removed

I had thought I found another form of the same issue, but the issue I was tracking was a flaw in the JSON Schema RDF context, which has since been corrected.

It's falling afoul of logic in the Term Selection algorithm step 9 where it things that var should be an IRI or IRI reference. Something is needed for this case, and it's odd that it didn't come up before now.

I'd say that this is a spec error with an unknown resolution at this time.

Replacing "@index": "ex:name" with "@index": "http://example.org/ns/name" in the compaction context solves the problem.

If the problem is roundtripping, it seems to me that there's another related problem in this example: when compacting with the full URI as index property, the result is as follows (on the playground):

{
  "@context": {
    "ex": "http://example.org/ns/",
    "prop": {
      "@id": "ex:prop",
      "@container": "@index",
      "@index": "http://example.org/ns/name"
    }
  },
  "prop": {
    "@none": [
      {
        "@id": "ex:bar",
        "ex:name": "bar"
      },
      {
        "@id": "ex:foo",
        "ex:name": "foo"
      }
    ]
  }
}

The @none index key shouldn't be here. It's as if the index property is ignored when compacting. I've encountered the same problem on the WoT Thing Description context (see w3c/wot-thing-description#1077).

Summary:

The value of @index on a property map can be a Compact IRI in addition to an IRI. The Compaction Algorithm is updated to first expand this value before re-compacting it. Added additional language to the Compaction Algorithm step 12.8.9.6.1 to first expand container key before compacting it.

@iherman this may or may not be considered an editorial change. Also, the description doesn't render properly in the errata doc.

I've updated my implementation and deployed it on the Ruby RDF Distiller.