There are issues that go beyond implementation difficulty. As you suggest, intimate awareness of schema intricacy cannot be blithely assumed. Hence, just dropping indexes and constraints because they mention a dropped column may be a disfavor; they may be valued DB features. With the present behavior, those intricate dependencies are at least brought to the DBA's attention, who can then decide whether some preliminary DROP operations are needed. I would not want to argue the position that the DDL implementing code can reliably make a better judgment on such cascade actions than an alerted human being. The idea of a table-like pragma listing drop-blocking dependencies is worth consideration. The library's evolution will undoubtedly be done carefully, with due consideration to the myriad use cases that can be anticipated.