Date: Mon, 21 Sep 1998 20:03:57 +0200 From: Antoine Leca <Antoine.Leca@renault.fr> - I am not really happy that xtime_get does not return the value of the readed clock I agree. I'm working on a set of proposed changes to Markus Kuhn's spec. One part of the proposal changes xtime_get's signature to be the following: xtime_t xtime_get(int clock_type); This is simpler, and will make xtime_get easier to use in practice, as the programmer won't have to worry about taking addresses of time objects. Avoiding addresses also let things run a bit faster on modern architectures; this can be helpful for fine-grained timing measurements. - similar with tz_error, there might be a need for xtime_conv_error, I think it's simpler, and more in accordance with existing practice, to let strerror and perror handle error numbers. We shouldn't need a separate error-string-generator just for times. That's too complicated; such a generator would also has to worry about locales and the like. Just let strerror do it -- that's what strerror is for. (I don't mean to suggest that we should use errno; I'd rather avoid errno, and I hope my proposed changes won't use errno.) - there should be a way to determine how to size the buffer for strfxtime Absolutely. strftime botches this -- e.g. there's no portable way to distinguish an error return from a format that generates an empty string! This must get fixed in any rewrite. The only remaining point I have if about procedures. I think the best way to do is to formally present this proposal at the WG14 meeting, at Santa Cruz, California, on October 5-9. I think it's too soon to present such a proposal; it hasn't been reviewed enough yet, and it hasn't been implemented (preferably by two independent implementations), tested, and used. Also, I suspect that the working-group schedule is such that it's not feasible to introduce this proposal for C9x. I think we're better off proposing it for C0x, or as a normative addendum to C9x sometime between C9x and C0x.