As per #5042 (comment)

• setkey.R::getindex(). Adding the selfrefok() check there was my first attempt for don't use index when selfref detects prior copy by another package #5084 but that didn't work because shallow() in bmerge() had already set selfref to ok but left the invalid index intact, hence fixing shallow. Hopefully that was a one-off and everything does now go through getindex() in one place.
• add verbose message(s) when indexes aren't being used due to selfref
• [.data.table at the start so as to remove invalid indexes as soon as possible? That way, if it generates a warning optionally too, that can be generated as close as possible after the external package call that created the invalid index, to help debugging performance issues caused by indexes being dropped.
• add tests preferably using base R like the test in setDF deletes the index attribute #4893 so they can be in the main tests.Rraw rather than other.Rraw