Map a lane's existing IndexType to the dialect its filter is written in.
§5.5 proposed a new MetadataFilterKind column for this. It turned out to be unnecessary:
IndexType already carries the information — every provider dispatches on it to decide whether a
lane is even theirs — so a second column would be a redundant field to keep in sync, wrong on
every pre-existing row until backfilled, and a new thing for authors to get wrong. Deriving is
strictly better. If a deployment ever needs to override the derivation, THEN add the column as an
override rather than as the source of truth.
Other maps to json rather than to none: an unknown external index is the case we know least
about, so it gets escaping rather than a pass. Choosing none there would mean the least-understood
lane is the only unprotected one.
Map a lane's existing
IndexTypeto the dialect its filter is written in.§5.5 proposed a new
MetadataFilterKindcolumn for this. It turned out to be unnecessary:IndexTypealready carries the information — every provider dispatches on it to decide whether a lane is even theirs — so a second column would be a redundant field to keep in sync, wrong on every pre-existing row until backfilled, and a new thing for authors to get wrong. Deriving is strictly better. If a deployment ever needs to override the derivation, THEN add the column as an override rather than as the source of truth.Othermaps tojsonrather than tonone: an unknown external index is the case we know least about, so it gets escaping rather than a pass. Choosingnonethere would mean the least-understood lane is the only unprotected one.