Is this a correct use of JSON-LD Index Maps?

I want to create a collection of people in json-ld format but I need to keep some keys instead of using arrays so I tried this first:

{
  "@context" : {
    "@base" : "http://www.example.com/data/",
    "@vocab" : "http://www.example.com/vocab#",
    "name" : "schema:name",
    "people" : {
      "@container" : "@index",
      "@id" : "people"
    },
    "schema" : "http://schema.org/"
  },
  "@id" : "http://www.example.com",
  "people" : {
    "person1" : {
      "@id" : "people/person1",
      "name" : "Person 1"
    },
    "person2" : {
      "@id" : "people/person2",
      "name" : "Person 2"
    },
    "person3" : {
      "@id" : "people/person3",
      "name" : "Person 3"
    },
    "person4" : {
      "@id" : "people/person4",
      "name" : "Person 4"
    }
  }
}

Testing it in the JSON-LD Playground everything looks fine and the N-Quads result show something as expected like this:

<http://www.example.com/data/people/person1> <http://schema.org/name> "Person 1" .
<http://www.example.com/data/people/person2> <http://schema.org/name> "Person 2" .
<http://www.example.com/data/people/person3> <http://schema.org/name> "Person 3" .
<http://www.example.com/data/people/person4> <http://schema.org/name> "Person 4" .
<http://www.example.com> <http://www.example.com/vocab#people> <http://www.example.com/data/people/person1> .
<http://www.example.com> <http://www.example.com/vocab#people> <http://www.example.com/data/people/person2> .
<http://www.example.com> <http://www.example.com/vocab#people> <http://www.example.com/data/people/person3> .
<http://www.example.com> <http://www.example.com/vocab#people> <http://www.example.com/data/people/person4> .

Then I wanted to add a schema:knows property also keeping the keys and not using arrays like this:

{
  "@context" : {
    "@base" : "http://www.example.com/data/",
    "@vocab" : "http://www.example.com/vocab#",
    "knows" : {
      "@container" : "@index",
      "@id" : "schema:knows",
      "@type" : "@id"
    },
    "name" : "schema:name",
    "people" : {
      "@container" : "@index",
      "@id" : "people"
    },
    "schema" : "http://schema.org/"
  },
  "@id" : "http://www.example.com",
  "people" : {
    "person1" : {
      "@id" : "people/person1",
      "knows" : {
        "person2" : "people/person2",
        "person3" : "people/person3"
      },
      "name" : "Person 1"
    },
    "person2" : {
      "@id" : "people/person2",
      "name" : "Person 2"
    },
    "person3" : {
      "@id" : "people/person3",
      "name" : "Person 3"
    },
    "person4" : {
      "@id" : "people/person4",
      "name" : "Person 4"
    }
  }
}

This time when I tested in the playground it return the following error:

{ "name": "jsonld.SyntaxError", "message": "Invalid JSON-LD syntax; conflicting @index property detected.", "details": { "code": "conflicting indexes", "subject": { "@id": "http://www.example.com/data/people/person2", "@index": "person2" } } }

So what is an alternative to keep the keys and never use arrays or the correct way to use Index Maps?

Regards

This is correct. You've discovered a bug in jsonld.js (which I already reported). You can try to run the example in my playground or by using Gregg Kellogg's RDF distiller service. Both return the triples you'd expect.