[kibwen]

To elaborate, the original borrow checker that launched with Rust 1.0 had edge cases that could only be resolved with a drastically different analysis pipeline. As a result, it wasn't able to properly reject some patterns that were disallowed by the conceptual model. Over the course of several years this new analysis pipeline was developed and the borrow checker was rewritten to use it, which now properly rejects these patterns. But three years is plenty of time for code in the wild to begin relying on edge cases, so there's been a year-long deprecation cycle to warn anyone who might not have updated their code by now.

[csande17]

Right, I guess I had it in my head that the Rust 2015-vs-2018 divide was that deprecation cycle.

[kibwen]

The difference between Rust 2015 and 2018 is less than most people think. Originally, yes, Rust 2018 was used to push the new borrow checker without affecting code using Rust 2015. But unlike a "version" that gets deprecated and left behind, a Rust edition perpetually continues to be supported and benefit from improvements with every compiler release. But this means that in order to support different borrow checkers in different editions the compiler had to continue to ship with both borrow checkers, which, considering how much code that is and the aforementioned bugs, was an intractable proposition in the long term.