3 ms·

As mentioned here [1], NLL doesn't just provide ergonomic benefits, it actually fixed soundness bugs w/ the previous borrow checker. [1]: http://smallcultfollo

As mentioned here [1], NLL doesn't just provide ergonomic benefits, it actually fixed soundness bugs w/ the previous borrow checker. [1]: http://smallcultfollowing.com/babysteps/blog/2018/10/31/mir-based-borrowck-is-almost-here/#diagnostics-migration-and-performance http://smallcultfollowing.com/babysteps/blog/2018/10/31/mir-...

foldr 8y ago
Is that because the lexical model was fundamentally unsound? Or just because various bugs were turned up in the course of implementing NLL?
- kibwen 8y ago
  It's because the previous implementation didn't track enough information to properly cover some edge cases of the theoretical model. Extending the old implementation to track such information would have been congruent to implementing the new borrow checker, with all the additional overhead that implies (or even more, considering that the new implementation has been designed with this in mind from the ground up).
  - foldr 8y ago
    So this is actually a major bug fix rather than an ergonomic improvement?
    - steveklabnik 8y ago
      It’s both.
      - foldr 8y ago
        Ok, good to know. That's definitely not how it's presented in the blog post.
        steveklabnik 8y ago
        Most people care way more about the ergonomic improvement than they do the soundness fixes, because the soundness issues are very, very rare, but the ergonomic pain is very present.
        foldr 8y ago
        No-one would suggest that the dev team ought to hold off on a soundness fix for performance reasons. It seems like it would be worth mentioning that aspect of NLL when raising the issue of compile times. I certainly would not have made my original comment if the blog post had made that clear.
        Twirrim 8y ago
        As mentioned in the first line of the blog post (and linked) it was essentially a follow-up to another blog post, http://smallcultfollowing.com/babysteps/blog/2018/10/31/mir-based-borrowck-is-almost-here/ http://smallcultfollowing.com/babysteps/blog/2018/10/31/mir-... which describes the state of NLL, which also gives an example of bugs fixed by NLL and why it's desirable. This one was purely focussing on the performance side of things mentioned on that previous blog post.
        foldr 8y ago
        That post briefly mentions bug fixes in the text, but doesn't actually give any examples.
        kibwen 8y ago
        Here's a runnable example of a bug that NLL fixes; remove the feature flag up top to observe the use-after-free: https://play.rust-lang.org/?gist=761478ea89b3cd1df50ee274cb7a909d&version=nightly https://play.rust-lang.org/?gist=761478ea89b3cd1df50ee274cb7...
- eximius 8y ago
  I believe the latter. IIRC, NLL is strictly more expressive than LL. I believe NLL can only shorten lifetimes, not extend them, by comparison.

foldr 8y ago

Is that because the lexical model was fundamentally unsound? Or just because various bugs were turned up in the course of implementing NLL?

kibwen 8y ago
It's because the previous implementation didn't track enough information to properly cover some edge cases of the theoretical model. Extending the old implementation to track such information would have been congruent to implementing the new borrow checker, with all the additional overhead that implies (or even more, considering that the new implementation has been designed with this in mind from the ground up).
- foldr 8y ago
  So this is actually a major bug fix rather than an ergonomic improvement?
  - steveklabnik 8y ago
    It’s both.
    - foldr 8y ago
      Ok, good to know. That's definitely not how it's presented in the blog post.
      - steveklabnik 8y ago
        Most people care way more about the ergonomic improvement than they do the soundness fixes, because the soundness issues are very, very rare, but the ergonomic pain is very present.
        foldr 8y ago
        No-one would suggest that the dev team ought to hold off on a soundness fix for performance reasons. It seems like it would be worth mentioning that aspect of NLL when raising the issue of compile times. I certainly would not have made my original comment if the blog post had made that clear.
      - Twirrim 8y ago
        As mentioned in the first line of the blog post (and linked) it was essentially a follow-up to another blog post, http://smallcultfollowing.com/babysteps/blog/2018/10/31/mir-based-borrowck-is-almost-here/ http://smallcultfollowing.com/babysteps/blog/2018/10/31/mir-... which describes the state of NLL, which also gives an example of bugs fixed by NLL and why it's desirable. This one was purely focussing on the performance side of things mentioned on that previous blog post.
        foldr 8y ago
        That post briefly mentions bug fixes in the text, but doesn't actually give any examples.
        kibwen 8y ago
        Here's a runnable example of a bug that NLL fixes; remove the feature flag up top to observe the use-after-free: https://play.rust-lang.org/?gist=761478ea89b3cd1df50ee274cb7a909d&version=nightly https://play.rust-lang.org/?gist=761478ea89b3cd1df50ee274cb7...
eximius 8y ago
I believe the latter. IIRC, NLL is strictly more expressive than LL. I believe NLL can only shorten lifetimes, not extend them, by comparison.