Another unanchored thought I've had on these matters is coming around to viewing namespace collision as a feature, not a bug. If a programming system is designed to allow you to link against a given module and more or less ignore any and every transitive dependency that this will bring in—because the programming environment makes it excessively easy to do so—then that's a pretty strong reason to consider whether or not that approach to information hiding is actually an anti-feature.

On the other hand, if during your work on a program you have to reconcile the names used within the system (i.e., such that no two module names collide), then it subtly encourages you to make sure you are able to account for every module that your program depends on.

People reflexively assume that this would make it cumbersome (or even intractable) to work on a program any larger than a toy, but empiraclly we can observe that a single, shared namespace can, by and large, scale way better than these protests would lead us to believe. It's not out of the question that a project might reach, say, 100kloc with very little friction arising as a consequence of this sort of forced reckoning. (And it's worth considering of the friction that it does impose: is it any worse than the costs we've observed over the last ~10 years from the systems that set out to solve this problem?)