aws
+aws_security_group.firewall
+  aws_security_group.firewall (destroy)
+aws_security_group.firewall (destroy)
+  provider.aws
+module.consul (expanded)
+  aws_security_group.firewall
+  provider.aws
+provider.aws
+`

terraform/graph_config_node.go

@@ -0,0 +1,546 @@
+package terraform
+
+import (
+	"fmt"
+	"strings"
+
+	"github.com/hashicorp/terraform/config"
+	"github.com/hashicorp/terraform/config/module"
+	"github.com/hashicorp/terraform/dag"
+)
+
+// graphNodeConfig is an interface that all graph nodes for the
+// configuration graph need to implement in order to build the variable
+// dependencies properly.
+type graphNodeConfig interface {
+	dag.NamedVertex
+
+	// All graph nodes should be dependent on other things, and able to
+	// be depended on.
+	GraphNodeDependable
+	GraphNodeDependent
+}
+
+// GraphNodeConfigModule represents a module within the configuration graph.
+type GraphNodeConfigModule struct {
+	Path   []string
+	Module *config.Module
+	Tree   *module.Tree
+}
+
+func (n *GraphNodeConfigModule) DependableName() []string {
+	return []string{n.Name()}
+}
+
+func (n *GraphNodeConfigModule) DependentOn() []string {
+	vars := n.Module.RawConfig.Variables
+	result := make([]string, 0, len(vars))
+	for _, v := range vars {
+		if vn := varNameForVar(v); vn != "" {
+			result = append(result, vn)
+		}
+	}
+
+	return result
+}
+
+func (n *GraphNodeConfigModule) Name() string {
+	return fmt.Sprintf("module.%s", n.Module.Name)
+}
+
+// GraphNodeExpandable
+func (n *GraphNodeConfigModule) Expand(b GraphBuilder) (GraphNodeSubgraph, error) {
+	// Build the graph first
+	graph, err := b.Build(n.Path)
+	if err != nil {
+		return nil, err
+	}
+
+	// Add the parameters node to the module
+	t := &ModuleInputTransformer{Variables: make(map[string]string)}
+	if err := t.Transform(graph); err != nil {
+		return nil, err
+	}
+
+	// Build the actual subgraph node
+	return &graphNodeModuleExpanded{
+		Original:    n,
+		Graph:       graph,
+		InputConfig: n.Module.RawConfig,
+		Variables:   t.Variables,
+	}, nil
+}
+
+// GraphNodeExpandable
+func (n *GraphNodeConfigModule) ProvidedBy() []string {
+	// Build up the list of providers by simply going over our configuration
+	// to find the providers that are configured there as well as the
+	// providers that the resources use.
+	config := n.Tree.Config()
+	providers := make(map[string]struct{})
+	for _, p := range config.ProviderConfigs {
+		providers[p.Name] = struct{}{}
+	}
+	for _, r := range config.Resources {
+		providers[resourceProvider(r.Type)] = struct{}{}
+	}
+
+	// Turn the map into a string. This makes sure that the list is
+	// de-dupped since we could be going over potentially many resources.
+	result := make([]string, 0, len(providers))
+	for p, _ := range providers {
+		result = append(result, p)
+	}
+
+	return result
+}
+
+// GraphNodeConfigOutput represents an output configured within the
+// configuration.
+type GraphNodeConfigOutput struct {
+	Output *config.Output
+}
+
+func (n *GraphNodeConfigOutput) Name() string {
+	return fmt.Sprintf("output.%s", n.Output.Name)
+}
+
+func (n *GraphNodeConfigOutput) DependableName() []string {
+	return []string{n.Name()}
+}
+
+func (n *GraphNodeConfigOutput) DependentOn() []string {
+	vars := n.Output.RawConfig.Variables
+	result := make([]string, 0, len(vars))
+	for _, v := range vars {
+		if vn := varNameForVar(v); vn != "" {
+			result = append(result, vn)
+		}
+	}
+
+	return result
+}
+
+// GraphNodeEvalable impl.
+func (n *GraphNodeConfigOutput) EvalTree() EvalNode {
+	return &EvalOpFilter{
+		Ops: []walkOperation{walkRefresh, walkPlan, walkApply},
+		Node: &EvalSequence{
+			Nodes: []EvalNode{
+				&EvalWriteOutput{
+					Name:  n.Output.Name,
+					Value: n.Output.RawConfig,
+				},
+			},
+		},
+	}
+}
+
+// GraphNodeConfigProvider represents a configured provider within the
+// configuration graph. These are only immediately in the graph when an
+// explicit `provider` configuration block is in the configuration.
+type GraphNodeConfigProvider struct {
+	Provider *config.ProviderConfig
+}
+
+func (n *GraphNodeConfigProvider) Name() string {
+	return fmt.Sprintf("provider.%s", n.Provider.Name)
+}
+
+func (n *GraphNodeConfigProvider) DependableName() []string {
+	return []string{n.Name()}
+}
+
+func (n *GraphNodeConfigProvider) DependentOn() []string {
+	vars := n.Provider.RawConfig.Variables
+	result := make([]string, 0, len(vars))
+	for _, v := range vars {
+		if vn := varNameForVar(v); vn != "" {
+			result = append(result, vn)
+		}
+	}
+
+	return result
+}
+
+// GraphNodeEvalable impl.
+func (n *GraphNodeConfigProvider) EvalTree() EvalNode {
+	return ProviderEvalTree(n.Provider.Name, n.Provider.RawConfig)
+}
+
+// GraphNodeProvider implementation
+func (n *GraphNodeConfigProvider) ProviderName() string {
+	return n.Provider.Name
+}
+
+// GraphNodeDotter impl.
+func (n *GraphNodeConfigProvider) Dot(name string) string {
+	return fmt.Sprintf(
+		"\"%s\" [\n"+
+			"\tlabel=\"%s\"\n"+
+			"\tshape=diamond\n"+
+			"];",
+		name,
+		n.Name())
+}
+
+// GraphNodeConfigResource represents a resource within the config graph.
+type GraphNodeConfigResource struct {
+	Resource *config.Resource
+
+	// If this is set to anything other than destroyModeNone, then this
+	// resource represents a resource that will be destroyed in some way.
+	DestroyMode GraphNodeDestroyMode
+}
+
+func (n *GraphNodeConfigResource) DependableName() []string {
+	return []string{n.Resource.Id()}
+}
+
+// GraphNodeDependent impl.
+func (n *GraphNodeConfigResource) DependentOn() []string {
+	result := make([]string, len(n.Resource.DependsOn),
+		(len(n.Resource.RawCount.Variables)+
+			len(n.Resource.RawConfig.Variables)+
+			len(n.Resource.DependsOn))*2)
+	copy(result, n.Resource.DependsOn)
+
+	for _, v := range n.Resource.RawCount.Variables {
+		if vn := varNameForVar(v); vn != "" {
+			result = append(result, vn)
+		}
+	}
+	for _, v := range n.Resource.RawConfig.Variables {
+		if vn := varNameForVar(v); vn != "" {
+			result = append(result, vn)
+		}
+	}
+	for _, p := range n.Resource.Provisioners {
+		for _, v := range p.ConnInfo.Variables {
+			if vn := varNameForVar(v); vn != "" && vn != n.Resource.Id() {
+				result = append(result, vn)
+			}
+		}
+		for _, v := range p.RawConfig.Variables {
+			if vn := varNameForVar(v); vn != "" && vn != n.Resource.Id() {
+				result = append(result, vn)
+			}
+		}
+	}
+
+	return result
+}
+
+func (n *GraphNodeConfigResource) Name() string {
+	result := n.Resource.Id()
+	switch n.DestroyMode {
+	case DestroyNone:
+	case DestroyPrimary:
+		result += " (destroy)"
+	case DestroyTainted:
+		result += " (destroy tainted)"
+	default:
+		result += " (unknown destroy type)"
+	}
+
+	return result
+}
+
+// GraphNodeDotter impl.
+func (n *GraphNodeConfigResource) Dot(name string) string {
+	if n.DestroyMode != DestroyNone {
+		return ""
+	}
+
+	return fmt.Sprintf(
+		"\"%s\" [\n"+
+			"\tlabel=\"%s\"\n"+
+			"\tshape=box\n"+
+			"];",
+		name,
+		n.Name())
+}
+
+// GraphNodeDynamicExpandable impl.
+func (n *GraphNodeConfigResource) DynamicExpand(ctx EvalContext) (*Graph, error) {
+	state, lock := ctx.State()
+	lock.RLock()
+	defer lock.RUnlock()
+
+	// Start creating the steps
+	steps := make([]GraphTransformer, 0, 5)
+
+	// Primary and non-destroy modes are responsible for creating/destroying
+	// all the nodes, expanding counts.
+	switch n.DestroyMode {
+	case DestroyNone:
+		fallthrough
+	case DestroyPrimary:
+		steps = append(steps, &ResourceCountTransformer{
+			Resource: n.Resource,
+			Destroy:  n.DestroyMode != DestroyNone,
+		})
+	}
+
+	// Additional destroy modifications.
+	switch n.DestroyMode {
+	case DestroyPrimary:
+		// If we're destroying the primary instance, then we want to
+		// expand orphans, which have all the same semantics in a destroy
+		// as a primary.
+		steps = append(steps, &OrphanTransformer{
+			State: state,
+			View:  n.Resource.Id(),
+		})
+
+		// If we're only destroying tainted resources, then we only
+		// want to find tainted resources and destroy them here.
+		steps = append(steps, &TaintedTransformer{
+			State:          state,
+			View:           n.Resource.Id(),
+			Deposed:        n.Resource.Lifecycle.CreateBeforeDestroy,
+			DeposedInclude: true,
+		})
+	case DestroyTainted:
+		// If we're only destroying tainted resources, then we only
+		// want to find tainted resources and destroy them here.
+		steps = append(steps, &TaintedTransformer{
+			State:          state,
+			View:           n.Resource.Id(),
+			Deposed:        n.Resource.Lifecycle.CreateBeforeDestroy,
+			DeposedInclude: false,
+		})
+	}
+
+	// Always end with the root being added
+	steps = append(steps, &RootTransformer{})
+
+	// Build the graph
+	b := &BasicGraphBuilder{Steps: steps}
+	return b.Build(ctx.Path())
+}
+
+// GraphNodeEvalable impl.
+func (n *GraphNodeConfigResource) EvalTree() EvalNode {
+	return &EvalSequence{
+		Nodes: []EvalNode{
+			&EvalInterpolate{Config: n.Resource.RawCount},
+			&EvalOpFilter{
+				Ops:  []walkOperation{walkValidate},
+				Node: &EvalValidateCount{Resource: n.Resource},
+			},
+			&EvalCountFixZeroOneBoundary{Resource: n.Resource},
+		},
+	}
+}
+
+// GraphNodeProviderConsumer
+func (n *GraphNodeConfigResource) ProvidedBy() []string {
+	return []string{resourceProvider(n.Resource.Type)}
+}
+
+// GraphNodeProvisionerConsumer
+func (n *GraphNodeConfigResource) ProvisionedBy() []string {
+	result := make([]string, len(n.Resource.Provisioners))
+	for i, p := range n.Resource.Provisioners {
+		result[i] = p.Type
+	}
+
+	return result
+}
+
+// GraphNodeDestroyable
+func (n *GraphNodeConfigResource) DestroyNode(mode GraphNodeDestroyMode) GraphNodeDestroy {
+	// If we're already a destroy node, then don't do anything
+	if n.DestroyMode != DestroyNone {
+		return nil
+	}
+
+	result := &graphNodeResourceDestroy{
+		GraphNodeConfigResource: *n,
+		Original:                n,
+	}
+	result.DestroyMode = mode
+	return result
+}
+
+// graphNodeResourceDestroy represents the logical destruction of a
+// resource. This node doesn't mean it will be destroyed for sure, but
+// instead that if a destroy were to happen, it must happen at this point.
+type graphNodeResourceDestroy struct {
+	GraphNodeConfigResource
+	Original *GraphNodeConfigResource
+}
+
+func (n *graphNodeResourceDestroy) CreateBeforeDestroy() bool {
+	// CBD is enabled if the resource enables it in addition to us
+	// being responsible for destroying the primary state. The primary
+	// state destroy node is the only destroy node that needs to be
+	// "shuffled" according to the CBD rules, since tainted resources
+	// don't have the same inverse dependencies.
+	return n.Original.Resource.Lifecycle.CreateBeforeDestroy &&
+		n.DestroyMode == DestroyPrimary
+}
+
+func (n *graphNodeResourceDestroy) CreateNode() dag.Vertex {
+	return n.Original
+}
+
+func (n *graphNodeResourceDestroy) DestroyInclude(d *ModuleDiff, s *ModuleState) bool {
+	// Always include anything other than the primary destroy
+	if n.DestroyMode != DestroyPrimary {
+		return true
+	}
+
+	// Get the count, and specifically the raw value of the count
+	// (with interpolations and all). If the count is NOT a static "1",
+	// then we keep the destroy node no matter what.
+	//
+	// The reasoning for this is complicated and not intuitively obvious,
+	// but I attempt to explain it below.
+	//
+	// The destroy transform works by generating the worst case graph,
+	// with worst case being the case that every resource already exists
+	// and needs to be destroy/created (force-new). There is a single important
+	// edge case where this actually results in a real-life cycle: if a
+	// create-before-destroy (CBD) resource depends on a non-CBD resource.
+	// Imagine a EC2 instance "foo" with CBD depending on a security
+	// group "bar" without CBD, and conceptualize the worst case destroy
+	// order:
+	//
+	//   1.) SG must be destroyed (non-CBD)
+	//   2.) SG must be created/updated
+	//   3.) EC2 instance must be created (CBD, requires the SG be made)
+	//   4.) EC2 instance must be destroyed (requires SG be destroyed)
+	//
+	// Except, #1 depends on #4, since the SG can't be destroyed while
+	// an EC2 instance is using it (AWS API requirements). As you can see,
+	// this is a real life cycle that can't be automatically reconciled
+	// except under two conditions:
+	//
+	//   1.) SG is also CBD. This doesn't work 100% of the time though
+	//       since the non-CBD resource might not support CBD. To make matters
+	//       worse, the entire transitive closure of dependencies must be
+	//       CBD (if the SG depends on a VPC, you have the same problem).
+	//   2.) EC2 must not CBD. This can't happen automatically because CBD
+	//       is used as a way to ensure zero (or minimal) downtime Terraform
+	//       applies, and it isn't acceptable for TF to ignore this request,
+	//       since it can result in unexpected downtime.
+	//
+	// Therefore, we compromise with this edge case here: if there is
+	// a static count of "1", we prune the diff to remove cycles during a
+	// graph optimization path if we don't see the resource in the diff.
+	// If the count is set to ANYTHING other than a static "1" (variable,
+	// computed attribute, static number greater than 1), then we keep the
+	// destroy, since it is required for dynamic graph expansion to find
+	// orphan/tainted count objects.
+	//
+	// This isn't ideal logic, but its strictly better without introducing
+	// new impossibilities. It breaks the cycle in practical cases, and the
+	// cycle comes back in no cases we've found to be practical, but just
+	// as the cycle would already exist without this anyways.
+	count := n.Original.Resource.RawCount
+	if raw := count.Raw[count.Key]; raw != "1" {
+		return true
+	}
+
+	// Okay, we're dealing with a static count. There are a few ways
+	// to include this resource.
+	prefix := n.Original.Resource.Id()
+
+	// If we're present in the diff proper, then keep it.
+	if d != nil {
+		for k, _ := range d.Resources {
+			if strings.HasPrefix(k, prefix) {
+				return true
+			}
+		}
+	}
+
+	// If we're in the state as a primary in any form, then keep it.
+	// This does a prefix check so it will also catch orphans on count
+	// decreases to "1".
+	if s != nil {
+		for k, v := range s.Resources {
+			if !strings.HasPrefix(k, prefix) {
+				continue
+			}
+
+			// Ignore exact matches and the 0'th index. We only care
+			// about if there is a decrease in count.
+			if k == prefix {
+				continue
+			}
+			if k == prefix+".0" {
+				continue
+			}
+
+			if v.Primary != nil {
+				return true
+			}
+		}
+	}
+
+	return false
+}
+
+// graphNodeModuleExpanded represents a module where the graph has
+// been expanded. It stores the graph of the module as well as a reference
+// to the map of variables.
+type graphNodeModuleExpanded struct {
+	Original    dag.Vertex
+	Graph       *Graph
+	InputConfig *config.RawConfig
+
+	// Variables is a map of the input variables. This reference should
+	// be shared with ModuleInputTransformer in order to create a connection
+	// where the variables are set properly.
+	Variables map[string]string
+}
+
+func (n *graphNodeModuleExpanded) Name() string {
+	return fmt.Sprintf("%s (expanded)", dag.VertexName(n.Original))
+}
+
+// GraphNodeDotter impl.
+func (n *graphNodeModuleExpanded) Dot(name string) string {
+	return fmt.Sprintf(
+		"\"%s\" [\n"+
+			"\tlabel=\"%s\"\n"+
+			"\tshape=component\n"+
+			"];",
+		name,
+		dag.VertexName(n.Original))
+}
+
+// GraphNodeEvalable impl.
+func (n *graphNodeModuleExpanded) EvalTree() EvalNode {
+	var resourceConfig *ResourceConfig
+	return &EvalSequence{
+		Nodes: []EvalNode{
+			&EvalInterpolate{
+				Config: n.InputConfig,
+				Output: &resourceConfig,
+			},
+
+			&EvalVariableBlock{
+				Config:    &resourceConfig,
+				Variables: n.Variables,
+			},
+
+			&EvalOpFilter{
+				Ops: []walkOperation{walkPlanDestroy},
+				Node: &EvalSequence{
+					Nodes: []EvalNode{
+						&EvalDiffDestroyModule{Path: n.Graph.Path},
+					},
+				},
+			},
+		},
+	}
+}
+
+// GraphNodeSubgraph impl.
+func (n *graphNodeModuleExpanded) Subgraph() *Graph {
+	return n.Graph
+}

terraform/graph_config_node_test.go

@@ -0,0 +1,103 @@
+package terraform
+
+import (
+	"reflect"
+	"strings"
+	"testing"
+
+	"github.com/hashicorp/terraform/config"
+	"github.com/hashicorp/terraform/dag"
+)
+
+func TestGraphNodeConfigModule_impl(t *testing.T) {
+	var _ dag.Vertex = new(GraphNodeConfigModule)
+	var _ dag.NamedVertex = new(GraphNodeConfigModule)
+	var _ graphNodeConfig = new(GraphNodeConfigModule)
+	var _ GraphNodeExpandable = new(GraphNodeConfigModule)
+}
+
+func TestGraphNodeConfigModuleExpand(t *testing.T) {
+	mod := testModule(t, "graph-node-module-expand")
+
+	node := &GraphNodeConfigModule{
+		Path:   []string{RootModuleName, "child"},
+		Module: &config.Module{},
+		Tree:   nil,
+	}
+
+	g, err := node.Expand(&BasicGraphBuilder{
+		Steps: []GraphTransformer{
+			&ConfigTransformer{Module: mod},
+		},
+	})
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual := strings.TrimSpace(g.Subgraph().String())
+	expected := strings.TrimSpace(testGraphNodeModuleExpandStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestGraphNodeConfigProvider_impl(t *testing.T) {
+	var _ dag.Vertex = new(GraphNodeConfigProvider)
+	var _ dag.NamedVertex = new(GraphNodeConfigProvider)
+	var _ graphNodeConfig = new(GraphNodeConfigProvider)
+	var _ GraphNodeProvider = new(GraphNodeConfigProvider)
+}
+
+func TestGraphNodeConfigProvider_ProviderName(t *testing.T) {
+	n := &GraphNodeConfigProvider{
+		Provider: &config.ProviderConfig{Name: "foo"},
+	}
+
+	if v := n.ProviderName(); v != "foo" {
+		t.Fatalf("bad: %#v", v)
+	}
+}
+
+func TestGraphNodeConfigResource_impl(t *testing.T) {
+	var _ dag.Vertex = new(GraphNodeConfigResource)
+	var _ dag.NamedVertex = new(GraphNodeConfigResource)
+	var _ graphNodeConfig = new(GraphNodeConfigResource)
+	var _ GraphNodeProviderConsumer = new(GraphNodeConfigResource)
+	var _ GraphNodeProvisionerConsumer = new(GraphNodeConfigResource)
+}
+
+func TestGraphNodeConfigResource_ProvidedBy(t *testing.T) {
+	n := &GraphNodeConfigResource{
+		Resource: &config.Resource{Type: "aws_instance"},
+	}
+
+	if v := n.ProvidedBy(); v[0] != "aws" {
+		t.Fatalf("bad: %#v", v)
+	}
+}
+
+func TestGraphNodeConfigResource_ProvisionedBy(t *testing.T) {
+	n := &GraphNodeConfigResource{
+		Resource: &config.Resource{
+			Type: "aws_instance",
+			Provisioners: []*config.Provisioner{
+				&config.Provisioner{Type: "foo"},
+				&config.Provisioner{Type: "bar"},
+			},
+		},
+	}
+
+	expected := []string{"foo", "bar"}
+	actual := n.ProvisionedBy()
+	if !reflect.DeepEqual(actual, expected) {
+		t.Fatalf("bad: %#v", actual)
+	}
+}
+
+const testGraphNodeModuleExpandStr = `
+aws_instance.bar
+  aws_instance.foo
+aws_instance.foo
+  module inputs
+module inputs
+`

terraform/graph_dot.go

@@ -6,371 +6,66 @@ import (
 	"fmt"
 	"strings"
 
-	"github.com/hashicorp/terraform/depgraph"
+	"github.com/hashicorp/terraform/dag"
 )
 
-// GraphDotOpts are options for turning a graph into dot format.
-type GraphDotOpts struct {
-	// ModuleDepth is the depth of modules to expand. Zero is no expansion,
-	// one expands the first set of modules, etc. If this is set to -1, then
-	// all modules are expanded.
-	ModuleDepth int
-
-	// Depth is an internal track of what depth we're at within
-	// the graph, used to control indentation and other such things.
-	depth int
+// GraphNodeDotter can be implemented by a node to cause it to be included
+// in the dot graph. The Dot method will be called which is expected to
+// return a representation of this node.
+type GraphNodeDotter interface {
+	// Dot is called to return the dot formatting for the node.
+	// The parameter must be the title of the node.
+	Dot(string) string
 }
 
+// GraphDotOpts are the options for generating a dot formatted Graph.
+type GraphDotOpts struct{}
+
 // GraphDot returns the dot formatting of a visual representation of
 // the given Terraform graph.
-func GraphDot(g *depgraph.Graph, opts *GraphDotOpts) string {
+func GraphDot(g *Graph, opts *GraphDotOpts) string {
 	buf := new(bytes.Buffer)
 
-	if opts.depth == 0 {
-		buf.WriteString("digraph {\n")
-		buf.WriteString("\tcompound = true;\n")
+	// Start the graph
+	buf.WriteString("digraph {\n")
+	buf.WriteString("\tcompound = true;\n")
+
+	// Go through all the vertices and draw it
+	vertices := g.Vertices()
+	dotVertices := make(map[dag.Vertex]struct{}, len(vertices))
+	for _, v := range vertices {
+		if dn, ok := v.(GraphNodeDotter); !ok {
+			continue
+		} else if dn.Dot("fake") == "" {
+			continue
+		}
+
+		dotVertices[v] = struct{}{}
 	}
 
-	// Determine and add the title
-	// graphDotTitle(buf, g)
+	for v, _ := range dotVertices {
+		dn := v.(GraphNodeDotter)
+		scanner := bufio.NewScanner(strings.NewReader(
+			dn.Dot(dag.VertexName(v))))
+		for scanner.Scan() {
+			buf.WriteString("\t" + scanner.Text() + "\n")
+		}
 
-	// Add all the resource.
-	graphDotAddResources(buf, g, opts)
+		// Draw all the edges
+		for _, t := range g.DownEdges(v).List() {
+			target := t.(dag.Vertex)
+			if _, ok := dotVertices[target]; !ok {
+				continue
+			}
 
-	// Add all the resource providers
-	graphDotAddResourceProviders(buf, g, opts)
-
-	// Add all the modules
-	graphDotAddModules(buf, g, opts)
-
-	if opts.depth == 0 {
-		buf.WriteString("}\n")
+			buf.WriteString(fmt.Sprintf(
+				"\t\"%s\" -> \"%s\";\n",
+				dag.VertexName(v),
+				dag.VertexName(target)))
+		}
 	}
 
+	// End the graph
+	buf.WriteString("}\n")
 	return buf.String()
 }
-
-func graphDotAddRoot(buf *bytes.Buffer, n *depgraph.Noun) {
-	buf.WriteString(fmt.Sprintf("\t\"%s\" [shape=circle];\n", "root"))
-
-	for _, e := range n.Edges() {
-		target := e.Tail()
-		buf.WriteString(fmt.Sprintf(
-			"\t\"%s\" -> \"%s\";\n",
-			"root",
-			target))
-	}
-}
-
-func graphDotAddModules(buf *bytes.Buffer, g *depgraph.Graph, opts *GraphDotOpts) {
-	for _, n := range g.Nouns {
-		_, ok := n.Meta.(*GraphNodeModule)
-		if !ok {
-			continue
-		}
-
-		if graphExpand(opts) {
-			// We're expanding
-			graphDotAddModuleExpand(buf, n, opts)
-		} else {
-			// We're not expanding, so just add the module on its own
-			graphDotAddModuleSingle(buf, n, opts)
-		}
-
-		graphWriteEdges(buf, n, opts)
-	}
-}
-
-func graphDotAddModuleExpand(
-	buf *bytes.Buffer, n *depgraph.Noun, opts *GraphDotOpts) {
-	m := n.Meta.(*GraphNodeModule)
-	tab := strings.Repeat("\t", opts.depth+1)
-	uniqueName := graphUniqueName(n, opts)
-
-	// Wrap ourselves in a subgraph
-	buf.WriteString(fmt.Sprintf("%ssubgraph \"cluster_%s\" {\n", tab, uniqueName))
-	defer buf.WriteString(fmt.Sprintf("%s}\n", tab))
-
-	// Add our label so that we have the proper name.
-	buf.WriteString(fmt.Sprintf("%s\tlabel = \"%s\";\n", tab, n))
-
-	// Add a hidden name for edges to point from/to
-	buf.WriteString(fmt.Sprintf("%s\t\"%s_hidden\" [fixedsize=true,width=0,height=0,label=\"\",style=invisible];\n", tab, uniqueName))
-
-	// Graph the subgraph just as we would any other graph
-	subOpts := *opts
-	subOpts.depth++
-	subStr := GraphDot(m.Graph, &subOpts)
-
-	// Tab all the lines of the subgraph
-	s := bufio.NewScanner(strings.NewReader(subStr))
-	for s.Scan() {
-		buf.WriteString(fmt.Sprintf("%s%s\n", tab, s.Text()))
-	}
-}
-
-func graphDotAddModuleSingle(
-	buf *bytes.Buffer, n *depgraph.Noun, opts *GraphDotOpts) {
-	tab := strings.Repeat("\t", opts.depth+1)
-	uniqueName := graphUniqueName(n, opts)
-
-	// Create this node.
-	buf.WriteString(fmt.Sprintf("%s\"%s\" [\n", tab, uniqueName))
-	buf.WriteString(fmt.Sprintf("%s\tlabel=\"%s\"\n", tab, n))
-	buf.WriteString(fmt.Sprintf("%s\tshape=component\n", tab))
-	buf.WriteString(fmt.Sprintf("%s];\n", tab))
-}
-
-func graphDotAddResources(
-	buf *bytes.Buffer, g *depgraph.Graph, opts *GraphDotOpts) {
-	// Determine if we have diffs. If we do, then we're graphing a
-	// plan, which alters our graph a bit.
-	hasDiff := false
-	for _, n := range g.Nouns {
-		rn, ok := n.Meta.(*GraphNodeResource)
-		if !ok {
-			continue
-		}
-		if rn.Resource.Diff != nil && !rn.Resource.Diff.Empty() {
-			hasDiff = true
-			break
-		}
-	}
-
-	var edgeBuf bytes.Buffer
-	// Do all the non-destroy resources
-	buf.WriteString("\tsubgraph {\n")
-	for _, n := range g.Nouns {
-		rn, ok := n.Meta.(*GraphNodeResource)
-		if !ok {
-			continue
-		}
-		if rn.Resource.Diff != nil && rn.Resource.Diff.Destroy {
-			continue
-		}
-
-		// If we have diffs then we're graphing a plan. If we don't have
-		// have a diff on this resource, don't graph anything, since the
-		// plan wouldn't do anything to this resource.
-		if hasDiff {
-			if rn.Resource.Diff == nil || rn.Resource.Diff.Empty() {
-				continue
-			}
-		}
-
-		// Determine the colors. White = no change, yellow = change,
-		// green = create. Destroy is in the next section.
-		var color, fillColor string
-		if rn.Resource.Diff != nil && !rn.Resource.Diff.Empty() {
-			if rn.Resource.State != nil && rn.Resource.State.ID != "" {
-				color = "#FFFF00"
-				fillColor = "#FFFF94"
-			} else {
-				color = "#00FF00"
-				fillColor = "#9EFF9E"
-			}
-		}
-
-		uniqueName := fmt.Sprintf("%d_%s", opts.depth, n)
-
-		// Create this node.
-		buf.WriteString(fmt.Sprintf("\t\t\"%s\" [\n", uniqueName))
-		buf.WriteString(fmt.Sprintf("\t\t\tlabel=\"%s\"\n", n))
-		buf.WriteString("\t\t\tshape=box\n")
-		if color != "" {
-			buf.WriteString("\t\t\tstyle=filled\n")
-			buf.WriteString(fmt.Sprintf("\t\t\tcolor=\"%s\"\n", color))
-			buf.WriteString(fmt.Sprintf("\t\t\tfillcolor=\"%s\"\n", fillColor))
-		}
-		buf.WriteString("\t\t];\n")
-
-		// Build up all the edges in a separate buffer so they're not in the
-		// subgraph.
-		graphWriteEdges(&edgeBuf, n, opts)
-	}
-	buf.WriteString("\t}\n\n")
-	if edgeBuf.Len() > 0 {
-		buf.WriteString(edgeBuf.String())
-		buf.WriteString("\n")
-	}
-
-	// Do all the destroy resources
-	edgeBuf.Reset()
-	buf.WriteString("\tsubgraph {\n")
-	for _, n := range g.Nouns {
-		rn, ok := n.Meta.(*GraphNodeResource)
-		if !ok {
-			continue
-		}
-		if rn.Resource.Diff == nil || !rn.Resource.Diff.Destroy {
-			continue
-		}
-
-		uniqueName := fmt.Sprintf("%d_%s", opts.depth, n)
-
-		buf.WriteString(fmt.Sprintf(
-			"\t\t\"%s\" [label=\"%s\",shape=box,style=filled,color=\"#FF0000\",fillcolor=\"#FF9494\"];\n", uniqueName, n))
-
-		graphWriteEdges(&edgeBuf, n, opts)
-	}
-	buf.WriteString("\t}\n\n")
-	if edgeBuf.Len() > 0 {
-		buf.WriteString(edgeBuf.String())
-		buf.WriteString("\n")
-	}
-
-	// Handle the meta resources
-	/*
-		edgeBuf.Reset()
-		for _, n := range g.Nouns {
-			_, ok := n.Meta.(*GraphNodeResourceMeta)
-			if !ok {
-				continue
-			}
-
-			// Determine which edges to add
-			var edges []digraph.Edge
-			if hasDiff {
-				for _, e := range n.Edges() {
-					rn, ok := e.Tail().(*depgraph.Noun).Meta.(*GraphNodeResource)
-					if !ok {
-						continue
-					}
-					if rn.Resource.Diff == nil || rn.Resource.Diff.Empty() {
-						continue
-					}
-					edges = append(edges, e)
-				}
-			} else {
-				edges = n.Edges()
-			}
-
-			// Do not draw if we have no edges
-			if len(edges) == 0 {
-				continue
-			}
-
-			uniqueName := fmt.Sprintf("%d_%s", opts.depth, n)
-			for _, e := range edges {
-				target := e.Tail()
-				uniqueTarget := fmt.Sprintf("%d_%s", opts.depth, target)
-				edgeBuf.WriteString(fmt.Sprintf(
-					"\t\"%s\" -> \"%s\";\n",
-					uniqueName,
-					uniqueTarget))
-			}
-		}
-		if edgeBuf.Len() > 0 {
-			buf.WriteString(edgeBuf.String())
-			buf.WriteString("\n")
-		}
-	*/
-}
-
-func graphDotAddResourceProviders(
-	buf *bytes.Buffer, g *depgraph.Graph, opts *GraphDotOpts) {
-	var edgeBuf bytes.Buffer
-	buf.WriteString("\tsubgraph {\n")
-	for _, n := range g.Nouns {
-		_, ok := n.Meta.(*GraphNodeResourceProvider)
-		if !ok {
-			continue
-		}
-
-		uniqueName := fmt.Sprintf("%d_%s", opts.depth, n)
-
-		// Create this node.
-		buf.WriteString(fmt.Sprintf("\t\t\"%s\" [\n", uniqueName))
-		buf.WriteString(fmt.Sprintf("\t\t\tlabel=\"%s\"\n", n))
-		buf.WriteString("\t\t\tshape=diamond\n")
-		buf.WriteString("\t\t];\n")
-
-		// Build up all the edges in a separate buffer so they're not in the
-		// subgraph.
-		graphWriteEdges(&edgeBuf, n, opts)
-	}
-	buf.WriteString("\t}\n\n")
-	if edgeBuf.Len() > 0 {
-		buf.WriteString(edgeBuf.String())
-		buf.WriteString("\n")
-	}
-}
-
-func graphDotTitle(buf *bytes.Buffer, g *depgraph.Graph) {
-	// Determine if we have diffs. If we do, then we're graphing a
-	// plan, which alters our graph a bit.
-	hasDiff := false
-	for _, n := range g.Nouns {
-		rn, ok := n.Meta.(*GraphNodeResource)
-		if !ok {
-			continue
-		}
-		if rn.Resource.Diff != nil && !rn.Resource.Diff.Empty() {
-			hasDiff = true
-			break
-		}
-	}
-
-	graphType := "Configuration"
-	if hasDiff {
-		graphType = "Plan"
-	}
-	title := fmt.Sprintf("Terraform %s Resource Graph", graphType)
-
-	buf.WriteString(fmt.Sprintf("\tlabel=\"%s\\n\\n\\n\";\n", title))
-	buf.WriteString("\tlabelloc=\"t\";\n\n")
-}
-
-func graphExpand(opts *GraphDotOpts) bool {
-	return opts.ModuleDepth > opts.depth || opts.ModuleDepth == -1
-}
-
-func graphUniqueName(n *depgraph.Noun, opts *GraphDotOpts) string {
-	return fmt.Sprintf("%d_%s", opts.depth, n)
-}
-
-func graphWriteEdges(
-	buf *bytes.Buffer, n *depgraph.Noun, opts *GraphDotOpts) {
-	tab := strings.Repeat("\t", opts.depth+1)
-
-	uniqueName := graphUniqueName(n, opts)
-	var ltail string
-	if _, ok := n.Meta.(*GraphNodeModule); ok && graphExpand(opts) {
-		ltail = "cluster_" + uniqueName
-		uniqueName = uniqueName + "_hidden"
-	}
-
-	for _, e := range n.Edges() {
-		target := e.Tail()
-		targetN := target.(*depgraph.Noun)
-		uniqueTarget := graphUniqueName(targetN, opts)
-
-		var lhead string
-		if _, ok := targetN.Meta.(*GraphNodeModule); ok && graphExpand(opts) {
-			lhead = "cluster_" + uniqueTarget
-			uniqueTarget = uniqueTarget + "_hidden"
-		}
-
-		var attrs string
-		if lhead != "" || ltail != "" {
-			var attrList []string
-			if lhead != "" {
-				attrList = append(attrList, fmt.Sprintf(
-					"lhead=\"%s\"", lhead))
-			}
-			if ltail != "" {
-				attrList = append(attrList, fmt.Sprintf(
-					"ltail=\"%s\"", ltail))
-			}
-
-			attrs = fmt.Sprintf(" [%s]", strings.Join(attrList, ","))
-		}
-
-		buf.WriteString(fmt.Sprintf(
-			"%s\"%s\" -> \"%s\"%s;\n",
-			tab,
-			uniqueName,
-			uniqueTarget,
-			attrs))
-	}
-}

terraform/graph_walk.go

@@ -0,0 +1,30 @@
+package terraform
+
+import (
+	"github.com/hashicorp/terraform/dag"
+)
+
+// GraphWalker is an interface that can be implemented that when used
+// with Graph.Walk will invoke the given callbacks under certain events.
+type GraphWalker interface {
+	EnterGraph(*Graph) EvalContext
+	ExitGraph(*Graph)
+	EnterVertex(dag.Vertex)
+	ExitVertex(dag.Vertex, error)
+	EnterEvalTree(dag.Vertex, EvalNode) EvalNode
+	ExitEvalTree(dag.Vertex, interface{}, error) error
+}
+
+// NullGraphWalker is a GraphWalker implementation that does nothing.
+// This can be embedded within other GraphWalker implementations for easily
+// implementing all the required functions.
+type NullGraphWalker struct{}
+
+func (NullGraphWalker) EnterGraph(*Graph) EvalContext                   { return nil }
+func (NullGraphWalker) ExitGraph(*Graph)                                {}
+func (NullGraphWalker) EnterVertex(dag.Vertex)                          {}
+func (NullGraphWalker) ExitVertex(dag.Vertex, error)                    {}
+func (NullGraphWalker) EnterEvalTree(v dag.Vertex, n EvalNode) EvalNode { return n }
+func (NullGraphWalker) ExitEvalTree(dag.Vertex, interface{}, error) error {
+	return nil
+}

terraform/graph_walk_context.go

@@ -0,0 +1,134 @@
+package terraform
+
+import (
+	"fmt"
+	"sync"
+
+	"github.com/hashicorp/errwrap"
+	"github.com/hashicorp/terraform/dag"
+)
+
+// ContextGraphWalker is the GraphWalker implementation used with the
+// Context struct to walk and evaluate the graph.
+type ContextGraphWalker struct {
+	NullGraphWalker
+
+	// Configurable values
+	Context   *Context
+	Operation walkOperation
+
+	// Outputs, do not set these. Do not read these while the graph
+	// is being walked.
+	ValidationWarnings []string
+	ValidationErrors   []error
+
+	errorLock           sync.Mutex
+	once                sync.Once
+	contexts            map[string]*BuiltinEvalContext
+	contextLock         sync.Mutex
+	providerCache       map[string]ResourceProvider
+	providerConfigCache map[string]*ResourceConfig
+	providerLock        sync.Mutex
+	provisionerCache    map[string]ResourceProvisioner
+	provisionerLock     sync.Mutex
+}
+
+func (w *ContextGraphWalker) EnterGraph(g *Graph) EvalContext {
+	w.once.Do(w.init)
+
+	w.contextLock.Lock()
+	defer w.contextLock.Unlock()
+
+	// If we already have a context for this path cached, use that
+	key := PathCacheKey(g.Path)
+	if ctx, ok := w.contexts[key]; ok {
+		return ctx
+	}
+
+	// Variables should be our context variables, but these only apply
+	// to the root module. As we enter subgraphs, we don't want to set
+	// variables, which is set by the SetVariables EvalContext function.
+	variables := w.Context.variables
+	if len(g.Path) > 1 {
+		// We're in a submodule, the variables should be empty
+		variables = make(map[string]string)
+	}
+
+	ctx := &BuiltinEvalContext{
+		PathValue:           g.Path,
+		Hooks:               w.Context.hooks,
+		InputValue:          w.Context.uiInput,
+		Providers:           w.Context.providers,
+		ProviderCache:       w.providerCache,
+		ProviderConfigCache: w.providerConfigCache,
+		ProviderInputConfig: w.Context.providerInputConfig,
+		ProviderLock:        &w.providerLock,
+		Provisioners:        w.Context.provisioners,
+		ProvisionerCache:    w.provisionerCache,
+		ProvisionerLock:     &w.provisionerLock,
+		DiffValue:           w.Context.diff,
+		DiffLock:            &w.Context.diffLock,
+		StateValue:          w.Context.state,
+		StateLock:           &w.Context.stateLock,
+		Interpolater: &Interpolater{
+			Operation: w.Operation,
+			Module:    w.Context.module,
+			State:     w.Context.state,
+			StateLock: &w.Context.stateLock,
+			Variables: variables,
+		},
+	}
+
+	w.contexts[key] = ctx
+	return ctx
+}
+
+func (w *ContextGraphWalker) EnterEvalTree(v dag.Vertex, n EvalNode) EvalNode {
+	// Acquire a lock on the semaphore
+	w.Context.parallelSem.Acquire()
+
+	// We want to filter the evaluation tree to only include operations
+	// that belong in this operation.
+	return EvalFilter(n, EvalNodeFilterOp(w.Operation))
+}
+
+func (w *ContextGraphWalker) ExitEvalTree(
+	v dag.Vertex, output interface{}, err error) error {
+	// Release the semaphore
+	w.Context.parallelSem.Release()
+
+	if err == nil {
+		return nil
+	}
+
+	// Acquire the lock because anything is going to require a lock.
+	w.errorLock.Lock()
+	defer w.errorLock.Unlock()
+
+	// Try to get a validation error out of it. If its not a validation
+	// error, then just record the normal error.
+	verr, ok := err.(*EvalValidateError)
+	if !ok {
+		return err
+	}
+
+	for _, msg := range verr.Warnings {
+		w.ValidationWarnings = append(
+			w.ValidationWarnings,
+			fmt.Sprintf("%s: %s", dag.VertexName(v), msg))
+	}
+	for _, e := range verr.Errors {
+		w.ValidationErrors = append(
+			w.ValidationErrors,
+			errwrap.Wrapf(fmt.Sprintf("%s: {{err}}", dag.VertexName(v)), e))
+	}
+
+	return nil
+}
+
+func (w *ContextGraphWalker) init() {
+	w.contexts = make(map[string]*BuiltinEvalContext, 5)
+	w.providerCache = make(map[string]ResourceProvider, 5)
+	w.providerConfigCache = make(map[string]*ResourceConfig, 5)
+	w.provisionerCache = make(map[string]ResourceProvisioner, 5)
+}

terraform/graph_walk_operation.go

@@ -0,0 +1,16 @@
+package terraform
+
+//go:generate stringer -type=walkOperation graph_walk_operation.go
+
+// walkOperation is an enum which tells the walkContext what to do.
+type walkOperation byte
+
+const (
+	walkInvalid walkOperation = iota
+	walkInput
+	walkApply
+	walkPlan
+	walkPlanDestroy
+	walkRefresh
+	walkValidate
+)

terraform/graph_walk_test.go

@@ -0,0 +1,9 @@
+package terraform
+
+import (
+	"testing"
+)
+
+func TestNullGraphWalker_impl(t *testing.T) {
+	var _ GraphWalker = NullGraphWalker{}
+}

terraform/interpolate.go

@@ -0,0 +1,440 @@
+package terraform
+
+import (
+	"fmt"
+	"os"
+	"strings"
+	"sync"
+
+	"github.com/hashicorp/terraform/config"
+	"github.com/hashicorp/terraform/config/lang/ast"
+	"github.com/hashicorp/terraform/config/module"
+)
+
+// Interpolater is the structure responsible for determining the values
+// for interpolations such as `aws_instance.foo.bar`.
+type Interpolater struct {
+	Operation walkOperation
+	Module    *module.Tree
+	State     *State
+	StateLock *sync.RWMutex
+	Variables map[string]string
+}
+
+// InterpolationScope is the current scope of execution. This is required
+// since some variables which are interpolated are dependent on what we're
+// operating on and where we are.
+type InterpolationScope struct {
+	Path     []string
+	Resource *Resource
+}
+
+// Values returns the values for all the variables in the given map.
+func (i *Interpolater) Values(
+	scope *InterpolationScope,
+	vars map[string]config.InterpolatedVariable) (map[string]ast.Variable, error) {
+	result := make(map[string]ast.Variable, len(vars))
+
+	// Copy the default variables
+	if i.Module != nil && scope != nil {
+		mod := i.Module
+		if len(scope.Path) > 1 {
+			mod = i.Module.Child(scope.Path[1:])
+		}
+		for _, v := range mod.Config().Variables {
+			for k, val := range v.DefaultsMap() {
+				result[k] = ast.Variable{
+					Value: val,
+					Type:  ast.TypeString,
+				}
+			}
+		}
+	}
+
+	for n, rawV := range vars {
+		var err error
+		switch v := rawV.(type) {
+		case *config.CountVariable:
+			err = i.valueCountVar(scope, n, v, result)
+		case *config.ModuleVariable:
+			err = i.valueModuleVar(scope, n, v, result)
+		case *config.PathVariable:
+			err = i.valuePathVar(scope, n, v, result)
+		case *config.ResourceVariable:
+			err = i.valueResourceVar(scope, n, v, result)
+		case *config.UserVariable:
+			err = i.valueUserVar(scope, n, v, result)
+		default:
+			err = fmt.Errorf("%s: unknown variable type: %T", n, rawV)
+		}
+
+		if err != nil {
+			return nil, err
+		}
+	}
+
+	return result, nil
+}
+
+func (i *Interpolater) valueCountVar(
+	scope *InterpolationScope,
+	n string,
+	v *config.CountVariable,
+	result map[string]ast.Variable) error {
+	switch v.Type {
+	case config.CountValueIndex:
+		result[n] = ast.Variable{
+			Value: scope.Resource.CountIndex,
+			Type:  ast.TypeInt,
+		}
+		return nil
+	default:
+		return fmt.Errorf("%s: unknown count type: %#v", n, v.Type)
+	}
+}
+
+func (i *Interpolater) valueModuleVar(
+	scope *InterpolationScope,
+	n string,
+	v *config.ModuleVariable,
+	result map[string]ast.Variable) error {
+	// If we're computing all dynamic fields, then module vars count
+	// and we mark it as computed.
+	if i.Operation == walkValidate {
+		result[n] = ast.Variable{
+			Value: config.UnknownVariableValue,
+			Type:  ast.TypeString,
+		}
+		return nil
+	}
+
+	// Build the path to the child module we want
+	path := make([]string, len(scope.Path), len(scope.Path)+1)
+	copy(path, scope.Path)
+	path = append(path, v.Name)
+
+	// Grab the lock so that if other interpolations are running or
+	// state is being modified, we'll be safe.
+	i.StateLock.RLock()
+	defer i.StateLock.RUnlock()
+
+	// Get the module where we're looking for the value
+	var value string
+	mod := i.State.ModuleByPath(path)
+	if mod == nil {
+		// If the module doesn't exist, then we can return an empty string.
+		// This happens usually only in Refresh() when we haven't populated
+		// a state. During validation, we semantically verify that all
+		// modules reference other modules, and graph ordering should
+		// ensure that the module is in the state, so if we reach this
+		// point otherwise it really is a panic.
+		value = config.UnknownVariableValue
+	} else {
+		// Get the value from the outputs
+		var ok bool
+		value, ok = mod.Outputs[v.Field]
+		if !ok {
+			// Same reasons as the comment above.
+			value = config.UnknownVariableValue
+		}
+	}
+
+	result[n] = ast.Variable{
+		Value: value,
+		Type:  ast.TypeString,
+	}
+	return nil
+}
+
+func (i *Interpolater) valuePathVar(
+	scope *InterpolationScope,
+	n string,
+	v *config.PathVariable,
+	result map[string]ast.Variable) error {
+	switch v.Type {
+	case config.PathValueCwd:
+		wd, err := os.Getwd()
+		if err != nil {
+			return fmt.Errorf(
+				"Couldn't get cwd for var %s: %s",
+				v.FullKey(), err)
+		}
+
+		result[n] = ast.Variable{
+			Value: wd,
+			Type:  ast.TypeString,
+		}
+	case config.PathValueModule:
+		if t := i.Module.Child(scope.Path[1:]); t != nil {
+			result[n] = ast.Variable{
+				Value: t.Config().Dir,
+				Type:  ast.TypeString,
+			}
+		}
+	case config.PathValueRoot:
+		result[n] = ast.Variable{
+			Value: i.Module.Config().Dir,
+			Type:  ast.TypeString,
+		}
+	default:
+		return fmt.Errorf("%s: unknown path type: %#v", n, v.Type)
+	}
+
+	return nil
+
+}
+
+func (i *Interpolater) valueResourceVar(
+	scope *InterpolationScope,
+	n string,
+	v *config.ResourceVariable,
+	result map[string]ast.Variable) error {
+	// If we're computing all dynamic fields, then module vars count
+	// and we mark it as computed.
+	if i.Operation == walkValidate || i.Operation == walkRefresh {
+		result[n] = ast.Variable{
+			Value: config.UnknownVariableValue,
+			Type:  ast.TypeString,
+		}
+		return nil
+	}
+
+	var attr string
+	var err error
+	if v.Multi && v.Index == -1 {
+		attr, err = i.computeResourceMultiVariable(scope, v)
+	} else {
+		attr, err = i.computeResourceVariable(scope, v)
+	}
+	if err != nil {
+		return err
+	}
+
+	result[n] = ast.Variable{
+		Value: attr,
+		Type:  ast.TypeString,
+	}
+	return nil
+}
+
+func (i *Interpolater) valueUserVar(
+	scope *InterpolationScope,
+	n string,
+	v *config.UserVariable,
+	result map[string]ast.Variable) error {
+	val, ok := i.Variables[v.Name]
+	if ok {
+		result[n] = ast.Variable{
+			Value: val,
+			Type:  ast.TypeString,
+		}
+		return nil
+	}
+
+	if _, ok := result[n]; !ok && i.Operation == walkValidate {
+		result[n] = ast.Variable{
+			Value: config.UnknownVariableValue,
+			Type:  ast.TypeString,
+		}
+		return nil
+	}
+
+	// Look up if we have any variables with this prefix because
+	// those are map overrides. Include those.
+	for k, val := range i.Variables {
+		if strings.HasPrefix(k, v.Name+".") {
+			result["var."+k] = ast.Variable{
+				Value: val,
+				Type:  ast.TypeString,
+			}
+		}
+	}
+
+	return nil
+}
+
+func (i *Interpolater) computeResourceVariable(
+	scope *InterpolationScope,
+	v *config.ResourceVariable) (string, error) {
+	id := v.ResourceId()
+	if v.Multi {
+		id = fmt.Sprintf("%s.%d", id, v.Index)
+	}
+
+	i.StateLock.RLock()
+	defer i.StateLock.RUnlock()
+
+	// Get the information about this resource variable, and verify
+	// that it exists and such.
+	module, _, err := i.resourceVariableInfo(scope, v)
+	if err != nil {
+		return "", err
+	}
+
+	// If we have no module in the state yet or count, return empty
+	if module == nil || len(module.Resources) == 0 {
+		return "", nil
+	}
+
+	// Get the resource out from the state. We know the state exists
+	// at this point and if there is a state, we expect there to be a
+	// resource with the given name.
+	r, ok := module.Resources[id]
+	if !ok && v.Multi && v.Index == 0 {
+		r, ok = module.Resources[v.ResourceId()]
+	}
+	if !ok {
+		r = nil
+	}
+	if r == nil {
+		return "", fmt.Errorf(
+			"Resource '%s' not found for variable '%s'",
+			id,
+			v.FullKey())
+	}
+
+	if r.Primary == nil {
+		goto MISSING
+	}
+
+	if attr, ok := r.Primary.Attributes[v.Field]; ok {
+		return attr, nil
+	}
+
+	// At apply time, we can't do the "maybe has it" check below
+	// that we need for plans since parent elements might be computed.
+	// Therefore, it is an error and we're missing the key.
+	//
+	// TODO: test by creating a state and configuration that is referencing
+	// a non-existent variable "foo.bar" where the state only has "foo"
+	// and verify plan works, but apply doesn't.
+	if i.Operation == walkApply {
+		goto MISSING
+	}
+
+	// We didn't find the exact field, so lets separate the dots
+	// and see if anything along the way is a computed set. i.e. if
+	// we have "foo.0.bar" as the field, check to see if "foo" is
+	// a computed list. If so, then the whole thing is computed.
+	if parts := strings.Split(v.Field, "."); len(parts) > 1 {
+		for i := 1; i < len(parts); i++ {
+			// Lists and sets make this
+			key := fmt.Sprintf("%s.#", strings.Join(parts[:i], "."))
+			if attr, ok := r.Primary.Attributes[key]; ok {
+				return attr, nil
+			}
+
+			// Maps make this
+			key = fmt.Sprintf("%s", strings.Join(parts[:i], "."))
+			if attr, ok := r.Primary.Attributes[key]; ok {
+				return attr, nil
+			}
+		}
+	}
+
+MISSING:
+	return "", fmt.Errorf(
+		"Resource '%s' does not have attribute '%s' "+
+			"for variable '%s'",
+		id,
+		v.Field,
+		v.FullKey())
+}
+
+func (i *Interpolater) computeResourceMultiVariable(
+	scope *InterpolationScope,
+	v *config.ResourceVariable) (string, error) {
+	i.StateLock.RLock()
+	defer i.StateLock.RUnlock()
+
+	// Get the information about this resource variable, and verify
+	// that it exists and such.
+	module, cr, err := i.resourceVariableInfo(scope, v)
+	if err != nil {
+		return "", err
+	}
+
+	// Get the count so we know how many to iterate over
+	count, err := cr.Count()
+	if err != nil {
+		return "", fmt.Errorf(
+			"Error reading %s count: %s",
+			v.ResourceId(),
+			err)
+	}
+
+	// If we have no module in the state yet or count, return empty
+	if module == nil || len(module.Resources) == 0 || count == 0 {
+		return "", nil
+	}
+
+	var values []string
+	for i := 0; i < count; i++ {
+		id := fmt.Sprintf("%s.%d", v.ResourceId(), i)
+
+		// If we're dealing with only a single resource, then the
+		// ID doesn't have a trailing index.
+		if count == 1 {
+			id = v.ResourceId()
+		}
+
+		r, ok := module.Resources[id]
+		if !ok {
+			continue
+		}
+
+		if r.Primary == nil {
+			continue
+		}
+
+		attr, ok := r.Primary.Attributes[v.Field]
+		if !ok {
+			continue
+		}
+
+		values = append(values, attr)
+	}
+
+	if len(values) == 0 {
+		return "", fmt.Errorf(
+			"Resource '%s' does not have attribute '%s' "+
+				"for variable '%s'",
+			v.ResourceId(),
+			v.Field,
+			v.FullKey())
+	}
+
+	return strings.Join(values, config.InterpSplitDelim), nil
+}
+
+func (i *Interpolater) resourceVariableInfo(
+	scope *InterpolationScope,
+	v *config.ResourceVariable) (*ModuleState, *config.Resource, error) {
+	// Get the module tree that contains our current path. This is
+	// either the current module (path is empty) or a child.
+	modTree := i.Module
+	if len(scope.Path) > 1 {
+		modTree = i.Module.Child(scope.Path[1:])
+	}
+
+	// Get the resource from the configuration so we can verify
+	// that the resource is in the configuration and so we can access
+	// the configuration if we need to.
+	var cr *config.Resource
+	for _, r := range modTree.Config().Resources {
+		if r.Id() == v.ResourceId() {
+			cr = r
+			break
+		}
+	}
+	if cr == nil {
+		return nil, nil, fmt.Errorf(
+			"Resource '%s' not found for variable '%s'",
+			v.ResourceId(),
+			v.FullKey())
+	}
+
+	// Get the relevant module
+	module := i.State.ModuleByPath(scope.Path)
+	return module, cr, nil
+}

terraform/interpolate_test.go

@@ -0,0 +1,135 @@
+package terraform
+
+import (
+	"os"
+	"reflect"
+	"sync"
+	"testing"
+
+	"github.com/hashicorp/terraform/config"
+	"github.com/hashicorp/terraform/config/lang/ast"
+)
+
+func TestInterpolater_countIndex(t *testing.T) {
+	i := &Interpolater{}
+
+	scope := &InterpolationScope{
+		Path:     rootModulePath,
+		Resource: &Resource{CountIndex: 42},
+	}
+
+	testInterpolate(t, i, scope, "count.index", ast.Variable{
+		Value: 42,
+		Type:  ast.TypeInt,
+	})
+}
+
+func TestInterpolater_moduleVariable(t *testing.T) {
+	lock := new(sync.RWMutex)
+	state := &State{
+		Modules: []*ModuleState{
+			&ModuleState{
+				Path: rootModulePath,
+				Resources: map[string]*ResourceState{
+					"aws_instance.web": &ResourceState{
+						Type: "aws_instance",
+						Primary: &InstanceState{
+							ID: "bar",
+						},
+					},
+				},
+			},
+			&ModuleState{
+				Path: []string{RootModuleName, "child"},
+				Outputs: map[string]string{
+					"foo": "bar",
+				},
+			},
+		},
+	}
+
+	i := &Interpolater{
+		State:     state,
+		StateLock: lock,
+	}
+
+	scope := &InterpolationScope{
+		Path: rootModulePath,
+	}
+
+	testInterpolate(t, i, scope, "module.child.foo", ast.Variable{
+		Value: "bar",
+		Type:  ast.TypeString,
+	})
+}
+
+func TestInterpolater_pathCwd(t *testing.T) {
+	i := &Interpolater{}
+	scope := &InterpolationScope{}
+
+	expected, err := os.Getwd()
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	testInterpolate(t, i, scope, "path.cwd", ast.Variable{
+		Value: expected,
+		Type:  ast.TypeString,
+	})
+}
+
+func TestInterpolater_pathModule(t *testing.T) {
+	mod := testModule(t, "interpolate-path-module")
+	i := &Interpolater{
+		Module: mod,
+	}
+	scope := &InterpolationScope{
+		Path: []string{RootModuleName, "child"},
+	}
+
+	path := mod.Child([]string{"child"}).Config().Dir
+	testInterpolate(t, i, scope, "path.module", ast.Variable{
+		Value: path,
+		Type:  ast.TypeString,
+	})
+}
+
+func TestInterpolater_pathRoot(t *testing.T) {
+	mod := testModule(t, "interpolate-path-module")
+	i := &Interpolater{
+		Module: mod,
+	}
+	scope := &InterpolationScope{
+		Path: []string{RootModuleName, "child"},
+	}
+
+	path := mod.Config().Dir
+	testInterpolate(t, i, scope, "path.root", ast.Variable{
+		Value: path,
+		Type:  ast.TypeString,
+	})
+}
+
+func testInterpolate(
+	t *testing.T, i *Interpolater,
+	scope *InterpolationScope,
+	n string, expectedVar ast.Variable) {
+	v, err := config.NewInterpolatedVariable(n)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual, err := i.Values(scope, map[string]config.InterpolatedVariable{
+		"foo": v,
+	})
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	expected := map[string]ast.Variable{
+		"foo": expectedVar,
+	}
+	if !reflect.DeepEqual(actual, expected) {
+		t.Fatalf("bad: %#v", actual)
+	}
+}

terraform/path.go

@@ -0,0 +1,24 @@
+package terraform
+
+import (
+	"crypto/md5"
+	"encoding/hex"
+)
+
+// PathCacheKey returns a cache key for a module path.
+//
+// TODO: test
+func PathCacheKey(path []string) string {
+	// There is probably a better way to do this, but this is working for now.
+	// We just create an MD5 hash of all the MD5 hashes of all the path
+	// elements. This gets us the property that it is unique per ordering.
+	hash := md5.New()
+	for _, p := range path {
+		single := md5.Sum([]byte(p))
+		if _, err := hash.Write(single[:]); err != nil {
+			panic(err)
+		}
+	}
+
+	return hex.EncodeToString(hash.Sum(nil))
+}

terraform/resource.go

@@ -93,7 +93,7 @@ type ResourceConfig struct {
 // NewResourceConfig creates a new ResourceConfig from a config.RawConfig.
 func NewResourceConfig(c *config.RawConfig) *ResourceConfig {
 	result := &ResourceConfig{raw: c}
-	result.interpolate(nil, nil)
+	result.interpolateForce()
 	return result
 }
 
@@ -201,28 +201,19 @@ func (c *ResourceConfig) get(
 	return current, true
 }
 
-func (c *ResourceConfig) interpolate(
-	ctx *walkContext, r *Resource) error {
-	if c == nil {
-		return nil
-	}
-
-	if ctx != nil {
-		if err := ctx.computeVars(c.raw, r); err != nil {
-			return err
-		}
-	}
-
+// interpolateForce is a temporary thing. We want to get rid of interpolate
+// above and likewise this, but it can only be done after the f-ast-graph
+// refactor is complete.
+func (c *ResourceConfig) interpolateForce() {
 	if c.raw == nil {
 		var err error
 		c.raw, err = config.NewRawConfig(make(map[string]interface{}))
 		if err != nil {
-			return err
+			panic(err)
 		}
 	}
 
 	c.ComputedKeys = c.raw.UnknownKeys()
 	c.Raw = c.raw.Raw
 	c.Config = c.raw.Config()
-	return nil
 }

terraform/resource_test.go

@@ -5,6 +5,7 @@ import (
 	"testing"
 
 	"github.com/hashicorp/terraform/config"
+	"github.com/hashicorp/terraform/config/lang/ast"
 )
 
 func TestInstanceInfo(t *testing.T) {
@@ -99,20 +100,33 @@ func TestResourceConfigGet(t *testing.T) {
 			}
 		}
 
-		rc := NewResourceConfig(rawC)
 		if tc.Vars != nil {
-			ctx := NewContext(&ContextOpts{Variables: tc.Vars})
-			err := rc.interpolate(
-				ctx.walkContext(walkInvalid, rootModulePath),
-				nil)
-			if err != nil {
+			vs := make(map[string]ast.Variable)
+			for k, v := range tc.Vars {
+				vs["var."+k] = ast.Variable{Value: v, Type: ast.TypeString}
+			}
+
+			if err := rawC.Interpolate(vs); err != nil {
 				t.Fatalf("err: %s", err)
 			}
 		}
 
+		rc := NewResourceConfig(rawC)
+		rc.interpolateForce()
+
 		v, _ := rc.Get(tc.Key)
 		if !reflect.DeepEqual(v, tc.Value) {
 			t.Fatalf("%d bad: %#v", i, v)
 		}
 	}
 }
+
+func testResourceConfig(
+	t *testing.T, c map[string]interface{}) *ResourceConfig {
+	raw, err := config.NewRawConfig(c)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	return NewResourceConfig(raw)
+}

terraform/semantics.go

@@ -3,9 +3,70 @@ package terraform
 import (
 	"fmt"
 
+	"github.com/hashicorp/go-multierror"
 	"github.com/hashicorp/terraform/config"
+	"github.com/hashicorp/terraform/dag"
 )
 
+// GraphSemanticChecker is the interface that semantic checks across
+// the entire Terraform graph implement.
+//
+// The graph should NOT be modified by the semantic checker.
+type GraphSemanticChecker interface {
+	Check(*dag.Graph) error
+}
+
+// UnorderedSemanticCheckRunner is an implementation of GraphSemanticChecker
+// that runs a list of SemanticCheckers against the vertices of the graph
+// in no specified order.
+type UnorderedSemanticCheckRunner struct {
+	Checks []SemanticChecker
+}
+
+func (sc *UnorderedSemanticCheckRunner) Check(g *dag.Graph) error {
+	var err error
+	for _, v := range g.Vertices() {
+		for _, check := range sc.Checks {
+			if e := check.Check(g, v); e != nil {
+				err = multierror.Append(err, e)
+			}
+		}
+	}
+
+	return err
+}
+
+// SemanticChecker is the interface that semantic checks across the
+// Terraform graph implement. Errors are accumulated. Even after an error
+// is returned, child vertices in the graph will still be visited.
+//
+// The graph should NOT be modified by the semantic checker.
+//
+// The order in which vertices are visited is left unspecified, so the
+// semantic checks should not rely on that.
+type SemanticChecker interface {
+	Check(*dag.Graph, dag.Vertex) error
+}
+
+// SemanticCheckModulesExist is an implementation of SemanticChecker that
+// verifies that all the modules that are referenced in the graph exist.
+type SemanticCheckModulesExist struct{}
+
+// TODO: test
+func (*SemanticCheckModulesExist) Check(g *dag.Graph, v dag.Vertex) error {
+	mn, ok := v.(*GraphNodeConfigModule)
+	if !ok {
+		return nil
+	}
+
+	if mn.Tree == nil {
+		return fmt.Errorf(
+			"module '%s' not found", mn.Module.Name)
+	}
+
+	return nil
+}
+
 // smcUserVariables does all the semantic checks to verify that the
 // variables given satisfy the configuration itself.
 func smcUserVariables(c *config.Config, vs map[string]string) []error {

terraform/state.go

@@ -100,6 +100,31 @@ func (s *State) ModuleByPath(path []string) *ModuleState {
 	return nil
 }
 
+// ModuleOrphans returns all the module orphans in this state by
+// returning their full paths. These paths can be used with ModuleByPath
+// to return the actual state.
+func (s *State) ModuleOrphans(path []string, c *config.Config) [][]string {
+	childrenKeys := make(map[string]struct{})
+	if c != nil {
+		for _, m := range c.Modules {
+			childrenKeys[m.Name] = struct{}{}
+		}
+	}
+
+	// Go over the direct children and find any that aren't in our
+	// keys.
+	var orphans [][]string
+	for _, m := range s.Children(path) {
+		if _, ok := childrenKeys[m.Path[len(m.Path)-1]]; ok {
+			continue
+		}
+
+		orphans = append(orphans, m.Path)
+	}
+
+	return orphans
+}
+
 // RootModule returns the ModuleState for the root module
 func (s *State) RootModule() *ModuleState {
 	root := s.ModuleByPath(rootModulePath)
@@ -278,12 +303,14 @@ func (m *ModuleState) Orphans(c *config.Config) []string {
 		keys[k] = struct{}{}
 	}
 
-	for _, r := range c.Resources {
-		delete(keys, r.Id())
+	if c != nil {
+		for _, r := range c.Resources {
+			delete(keys, r.Id())
 
-		for k, _ := range keys {
-			if strings.HasPrefix(k, r.Id()+".") {
-				delete(keys, k)
+			for k, _ := range keys {
+				if strings.HasPrefix(k, r.Id()+".") {
+					delete(keys, k)
+				}
 			}
 		}
 	}
@@ -346,10 +373,17 @@ func (m *ModuleState) deepcopy() *ModuleState {
 func (m *ModuleState) prune() {
 	for k, v := range m.Resources {
 		v.prune()
+
 		if (v.Primary == nil || v.Primary.ID == "") && len(v.Tainted) == 0 {
 			delete(m.Resources, k)
 		}
 	}
+
+	for k, v := range m.Outputs {
+		if v == config.UnknownVariableValue {
+			delete(m.Outputs, k)
+		}
+	}
 }
 
 func (m *ModuleState) sort() {
@@ -517,12 +551,14 @@ func (r *ResourceState) prune() {
 	n := len(r.Tainted)
 	for i := 0; i < n; i++ {
 		inst := r.Tainted[i]
-		if inst.ID == "" {
+		if inst == nil || inst.ID == "" {
 			copy(r.Tainted[i:], r.Tainted[i+1:])
 			r.Tainted[n-1] = nil
 			n--
+			i--
 		}
 	}
+
 	r.Tainted = r.Tainted[:n]
 }
 

terraform/state_test.go

@@ -59,6 +59,58 @@ func TestStateAddModule(t *testing.T) {
 	}
 }
 
+func TestStateModuleOrphans(t *testing.T) {
+	state := &State{
+		Modules: []*ModuleState{
+			&ModuleState{
+				Path: RootModulePath,
+			},
+			&ModuleState{
+				Path: []string{RootModuleName, "foo"},
+			},
+			&ModuleState{
+				Path: []string{RootModuleName, "bar"},
+			},
+		},
+	}
+
+	config := testModule(t, "state-module-orphans").Config()
+	actual := state.ModuleOrphans(RootModulePath, config)
+	expected := [][]string{
+		[]string{RootModuleName, "foo"},
+	}
+
+	if !reflect.DeepEqual(actual, expected) {
+		t.Fatalf("bad: %#v", actual)
+	}
+}
+
+func TestStateModuleOrphans_nilConfig(t *testing.T) {
+	state := &State{
+		Modules: []*ModuleState{
+			&ModuleState{
+				Path: RootModulePath,
+			},
+			&ModuleState{
+				Path: []string{RootModuleName, "foo"},
+			},
+			&ModuleState{
+				Path: []string{RootModuleName, "bar"},
+			},
+		},
+	}
+
+	actual := state.ModuleOrphans(RootModulePath, nil)
+	expected := [][]string{
+		[]string{RootModuleName, "foo"},
+		[]string{RootModuleName, "bar"},
+	}
+
+	if !reflect.DeepEqual(actual, expected) {
+		t.Fatalf("bad: %#v", actual)
+	}
+}
+
 func TestInstanceState_MergeDiff(t *testing.T) {
 	is := InstanceState{
 		ID: "foo",

terraform/terraform_test.go

@@ -200,6 +200,13 @@ aws_instance.bar:
   type = aws_instance
 `
 
+const testTerraformApplyCreateBeforeUpdateStr = `
+aws_instance.bar:
+  ID = foo
+  foo = baz
+  type = aws_instance
+`
+
 const testTerraformApplyCancelStr = `
 aws_instance.foo:
   ID = foo
@@ -233,7 +240,7 @@ aws_instance.foo.1:
 `
 
 const testTerraformApplyCountDecToOneStr = `
-aws_instance.foo.0:
+aws_instance.foo:
   ID = bar
   foo = foo
   type = aws_instance
@@ -278,6 +285,17 @@ module.child:
     type = aws_instance
 `
 
+const testTerraformApplyMultiProviderStr = `
+aws_instance.bar:
+  ID = foo
+  foo = bar
+  type = aws_instance
+do_instance.foo:
+  ID = foo
+  num = 2
+  type = do_instance
+`
+
 const testTerraformApplyProvisionerStr = `
 aws_instance.bar:
   ID = foo

terraform/test-fixtures/apply-good-create-before-update/main.tf

@@ -0,0 +1,7 @@
+resource "aws_instance" "bar" {
+    foo = "baz"
+
+    lifecycle {
+        create_before_destroy = true
+    }
+}

terraform/test-fixtures/apply-multi-provider/main.tf

@@ -0,0 +1,7 @@
+resource "do_instance" "foo" {
+    num = "2"
+}
+
+resource "aws_instance" "bar" {
+    foo = "bar"
+}

terraform/test-fixtures/graph-builder-basic/main.tf

@@ -0,0 +1,5 @@
+provider "aws" {}
+resource "aws_instance" "db" {}
+resource "aws_instance" "web" {
+    foo = "${aws_instance.db.id}"
+}

terraform/test-fixtures/graph-builder-cbd-non-cbd/main.tf

@@ -0,0 +1,9 @@
+provider "aws" {}
+
+resource "aws_lc" "foo" {}
+
+resource "aws_asg" "foo" {
+    lc = "${aws_lc.foo.id}"
+
+    lifecycle { create_before_destroy = true }
+}

terraform/test-fixtures/graph-builder-modules/consul/main.tf

@@ -0,0 +1,2 @@
+provider "aws" {}
+resource "aws_instance" "server" {}

terraform/test-fixtures/graph-builder-modules/main.tf

@@ -0,0 +1,16 @@
+module "consul" {
+    foo = "${aws_security_group.firewall.foo}"
+    source = "./consul"
+}
+
+provider "aws" {}
+
+resource "aws_security_group" "firewall" {}
+
+resource "aws_instance" "web" {
+    security_groups = [
+        "foo",
+        "${aws_security_group.firewall.foo}",
+        "${module.consul.security_group}"
+    ]
+}

terraform/test-fixtures/graph-missing-deps/main.tf

@@ -0,0 +1,5 @@
+resource "aws_instance" "db" {}
+
+resource "aws_instance" "web" {
+    foo = "${aws_instance.lb.id}"
+}

terraform/test-fixtures/graph-node-module-expand/child/main.tf

@@ -0,0 +1,4 @@
+resource "aws_instance" "foo" {}
+resource "aws_instance" "bar" {
+    var = "${aws_instance.foo.whatever}"
+}

terraform/test-fixtures/graph-node-module-expand/main.tf

@@ -0,0 +1,3 @@
+module "child" {
+    source = "./child"
+}

terraform/test-fixtures/graph-outputs/main.tf

@@ -0,0 +1,5 @@
+resource "aws_instance" "foo" {}
+
+output "foo" {
+    value = "${aws_instance.foo.value}"
+}

terraform/test-fixtures/interpolate-path-module/main.tf

@@ -0,0 +1,3 @@
+module "child" {
+    source = "./child"
+}

terraform/test-fixtures/state-module-orphans/bar/main.tf

@@ -0,0 +1 @@
+# Nothing

terraform/test-fixtures/state-module-orphans/main.tf

@@ -0,0 +1,3 @@
+module "bar" {
+    source = "./bar"
+}

terraform/test-fixtures/transform-create-before-destroy-basic/main.tf

@@ -0,0 +1,9 @@
+resource "aws_instance" "web" {
+    lifecycle {
+        create_before_destroy = true
+    }
+}
+
+resource "aws_load_balancer" "lb" {
+    member = "${aws_instance.web.id}"
+}

terraform/test-fixtures/transform-create-before-destroy-twice/main.tf

@@ -0,0 +1,9 @@
+resource "aws_lc" "foo" {
+    lifecycle { create_before_destroy = true }
+}
+
+resource "aws_autoscale" "bar" {
+    lc = "${aws_lc.foo.id}"
+
+    lifecycle { create_before_destroy = true }
+}

terraform/test-fixtures/transform-destroy-basic/main.tf

@@ -0,0 +1,5 @@
+resource "aws_instance" "foo" {}
+
+resource "aws_instance" "bar" {
+    value = "${aws_instance.foo.value}"
+}

terraform/test-fixtures/transform-destroy-deps/main.tf

@@ -0,0 +1,5 @@
+resource "aws_lc" "foo" {}
+
+resource "aws_asg" "bar" {
+  lc = "${aws_lc.foo.id}"
+}

terraform/test-fixtures/transform-destroy-prune-count/main.tf

@@ -0,0 +1,6 @@
+resource "aws_instance" "foo" {}
+
+resource "aws_instance" "bar" {
+    value = "${aws_instance.foo.value}"
+    count = "5"
+}

terraform/test-fixtures/transform-orphan-basic/main.tf

@@ -0,0 +1 @@
+resource "aws_instance" "web" {}

terraform/test-fixtures/transform-orphan-modules/main.tf

@@ -0,0 +1 @@
+resource "aws_instance" "foo" {}

terraform/test-fixtures/transform-provider-basic/main.tf

@@ -0,0 +1,2 @@
+provider "aws" {}
+resource "aws_instance" "web" {}

terraform/test-fixtures/transform-provider-prune/main.tf

@@ -0,0 +1,2 @@
+provider "aws" {}
+resource "foo_instance" "web" {}

terraform/test-fixtures/transform-provisioner-basic/main.tf

@@ -0,0 +1,3 @@
+resource "aws_instance" "web" {
+    provisioner "shell" {}
+}

terraform/test-fixtures/transform-provisioner-prune/main.tf

@@ -0,0 +1,3 @@
+resource "aws_instance" "web" {
+    provisioner "foo" {}
+}

terraform/test-fixtures/transform-resource-count-basic/main.tf

@@ -0,0 +1,4 @@
+resource "aws_instance" "foo" {
+    count = 3
+    value = "${aws_instance.foo.0.value}"
+}

terraform/test-fixtures/transform-resource-count-negative/main.tf

@@ -0,0 +1,4 @@
+resource "aws_instance" "foo" {
+    count = -5
+    value = "${aws_instance.foo.0.value}"
+}

terraform/test-fixtures/transform-root-basic/main.tf

@@ -0,0 +1,5 @@
+provider "aws" {}
+resource "aws_instance" "foo" {}
+
+provider "do" {}
+resource "do_droplet" "bar" {}

terraform/test-fixtures/transform-tainted-basic/main.tf

@@ -0,0 +1 @@
+resource "aws_instance" "web" {}

terraform/test-fixtures/validate-bad-prov-conf/main.tf

@@ -3,5 +3,7 @@ provider "aws" {
 }
 
 resource "aws_instance" "test" {
-    provisioner "shell" {}
+    provisioner "shell" {
+        command = "foo"
+    }
 }

terraform/transform.go

@@ -0,0 +1,21 @@
+package terraform
+
+import (
+	"github.com/hashicorp/terraform/dag"
+)
+
+// GraphTransformer is the interface that transformers implement. This
+// interface is only for transforms that need entire graph visibility.
+type GraphTransformer interface {
+	Transform(*Graph) error
+}
+
+// GraphVertexTransformer is an interface that transforms a single
+// Vertex within with graph. This is a specialization of GraphTransformer
+// that makes it easy to do vertex replacement.
+//
+// The GraphTransformer that runs through the GraphVertexTransformers is
+// VertexTransformer.
+type GraphVertexTransformer interface {
+	Transform(dag.Vertex) (dag.Vertex, error)
+}

terraform/transform_config.go

@@ -0,0 +1,105 @@
+package terraform
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/hashicorp/go-multierror"
+	"github.com/hashicorp/terraform/config"
+	"github.com/hashicorp/terraform/config/module"
+)
+
+// ConfigTransformer is a GraphTransformer that adds the configuration
+// to the graph. The module used to configure this transformer must be
+// the root module. We'll look up the child module by the Path in the
+// Graph.
+type ConfigTransformer struct {
+	Module *module.Tree
+}
+
+func (t *ConfigTransformer) Transform(g *Graph) error {
+	// A module is required and also must be completely loaded.
+	if t.Module == nil {
+		return errors.New("module must not be nil")
+	}
+	if !t.Module.Loaded() {
+		return errors.New("module must be loaded")
+	}
+
+	// Get the module we care about
+	module := t.Module.Child(g.Path[1:])
+	if module == nil {
+		return nil
+	}
+
+	// Get the configuration for this module
+	config := module.Config()
+
+	// Create the node list we'll use for the graph
+	nodes := make([]graphNodeConfig, 0,
+		(len(config.ProviderConfigs)+len(config.Modules)+len(config.Resources))*2)
+
+	// Write all the provider configs out
+	for _, pc := range config.ProviderConfigs {
+		nodes = append(nodes, &GraphNodeConfigProvider{Provider: pc})
+	}
+
+	// Write all the resources out
+	for _, r := range config.Resources {
+		nodes = append(nodes, &GraphNodeConfigResource{Resource: r})
+	}
+
+	// Write all the modules out
+	children := module.Children()
+	for _, m := range config.Modules {
+		path := make([]string, len(g.Path), len(g.Path)+1)
+		copy(path, g.Path)
+		path = append(path, m.Name)
+
+		nodes = append(nodes, &GraphNodeConfigModule{
+			Path:   path,
+			Module: m,
+			Tree:   children[m.Name],
+		})
+	}
+
+	// Write all the outputs out
+	for _, o := range config.Outputs {
+		nodes = append(nodes, &GraphNodeConfigOutput{Output: o})
+	}
+
+	// Err is where the final error value will go if there is one
+	var err error
+
+	// Build the graph vertices
+	for _, n := range nodes {
+		g.Add(n)
+	}
+
+	// Build up the dependencies. We have to do this outside of the above
+	// loop since the nodes need to be in place for us to build the deps.
+	for _, n := range nodes {
+		if missing := g.ConnectDependent(n); len(missing) > 0 {
+			for _, m := range missing {
+				err = multierror.Append(err, fmt.Errorf(
+					"%s: missing dependency: %s", n.Name(), m))
+			}
+		}
+	}
+
+	return err
+}
+
+// varNameForVar returns the VarName value for an interpolated variable.
+// This value is compared to the VarName() value for the nodes within the
+// graph to build the graph edges.
+func varNameForVar(raw config.InterpolatedVariable) string {
+	switch v := raw.(type) {
+	case *config.ModuleVariable:
+		return fmt.Sprintf("module.%s", v.Name)
+	case *config.ResourceVariable:
+		return v.ResourceId()
+	default:
+		return ""
+	}
+}

terraform/transform_config_test.go

@@ -0,0 +1,128 @@
+package terraform
+
+import (
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/hashicorp/terraform/config/module"
+)
+
+func TestConfigTransformer_nilModule(t *testing.T) {
+	g := Graph{Path: RootModulePath}
+	tf := &ConfigTransformer{}
+	if err := tf.Transform(&g); err == nil {
+		t.Fatal("should error")
+	}
+}
+
+func TestConfigTransformer_unloadedModule(t *testing.T) {
+	mod, err := module.NewTreeModule(
+		"", filepath.Join(fixtureDir, "graph-basic"))
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	g := Graph{Path: RootModulePath}
+	tf := &ConfigTransformer{Module: mod}
+	if err := tf.Transform(&g); err == nil {
+		t.Fatal("should error")
+	}
+}
+
+func TestConfigTransformer(t *testing.T) {
+	g := Graph{Path: RootModulePath}
+	tf := &ConfigTransformer{Module: testModule(t, "graph-basic")}
+	if err := tf.Transform(&g); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphBasicStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestConfigTransformer_dependsOn(t *testing.T) {
+	g := Graph{Path: RootModulePath}
+	tf := &ConfigTransformer{Module: testModule(t, "graph-depends-on")}
+	if err := tf.Transform(&g); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphDependsOnStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestConfigTransformer_modules(t *testing.T) {
+	g := Graph{Path: RootModulePath}
+	tf := &ConfigTransformer{Module: testModule(t, "graph-modules")}
+	if err := tf.Transform(&g); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphModulesStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestConfigTransformer_outputs(t *testing.T) {
+	g := Graph{Path: RootModulePath}
+	tf := &ConfigTransformer{Module: testModule(t, "graph-outputs")}
+	if err := tf.Transform(&g); err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testGraphOutputsStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestConfigTransformer_errMissingDeps(t *testing.T) {
+	g := Graph{Path: RootModulePath}
+	tf := &ConfigTransformer{Module: testModule(t, "graph-missing-deps")}
+	if err := tf.Transform(&g); err == nil {
+		t.Fatalf("err: %s", err)
+	}
+}
+
+const testGraphBasicStr = `
+aws_instance.web
+  aws_security_group.firewall
+aws_load_balancer.weblb
+  aws_instance.web
+aws_security_group.firewall
+openstack_floating_ip.random
+provider.aws
+  openstack_floating_ip.random
+`
+
+const testGraphDependsOnStr = `
+aws_instance.db
+  aws_instance.web
+aws_instance.web
+`
+
+const testGraphModulesStr = `
+aws_instance.web
+  aws_security_group.firewall
+  module.consul
+aws_security_group.firewall
+module.consul
+  aws_security_group.firewall
+provider.aws
+`
+
+const testGraphOutputsStr = `
+aws_instance.foo
+output.foo
+  aws_instance.foo
+`

terraform/transform_destroy.go

@@ -0,0 +1,230 @@
+package terraform
+
+import (
+	"github.com/hashicorp/terraform/dag"
+)
+
+type GraphNodeDestroyMode byte
+
+const (
+	DestroyNone    GraphNodeDestroyMode = 0
+	DestroyPrimary GraphNodeDestroyMode = 1 << iota
+	DestroyTainted
+)
+
+// GraphNodeDestroyable is the interface that nodes that can be destroyed
+// must implement. This is used to automatically handle the creation of
+// destroy nodes in the graph and the dependency ordering of those destroys.
+type GraphNodeDestroyable interface {
+	// DestroyNode returns the node used for the destroy with the given
+	// mode. If this returns nil, then a destroy node for that mode
+	// will not be added.
+	DestroyNode(GraphNodeDestroyMode) GraphNodeDestroy
+}
+
+// GraphNodeDestroy is the interface that must implemented by
+// nodes that destroy.
+type GraphNodeDestroy interface {
+	dag.Vertex
+
+	// CreateBeforeDestroy is called to check whether this node
+	// should be created before it is destroyed. The CreateBeforeDestroy
+	// transformer uses this information to setup the graph.
+	CreateBeforeDestroy() bool
+
+	// CreateNode returns the node used for the create side of this
+	// destroy. This must already exist within the graph.
+	CreateNode() dag.Vertex
+}
+
+// GraphNodeDestroyPrunable is the interface that can be implemented to
+// signal that this node can be pruned depending on state.
+type GraphNodeDestroyPrunable interface {
+	// DestroyInclude is called to check if this node should be included
+	// with the given state. The state and diff must NOT be modified.
+	DestroyInclude(*ModuleDiff, *ModuleState) bool
+}
+
+// DestroyTransformer is a GraphTransformer that creates the destruction
+// nodes for things that _might_ be destroyed.
+type DestroyTransformer struct{}
+
+func (t *DestroyTransformer) Transform(g *Graph) error {
+	var connect, remove []dag.Edge
+
+	modes := []GraphNodeDestroyMode{DestroyPrimary, DestroyTainted}
+	for _, m := range modes {
+		connectMode, removeMode, err := t.transform(g, m)
+		if err != nil {
+			return err
+		}
+
+		connect = append(connect, connectMode...)
+		remove = append(remove, removeMode...)
+	}
+
+	// Atomatically add/remove the edges
+	for _, e := range connect {
+		g.Connect(e)
+	}
+	for _, e := range remove {
+		g.RemoveEdge(e)
+	}
+
+	return nil
+}
+
+func (t *DestroyTransformer) transform(
+	g *Graph, mode GraphNodeDestroyMode) ([]dag.Edge, []dag.Edge, error) {
+	var connect, remove []dag.Edge
+	nodeToCn := make(map[dag.Vertex]dag.Vertex, len(g.Vertices()))
+	nodeToDn := make(map[dag.Vertex]dag.Vertex, len(g.Vertices()))
+	for _, v := range g.Vertices() {
+		// If it is not a destroyable, we don't care
+		cn, ok := v.(GraphNodeDestroyable)
+		if !ok {
+			continue
+		}
+
+		// Grab the destroy side of the node and connect it through
+		n := cn.DestroyNode(mode)
+		if n == nil {
+			continue
+		}
+
+		// Store it
+		nodeToCn[n] = cn
+		nodeToDn[cn] = n
+
+		// Add it to the graph
+		g.Add(n)
+
+		// Inherit all the edges from the old node
+		downEdges := g.DownEdges(v).List()
+		for _, edgeRaw := range downEdges {
+			g.Connect(dag.BasicEdge(n, edgeRaw.(dag.Vertex)))
+		}
+
+		// Add a new edge to connect the node to be created to
+		// the destroy node.
+		connect = append(connect, dag.BasicEdge(v, n))
+	}
+
+	// Go through the nodes we added and determine if they depend
+	// on any nodes with a destroy node. If so, depend on that instead.
+	for n, _ := range nodeToCn {
+		for _, downRaw := range g.DownEdges(n).List() {
+			target := downRaw.(dag.Vertex)
+			cn2, ok := target.(GraphNodeDestroyable)
+			if !ok {
+				continue
+			}
+
+			newTarget := nodeToDn[cn2]
+			if newTarget == nil {
+				continue
+			}
+
+			// Make the new edge and transpose
+			connect = append(connect, dag.BasicEdge(newTarget, n))
+
+			// Remove the old edge
+			remove = append(remove, dag.BasicEdge(n, target))
+		}
+	}
+
+	return connect, remove, nil
+}
+
+// CreateBeforeDestroyTransformer is a GraphTransformer that modifies
+// the destroys of some nodes so that the creation happens before the
+// destroy.
+type CreateBeforeDestroyTransformer struct{}
+
+func (t *CreateBeforeDestroyTransformer) Transform(g *Graph) error {
+	// We "stage" the edge connections/destroys in these slices so that
+	// while we're doing the edge transformations (transpositions) in
+	// the graph, we're not affecting future edge transpositions. These
+	// slices let us stage ALL the changes that WILL happen so that all
+	// of the transformations happen atomically.
+	var connect, destroy []dag.Edge
+
+	for _, v := range g.Vertices() {
+		// We only care to use the destroy nodes
+		dn, ok := v.(GraphNodeDestroy)
+		if !ok {
+			continue
+		}
+
+		// If the node doesn't need to create before destroy, then continue
+		if !dn.CreateBeforeDestroy() {
+			continue
+		}
+
+		// Get the creation side of this node
+		cn := dn.CreateNode()
+
+		// Take all the things which depend on the creation node and
+		// make them dependencies on the destruction. Clarifying this
+		// with an example: if you have a web server and a load balancer
+		// and the load balancer depends on the web server, then when we
+		// do a create before destroy, we want to make sure the steps are:
+		//
+		// 1.) Create new web server
+		// 2.) Update load balancer
+		// 3.) Delete old web server
+		//
+		// This ensures that.
+		for _, sourceRaw := range g.UpEdges(cn).List() {
+			source := sourceRaw.(dag.Vertex)
+			connect = append(connect, dag.BasicEdge(dn, source))
+		}
+
+		// Swap the edge so that the destroy depends on the creation
+		// happening...
+		connect = append(connect, dag.BasicEdge(dn, cn))
+		destroy = append(destroy, dag.BasicEdge(cn, dn))
+	}
+
+	for _, edge := range connect {
+		g.Connect(edge)
+	}
+	for _, edge := range destroy {
+		g.RemoveEdge(edge)
+	}
+
+	return nil
+}
+
+// PruneDestroyTransformer is a GraphTransformer that removes the destroy
+// nodes that aren't in the diff.
+type PruneDestroyTransformer struct {
+	Diff  *Diff
+	State *State
+}
+
+func (t *PruneDestroyTransformer) Transform(g *Graph) error {
+	var modDiff *ModuleDiff
+	var modState *ModuleState
+	if t.Diff != nil {
+		modDiff = t.Diff.ModuleByPath(g.Path)
+	}
+	if t.State != nil {
+		modState = t.State.ModuleByPath(g.Path)
+	}
+
+	for _, v := range g.Vertices() {
+		// If it is not a destroyer, we don't care
+		dn, ok := v.(GraphNodeDestroyPrunable)
+		if !ok {
+			continue
+		}
+
+		// Remove it if we should
+		if !dn.DestroyInclude(modDiff, modState) {
+			g.Remove(v)
+		}
+	}
+
+	return nil
+}

terraform/transform_destroy_test.go

@@ -0,0 +1,411 @@
+package terraform
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestDestroyTransformer(t *testing.T) {
+	mod := testModule(t, "transform-destroy-basic")
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformDestroyBasicStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestCreateBeforeDestroyTransformer(t *testing.T) {
+	mod := testModule(t, "transform-create-before-destroy-basic")
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &CreateBeforeDestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformCreateBeforeDestroyBasicStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestCreateBeforeDestroyTransformer_twice(t *testing.T) {
+	mod := testModule(t, "transform-create-before-destroy-twice")
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &CreateBeforeDestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformCreateBeforeDestroyTwiceStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestPruneDestroyTransformer(t *testing.T) {
+	var diff *Diff
+	mod := testModule(t, "transform-destroy-basic")
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &PruneDestroyTransformer{Diff: diff}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformPruneDestroyBasicStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestPruneDestroyTransformer_diff(t *testing.T) {
+	mod := testModule(t, "transform-destroy-basic")
+
+	diff := &Diff{
+		Modules: []*ModuleDiff{
+			&ModuleDiff{
+				Path: RootModulePath,
+				Resources: map[string]*InstanceDiff{
+					"aws_instance.bar": &InstanceDiff{},
+				},
+			},
+		},
+	}
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &PruneDestroyTransformer{Diff: diff}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformPruneDestroyBasicDiffStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestPruneDestroyTransformer_count(t *testing.T) {
+	mod := testModule(t, "transform-destroy-prune-count")
+
+	diff := &Diff{}
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &PruneDestroyTransformer{Diff: diff}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformPruneDestroyCountStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestPruneDestroyTransformer_countDec(t *testing.T) {
+	mod := testModule(t, "transform-destroy-basic")
+
+	diff := &Diff{}
+	state := &State{
+		Modules: []*ModuleState{
+			&ModuleState{
+				Path: RootModulePath,
+				Resources: map[string]*ResourceState{
+					"aws_instance.bar.1": &ResourceState{
+						Primary: &InstanceState{},
+					},
+					"aws_instance.bar.2": &ResourceState{
+						Primary: &InstanceState{},
+					},
+				},
+			},
+		},
+	}
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &PruneDestroyTransformer{Diff: diff, State: state}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformPruneDestroyCountDecStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+func TestPruneDestroyTransformer_countState(t *testing.T) {
+	mod := testModule(t, "transform-destroy-basic")
+
+	diff := &Diff{}
+	state := &State{
+		Modules: []*ModuleState{
+			&ModuleState{
+				Path: RootModulePath,
+				Resources: map[string]*ResourceState{
+					"aws_instance.bar": &ResourceState{
+						Primary: &InstanceState{},
+					},
+				},
+			},
+		},
+	}
+
+	g := Graph{Path: RootModulePath}
+	{
+		tf := &ConfigTransformer{Module: mod}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &DestroyTransformer{}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	{
+		tf := &PruneDestroyTransformer{Diff: diff, State: state}
+		if err := tf.Transform(&g); err != nil {
+			t.Fatalf("err: %s", err)
+		}
+	}
+
+	actual := strings.TrimSpace(g.String())
+	expected := strings.TrimSpace(testTransformPruneDestroyCountStateStr)
+	if actual != expected {
+		t.Fatalf("bad:\n\n%s", actual)
+	}
+}
+
+const testTransformDestroyBasicStr = `
+aws_instance.bar
+  aws_instance.bar (destroy tainted)
+  aws_instance.bar (destroy)
+  aws_instance.foo
+aws_instance.bar (destroy tainted)
+aws_instance.bar (destroy)
+aws_instance.foo
+  aws_instance.foo (destroy tainted)
+  aws_instance.foo (destroy)
+aws_instance.foo (destroy tainted)
+  aws_instance.bar (destroy tainted)
+aws_instance.foo (destroy)
+  aws_instance.bar (destroy)
+`
+
+const testTransformPruneDestroyBasicStr = `
+aws_instance.bar
+  aws_instance.bar (destroy tainted)
+  aws_instance.foo
+aws_instance.bar (destroy tainted)
+aws_instance.foo
+  aws_instance.foo (destroy tainted)
+aws_instance.foo (destroy tainted)
+  aws_instance.bar (destroy tainted)
+`
+
+const testTransformPruneDestroyBasicDiffStr = `
+aws_instance.bar
+  aws_instance.bar (destroy tainted)
+  aws_instance.bar (destroy)
+  aws_instance.foo
+aws_instance.bar (destroy tainted)
+aws_instance.bar (destroy)
+aws_instance.foo
+  aws_instance.foo (destroy tainted)
+aws_instance.foo (destroy tainted)
+  aws_instance.bar (destroy tainted)
+`
+
+const testTransformPruneDestroyCountStr = `
+aws_instance.bar
+  aws_instance.bar (destroy tainted)
+  aws_instance.bar (destroy)
+  aws_instance.foo
+aws_instance.bar (destroy tainted)
+aws_instance.bar (destroy)
+aws_instance.foo
+  aws_instance.foo (destroy tainted)
+aws_instance.foo (destroy tainted)
+  aws_instance.bar (destroy tainted)
+`
+
+const testTransformPruneDestroyCountDecStr = `
+aws_instance.bar
+  aws_instance.bar (destroy tainted)
+  aws_instance.bar (destroy)
+  aws_instance.foo
+aws_instance.bar (destroy tainted)
+aws_instance.bar (destroy)
+aws_instance.foo
+  aws_instance.foo (destroy tainted)
+aws_instance.foo (destroy tainted)
+  aws_instance.bar (destroy tainted)
+`
+
+const testTransformPruneDestroyCountStateStr = `
+aws_instance.bar
+  aws_instance.bar (destroy tainted)
+  aws_instance.foo
+aws_instance.bar (destroy tainted)
+aws_instance.foo
+  aws_instance.foo (destroy tainted)
+aws_instance.foo (destroy tainted)
+  aws_instance.bar (destroy tainted)
+`
+
+const testTransformCreateBeforeDestroyBasicStr = `
+aws_instance.web
+  aws_instance.web (destroy tainted)
+aws_instance.web (destroy tainted)
+  aws_load_balancer.lb (destroy tainted)
+aws_instance.web (destroy)
+  aws_instance.web
+  aws_load_balancer.lb
+  aws_load_balancer.lb (destroy)
+aws_load_balancer.lb
+  aws_instance.web
+  aws_load_balancer.lb (destroy tainted)
+  aws_load_balancer.lb (destroy)
+aws_load_balancer.lb (destroy tainted)
+aws_load_balancer.lb (destroy)
+`
+
+const testTransformCreateBeforeDestroyTwiceStr = `
+aws_autoscale.bar
+  aws_autoscale.bar (destroy tainted)
+  aws_lc.foo
+aws_autoscale.bar (destroy tainted)
+aws_autoscale.bar (destroy)
+  aws_autoscale.bar
+aws_lc.foo
+  aws_lc.foo (destroy tainted)
+aws_lc.foo (destroy tainted)
+  aws_autoscale.bar (destroy tainted)
+aws_lc.foo (destroy)
+  aws_autoscale.bar
+  aws_autoscale.bar (destroy)
+  aws_lc.foo
+`

terraform/transform_expand.go

@@ -0,0 +1,61 @@
+package terraform
+
+import (
+	"log"
+
+	"github.com/hashicorp/terraform/dag"
+)
+
+// GraphNodeExapndable is an interface that nodes can implement to
+// signal that they can be expanded. Expanded nodes turn into
+// GraphNodeSubgraph nodes within the graph.
+type GraphNodeExpandable interface {
+	Expand(GraphBuilder) (GraphNodeSubgraph, error)
+}
+
+// GraphNodeDynamicExpandable is an interface that nodes can implement
+// to signal that they can be expanded at eval-time (hence dynamic).
+// These nodes are given the eval context and are expected to return
+// a new subgraph.
+type GraphNodeDynamicExpandable interface {
+	DynamicExpand(EvalContext) (*Graph, error)
+}
+
+// GraphNodeSubgraph is an interface a node can implement if it has
+// a larger subgraph that should be walked.
+type GraphNodeSubgraph interface {
+	Subgraph() *Graph
+}
+
+// ExpandTransform is a transformer that does a subgraph expansion
+// at graph transform time (vs. at eval time). The benefit of earlier
+// subgraph expansion is that errors with the graph build can be detected
+// at an earlier stage.
+type ExpandTransform struct {
+	Builder GraphBuilder
+}
+
+func (t *ExpandTransform) Transform(v dag.Vertex) (dag.Vertex, error) {
+	ev, ok := v.(GraphNodeExpandable)
+	if !ok {
+		// This isn't an expandable vertex, so just ignore it.
+		return v, nil
+	}
+
+	// Expand the subgraph!
+	log.Printf("[DEBUG] vertex %s: static expanding", dag.VertexName(ev))
+	return ev.Expand(t.Builder)
+}
+
+type GraphNodeBasicSubgraph struct {
+	NameValue string
+	Graph     *Graph
+}
+
+func (n *GraphNodeBasicSubgraph) Name() string {
+	return n.NameValue
+}
+
+func (n *GraphNodeBasicSubgraph) Subgraph() *Graph {
+	return n.Graph
+}

terraform/transform_expand_test.go

@@ -0,0 +1,77 @@
+package terraform
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/hashicorp/terraform/dag"
+)
+
+func TestExpandTransform_impl(t *testing.T) {
+	var _ GraphVertexTransformer = new(ExpandTransform)
+}
+
+func TestExpandTransform(t *testing.T) {
+	var g Graph
+	g.Add(1)
+	g.Add(2)
+	g.Connect(dag.BasicEdge(1, 2))
+
+	tf := &ExpandTransform{}
+	out, err := tf.Transform(&testExpandable{
+		Result: &g,
+	})
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+
+	sn, ok := out.(GraphNodeSubgraph)
+	if !ok {
+		t.Fatalf("not subgraph: %#v", out)
+	}
+
+	actual := strings.TrimSpace(sn.Subgraph().String())
+	expected := strings.TrimSpace(testExpandTransformStr)
+	if actual != expected {
+		t.Fatalf("bad: %s", actual)
+	}
+}
+
+func TestExpandTransform_nonExpandable(t *testing.T) {
+	tf := &ExpandTransform{}
+	out, err := tf.Transform(42)
+	if err != nil {
+		t.Fatalf("err: %s", err)
+	}
+	if out != 42 {
+		t.Fatalf("bad: %#v", out)
+	}
+}
+
+type testExpandable struct {
+	// Inputs
+	Result      *Graph
+	ResultError error
+
+	// Outputs
+	Builder GraphBuilder
+}
+
+func (n *testExpandable) Expand(b GraphBuilder) (GraphNodeSubgraph, error) {
+	n.Builder = b
+	return &testSubgraph{n.Result}, n.ResultError
+}
+
+type testSubgraph struct {
+	Graph *Graph
+}
+
+func (n *testSubgraph) Subgraph() *Graph {
+	return n.Graph
+}
+
+const testExpandTransformStr = `
+1
+  2
+2
+`