sangria-example.md

in this case the structure of the Relation goes like this:

Relation[<Entity type>, <Intermediate data-structure type that contains both: entity with ID and the relation ID that was used to load it >, <Relation ID>](<arbitrary relation name>, <intermediate data-structure> ⇒ <list of relation IDs>, <intermediate data-structure> ⇒ <entity>)

Since your are representing many-to-many relation, this would be the correct one:

val featuresByProductId = Relation[Feature, (Seq[Long /* ← this long is a product Id */], Feature), Long /* ← this long is a product Id */]("featuresByProductId", _._1, _._2)

// or

val featuresByProductId = Relation[Feature, (ProductFeature, Feature), Long /* ← this long is a product Id */]("featuresByProductId", tmp ⇒ Seq(tmp._1.productId), _._2)

The fetcher would probably look like this:

val featuresFetcher = Fetcher.relCaching(
  (repos: Context, ids: Seq[Long]) ⇒
    repos.featuresRepository.findFeaturesByIds(ids),

  (repos: Context, ids: RelationIds[Feature]) ⇒
    repos.featuresRepository.findFeaturesByProductId(ids(featuresByProductId))
)(HasId(_.id))

Assuming that findFeaturesByProductId will return this intermediate data-structure of type (ProductFeature, Feature) the purpose of this more complicated 3-argument relations is for cases where you have many-to-many relations and can't get the relation ID (productId) from the entity type (Feature). In this case you need to temporarily store the combination in some intermediate data-structure.