Path: utzoo!attcan!uunet!decwrl!ucbvax!THUMPER.BELLCORE.COM!nsb From: nsb@THUMPER.BELLCORE.COM (Nathaniel Borenstein) Newsgroups: comp.soft-sys.andrew Subject: Re: Dear Saint Andrew... Message-ID: Date: 19 Jul 90 12:25:44 GMT References: Sender: daemon@ucbvax.BERKELEY.EDU Organization: The Internet Lines: 21 Excerpts from info-andrew: 18-Jul-90 Re: Dear Saint Andrew... Bill Janssen@parc.xerox. (2504+0) > Maybe that's the way UNIX does quality control. None of the `standard' > applications have any error-checking code, they just expect the > underlying mechanisms to work, so the OS people have to *make* them > work. Doesn't work out too well for people running research versions of > the OS, though... I can't let this one slip by. Most of the critical system calls are DOCUMENTED to have conditions when they fail. A responsible system designer therefore has to consider the consequences when this happens. You can build, for example, the most reliable file system in the world, but it will still occasionally fail to store files -- what you expect, however, is that it will tell you whether or not it has succeeded, and let you respond accordingly. I don't think this sort of thing works much better for "industrial-strength" versions of UNIX than for "research" versions. The system calls typically have semantics that include reasonable failure cases, and you have to consider these, that's all.