Support for row-level deletes will allow Iceberg to be used for several new use cases, like UPSERT, MERGE INTO, etc.

The approach is to add two ways to encode deletes in the Iceberg spec so that engines can delete rows in existing files and append new data to replace those rows.