Rust: Tracking issue for RFC 2229, "Closures Capture Disjoint Fields"

[Centril]

This is a tracking issue for the RFC "Closures Capture Disjoint Fields" (rust-lang/rfcs#2229).

Status

This is under active development by the wg-rfc-2229 compiler-team working group. You can follow along in #t-compiler/wg-rfc-2229 stream of the rust-lang Zulip.

Implementation plan and status

• [x] https://github.com/rust-lang/rust/issues/56905 refactor the compiler to be less reliant on freevars, which are tied to the variables, and instead to use a map produced by the type-checker
• [x] Rename upvars query to upvars_mentioned https://github.com/rust-lang/rust/pull/72544
• [x] Rename upvars_list to closure_captures https://github.com/rust-lang/rust/pull/72591
• [ ] Refactor uses of the upvars_mentioned query to extract information from the typeck tables (https://github.com/sexxi-goose/rust/issues/4)
• [ ] Refactor hir::Place to include full detail on projections
• [ ] Refactor hir::Place to separate out the hir_id and Span (that is a PlaceReference)
• [ ] Change closure captures to use a hir::Place instead of just an upvar
• [ ] Update the closure analysis to compute minimal hir::Place captures

At this point, the implementation will capture paths and not just variables, but this will be a breaking change. Remaining work would be to:

• [ ] analyze the size of the closures that result
• [ ] write a lint that detects when destructors might run at different times and report a warning, including some idea of heuristics for things we don't care about
• [ ] use that to analyze impact on the ecosystem and to prepare a "transition lint"

Links, see also

• rust-lang/compiler-team#292, a Major Change Proposal outlining an initial plan

Unresolved questions:

• [ ] How to optimize pointers. Can borrows that all reference parts of the same object be stored as a single pointer? How should this optimization be implemented (e.g. a special repr, refinement typing)?
• [ ] How to signal that a function is pure. Is this even needed/wanted? Any other places where the language could benefit?
• [ ] Should Box be special?
• [ ] Drop order can change as a result of this RFC, is this a real stability problem? How should this be resolved?

This issue has been assigned to @arora-aman via this comment.

[eddyb]

IIRC, this needs to wait for the new MIR (NLL) borrow-checker to replace the old one.
cc @rust-lang/wg-compiler-nll

[samsartor]

I would love to follow this through and give the code a shot, but it's probably too big of a first contribution to the compiler. Maybe I'll take a swing at some starter issues and then come back if no one else has gotten it done.

@eddyb Yes, this definitely should be built on NLL.

[ghedo]

@blitzerr could you please provide a status update on this? it looks like the related issues (#57482 and #60205) had some work done, but they look stuck now. Anything that needs help with? Not sure if I can do anything, but I can try :)

[lzutao]

Concern: How does this feature interact with self-referential struct?

[RustyYato]

@lzutao Rust has no native support for self referential types, so you will just have to be careful with the unsafe code you write.

[lzutao]

I know. But with polonius, there might be support for self-referential types in the future. So for future design space we need to care about how to interact this feature with self-referential types.

[Ixrec]

This is the first I've heard of Polonius having anything to do with self-referential types, despite having read all the blog posts and docs behind your link. Is this interaction spelled out somewhere?

[KiChjang]

@Ixrec I do recall that there was a presentation from Niko that says it could maybe help with self-referential structs somehow, but he did preface it by saying that it's purely speculative: https://youtu.be/_agDeiWek8w?t=1900.

[nikomatsakis]

I'm not really aware of any particular interaction between this feature and self-referential types. Am I missing something?

[nikomatsakis]

@rustbot assign arora-aman

[rustbot]

Error: Parsing assign command in comment failed: ...'arora-aman' | error: user should start with @ at >| ''...

Please let @rust-lang/release know if you're having trouble with this bot.

[varkor]

@rustbot assign @arora-aman

[arora-aman]

@rustbot claim