Field limits for objects vs flattened documents

[acmatt (Matt)]

I recently ran into an issue when creating indexes and wanted to confirm if my understanding was correct.

When I attempt to create a new index, with a total_fields.limit of 1, and a single object with a single field I receive an illegal_argument_exception

request

    PUT /test-index-1
    {
    	"settings": {
    		"index.mapping.total_fields.limit": 1
    	},
    	"mappings": {
    		"properties": {
    			"foo": {
                    "properties": {
                        "bar": {
                              "type": "keyword"
                        }
                    }
                }
            }
        }
    }

response

    {
        "error": {
            "root_cause": [
                {
                    "type": "illegal_argument_exception",
                    "reason": "Limit of total fields [1] in index [test-index-1] has been exceeded"
                }
            ],
            "type": "illegal_argument_exception",
            "reason": "Limit of total fields [1] in index [test-index-1] has been exceeded"
        },
        "status": 400
    }

However, if I flatten the object I do not receive the same error
request

    PUT /test-index-2
    {
    	"settings": {
    		"index.mapping.total_fields.limit": 1
    	},
    	"mappings": {
    		"properties": {
    			"foo_bar": {"type": "keyword"}
            }
        }
    }

response

    {
	    "acknowledged": true,
	    "shards_acknowledged": true,
	    "index": "test-index-2"
    }

Based on these examples I have a couple of question:

1. Does creating an object with a single property really create more than a single field? My understanding is that Elasticsearch will flatten these objects so they two examples should be equivalent.
   I increased index.mapping.total_fields.limit to 2, and inserted a few identical documents into both test-index-1 and test-index-2. The sizes of both Elasticsearch indexes are the same, and the Lucene data in the Elasticsearch data directory on disk looks the same.

2. Is this just a bug in the way the index PUT endpoint counts fields?

Thanks in advance for any replies. Interested in learning more about how this works.