2025-11-03— Last Week in Carbon

[josh11b]

News

• Alina Sbirlea gave a quick talk at the 2025 LLVM Developers' Meeting about her work on merging equivalent instantiations of generic functions in Carbon: Optimizing generic code lowering to LLVM-IR through function equivalence coalescing (slides)

Updates on leads issues

• Decisions
  • (none)
• New issues
  • #6285: Can Self appear in the where expression of require?
    • 10-30: concluded "yes", with the understanding that require declarations produce a generic in an interface or constraint.
  • #6291: C++ Interop: How to map reference types (T&) vs. pointer types (T*)?
• Active discussions
  • #6275: C++ Interop: How to model distinct C++ integer types (long, long long) and their operations?
  • #6166: C++ Interop: How should C++ operators interact with Carbon operator interfaces?
    • Support for "option 4," the most complex and capable option, with a modification to allow Carbon impls to hide ambiguous C++ cases

Updates on proposals

• Decisions
  • (none)
• New RFC proposals
  • (none)
• Active discussions
  • #5913: Reworking unformed state
    • Discussion about alternatives to using unsafe as T to get a reference to an uninitialized variable of type T bypassing MaybeUnformed.

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

• Moderation and conduct teams are still the same size, available in theory from UTC-7 until UTC-5.5: 10 trained members altogether + community lead (hands-off trainer, coach, guide)
  • WIP: Discussion of insights from interviews of C# community leaders about C#’s community journey among the moderation team
  • PENDING: Providing resources to help close our skill gaps as identified by the team itself
• Activity in the past week:
  • Spam on Discord was reported, deleted and the spamming account was banned
  • In the last 7 days in our repo: 7 authors pushed 28 commits to trunk, we changed 577 files with 37,938 additions and 28,297 deletions, which means we merged 28 PRs, opened 6 PRs, we didn’t close any issue and opened 4 new ones.

Activity on toolchain

• #toolchain ref in argument as instruction
• #toolchain Require instructions and the interaction with specifics
• #toolchain #6290 Deleting ImportCpp until we have a use
• #toolchain Conversion of {} to a facet
• #toolchain Difference between identifying and resolving/completing a facet type
• PRs:
  • unit tagging in #6272 #6271 #6288 #6298
  • thunks in #6276 #6277
  • C++ interop in #6268 #6234 #6279 #6284 #6296 #6290
  • ref in #6283
  • #toolchain singleton types in #6289 #6301 #6300
  • generics in #6294 #6292 #6297
  • require declarations in #6286
  • named constraints in #6304 #6305 #6308
  • others in #6281 #6287 #6295 #6299

Activity on C++ interop

• #interop How to handle `MaybeUnformed`.
• #interop State of nullable and const pointers support.

Activity on memory safety

• #safety Safety Unit No. 16: Paths into sum types
  • Proposes a field-like rather than type-like syntax for enums, so that we can concisely express paths into them.
  • Anonymous enums can restrict our ability to generate efficient representations, but they don't have to be anonymous in that sense.
• #safety Separation logic could be interesting for proving safety.
  • Original paper is a good read. This citation is also interesting.
• #safety The Scala community is working on problems very similar to Safety unit no. 11: interface effects.
  • A bunch of papers were published this month (links in the Discord discussion)
  • Scala capture types seem close to our alias sets.
• #safety Interesting and relevant new paper Place and Capability Graphs in OOPSLA '25.
• #safety "External uniqueness is unique enough" by Dave Clarke is also interesting.
  • Would we invalidate external pointers into Box(T) on move?
  • Some C++ use cases rely on moving unique_ptr being non-invalidating.
  • Maybe for non-returning functions, no annotations means you get the effects that an empty body would have.
    • If so, moving Box(T) into a function invalidates by default. If it's captured into the return, that would need an explicit annotation.
    • But explicitly annotating that it's not freed could be awkward.
  • Should we allow taking multiple references to a Box(T)?
    • Function arguments are presumed to be allowed to alias unless they say otherwise.
    • That might be the wrong default for new code, though.
• #safety Safety Unit No. 18: an object graph model
  • Relates to previous discussions of how to make a partially-invalid struct valid again.
  • This is about invalidated rather than uninitialized pointers, but it would be nice if we could reuse MaybeUnformed.

Other active discussions

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

• #pattern-matching Continued discussion of Ante is as an alternative to if let.
  • Decouples pattern matching from control flow, but at the cost of more complex lexical scopes (including within expressions)
• #generics-and-templates Considering removing .Self from the language, and instead explicitly use the name that's being declared.
  • We do still need Self, but it's symbolic, unlike .Self which is concrete.
• #typesystem #syntax 10-30 How to handle disambiguation of the two different meanings of ref in expressions.
  • ref in an argument list marks an argument to a ref parameter, but ref in a return type expression marks a by-reference return form.
  • In a struct literal, : vs. = may give a conflicting signal about which meaning of ref is intended. That doesn't affect correct code, but affects where and how we diagnose.
  • In theory something like fn I() -> (ref a, ref b) * 3; might be valid someday, with ref having the argument meaning, so we probably shouldn't try to disambiguate within a return type during parsing.
  • However, we can probably disambiguate outside a return type during parsing.
  • Another option would be to remove the ambiguity, and use something like out or mut for the argument meaning.
• #generics-and-templates Does this require A to be identified or complete?

interface A;
constraint B {
  require impls A where .X = {};
}

• Accessing a facet type member requires it to be complete.
• #contributing-help How to inspect a Map or Set in debugger.
  • No good way yet.
• #contributing-help Implementation strategies for string indexing

RSS Feed for "Last Week in Carbon" here