terraform/config/config.go

1201 lines
29 KiB
Go
Raw Normal View History

2014-05-24 20:36:22 +02:00
// The config package is responsible for loading and validating the
// configuration.
2014-05-23 01:56:28 +02:00
package config
import (
2014-05-24 22:57:51 +02:00
"fmt"
2014-10-09 00:06:04 +02:00
"regexp"
2014-10-02 20:34:08 +02:00
"strconv"
"strings"
2014-07-03 19:14:17 +02:00
hcl2 "github.com/hashicorp/hcl2/hcl"
2016-01-31 08:38:37 +01:00
"github.com/hashicorp/hil/ast"
"github.com/hashicorp/terraform/helper/hilmapstructure"
"github.com/hashicorp/terraform/plugin/discovery"
"github.com/hashicorp/terraform/tfdiags"
"github.com/mitchellh/reflectwalk"
)
2014-10-09 00:06:04 +02:00
// NameRegexp is the regular expression that all names (modules, providers,
// resources, etc.) must follow.
var NameRegexp = regexp.MustCompile(`(?i)\A[A-Z0-9_][A-Z0-9\-\_]*\z`)
2014-10-09 00:06:04 +02:00
2014-05-23 01:56:28 +02:00
// Config is the configuration that comes from loading a collection
// of Terraform templates.
type Config struct {
// Dir is the path to the directory where this configuration was
// loaded from. If it is blank, this configuration wasn't loaded from
// any meaningful directory.
Dir string
Terraform *Terraform
2015-03-05 21:56:31 +01:00
Atlas *AtlasConfig
2014-09-12 00:58:30 +02:00
Modules []*Module
2014-07-20 01:05:48 +02:00
ProviderConfigs []*ProviderConfig
2014-05-26 03:05:18 +02:00
Resources []*Resource
2014-07-19 02:48:30 +02:00
Variables []*Variable
Locals []*Local
2014-07-19 07:21:52 +02:00
Outputs []*Output
2014-07-19 01:00:21 +02:00
// The fields below can be filled in by loaders for validation
// purposes.
unknownKeys []string
2014-05-26 03:05:18 +02:00
}
2015-03-05 21:56:31 +01:00
// AtlasConfig is the configuration for building in HashiCorp's Atlas.
type AtlasConfig struct {
Name string
Include []string
Exclude []string
}
2014-09-12 00:58:30 +02:00
// Module is a module used within a configuration.
//
// This does not represent a module itself, this represents a module
// call-site within an existing configuration.
type Module struct {
Name string
Source string
Version string
Providers map[string]string
2014-09-12 00:58:30 +02:00
RawConfig *RawConfig
}
2014-05-26 03:05:18 +02:00
// ProviderConfig is the configuration for a resource provider.
//
// For example, Terraform needs to set the AWS access keys for the AWS
// resource provider.
type ProviderConfig struct {
2014-07-20 01:05:48 +02:00
Name string
Alias string
Version string
RawConfig *RawConfig
2014-05-23 01:56:28 +02:00
}
// A resource represents a single Terraform resource in the configuration.
// A Terraform resource is something that supports some or all of the
// usual "create, read, update, delete" operations, depending on
// the given Mode.
2014-05-23 01:56:28 +02:00
type Resource struct {
Mode ResourceMode // which operations the resource supports
Name string
Type string
2014-10-02 20:34:08 +02:00
RawCount *RawConfig
RawConfig *RawConfig
Provisioners []*Provisioner
Provider string
DependsOn []string
Lifecycle ResourceLifecycle
}
// Copy returns a copy of this Resource. Helpful for avoiding shared
// config pointers across multiple pieces of the graph that need to do
// interpolation.
func (r *Resource) Copy() *Resource {
n := &Resource{
Mode: r.Mode,
Name: r.Name,
Type: r.Type,
RawCount: r.RawCount.Copy(),
RawConfig: r.RawConfig.Copy(),
Provisioners: make([]*Provisioner, 0, len(r.Provisioners)),
Provider: r.Provider,
DependsOn: make([]string, len(r.DependsOn)),
Lifecycle: *r.Lifecycle.Copy(),
}
for _, p := range r.Provisioners {
n.Provisioners = append(n.Provisioners, p.Copy())
}
copy(n.DependsOn, r.DependsOn)
return n
}
// ResourceLifecycle is used to store the lifecycle tuning parameters
// to allow customized behavior
type ResourceLifecycle struct {
CreateBeforeDestroy bool `mapstructure:"create_before_destroy"`
PreventDestroy bool `mapstructure:"prevent_destroy"`
IgnoreChanges []string `mapstructure:"ignore_changes"`
}
// Copy returns a copy of this ResourceLifecycle
func (r *ResourceLifecycle) Copy() *ResourceLifecycle {
n := &ResourceLifecycle{
CreateBeforeDestroy: r.CreateBeforeDestroy,
PreventDestroy: r.PreventDestroy,
IgnoreChanges: make([]string, len(r.IgnoreChanges)),
}
copy(n.IgnoreChanges, r.IgnoreChanges)
return n
}
// Provisioner is a configured provisioner step on a resource.
type Provisioner struct {
Type string
RawConfig *RawConfig
ConnInfo *RawConfig
When ProvisionerWhen
OnFailure ProvisionerOnFailure
2014-05-23 01:56:28 +02:00
}
// Copy returns a copy of this Provisioner
func (p *Provisioner) Copy() *Provisioner {
return &Provisioner{
Type: p.Type,
RawConfig: p.RawConfig.Copy(),
ConnInfo: p.ConnInfo.Copy(),
When: p.When,
OnFailure: p.OnFailure,
}
}
// Variable is a module argument defined within the configuration.
2014-05-23 01:56:28 +02:00
type Variable struct {
Name string
DeclaredType string `mapstructure:"type"`
Default interface{}
Description string
2014-05-23 01:56:28 +02:00
}
// Local is a local value defined within the configuration.
type Local struct {
Name string
RawConfig *RawConfig
}
2014-07-04 19:43:06 +02:00
// Output is an output defined within the configuration. An output is
// resulting data that is highlighted by Terraform when finished. An
// output marked Sensitive will be output in a masked form following
// application, but will still be available in state.
2014-07-04 19:43:06 +02:00
type Output struct {
Name string
DependsOn []string
Description string
Sensitive bool
RawConfig *RawConfig
2014-07-04 19:43:06 +02:00
}
2014-07-22 17:43:16 +02:00
// VariableType is the type of value a variable is holding, and returned
// by the Type() function on variables.
2014-07-22 16:41:55 +02:00
type VariableType byte
const (
VariableTypeUnknown VariableType = iota
VariableTypeString
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
VariableTypeList
2014-07-22 16:41:55 +02:00
VariableTypeMap
)
func (v VariableType) Printable() string {
switch v {
case VariableTypeString:
return "string"
case VariableTypeMap:
return "map"
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
case VariableTypeList:
return "list"
default:
return "unknown"
}
}
2014-06-05 21:21:05 +02:00
// ProviderConfigName returns the name of the provider configuration in
// the given mapping that maps to the proper provider configuration
// for this resource.
2014-07-20 01:05:48 +02:00
func ProviderConfigName(t string, pcs []*ProviderConfig) string {
2014-06-05 21:21:05 +02:00
lk := ""
2014-07-20 01:05:48 +02:00
for _, v := range pcs {
k := v.Name
if strings.HasPrefix(t, k) && len(k) > len(lk) {
2014-06-05 21:21:05 +02:00
lk = k
}
}
return lk
}
2014-09-12 00:58:30 +02:00
// A unique identifier for this module.
func (r *Module) Id() string {
return fmt.Sprintf("%s", r.Name)
2014-09-12 00:58:30 +02:00
}
2014-10-02 20:34:08 +02:00
// Count returns the count of this resource.
func (r *Resource) Count() (int, error) {
raw := r.RawCount.Value()
count, ok := r.RawCount.Value().(string)
if !ok {
return 0, fmt.Errorf(
"expected count to be a string or int, got %T", raw)
}
v, err := strconv.ParseInt(count, 0, 0)
2014-10-02 20:34:08 +02:00
if err != nil {
return 0, fmt.Errorf(
"cannot parse %q as an integer",
count,
)
2014-10-02 20:34:08 +02:00
}
return int(v), nil
}
// A unique identifier for this resource.
func (r *Resource) Id() string {
switch r.Mode {
case ManagedResourceMode:
return fmt.Sprintf("%s.%s", r.Type, r.Name)
case DataResourceMode:
return fmt.Sprintf("data.%s.%s", r.Type, r.Name)
default:
panic(fmt.Errorf("unknown resource mode %s", r.Mode))
}
}
// ProviderFullName returns the full name of the provider for this resource,
// which may either be specified explicitly using the "provider" meta-argument
// or implied by the prefix on the resource type name.
func (r *Resource) ProviderFullName() string {
return ResourceProviderFullName(r.Type, r.Provider)
}
// ResourceProviderFullName returns the full (dependable) name of the
// provider for a hypothetical resource with the given resource type and
// explicit provider string. If the explicit provider string is empty then
// the provider name is inferred from the resource type name.
func ResourceProviderFullName(resourceType, explicitProvider string) string {
if explicitProvider != "" {
// check for an explicit provider name, or return the original
parts := strings.SplitAfter(explicitProvider, "provider.")
return parts[len(parts)-1]
}
idx := strings.IndexRune(resourceType, '_')
if idx == -1 {
// If no underscores, the resource name is assumed to be
// also the provider name, e.g. if the provider exposes
// only a single resource of each type.
return resourceType
}
return resourceType[:idx]
}
// Validate does some basic semantic checking of the configuration.
func (c *Config) Validate() tfdiags.Diagnostics {
2014-09-25 00:48:46 +02:00
if c == nil {
return nil
}
var diags tfdiags.Diagnostics
2014-07-19 01:00:21 +02:00
for _, k := range c.unknownKeys {
diags = diags.Append(
fmt.Errorf("Unknown root level key: %s", k),
)
2014-07-19 01:00:21 +02:00
}
// Validate the Terraform config
if tf := c.Terraform; tf != nil {
errs := c.Terraform.Validate()
for _, err := range errs {
diags = diags.Append(err)
}
}
vars := c.InterpolatedVariables()
2014-07-19 02:48:30 +02:00
varMap := make(map[string]*Variable)
for _, v := range c.Variables {
2016-08-25 23:51:49 +02:00
if _, ok := varMap[v.Name]; ok {
diags = diags.Append(fmt.Errorf(
2016-08-25 23:51:49 +02:00
"Variable '%s': duplicate found. Variable names must be unique.",
v.Name,
))
2016-08-25 23:51:49 +02:00
}
2014-07-19 02:48:30 +02:00
varMap[v.Name] = v
}
for k, _ := range varMap {
if !NameRegexp.MatchString(k) {
diags = diags.Append(fmt.Errorf(
"variable %q: variable name must match regular expression %s",
k, NameRegexp,
))
}
}
for _, v := range c.Variables {
if v.Type() == VariableTypeUnknown {
diags = diags.Append(fmt.Errorf(
"Variable '%s': must be a string or a map",
v.Name,
))
continue
}
interp := false
fn := func(n ast.Node) (interface{}, error) {
// LiteralNode is a literal string (outside of a ${ ... } sequence).
// interpolationWalker skips most of these. but in particular it
// visits those that have escaped sequences (like $${foo}) as a
// signal that *some* processing is required on this string. For
// our purposes here though, this is fine and not an interpolation.
if _, ok := n.(*ast.LiteralNode); !ok {
interp = true
}
return "", nil
}
w := &interpolationWalker{F: fn}
2014-08-12 06:55:47 +02:00
if v.Default != nil {
if err := reflectwalk.Walk(v.Default, w); err == nil {
if interp {
diags = diags.Append(fmt.Errorf(
"variable %q: default may not contain interpolations",
v.Name,
))
2014-08-12 06:55:47 +02:00
}
}
}
}
2014-07-21 17:34:44 +02:00
// Check for references to user variables that do not actually
// exist and record those errors.
for source, vs := range vars {
for _, v := range vs {
uv, ok := v.(*UserVariable)
if !ok {
continue
}
2014-07-19 02:48:30 +02:00
if _, ok := varMap[uv.Name]; !ok {
diags = diags.Append(fmt.Errorf(
"%s: unknown variable referenced: '%s'; define it with a 'variable' block",
source,
uv.Name,
))
}
}
}
// Check that all count variables are valid.
for source, vs := range vars {
2014-10-08 03:03:11 +02:00
for _, rawV := range vs {
switch v := rawV.(type) {
case *CountVariable:
if v.Type == CountValueInvalid {
diags = diags.Append(fmt.Errorf(
2014-10-08 03:03:11 +02:00
"%s: invalid count variable: %s",
source,
v.FullKey(),
))
2014-10-08 03:03:11 +02:00
}
case *PathVariable:
if v.Type == PathValueInvalid {
diags = diags.Append(fmt.Errorf(
2014-10-08 03:03:11 +02:00
"%s: invalid path variable: %s",
source,
v.FullKey(),
))
2014-10-08 03:03:11 +02:00
}
}
}
}
// Check that providers aren't declared multiple times and that their
// version constraints, where present, are syntactically valid.
providerSet := make(map[string]bool)
for _, p := range c.ProviderConfigs {
name := p.FullName()
if _, ok := providerSet[name]; ok {
diags = diags.Append(fmt.Errorf(
"provider.%s: multiple configurations present; only one configuration is allowed per provider",
name,
))
continue
}
if p.Version != "" {
_, err := discovery.ConstraintStr(p.Version).Parse()
if err != nil {
diags = diags.Append(&hcl2.Diagnostic{
Severity: hcl2.DiagError,
Summary: "Invalid provider version constraint",
Detail: fmt.Sprintf(
"The value %q given for provider.%s is not a valid version constraint.",
p.Version, name,
),
// TODO: include a "Subject" source reference in here,
// once the config loader is able to retain source
// location information.
})
}
}
providerSet[name] = true
}
2014-09-12 01:02:36 +02:00
// Check that all references to modules are valid
modules := make(map[string]*Module)
dupped := make(map[string]struct{})
for _, m := range c.Modules {
// Check for duplicates
2014-09-12 01:02:36 +02:00
if _, ok := modules[m.Id()]; ok {
if _, ok := dupped[m.Id()]; !ok {
dupped[m.Id()] = struct{}{}
diags = diags.Append(fmt.Errorf(
"module %q: module repeated multiple times",
m.Id(),
))
2014-09-12 01:02:36 +02:00
}
// Already seen this module, just skip it
continue
2014-09-12 01:02:36 +02:00
}
modules[m.Id()] = m
// Check that the source has no interpolations
rc, err := NewRawConfig(map[string]interface{}{
"root": m.Source,
})
if err != nil {
diags = diags.Append(fmt.Errorf(
"module %q: module source error: %s",
m.Id(), err,
))
} else if len(rc.Interpolations) > 0 {
diags = diags.Append(fmt.Errorf(
"module %q: module source cannot contain interpolations",
m.Id(),
))
}
// Check that the name matches our regexp
if !NameRegexp.Match([]byte(m.Name)) {
diags = diags.Append(fmt.Errorf(
"module %q: module name must be a letter or underscore followed by only letters, numbers, dashes, and underscores",
m.Id(),
))
}
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
// Check that the configuration can all be strings, lists or maps
raw := make(map[string]interface{})
for k, v := range m.RawConfig.Raw {
var strVal string
if err := hilmapstructure.WeakDecode(v, &strVal); err == nil {
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
raw[k] = strVal
continue
}
var mapVal map[string]interface{}
if err := hilmapstructure.WeakDecode(v, &mapVal); err == nil {
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
raw[k] = mapVal
continue
}
var sliceVal []interface{}
if err := hilmapstructure.WeakDecode(v, &sliceVal); err == nil {
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
raw[k] = sliceVal
continue
}
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
diags = diags.Append(fmt.Errorf(
"module %q: argument %s must have a string, list, or map value",
m.Id(), k,
))
}
// Check for invalid count variables
for _, v := range m.RawConfig.Variables {
switch v.(type) {
case *CountVariable:
diags = diags.Append(fmt.Errorf(
"module %q: count variables are only valid within resources",
m.Name,
))
case *SelfVariable:
diags = diags.Append(fmt.Errorf(
"module %q: self variables are only valid within resources",
m.Name,
))
}
}
// Update the raw configuration to only contain the string values
m.RawConfig, err = NewRawConfig(raw)
if err != nil {
diags = diags.Append(fmt.Errorf(
"%s: can't initialize configuration: %s",
m.Id(), err,
))
}
// check that all named providers actually exist
for _, p := range m.Providers {
if !providerSet[p] {
diags = diags.Append(fmt.Errorf(
"module %q: cannot pass non-existent provider %q",
m.Name, p,
))
}
}
2014-09-12 01:02:36 +02:00
}
dupped = nil
// Check that all variables for modules reference modules that
// exist.
for source, vs := range vars {
for _, v := range vs {
mv, ok := v.(*ModuleVariable)
if !ok {
continue
}
if _, ok := modules[mv.Name]; !ok {
diags = diags.Append(fmt.Errorf(
"%s: unknown module referenced: %s",
source, mv.Name,
))
}
}
}
// Check that all references to resources are valid
resources := make(map[string]*Resource)
2014-09-12 01:02:36 +02:00
dupped = make(map[string]struct{})
for _, r := range c.Resources {
2014-07-19 01:31:01 +02:00
if _, ok := resources[r.Id()]; ok {
if _, ok := dupped[r.Id()]; !ok {
dupped[r.Id()] = struct{}{}
diags = diags.Append(fmt.Errorf(
2014-07-19 01:31:01 +02:00
"%s: resource repeated multiple times",
r.Id(),
))
2014-07-19 01:31:01 +02:00
}
}
resources[r.Id()] = r
}
2014-07-19 01:31:32 +02:00
dupped = nil
2014-07-26 23:49:55 +02:00
// Validate resources
2014-07-23 02:16:48 +02:00
for n, r := range resources {
// Verify count variables
for _, v := range r.RawCount.Variables {
switch v.(type) {
case *CountVariable:
diags = diags.Append(fmt.Errorf(
"%s: resource count can't reference count variable: %s",
n, v.FullKey(),
))
case *SimpleVariable:
diags = diags.Append(fmt.Errorf(
"%s: resource count can't reference variable: %s",
n, v.FullKey(),
))
// Good
case *ModuleVariable:
case *ResourceVariable:
2017-03-14 00:38:54 +01:00
case *TerraformVariable:
case *UserVariable:
case *LocalVariable:
default:
diags = diags.Append(fmt.Errorf(
2017-03-14 00:38:54 +01:00
"Internal error. Unknown type in count var in %s: %T",
n, v,
))
}
}
if !r.RawCount.couldBeInteger() {
diags = diags.Append(fmt.Errorf(
"%s: resource count must be an integer", n,
))
2014-10-03 01:51:20 +02:00
}
r.RawCount.init()
// Validate DependsOn
for _, err := range c.validateDependsOn(n, r.DependsOn, resources, modules) {
diags = diags.Append(err)
}
// Verify provisioners
for _, p := range r.Provisioners {
// This validation checks that there are no splat variables
// referencing ourself. This currently is not allowed.
for _, v := range p.ConnInfo.Variables {
rv, ok := v.(*ResourceVariable)
if !ok {
continue
}
if rv.Multi && rv.Index == -1 && rv.Type == r.Type && rv.Name == r.Name {
diags = diags.Append(fmt.Errorf(
"%s: connection info cannot contain splat variable referencing itself",
n,
))
break
}
}
for _, v := range p.RawConfig.Variables {
rv, ok := v.(*ResourceVariable)
if !ok {
continue
}
if rv.Multi && rv.Index == -1 && rv.Type == r.Type && rv.Name == r.Name {
diags = diags.Append(fmt.Errorf(
"%s: connection info cannot contain splat variable referencing itself",
n,
))
break
}
}
// Check for invalid when/onFailure values, though this should be
// picked up by the loader we check here just in case.
if p.When == ProvisionerWhenInvalid {
diags = diags.Append(fmt.Errorf(
"%s: provisioner 'when' value is invalid", n,
))
}
if p.OnFailure == ProvisionerOnFailureInvalid {
diags = diags.Append(fmt.Errorf(
"%s: provisioner 'on_failure' value is invalid", n,
))
}
}
// Verify ignore_changes contains valid entries
for _, v := range r.Lifecycle.IgnoreChanges {
if strings.Contains(v, "*") && v != "*" {
diags = diags.Append(fmt.Errorf(
"%s: ignore_changes does not support using a partial string together with a wildcard: %s",
n, v,
))
}
}
// Verify ignore_changes has no interpolations
rc, err := NewRawConfig(map[string]interface{}{
"root": r.Lifecycle.IgnoreChanges,
})
if err != nil {
diags = diags.Append(fmt.Errorf(
"%s: lifecycle ignore_changes error: %s",
n, err,
))
} else if len(rc.Interpolations) > 0 {
diags = diags.Append(fmt.Errorf(
"%s: lifecycle ignore_changes cannot contain interpolations",
n,
))
}
// If it is a data source then it can't have provisioners
if r.Mode == DataResourceMode {
if _, ok := r.RawConfig.Raw["provisioner"]; ok {
diags = diags.Append(fmt.Errorf(
"%s: data sources cannot have provisioners",
n,
))
}
}
2014-07-23 02:16:48 +02:00
}
for source, vs := range vars {
for _, v := range vs {
rv, ok := v.(*ResourceVariable)
if !ok {
continue
}
id := rv.ResourceId()
if _, ok := resources[id]; !ok {
diags = diags.Append(fmt.Errorf(
"%s: unknown resource '%s' referenced in variable %s",
source,
id,
rv.FullKey(),
))
continue
}
}
}
// Check that all locals are valid
{
found := make(map[string]struct{})
for _, l := range c.Locals {
if _, ok := found[l.Name]; ok {
diags = diags.Append(fmt.Errorf(
"%s: duplicate local. local value names must be unique",
l.Name,
))
continue
}
found[l.Name] = struct{}{}
for _, v := range l.RawConfig.Variables {
if _, ok := v.(*CountVariable); ok {
diags = diags.Append(fmt.Errorf(
"local %s: count variables are only valid within resources", l.Name,
))
}
}
}
}
// Check that all outputs are valid
2016-08-25 23:43:57 +02:00
{
found := make(map[string]struct{})
for _, o := range c.Outputs {
// Verify the output is new
if _, ok := found[o.Name]; ok {
diags = diags.Append(fmt.Errorf(
"output %q: an output of this name was already defined",
o.Name,
))
continue
}
2016-08-25 23:43:57 +02:00
found[o.Name] = struct{}{}
var invalidKeys []string
valueKeyFound := false
for k := range o.RawConfig.Raw {
if k == "value" {
valueKeyFound = true
continue
}
2016-08-25 23:43:57 +02:00
if k == "sensitive" {
if sensitive, ok := o.RawConfig.config[k].(bool); ok {
if sensitive {
o.Sensitive = true
}
continue
}
diags = diags.Append(fmt.Errorf(
"output %q: value for 'sensitive' must be boolean",
o.Name,
))
2016-08-25 23:43:57 +02:00
continue
}
if k == "description" {
if desc, ok := o.RawConfig.config[k].(string); ok {
o.Description = desc
continue
}
diags = diags.Append(fmt.Errorf(
"output %q: value for 'description' must be string",
o.Name,
))
continue
}
2016-08-25 23:43:57 +02:00
invalidKeys = append(invalidKeys, k)
}
if len(invalidKeys) > 0 {
diags = diags.Append(fmt.Errorf(
"output %q: invalid keys: %s",
o.Name, strings.Join(invalidKeys, ", "),
))
}
2016-08-25 23:43:57 +02:00
if !valueKeyFound {
diags = diags.Append(fmt.Errorf(
"output %q: missing required 'value' argument", o.Name,
))
2016-08-25 23:43:57 +02:00
}
for _, v := range o.RawConfig.Variables {
if _, ok := v.(*CountVariable); ok {
diags = diags.Append(fmt.Errorf(
"output %q: count variables are only valid within resources",
o.Name,
))
2016-08-25 23:43:57 +02:00
}
}
// Detect a common mistake of using a "count"ed resource in
// an output value without using the splat or index form.
// Prior to 0.11 this error was silently ignored, but outputs
// now have their errors checked like all other contexts.
//
// TODO: Remove this in 0.12.
for _, v := range o.RawConfig.Variables {
rv, ok := v.(*ResourceVariable)
if !ok {
continue
}
// If the variable seems to be treating the referenced
// resource as a singleton (no count specified) then
// we'll check to make sure it is indeed a singleton.
// It's a warning if not.
if rv.Multi || rv.Index != 0 {
// This reference is treating the resource as a
// multi-resource, so the warning doesn't apply.
continue
}
for _, r := range c.Resources {
if r.Id() != rv.ResourceId() {
continue
}
// We test specifically for the raw string "1" here
// because we _do_ want to generate this warning if
// the user has provided an expression that happens
// to return 1 right now, to catch situations where
// a count might dynamically be set to something
// other than 1 and thus splat syntax is still needed
// to be safe.
if r.RawCount != nil && r.RawCount.Raw != nil && r.RawCount.Raw["count"] != "1" && rv.Field != "count" {
diags = diags.Append(tfdiags.SimpleWarning(fmt.Sprintf(
"output %q: must use splat syntax to access %s attribute %q, because it has \"count\" set; use %s.*.%s to obtain a list of the attributes across all instances",
o.Name,
r.Id(), rv.Field,
r.Id(), rv.Field,
)))
}
}
}
}
}
2015-02-23 23:43:14 +01:00
// Validate the self variable
for source, rc := range c.rawConfigs() {
// Ignore provisioners. This is a pretty brittle way to do this,
// but better than also repeating all the resources.
if strings.Contains(source, "provision") {
continue
}
for _, v := range rc.Variables {
if _, ok := v.(*SelfVariable); ok {
diags = diags.Append(fmt.Errorf(
"%s: cannot contain self-reference %s",
source, v.FullKey(),
))
2015-02-23 23:43:14 +01:00
}
}
}
return diags
}
// InterpolatedVariables is a helper that returns a mapping of all the interpolated
// variables within the configuration. This is used to verify references
// are valid in the Validate step.
func (c *Config) InterpolatedVariables() map[string][]InterpolatedVariable {
result := make(map[string][]InterpolatedVariable)
for source, rc := range c.rawConfigs() {
for _, v := range rc.Variables {
result[source] = append(result[source], v)
}
}
return result
}
// rawConfigs returns all of the RawConfigs that are available keyed by
// a human-friendly source.
func (c *Config) rawConfigs() map[string]*RawConfig {
result := make(map[string]*RawConfig)
for _, m := range c.Modules {
source := fmt.Sprintf("module '%s'", m.Name)
result[source] = m.RawConfig
}
for _, pc := range c.ProviderConfigs {
source := fmt.Sprintf("provider config '%s'", pc.Name)
result[source] = pc.RawConfig
}
for _, rc := range c.Resources {
source := fmt.Sprintf("resource '%s'", rc.Id())
result[source+" count"] = rc.RawCount
result[source+" config"] = rc.RawConfig
for i, p := range rc.Provisioners {
subsource := fmt.Sprintf(
"%s provisioner %s (#%d)",
source, p.Type, i+1)
result[subsource] = p.RawConfig
}
}
for _, o := range c.Outputs {
source := fmt.Sprintf("output '%s'", o.Name)
result[source] = o.RawConfig
}
return result
}
func (c *Config) validateDependsOn(
n string,
v []string,
resources map[string]*Resource,
modules map[string]*Module) []error {
// Verify depends on points to resources that all exist
var errs []error
for _, d := range v {
// Check if we contain interpolations
rc, err := NewRawConfig(map[string]interface{}{
"value": d,
})
if err == nil && len(rc.Variables) > 0 {
errs = append(errs, fmt.Errorf(
"%s: depends on value cannot contain interpolations: %s",
n, d))
continue
}
// If it is a module, verify it is a module
if strings.HasPrefix(d, "module.") {
name := d[len("module."):]
if _, ok := modules[name]; !ok {
errs = append(errs, fmt.Errorf(
"%s: resource depends on non-existent module '%s'",
n, name))
}
continue
}
// Check resources
if _, ok := resources[d]; !ok {
errs = append(errs, fmt.Errorf(
"%s: resource depends on non-existent resource '%s'",
n, d))
}
}
return errs
}
2014-09-12 04:54:02 +02:00
func (m *Module) mergerName() string {
return m.Id()
}
func (m *Module) mergerMerge(other merger) merger {
m2 := other.(*Module)
result := *m
result.Name = m2.Name
result.RawConfig = result.RawConfig.merge(m2.RawConfig)
if m2.Source != "" {
result.Source = m2.Source
}
return &result
}
2014-07-21 02:17:03 +02:00
func (o *Output) mergerName() string {
return o.Name
}
func (o *Output) mergerMerge(m merger) merger {
o2 := m.(*Output)
result := *o
result.Name = o2.Name
result.Description = o2.Description
2014-07-21 02:17:03 +02:00
result.RawConfig = result.RawConfig.merge(o2.RawConfig)
result.Sensitive = o2.Sensitive
result.DependsOn = o2.DependsOn
2014-07-21 02:17:03 +02:00
return &result
}
func (c *ProviderConfig) GoString() string {
return fmt.Sprintf("*%#v", *c)
}
func (c *ProviderConfig) FullName() string {
if c.Alias == "" {
return c.Name
}
return fmt.Sprintf("%s.%s", c.Name, c.Alias)
}
2014-07-21 02:17:03 +02:00
func (c *ProviderConfig) mergerName() string {
return c.Name
}
func (c *ProviderConfig) mergerMerge(m merger) merger {
c2 := m.(*ProviderConfig)
result := *c
result.Name = c2.Name
result.RawConfig = result.RawConfig.merge(c2.RawConfig)
if c2.Alias != "" {
result.Alias = c2.Alias
}
2014-07-21 02:17:03 +02:00
return &result
}
func (r *Resource) mergerName() string {
return r.Id()
2014-07-21 02:17:03 +02:00
}
func (r *Resource) mergerMerge(m merger) merger {
r2 := m.(*Resource)
result := *r
result.Mode = r2.Mode
2014-07-21 02:17:03 +02:00
result.Name = r2.Name
result.Type = r2.Type
result.RawConfig = result.RawConfig.merge(r2.RawConfig)
2014-10-02 20:34:08 +02:00
if r2.RawCount.Value() != "1" {
result.RawCount = r2.RawCount
2014-07-21 02:17:03 +02:00
}
if len(r2.Provisioners) > 0 {
result.Provisioners = r2.Provisioners
}
return &result
}
// Merge merges two variables to create a new third variable.
func (v *Variable) Merge(v2 *Variable) *Variable {
// Shallow copy the variable
result := *v
// The names should be the same, but the second name always wins.
result.Name = v2.Name
if v2.DeclaredType != "" {
result.DeclaredType = v2.DeclaredType
}
if v2.Default != nil {
2014-07-21 02:17:03 +02:00
result.Default = v2.Default
}
if v2.Description != "" {
result.Description = v2.Description
}
return &result
}
var typeStringMap = map[string]VariableType{
"string": VariableTypeString,
"map": VariableTypeMap,
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
"list": VariableTypeList,
}
// Type returns the type of variable this is.
2014-07-22 16:41:55 +02:00
func (v *Variable) Type() VariableType {
if v.DeclaredType != "" {
declaredType, ok := typeStringMap[v.DeclaredType]
if !ok {
return VariableTypeUnknown
}
return declaredType
2014-07-22 16:41:55 +02:00
}
return v.inferTypeFromDefault()
}
// ValidateTypeAndDefault ensures that default variable value is compatible
// with the declared type (if one exists), and that the type is one which is
// known to Terraform
func (v *Variable) ValidateTypeAndDefault() error {
// If an explicit type is declared, ensure it is valid
if v.DeclaredType != "" {
if _, ok := typeStringMap[v.DeclaredType]; !ok {
validTypes := []string{}
for k := range typeStringMap {
validTypes = append(validTypes, k)
}
return fmt.Errorf(
"Variable '%s' type must be one of [%s] - '%s' is not a valid type",
v.Name,
strings.Join(validTypes, ", "),
v.DeclaredType,
)
}
2014-07-22 16:41:55 +02:00
}
if v.DeclaredType == "" || v.Default == nil {
return nil
2014-07-22 16:41:55 +02:00
}
if v.inferTypeFromDefault() != v.Type() {
return fmt.Errorf("'%s' has a default value which is not of type '%s' (got '%s')",
v.Name, v.DeclaredType, v.inferTypeFromDefault().Printable())
}
return nil
2014-07-22 16:41:55 +02:00
}
2014-07-21 02:17:03 +02:00
func (v *Variable) mergerName() string {
return v.Name
}
func (v *Variable) mergerMerge(m merger) merger {
return v.Merge(m.(*Variable))
}
// Required tests whether a variable is required or not.
func (v *Variable) Required() bool {
return v.Default == nil
}
// inferTypeFromDefault contains the logic for the old method of inferring
// variable types - we can also use this for validating that the declared
// type matches the type of the default value
func (v *Variable) inferTypeFromDefault() VariableType {
if v.Default == nil {
return VariableTypeString
}
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
var s string
if err := hilmapstructure.WeakDecode(v.Default, &s); err == nil {
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
v.Default = s
return VariableTypeString
}
var m map[string]interface{}
if err := hilmapstructure.WeakDecode(v.Default, &m); err == nil {
v.Default = m
return VariableTypeMap
}
var l []interface{}
if err := hilmapstructure.WeakDecode(v.Default, &l); err == nil {
core: support native list variables in config This commit adds support for native list variables and outputs, building up on the previous change to state. Interpolation functions now return native lists in preference to StringList. List variables are defined like this: variable "test" { # This can also be inferred type = "list" default = ["Hello", "World"] } output "test_out" { value = "${var.a_list}" } This results in the following state: ``` ... "outputs": { "test_out": [ "hello", "world" ] }, ... ``` And the result of terraform output is as follows: ``` $ terraform output test_out = [ hello world ] ``` Using the output name, an xargs-friendly representation is output: ``` $ terraform output test_out hello world ``` The output command also supports indexing into the list (with appropriate range checking and no wrapping): ``` $ terraform output test_out 1 world ``` Along with maps, list outputs from one module may be passed as variables into another, removing the need for the `join(",", var.list_as_string)` and `split(",", var.list_as_string)` which was previously necessary in Terraform configuration. This commit also updates the tests and implementations of built-in interpolation functions to take and return native lists where appropriate. A backwards compatibility note: previously the concat interpolation function was capable of concatenating either strings or lists. The strings use case was deprectated a long time ago but still remained. Because we cannot return `ast.TypeAny` from an interpolation function, this use case is no longer supported for strings - `concat` is only capable of concatenating lists. This should not be a huge issue - the type checker picks up incorrect parameters, and the native HIL string concatenation - or the `join` function - can be used to replicate the missing behaviour.
2016-04-22 02:03:24 +02:00
v.Default = l
return VariableTypeList
}
return VariableTypeUnknown
}
func (m ResourceMode) Taintable() bool {
switch m {
case ManagedResourceMode:
return true
case DataResourceMode:
return false
default:
panic(fmt.Errorf("unsupported ResourceMode value %s", m))
}
}