`JSON::LD::API.toRdf`, when given a Hash as input, may mutate it

[ClearlyClaire]

When given a hash with an embedded @context, it can end up removing @context from the nested hash.

This seems to be because of

json-ld/lib/json/ld/expand.rb

Line 91 in 9fd51e0

context = context.parse(input.delete('@context'), base: @options[:base])

[gkellogg]

Yes, but I believe this is removing it from a copy of the original input, unless you have a use case that illustrates an actual issue created because of that. That said, the code would probably work using input['@context'] instead of input.delete('@context').

[ClearlyClaire]

Yes, I did run into this issue with an actual use case. Mastodon signs (some of) its activities by using Linked-Data Signatures, so it uses JSON::LD::API.toRdf to normalize its activities to be signed.

Because not all consumers handle JSON-LD, we ensure that our payloads can be processed as plain JSON, so we attach the signature (computed on the normalized object) to the original hash object.

A recent PR aiming to fix a bug in Mastodon introduces nested contexts, which strips a @context from the input object and cause its meaning to change and the signature to be wrong: mastodon/mastodon#18354

The signature is basically computed like this:

def canonicalize(json)
  graph = RDF::Graph.new << JSON::LD::API.toRdf(json, documentLoader: method(:load_jsonld_context))
  graph.dump(:normalize)
end

def hash(obj)
  Digest::SHA256.hexdigest(canonicalize(obj))
end

def sign!(creator, sign_with: nil)
  options = {
    'type'    => 'RsaSignature2017',
    'creator' => [ActivityPub::TagManager.instance.uri_for(creator), '#main-key'].join,
    'created' => Time.now.utc.iso8601,
  }

  options_hash  = hash(options.without('type', 'id', 'signatureValue').merge('@context' => CONTEXT))
  document_hash = hash(@json.without('signature'))
  to_be_signed  = options_hash + document_hash
  keypair       = sign_with.present? ? OpenSSL::PKey::RSA.new(sign_with) : creator.keypair

  signature = Base64.strict_encode64(keypair.sign(OpenSSL::Digest.new('SHA256'), to_be_signed))

  @json.merge('signature' => options.merge('signatureValue' => signature))
end

I can easily work around this issue by building my object differently or doing a deep_dup on the input, but this was fairly surprising behavior.

[gkellogg]

I'll get an update out shortly with a fix. Thanks for the use case.

[gkellogg]

I pushed a fix to the develop branch. If that solves your issues, I'll push a new release out.

[ClearlyClaire]

It does solve it! Thank you for the quick fix!