On Fri, Oct 15, 2010 at 4:35 AM, Branko Čibej <br...@xbc.nu> wrote: > On 14.10.2010 20:39, Hyrum K. Wright wrote: >> The following is a somewhat naïve implementation, but does it jive >> with your suggestion? > > Roughly yes, see the other comment. > On reflection, though, I like the suggestion of returning an > std::pair<std::string, bool>. Make a typedef of that so that users can > declare return-value variables, and use it where it's absolutely > necessary to know that it's not an empty string but a null string. Saves > a lot of trouble with the string subclassing, too. And better than > pulling in 90% of boost just to get a poor-man's replacement for null > references.
Yeah, after that little implementation exercise, and hearing here all the nuances of subclassing std::string, I'm leaning in this direction as well. Having a separate NullableString type would also help folks know which strings are guaranteed to have values and which ones are "optional" via the API. (I hope our C docs point this out explicitly, too.) -Hyrum
