Hey Nicolas, Instead of passing arguments to __clone(), I wondered about a new > __clone_with(array > $properties) that could be implemented next to __clone(), with the > following behavior: > > - if only __clone is implemented, same behavior as always > - if __clone_with is implemented, __clone is never called. Instead, the > engine calls __clone_with with the map of properties in the "with" (or > an empty array if no "with"). In addition and *before* the call, the > engine > sets the properties according to the "with" (potentially failing > visibility/type checks as expected) >
Interesting approach! The rules seem clear, and as far as I'm qualified to tell, the implementation is viable. I'm not sure though what should happen if one of the "clone with" assignments throws an exception? By default, __clone_with() would never be called in this case. Are we OK with this, right? Because this way, it's easy to circumvent any validation performed inside __clone_with() by triggering an exception (and then catching it). I don't think this is a huge problem, since visibility rules apply, so one cannot just blindly do silly things with someone else's code. Also, I'm wondering whether it's worth the complexity to add support for another magic method, or would we be fine without this feature and its benefits? To put it another way: how much complexity is it worth to fix the problem of the wasteful deep cloning? Does anyone have an informed answer/opinion?
