Hello, Our payload handling became broken since Lucene/Solr 7.2, we sometimes get 0.0 = AveragePayloadFunction.docScore() for some but not all query clauses. We only have payloads on some terms, to signal the similarity it needs to 'punish' the term, e.g. being a article or adjective.
I examined the tickets and it is not immediately clear on how to migrate from old Similarity style payload scoring, to the new method. Is really as simple as making a custom PayloadDecoder (that actually then houses the scoring logic we now have in the Similarity), and pass that new decoder to PayloadScoreQuery in our custom QParser? Additional question, why do some clauses receive 0? In the similarity, if there is no payload, i always return 1.0f. Also, is/was it known that user implementations could break with the 7.2 upgrade? It did not show up on our tests (probably due to it not failing for every clause). The problem only became visible during testing, but by accident. Many thanks, Markus
