Assuming SQLite has any internal counters or indexes, let alone B-trees or anything fancy, then this approach won't work. It will work for a raw record format (

Assuming SQLite has any internal counters or indexes, let alone B-trees or anything fancy, then this approach won't work. It will work for a raw record format (CSV, JSON, Protobuf, an mmapped array, etc.), but the author wants to actually interact with a real SQLite database. Generating a billion rows in some non-SQLite format still leaves problem to converting that format by loading it into SQLite, which isn't really a reduction of the original problem.