←back to thread

Stop making me memorize the borrow checker

(erikmcclure.com)
177 points signa11 | 1 comments | 16 Nov 24 23:29 UTC | HN request time: 0.21s | source
Show context
sestep ◴[17 Nov 24 00:28 UTC] No.42160860[source]▶
It is interesting that the borrow checker doesn't run until after typechecking succeeds. As far as I'm aware, rust-analyzer has its own builtin logic for doing typechecking, but it delegates to rustc for borrow checking. I wonder whether this is just a temporary situation due to lack of engineering resources to implement borrow checking in rust-analyzer; personally I doubt that, especially since gccrs is incorporating components of rustc wholesale and so I'd be a bit surprised if rust-analyzer moves in the opposite direction. In theory it seems possible to support borrow checking in the IDE for ill-typed programs, but having borrow checking as a separate analysis pass depending on successful typechecking is just such a nice abstraction boundary to have for maintaining the toolchain.
replies(1): >>42160870 #
dwattttt ◴[17 Nov 24 00:29 UTC] No.42160870[source]▶
borrow checking relies on types to be able to check; types are what carry borrows after all.
replies(1): >>42160918 #
sestep ◴[17 Nov 24 00:38 UTC] No.42160918[source]▶
This is true but misleading: analogously, typechecking depends on parsing, but IDEs typically make a best effort to typecheck syntactically ill-formed programs.
replies(1): >>42161044 #
1. estebank ◴[17 Nov 24 01:00 UTC] No.42161044[source]▶
rustc does its best to recover, continue and provide diagnostics from later stages. But at the same time it is better to provide a single early error and mark the entire node as recovered to avoid further errors at the cost of requiring more cycles of back and forth, over the alternative of tons of useless knock down errors that drown the underlying cause of the problem.

We are always on the lookout for improving in this area. Having examples of cases where we conceivably should have done better but didn't is useful. As mentioned already, the complexity here is that doing the right thing for the user requires architecting multiple separate stages of the compiler to talk to each other in way more complex ways than originally intended.