Flag for preventing fallback to default field when using quote_field_suffix

[pbvahlst]

As default "quote_field_suffix" falls back to the default field if the field generated by
adding the "quote_field_suffix" does not exist.

When you have a multi field like:
"title": { "fields": { "exact": {}, "folded": {} } }
it will search both title.exact and title.folded which is probably not what we want.

Could we have a flag for ignoring fields not found instead of using the original field?

[elasticmachine]

Pinging @elastic/es-search

[jimczi]

I am not sure I understand what the issue is. If you provide quote_field_suffix as exact and your search targets the title field the query_string will use title.exact for the quoted parts of the query. If title.exact does not exist the original field is used so I don't see what multi fields has to do with this. Can you provide an example (mappings + query) that would benefit from this feature ?

[pbvahlst]

Sorry forgot to mention that I'm using QueryStringQuery and I have set fields to ["title", "title.exact", "title.folded"] so when not searching for a phrase all three fields will be queried

[pbvahlst]

Here is a comple example

mapping

{
  "mappings": {
    "_doc": {
      "properties": {
        "title": { 
          "type": "text",
          "fields": {
            "exact": { 
              "type": "text",
              "analyzer": "english"
            },
            "folded": {
              "type": "text",
              "analyzer": "custom_folded"  
            }
          }
        }
      }
    }
  }
}

query

{
  "query": {
    "query_string" : {
      "fields" : ["title", "title.exact", "title.folded"],
      "query" : "\"john doe\"",
      "quote_field_suffix": ".exact"
    }
  }
}

When the parser adds qoute_field_suffix we get title.exact and title.folded.exact because title.folded.exact doesn't exist it falls back to title.folded and searches in title.exact and title.folded which is not what is expected.

[jimczi]

Thanks for explaining. However I am not sure that quote_field_suffix should be used to ignore fields in the query. Currently it can be used to redirect fields to an alternative that is better suited for a phrase query (with shingles for instance) so I don't understand why you need to use title.exact on non-phrase queries. Couldn't you set fields to ["title", "title.folded"] only ? The other alternative is to use a field query "query": "title:\"john doe\"" that ignores the provided fields.

[pbvahlst]

Thanks for a quick response.
The title.exact is not the problem but instead all the multi fields which doesn't match the name of the root field + qoute_field_suffix. If I set fields to ["title", "title.folded"] and qoute_field_suffix = ".exact" it would still search title.exact (which is correct) but also title.folded because there is no title.folded.exact. If it should be acceptable that quote_field_suffix will fall back to the original field if the field could not be found the usecase described at https://www.elastic.co/guide/en/elasticsearch/reference/current/mixing-exact-search-with-stemming.html seems misleading?

[jimczi]

Well the issue here is that you have more than 2 fields. I don't think that the documentation is misleading but it doesn't take into account the fact that you can have more than two variations on the same field. What should be the behavior if you search for "fields": ["title", "body"], "quote_field_suffix": ".exact" and only title.exact exists ? Should we ignore the body field if the query is quoted ? It might be more confusing than the current behavior but I got your point so I'll leave the issue open for others to comment. IMO we should not rely on a flag but rather ignore the field entirely if we think that it's the correct behavior for this feature.

[pbvahlst]

Seems reasonable. In my opinion it should ignore the field if it doesn't exist. By maybe a mention in the documentation about the fallback to the default field could help. It is only mentioned in the javadoc. From reading the documentation I a least got the idea that it would only match the fields it could find, but I understand your example as well.
My fix for now is to create a dummy field called title.folded.exact with no content.

[jimczi]

@slumx we discussed internally and opened #40975 to simplify the boosting/filtering of exact matches. It is complicated to do it right today especially when searching on multiple fields so we'd like to propose a better solution to hanlde scoring and filtering more efficiently. In the mean time I marked this issue with the >docs label because I agree that the behavior of the quote_field_suffix is unclear in the documentation.