I don't follow. How would fsyncing the rollback journal affect the durability of the actual database? Do you actually think that the database would reapply an already committed journal whose ID in the header already indicates that the transaction was committed, when the database is already consistent? I really think you should re-review the definition of durability of a database, especially before saying the creator of SQLite is incorrect about its implementation.
I don't follow. How would fsyncing the rollback journal affect the durability of the actual database? Do you actually think that the database would reapply an already committed journal whose ID in the header already indicates that the transaction was committed, when the database is already consistent? I really think you should re-review the definition of durability of a database, especially before saying the creator of SQLite is incorrect about its implementation.