Query DSL should disallow setting boosts on inner span queries

[jpountz]

This is a follow-up of https://discuss.elastic.co/t/troubles-with-complex-span-query-term-boosting/116106. It is a bit confusing that we are allowing boosts on inner span queries in spite of the fact that they will be ignored.

One idea to fix it is to make SpanBoostQuery a forbidden API (I'll bring up removing it from Lucene) and replacing usage of SpanBoostQuery with BoostQuery.

[s1monw]

++ to make the fix here first

[elasticmachine]

Pinging @elastic/es-search-aggs

[cbismuth]

I'd like to work on this one.

[cbismuth]

cURL recreation script created in this Gist.

[cbismuth]

After having a look at implementations of SpanQueryBuilder, here are below compound span queries along with their class attributes.

• SpanNearQueryBuilder
  • SpanNearQueryBuilder#clauses
• SpanContainingQueryBuilder
  • SpanContainingQueryBuilder#big
  • SpanContainingQueryBuilder#little
• SpanWithinQueryBuilder
  • SpanWithinQueryBuilder#big
  • SpanWithinQueryBuilder#little
• SpanOrQueryBuilder
  • SpanOrQueryBuilder#clauses
• SpanNotQueryBuilder
  • SpanNotQueryBuilder#include
  • SpanNotQueryBuilder#exclude

Here are below other span queries in which boost shouldn't be disallowed as they aren't compound.

• SpanTermQueryBuilder (uses a single SpanTermQueryBuilder in fromXContent)
• SpanMultiTermQueryBuilder (uses a single MultiTermQueryBuilder#multiTermQueryBuilder)
• FieldMaskingSpanQueryBuilder (uses a single SpanQueryBuilder#queryBuilder)
• SpanFirstQueryBuilder (uses a single SpanQueryBuilder#matchBuilder)

I'll try to raise an exception when boost field is parsed in compound span queries.

[cbismuth]

I've opened PR #34112.

Here is a sample output when running referenced Gist.

{
  "error" : {
    "root_cause" : [
      {
        "type" : "parsing_exception",
        "reason" : "spanOr [clauses] can't have boost value",
        "line" : 18,
        "col" : 15
      }
    ],
    "type" : "parsing_exception",
    "reason" : "spanOr [clauses] can't have boost value",
    "line" : 18,
    "col" : 15
  },
  "status" : 400
}

Notes

• Tests aren't developed yet,
• Exception checking can be centralized to be reused in other span query builders,
• Backward compatibility should be discussed.

But please let me know what do you think about this implementation.

@jpountz I've seen this line of code below in SpanMultiTermQueryBuilder, is it what you think about when you're talking about a forbidden API.

elasticsearch/server/src/main/java/org/elasticsearch/index/query/SpanMultiTermQueryBuilder.java

Line 196 in 269ae0b

assert subQuery instanceof SpanBoostQuery == false;