2025-06-23 — Last Week in Carbon

[josh11b]

News

RustConf talk title now up: Memory Safety Everywhere with Both Rust and Carbon

Updates on leads issues

• Decisions
  • #5523 Orthogonality in the binding design questions prompted by #5434
  • #5524 Semantic restrictions of immutable value (let) bindings
    • also from #5434
• New issues
  • (none)
• Active discussions
  • #5522 Immutable value semantic returns questions from #5434
    • 06-16 let -> val
    • potential direction, and some data about cost of keywords under consideration

Updates on proposals

• Decisions
  • #5670 Guidance on AI coding tools
• New RFC proposals
  • #5661 Progressive disclosure principle
• Active discussions
  • #5448 Carbon/C++ Interop: Primitive Types proposal
    • Updated based on comments
  • #5434 ref parameters, arguments, returns and val returns
    • 06-16 a b syntax of returns
    • Updated: New syntax, switched to val keywords, and other updates from leads issues

Updates on moderation across our community spaces (Discord and GitHub)

• 10 trained moderation & conduct team members + community lead, in theory available from UTC-7 to UTC+5.5.
  • PENDING: Summarizing conversation about the Zig community journey.
  • BLOCKED: resuming conversation about how we can be more intentional about providing good first issues for newcomers
  • No conduct related moderation intervention for over a month! 👏
• Last week’s stats across our spaces:
  • In the last 7 days in our repo: 7 authors pushed 27 commits to trunk, we changed 643 files with 11,712 additions and 18,305 deletions, 27 merged PRs and 14 open PRs, we closed 3 issues and opened 2 new ones.

Activity on toolchain

• #generics-and-templates #5672 rewrite resolution goes exponential
• 06-16 Idea to make generics work with eval didn't work with import
• 06-17 Rewrite constraints
• 06-17 What to work on next
• #toolchain PR #5653 made evaluating the prelude more expensive for tests. Switching many tests to use min_prelude to speed things up.
• #toolchain Reorganizing min_prelude options.
  • #toolchain Can now include none.carbon to get --no-prelude-import.
• #toolchain SemIR doesn't include even referenced constants from the prelude
• #toolchain Ordering of settings at the top of tests
• #toolchain #5678 is adding the destroy interface to replace destruction
  • Some discussion about minimize the amount of work done for tuples
• #toolchain Examples of using a rewritten associated constant value that should work
• PRs:
  • backtraces in #5651
  • add tests in #5677 #5674
  • api cleanup in #5679 #5701
  • test updates in #5683 #5676 #5680 #5681 #5630 #5682 #5695 #5690 #5645
  • rewrite constraints in #5686
  • test infrastructure in #5694
  • error handling in #5692
  • for in #5685
  • binding category in #5662
  • C++ struct interop in #5538
  • documentation in #5687 #5663 #5655
  • deps in #5700
  • other changes in #5654

Activity on C++ interop

• 06-17 Decided on a new type instruction for C++ overload set containing a list of Clang declarations. This is independent from Carbon overloading, which is not implemented yet.
• 06-17 Decided on having `Cpp.float` and `Cpp.double` as aliases to `f32` and `f64`.
• 06-17 In-flight Clang draft has been updated and now closely resembles how it should behave, needs another final polish before sent for review.

Other active discussions

These include Discord Chat, GitHub Discussions, or open discussion sessions.

• #style: Do we want "must not be null" comments on parameters, or is that implicit in not specifying the meaning of null?
  • Style guide currently requires that.
• #pattern-matching: How to unify sum type matching with dynamic dispatch?
  • Eventually converged on a Swift-like as Foo operator, with some sort of shorthand for binding patterns.
• #style: #5663 codifies existing practice for auto in the style guide.
  • Some discussion of whether it should be a proposal.
• #infra: Best practices for stacked pull requests
  • GitHub doesn't give us good options here, but the benefits of stacking are still sometimes worth the cost.
• #generics-and-templates: Updating the design of the indexing operator interfaces with generic forms
  • Problems arise since the IndexWith use case involves implementing IndexWithPrimitive twice: once for values and once for references, but we gave up on supporting implementing two interfaces with a single impl in issue #4566 and proposal #5168. There are some options, but they all have downsides.
  • Potentially similar issues with the comparison operator interface.
• #generics-and-templates: "Identical" in the design of rewrite constraints means semantic match, not syntactic. We only use syntactic match for redeclarations, and only within a single scope.
• #generics-and-templates: Does the design for user defined sum types still make sense in light of the current generics design?
  • It uses generics in some unconventional ways; it would be better to avoid that.
  • Forms and ref returns make it possible to avoid continuation-passing, but it's not clear how to make the API safe.
  • Planning to pursue an unsafe API for 0.1, with the intent of revisiting it later as part of a broader look at adding an effect system.
• #strategy: Draft of the next Carbon Copy newsletter is out for review.
• #infra: Reminder to avoid replying to PR comments in the main "Conversation" view because it leads to notification spam.
  • But that doesn't apply if you've started a review by commenting at least once in the "Files changed" view.
  • Also don't @ people in commit messages or PR descriptions.
• #interop: Concerns about having ValueStore canonicalization ignore some fields.
  • Continued from discussion last week, now with an implementation in #5538.
• #lambdas: => can't deduce the return form because ref and val require parameters to be marked bound.
  • It may make sense to work around this with some kind of deduction, but we don't need to resolve this in #5434.
• #pointers-and-references: How does the ref callsite annotation interact with parameters with a deduced form?
  • Does it affect the form of the parameter? Neither option seems good.
  • Also, do positional parameters have a deduced form or only a deduced type?

RSS Feed for "Last Week in Carbon" here