[Info-ingres] Is Ingres supposed to rollback in case of the connection dropping?

[nikosv]

In an ABF application I sometimes see transactions being half-done.For an insert to one table and then an insert to another table within the same ABF session, having the connection dropped (I guess in between), or some other Ingres internal exception happening, I find that the first insert completed successfully but the second not (as it should,since this point was never reached).Should Ingres had rolled back so that both the statements fail?

The session is run with autocommit off, the commit submitted only after both of the inserts.Also there's checking iierror with 
inquire_ingres (h_rowcount = rowcount, h_errorno = errorno), after each sql statement but I guess that only has an effect as long as there's no out of the ordinary exceptions ?