blob: 96e62a58d4983d237136f91bfe5b51344156bde6 [file] [log] [blame] [edit]
package hcl
import (
"fmt"
)
// MergeFiles combines the given files to produce a single body that contains
// configuration from all of the given files.
//
// The ordering of the given files decides the order in which contained
// elements will be returned. If any top-level attributes are defined with
// the same name across multiple files, a diagnostic will be produced from
// the Content and PartialContent methods describing this error in a
// user-friendly way.
func MergeFiles(files []*File) Body {
var bodies []Body
for _, file := range files {
bodies = append(bodies, file.Body)
}
return MergeBodies(bodies)
}
// MergeBodies is like MergeFiles except it deals directly with bodies, rather
// than with entire files.
func MergeBodies(bodies []Body) Body {
if len(bodies) == 0 {
// Swap out for our singleton empty body, to reduce the number of
// empty slices we have hanging around.
return emptyBody
}
// If any of the given bodies are already merged bodies, we'll unpack
// to flatten to a single mergedBodies, since that's conceptually simpler.
// This also, as a side-effect, eliminates any empty bodies, since
// empties are merged bodies with no inner bodies.
var newLen int
var flatten bool
for _, body := range bodies {
if children, merged := body.(mergedBodies); merged {
newLen += len(children)
flatten = true
} else {
newLen++
}
}
if !flatten { // not just newLen == len, because we might have mergedBodies with single bodies inside
return mergedBodies(bodies)
}
if newLen == 0 {
// Don't allocate a new empty when we already have one
return emptyBody
}
new := make([]Body, 0, newLen)
for _, body := range bodies {
if children, merged := body.(mergedBodies); merged {
new = append(new, children...)
} else {
new = append(new, body)
}
}
return mergedBodies(new)
}
var emptyBody = mergedBodies([]Body{})
// EmptyBody returns a body with no content. This body can be used as a
// placeholder when a body is required but no body content is available.
func EmptyBody() Body {
return emptyBody
}
type mergedBodies []Body
// Content returns the content produced by applying the given schema to all
// of the merged bodies and merging the result.
//
// Although required attributes _are_ supported, they should be used sparingly
// with merged bodies since in this case there is no contextual information
// with which to return good diagnostics. Applications working with merged
// bodies may wish to mark all attributes as optional and then check for
// required attributes afterwards, to produce better diagnostics.
func (mb mergedBodies) Content(schema *BodySchema) (*BodyContent, Diagnostics) {
// the returned body will always be empty in this case, because mergedContent
// will only ever call Content on the child bodies.
content, _, diags := mb.mergedContent(schema, false)
return content, diags
}
func (mb mergedBodies) PartialContent(schema *BodySchema) (*BodyContent, Body, Diagnostics) {
return mb.mergedContent(schema, true)
}
func (mb mergedBodies) JustAttributes() (Attributes, Diagnostics) {
attrs := make(map[string]*Attribute)
var diags Diagnostics
for _, body := range mb {
thisAttrs, thisDiags := body.JustAttributes()
if len(thisDiags) != 0 {
diags = append(diags, thisDiags...)
}
if thisAttrs != nil {
for name, attr := range thisAttrs {
if existing := attrs[name]; existing != nil {
diags = diags.Append(&Diagnostic{
Severity: DiagError,
Summary: "Duplicate argument",
Detail: fmt.Sprintf(
"Argument %q was already set at %s",
name, existing.NameRange.String(),
),
Subject: &attr.NameRange,
})
continue
}
attrs[name] = attr
}
}
}
return attrs, diags
}
func (mb mergedBodies) MissingItemRange() Range {
if len(mb) == 0 {
// Nothing useful to return here, so we'll return some garbage.
return Range{
Filename: "<empty>",
}
}
// arbitrarily use the first body's missing item range
return mb[0].MissingItemRange()
}
func (mb mergedBodies) mergedContent(schema *BodySchema, partial bool) (*BodyContent, Body, Diagnostics) {
// We need to produce a new schema with none of the attributes marked as
// required, since _any one_ of our bodies can contribute an attribute value.
// We'll separately check that all required attributes are present at
// the end.
mergedSchema := &BodySchema{
Blocks: schema.Blocks,
}
for _, attrS := range schema.Attributes {
mergedAttrS := attrS
mergedAttrS.Required = false
mergedSchema.Attributes = append(mergedSchema.Attributes, mergedAttrS)
}
var mergedLeftovers []Body
content := &BodyContent{
Attributes: map[string]*Attribute{},
}
var diags Diagnostics
for _, body := range mb {
var thisContent *BodyContent
var thisLeftovers Body
var thisDiags Diagnostics
if partial {
thisContent, thisLeftovers, thisDiags = body.PartialContent(mergedSchema)
} else {
thisContent, thisDiags = body.Content(mergedSchema)
}
if thisLeftovers != nil {
mergedLeftovers = append(mergedLeftovers, thisLeftovers)
}
if len(thisDiags) != 0 {
diags = append(diags, thisDiags...)
}
if thisContent.Attributes != nil {
for name, attr := range thisContent.Attributes {
if existing := content.Attributes[name]; existing != nil {
diags = diags.Append(&Diagnostic{
Severity: DiagError,
Summary: "Duplicate argument",
Detail: fmt.Sprintf(
"Argument %q was already set at %s",
name, existing.NameRange.String(),
),
Subject: &attr.NameRange,
})
continue
}
content.Attributes[name] = attr
}
}
if len(thisContent.Blocks) != 0 {
content.Blocks = append(content.Blocks, thisContent.Blocks...)
}
}
// Finally, we check for required attributes.
for _, attrS := range schema.Attributes {
if !attrS.Required {
continue
}
if content.Attributes[attrS.Name] == nil {
// We don't have any context here to produce a good diagnostic,
// which is why we warn in the Content docstring to minimize the
// use of required attributes on merged bodies.
diags = diags.Append(&Diagnostic{
Severity: DiagError,
Summary: "Missing required argument",
Detail: fmt.Sprintf(
"The argument %q is required, but was not set.",
attrS.Name,
),
})
}
}
leftoverBody := MergeBodies(mergedLeftovers)
return content, leftoverBody, diags
}