On Tuesday 17 August 2010 21:46:27 SHOO wrote: > > 2. As a user of this module, I would much prefer to have an exception > > throw when I call start or stop function out of order, instead of silent > > return. If the only returns, there semantic requirement for having stop > > function - it is has then same meaning as peek. > > There are three ways: > 1. enforce > 2. assert > 3. return; > > Which do you like? > I like #2 because I want to suppress influence to runtime.
I think that the typical thing to do at this point is to use assert internally to verify the state of the struct or class itself, and to use enforce to check input from the users of the module. As for calling stop() out of order, that's a bit harder to say. It's not checking input per se, but it's not exactly checking the internal consistency of the struct either... I think that I'm going to have to go with enforce() on this one simply because odds are that the users of the module are going to be using the phobos lib which will likely have been compiled in release mode where there are no assertions. As such, an assert would generally be useless, and you're just going to get silent failures. If it were primarily other Phobos modules using it that would be one thing, but since it's definitely intended for general use, I'd have to say use enforce(). Now, what you can do if you think that that could be a performance issue for a high-precision timer is you can have two versions of the function - one which has enforce() and one which silently fails on a mismatch. That one, the user can decide which to use. So, maybe stop() and checkedStop() or something similar. - Jonathan M Davis _______________________________________________ phobos mailing list phobos@puremagic.com http://lists.puremagic.com/mailman/listinfo/phobos
