isolate librustc_metadata regression using NLL #48623

Activity

1. nikomatsakis

   added

   A-NLLArea: Non-lexical lifetimes (NLL)Area: Non-lexical lifetimes (NLL)

   WG-compiler-nll

   on Feb 28, 2018
2. nikomatsakis

   modified the milestones: NLL: Invalid code does not compile, NLL: Valid code works on Feb 28, 2018
3. matthewjasper commented on Mar 2, 2018

   

   matthewjasper

   on Mar 2, 2018

   Last edited by matthewjasper

   Contributor

   More actions

   Got it down to this:

   #![feature(nll)]
   
   pub struct DATA;
   
   impl Drop for DATA {
       fn drop(&mut self) {}
   }
   
   pub fn record<'a: 'x, 'x>(
       s: &'a mut (),
       op: fn(&'x mut (), DATA),
       data: DATA,
   ) {
       move || {
           op(&mut *s, data);
       };
   }

   Removing the Drop implementation makes this compile. Suggesting that this is the issue solved by #47917, but for closure types.
   Removing DATA entirely causes this to fail on AST borrowck as well. I'm not sure what's happening there.

   edit: removing DATA causes the closure to try to implement FnMut, so the borrowck errors are correct.

   👍React with 👍2Reacted by Casey Primozic and David Wood
4. nikomatsakis commented on Mar 7, 2018

   

   nikomatsakis

   on Mar 7, 2018

   ContributorAuthor

   More actions

   BTW, @matthewjasper --- I never got around to writing a comment, but thanks a lot! that's super helpful. I have to look a bit more closely to decide if this error is legit or not, of course...
5. nikomatsakis

   added

   NLL-completeWorking towards the "valid code works" goalWorking towards the "valid code works" goal

   on Mar 14, 2018
6. nikomatsakis commented on Apr 3, 2018

   

   nikomatsakis

   on Apr 3, 2018

   ContributorAuthor

   More actions

   @matthewjasper

   Removing the Drop implementation makes this compile. Suggesting that this is the issue solved by #47917, but for closure types.

   Looking back at this, I think this hypothesis is very likely true. We could extend that fix readily enough (maybe you or @davidtwco wants to give that a go?) but I wonder now if there is a more general approach using the dropck-outlives code.
7. nikomatsakis

   assigned

   matthewjasper

   on Apr 10, 2018
8. bors

   closed this as completedin #49822on Apr 28, 2018