2026-02-02— Last Week in Carbon #6683

josh11b · 2026-02-03T00:34:25Z

josh11b
Feb 3, 2026
Maintainer

Decisions
- (none)
New issues
- (none)
Active discussions
- #5916 C++ interop: Mapping Carbon struct literals to a C++ type
  - Previous discussion concluded that Carbon struct literals should map to C++ list initialization.
  - But probably only when they perform aggregate initialization.

Decisions
- (none)
New RFC proposals
- #6668 C++ interop type mapping for integer and floating-point literals
- #6676 Carbon/C++ Interop: Importing C/C++ object-like macros
Active discussions
- #6641 Clang IRGen in Carbon
  - Proposal should probably focus narrowly on the high-level directional decision and move the details to toolchain/docs or defer them to a follow-up.
- #5990 Principle: The signature is the contract
  - Could we coin a term like "pact" to avoid confusion with other meanings of "contract"?

Conduct team update:
- Transparency report for Q4 2025 in the making
Moderation interventions last week:
- A Discord member was banned after repeatedly posting a generic job search
Activity in the past week:
- 9 authors have pushed 15 commits to trunk
- On trunk, 90 files have changed and there have been 16,407 additions and 3,151 deletions
- We merged 15 PRs, opened 7 additional PRs, closed 3 issues, and opened 3 new issues

#safety How do alias set parameters for class types work (continued from last week)?
01-27 01-29 Examples for working out const and alias set syntax
01-27 Invalidation
01-29 Should we require public methods to only return public alias sets?
01-29 Carbon's str has an alias set parameter, but is the same as C++ std::string_view which doesn't.
01-29 Is C++ adding any annotations that would allow us to understand disjointness?
01-30 Rust interop: can we express the contract Rust expects even though Carbon has a different approach. Rust: exclusivity, lifetimes, deep immutability. Carbon: disjointness, invalidation, shallow immutability
- Modelling a bitvector that has a find_unused method that returns an UnusedId instance that prevents modification of the bitvector until caller decides whether to use the bit (link to Rust example).
#safety Updated Safety Unit 32 based on 01-29 discussion, moving some content into new unit 33. Smaller updates to 31.

#generics-and-templates Inner Self in interface generics (continued from last week)
#infra FYI if you have edit access to Google Docs, collapsing a section heading is visible to all readers.
#syntax Struct syntaxes (continued from last week)
- Can we support defaulting individual struct elements in a function parameter?
#typesystem How to expose C++ conversions to reference types in Carbon (continued from last week)
#syntax Safety unit on alias set argument syntax
#generics-and-templates Completeness of associated constant types.
#proposals-prs-and-process Best practices for sending/receiving notifications during code review.
- Several people have had bad experiences with the "new experience" for GitHub reviews, but it's difficult to turn off.
#infra Some people are seeing Bazel re-download (and sometimes rebuild) dependencies more often than should be necessary.
#generics-and-templates How to fix a bug caused by accessing an unresolved specific?
- The problem is accessing a non-canonical specific. Possibly we should track the canonical counterpart of each non-canonical specific, and use that instead.
#contributing-help We don't have any good first issues marked in GitHub right now. We should make an effort to populate that.
#interop How do we represent C++ operator overloads in the toolchain? (continued from last week)
#interop Interop now exposes initialization from a std::initializer_list, e.g.

import Cpp library "<vector>"
fn Run() -> i32 {
  var v: Cpp.std.vector(i32) = (1, 2, 3);
  return v.size() as i32;
}

01-29 Should str and slices have the same defaults with respect to mutability?
- Is it OK for mutability and type to be separate parameters, e.g. slice(Foo, const), or do they need to be packaged together?
- Consensus seemed to lean toward packaging them together, probably as a typed alias set.
#syntax Do we have an approved syntax for slice types?
- Not yet, but we've mostly been writing slice(T) in examples.
#initialization How do we want var x: T; with no initializer to behave, particularly when T is a C++ type?
#generics-and-templates Confusing behavior with ImportRefs and local constants when importing generics.
#generics-and-templates Difficulties with getting rid of FunctionTypeWithSelf.