Statements w/ Blank Nodes can be duplicated in serialized output

[no-reply]

The definition of blank node equality runs afoul of keyspace definition for Ruby Hashes. This leads to repeated inserts of blank nodes that satisfy #==, and the same #id and `#hash, when they are not exactly the same object.

repo = RDF::Repository.new

repo.insert([RDF::Node.new('s'), RDF::URI('http://ex.org/p'), 'o'])
repo.insert([RDF::Node.new('s'), RDF::URI('http://ex.org/p'), 'o'])

repo.dump :ntriples
# => "_:s <http://ex.org/p> \"o\" .\n_:s <http://ex.org/p> \"o\" .\n"

A (proposed) failing test for this is at rdf-spec. An alternate solution is to continue to fail that test, but check that internally unique blank nodes are reflected as such in serialized output.

[gkellogg]

This is a serializer issue, isn't it? There's a :unique_bnodes option to Writer that re-issues BNode identifiers because of this reason; mostly, people expect to see the same BNode identifier used when re-serializing, so it's not the default.

You are inserting two different nodes, so they should be two different statements.

Alternatively, Writer could scan the graph/repo to look for BNodes with the same id which are different nodes and modify the id; however, Writer doesn't operate on graph/repo, but on each statement coming in, so it would probably need to keep a local memory and this could get ugly.

[no-reply]

The proposed solutions to this are broken down into #262 and #263. The issue is represented correctly by those tickets.

Closing.