Wildcard Query with Stop Words in Romanian Analyzer Generates MatchNoDocsQuery

[halilbulentorhon]

Description

Hi,
On the Romanian analyzer, using a wildcard for a stop word in a simple_query_string generates a MatchNoDocsQuery, which prevents matching any documents. The expected behavior is to remove the stop word from the query instead of generating a MatchNoDocsQuery. This behavior seems similar to issue #1272.

Environment

{
  "name" : "7f11a5b18353",
  "cluster_name" : "docker-cluster",
  "cluster_uuid" : "_XRGD9iLQP6BSlAcqjVV0A",
  "version" : {
    "number" : "8.1.3",
    "build_flavor" : "default",
    "build_type" : "docker",
    "build_hash" : "39afaa3c0fe7db4869a161985e240bd7182d7a07",
    "build_date" : "2022-04-19T08:13:25.444693396Z",
    "build_snapshot" : false,
    "lucene_version" : "9.0.0",
    "minimum_wire_compatibility_version" : "7.17.0",
    "minimum_index_compatibility_version" : "7.0.0"
  },
  "tagline" : "You Know, for Search"
}

Steps to Reproduce:

1. Create an index with the Romanian analyzer:

   PUT test_index
   {
     "settings": {
       "index": {
         "refresh_interval": "1s",
         "number_of_shards": "1",
         "analysis": {
           "analyzer": {
             "default": {
               "type": "romanian"
             }
           }
         },
         "number_of_replicas": "0"
       }
     },
     "mappings": {
       "properties": {
         "text": {
           "type": "text"
         }
       }
     }
   }
   

2. Index a document containing the stopword "Pentru":

   PUT test_index/_doc/1
   {
     "text": "Textile Pentru Casă"
   }
   

3. Search using simple_query_string with a wildcard:

   GET test_index/_search
   {
     "profile": true, 
     "query": {
       "simple_query_string": {
         "query": "(Textile*+Pentru*+Casă*)",
         "fields": ["text"],
         "analyze_wildcard": true
       }
     }
   }
   

Observed Behavior

The search result returns no matches, and the query generates a MatchNoDocsQuery. The profile output shows:

{
  "type": "MatchNoDocsQuery",
  "description": "MatchNoDocsQuery('empty BooleanQuery')"
}

Expected Behavior

The stop word should be removed from the query, and the document should match as expected.

[benwtrent]

I replicated on latest with a simplified format.

GET test_index/_search
{
  "query": {
    "query_string": {
      "query": "Textile* Pentru* Casă*",
      "fields": ["text"],
      "analyze_wildcard": true,
      "default_operator": "AND"
    }
  }
}

This works. I think this is a bug in simple_query_string and analyzing wildcards.

For example:

GET test_index/_search
{ 
  "query": {
    "simple_query_string": {
      "query": "Textile* Pentru* Casă*",
      "fields": ["text"],
      "analyze_wildcard": true,
      "default_operator": "AND"
    }
  }
}

This is basically the same query and should result in the same, but it gets transformed into a match-none.

FWIW, this is a general problem. Here is another failing example:

PUT test_index
{
  "mappings": {
    "properties": {
      "text": {
        "analyzer": "english",
        "type": "text"
      }
    }
  }
}

POST test_index/_doc
{
  "text": "The Times"
}

GET test_index/_search
{
  "query": {
    "simple_query_string": {
      "query": "the* time*",
      "fields": ["text"],
      "analyze_wildcard": true,
      "default_operator": "AND"
    }
  }
}

[elasticsearchmachine]

Pinging @elastic/es-search-relevance (Team:Search Relevance)

[halilbulentorhon]

Hi,
Thank you for reviewing. I’m willing to fix this bug ASAP and will send a PR shortly.

[benwtrent]

The bug was introduced in https://issues.apache.org/jira/browse/LUCENE-10022

It was fixed in query_string, but never in simple_query_string: #35756

[benwtrent]

test to replicate the bug

 public void testSimpleQueryStringWithAnalysisStopWords() throws Exception {
        String mapping = Strings.toString(
            XContentFactory.jsonBuilder()
                .startObject()
                .startObject("properties")
                .startObject("body")
                .field("type", "text")
                .field("analyzer", "stop")
                .endObject()
                .endObject()
                .endObject()
        );

        CreateIndexRequestBuilder mappingRequest = indicesAdmin().prepareCreate("test1").setMapping(mapping);
        mappingRequest.get();
        indexRandom(true, prepareIndex("test1").setId("1").setSource("body", "Some Text"));
        refresh();

        assertHitCount(prepareSearch().setQuery(simpleQueryStringQuery("the* text*").analyzeWildcard(true).defaultOperator(Operator.AND).field("body")), 1);
    }

In SimpleQueryStringIT

The fix should be like 3 loc.

[benwtrent]

@halilbulentorhon go ahead and open a PR :) the fix should be simple, add the test ^ above and 3 lines in the simple query string parser to return null if the number of disjuctions is empty instead of creating an empty DisMax query.