gotosocial/vendor/github.com/superseriousbusiness/activity/streams/util.go

223 lines
5.4 KiB
Go
Raw Permalink Normal View History

package streams
import (
"maps"
"slices"
2021-11-13 16:29:43 +00:00
"github.com/superseriousbusiness/activity/streams/vocab"
)
const (
// jsonLDContext is the key for the JSON-LD specification's context
// value. It contains the definitions of the types contained within the
// rest of the payload. Important for linked-data representations, but
// only applicable to go-fed at code-generation time.
jsonLDContext = "@context"
asNS = "https://www.w3.org/ns/activitystreams"
tootNS = "http://joinmastodon.org/ns"
schemaNS = "http://schema.org"
)
// Map of inlines @context entries that may need to be added
// when vocabs include "https://www.w3.org/ns/activitystreams".
var asInlines = map[string]any{
"Hashtag": "as:Hashtag",
"alsoKnownAs": "as:alsoKnownAs",
"manuallyApprovesFollowers": "as:manuallyApprovesFollowers",
"sensitive": "as:sensitive",
"movedTo": map[string]string{
"@id": "as:movedTo",
"@type": "@id",
},
}
// Map of inlines @context entries that may need to be
// added when vocabs include "http://joinmastodon.org/ns".
var tootInlines = map[string]any{
"Emoji": "toot:Emoji",
"blurhash": "toot:blurhash",
"discoverable": "toot:discoverable",
"indexable": "toot:indexable",
"memorial": "toot:memorial",
"suspended": "toot:suspended",
"votersCount": "toot:votersCount",
"featured": map[string]string{
"@id": "toot:featured",
"@type": "@id",
},
"featuredTags": map[string]string{
"@id": "toot:featuredTags",
"@type": "@id",
},
"focalPoint": map[string]string{
"@container": "@list",
"@id": "toot:focalPoint",
},
}
// Map of inlines @context entries that may need to
// be added when vocabs include "http://schema.org".
var schemaInlines = map[string]any{
"PropertyValue": "schema:PropertyValue",
"value": "schema:value",
}
// getLookup returns a lookup map of all interesting field names
// + type names on the given "in" map that may need to be inlined.
func getLookup(in map[string]any) map[string]struct{} {
out := make(map[string]struct{})
for k, v := range in {
// Pull out keys from any nested maps.
if nested, ok := v.(map[string]any); ok {
maps.Copy(out, getLookup(nested))
continue
}
// Pull out keys from any
// arrays of nested maps.
if nestedIs, ok := v.([]any); ok {
for _, nestedI := range nestedIs {
if nested, ok := nestedI.(map[string]any); ok {
maps.Copy(out, getLookup(nested))
continue
}
}
}
// For types, we actually care about
// the *value*, ie., the name of the
// type, not the type key itself.
if k == "type" {
out[v.(string)] = struct{}{}
continue
}
out[k] = struct{}{}
}
return out
}
func copyInlines(
src map[string]any,
dst map[string]any,
lookup map[string]struct{},
) {
for k, v := range src {
_, ok := lookup[k]
if ok {
dst[k] = v
}
}
}
// Serialize adds the context vocabularies contained within the type
// into the JSON-LD @context field, and aliases them appropriately.
func Serialize(a vocab.Type) (m map[string]any, e error) {
m, e = a.Serialize()
if e != nil {
return
}
var (
// Slice of vocab URIs
// used in this vocab.Type.
vocabs = a.JSONLDContext()
// Slice of vocab URIs to add
// to the base @context slice.
includeVocabs []string
// Object to inline as an extra
// entry in the @context slice.
inlinedContext = make(map[string]any)
)
// Get a lookup of all field and
// type names we need to care about.
lookup := getLookup(m)
// Go through each used vocab and see
// if we need to special case it.
for vocab := range vocabs {
switch vocab {
case asNS:
// ActivityStreams vocab.
//
// The namespace URI already points to
// a proper @context document but we
// need to add some extra inlines.
includeVocabs = append(includeVocabs, asNS)
copyInlines(asInlines, inlinedContext, lookup)
case schemaNS:
// Schema vocab.
//
// The URI doesn't point to a @context
// document so we need to inline everything.
inlinedContext["schema"] = schemaNS + "#"
copyInlines(schemaInlines, inlinedContext, lookup)
case tootNS:
// Toot/Mastodon vocab.
//
// The URI doesn't point to a @context
// document so we need to inline everything.
inlinedContext["toot"] = tootNS + "#"
copyInlines(tootInlines, inlinedContext, lookup)
default:
// No special case.
includeVocabs = append(includeVocabs, vocab)
}
}
// Sort used vocab entries alphabetically
// to make their ordering predictable.
slices.Sort(includeVocabs)
// Create final slice of @context
// entries we'll need to include.
contextEntries := make([]any, 0, len(includeVocabs)+1)
// Append each included vocab to the slice.
for _, vocab := range includeVocabs {
contextEntries = append(contextEntries, vocab)
}
// Append any inlinedContext to the slice.
if len(inlinedContext) != 0 {
contextEntries = append(contextEntries, inlinedContext)
}
// Include @context on the final output,
// using an array if there's more than
// one entry, just a property otherwise.
if len(contextEntries) != 1 {
m[jsonLDContext] = contextEntries
} else {
m[jsonLDContext] = contextEntries[0]
}
// Delete any existing `@context` in child maps.
var cleanFnRecur func(map[string]interface{})
cleanFnRecur = func(r map[string]interface{}) {
for _, v := range r {
if n, ok := v.(map[string]interface{}); ok {
delete(n, jsonLDContext)
cleanFnRecur(n)
}
}
}
cleanFnRecur(m)
return
}