Multi field with same name applies wrong analyzer

[jappievw]

When a multi-field property has a field with the same name, inquisitor just proposes to use the field.field to test analyzers. However, ElasticSearch then falls back to the default analyzer.

Please consider this example:

curl -XPOST http://127.0.0.1:9200/test_multifield_same_name -d '{
    "mappings" : {
        "type" : {
            "properties" : {
                "brand" : {
                    "type" : "multi_field",
                    "fields" : {
                        "brand" : { "type" : "string", "analyzer" : "whitespace" },
                        "untouched" : { "type" : "string", "index" : "not_analyzed" }
                    }
                }
            }
        }
    }
}'

Inquisitor suggests to test the analyzer by using the following request: curl -XPOST "http://127.0.0.1:9200/test_multifield_same_name/_analyze?pretty=true&field=brand.brand" -d 'This Is Just A Test With Capitalized Words'. This results in the standard analyzer to be used. The result is:

{
  "tokens" : [ {
    "token" : "just",
    "start_offset" : 8,
    "end_offset" : 12,
    "type" : "<ALPHANUM>",
    "position" : 3
  }, {
    "token" : "test",
    "start_offset" : 15,
    "end_offset" : 19,
    "type" : "<ALPHANUM>",
    "position" : 5
  }, {
    "token" : "capitalized",
    "start_offset" : 25,
    "end_offset" : 36,
    "type" : "<ALPHANUM>",
    "position" : 7
  }, {
    "token" : "words",
    "start_offset" : 37,
    "end_offset" : 42,
    "type" : "<ALPHANUM>",
    "position" : 8
  } ]
}

In order to test the analyzer for this brand field, Elasticsearch expects to test it like this: curl -XPOST "http://127.0.0.1:9200/test_multifield_same_name/_analyze?pretty=true&field=brand" -d 'This Is Just A Test With Capitalized Words'. This in fact results in the whitespace analyzer to be applied. The output is:

{
  "tokens" : [ {
    "token" : "This",
    "start_offset" : 0,
    "end_offset" : 4,
    "type" : "word",
    "position" : 1
  }, {
    "token" : "Is",
    "start_offset" : 5,
    "end_offset" : 7,
    "type" : "word",
    "position" : 2
  }, {
    "token" : "Just",
    "start_offset" : 8,
    "end_offset" : 12,
    "type" : "word",
    "position" : 3
  }, {
    "token" : "A",
    "start_offset" : 13,
    "end_offset" : 14,
    "type" : "word",
    "position" : 4
  }, {
    "token" : "Test",
    "start_offset" : 15,
    "end_offset" : 19,
    "type" : "word",
    "position" : 5
  }, {
    "token" : "With",
    "start_offset" : 20,
    "end_offset" : 24,
    "type" : "word",
    "position" : 6
  }, {
    "token" : "Capitalized",
    "start_offset" : 25,
    "end_offset" : 36,
    "type" : "word",
    "position" : 7
  }, {
    "token" : "Words",
    "start_offset" : 37,
    "end_offset" : 42,
    "type" : "word",
    "position" : 8
  } ]
}

The question is: when the property and the field name have exactly the same name, can you suggest it with just the property name?

[polyfractal]

Interesting, thanks for the fix. This actually looks like a bug in Elasticsearch itself, I'm going to open an issue over there to see if they can fix the underlying issue. Until then, your patch will work great.

Thanks!

[jappievw]

Yeap, it's inconsistent behaviour. I think it's caused by the fact that the multi_field property type has been added in a later release. A design goal was to make it fully backwards compatible, which is what they achieved with allowing brand instead of requiring brand.brand.