Bo Lin on SDF Blog

Yes, that ultimately comes down to how hot the code path is and the asymptotic complexity of the `clone` method itself, which are very much influenced by the nature of your application. What we're saying is that, in the context of "fast development", it's a beneficial mental model to treat cloning as cheap and expendable, *until proven otherwise*. Perhaps a direct corollary here is that it would generally be a good idea to design your datastructures *such that* clones are as cheap as possible (e.g. by making it shallow), since `Clone` is such an integral part of Rust ownership model. That way you set it up so you can use clone "guilt-free" to satisfy the borrow checker, while relegating the deep copy to an explicit method for those (much rarer) cicurmstances that needs it.