2026-03-23 — Last Week in Carbon

[josh11b]

Updates on leads issues

• Decisions
  • #5261 We should add `ref` bindings to Carbon, paralleling reference expressions: previously resolved by p5434.
• New issues
  • #6931 Update to the syntax for class members: functions, methods, fields, etc.
  • #6932 Should we move away from `:!` syntax for generics and templates?
• Active discussions
  • (none)

Updates on proposals

• Decisions
  • (none)
• New RFC proposals
  • (none)
• Active discussions
  • #6902 Identification of a named constraint during definition
    • #generics-and-templates Do we allow conversion between T:! N & J and T:! N & K when N is partially identified?

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

• Conduct team updates:
  • none
• Moderation interventions last week:
  • Numerous GitHub issues by one user were closed after we explained to the user that their posts were off-topic
  • WIP: Outreach for new interview round of neighbor language community
  • WIP: input about open vs. appropriate communication for the moderation team
• Activity in the past week:
  • 1 new contributor! 🎉
  • 10 authors have pushed 26 commits to trunk and 27 commits to all branches.
  • On trunk, 329 files have changed and there have been 7,764 additions and 5,029 deletions
  • We merged 26 PRs, opened 11 additional PRs, closed 10 issues, and opened 2 new issues.

Activity on toolchain

• #typesystem 03-17 #toolchain #6926 Does a discarded expression need to have a complete type? When do we require types to be complete?
  • One possible rule: all non-reference expressions must have complete types.
  • Alternatively, we could e.g. allow incomplete-type value expressions in generic code, so long as we know the type will be completed in the specific.
  • First rule is simpler, but looks harder to implement from where we are now.
• #generics-and-templates Crash due to interaction between impl lookup failure and monomorphization failure.
• #toolchain Associated types are treated like associated constants by the compiler
• #toolchain Returning a bool in a flag register in LLVM
• #toolchain Concrete value expressions will eventually have a complete type
• #toolchain Debug info in lower tests
• #toolchain Bazel-native bootstrapping working
• #toolchain SemIR to include in tests
• #toolchain Line numbering in split tests
• PRs:
  • constexpr pointers in #6907
  • integers in #6908 #6909 #6918 #6921 #6924 #6949
  • named constraints in #6922
  • macro constants in #6923
  • namespace reverse interop in #6935
  • complete & incomplete types in #6926 #6939
  • forms and pattern matching in #6927 #6928 #6929 #6936
  • var default initialization in #6934
  • associated types in #6937
  • deps in #6945 #6911
  • #toolchain agent skills in #6925
  • others in #6914 #6917 #6933 #6916 #6941
• New contributors:
  • @cuiweixie made their first contribution in #6949

Activity on memory safety

• #safety How to define the IndexWith interface in order to include an alias set in the Result return type. Are place groups ever part of the form? Yes, if part of the type.
• #safety Examples of and discussion around permissive code returning bad references (to strict code). Follows up in open discussions.
• 03-16 How to call unsafe C++ from Carbon, when C++ returns a pointer or a reference. What alias set is generated for Carbon’s perspective. ^permissive as a possible option. Heuristics to try make more C++ accessible to strict carbon without unsafe keywords.
• 03-18 Goals around strict Carbon interaction with permissive Carbon, and permissive Carbon with C++. Safe and unsafe place groups.
• #safety Introduce safety unit on syntax of effects. Followed up in open discussion.
• 03-19 Discussion of syntax for how to express effects and relationships between parameter alias sets in a function signature.
• #safety Clarify intended relationship between a mutex and its guarded vars, which is not through a place group.
• #safety Introduce safety unit on unstructured threading that prevents thread races and uses mutexes to guard mutable state.
• #safety Tuesday safety discussions will be every week instead of bi-weekly.

Other active discussions

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

• #infra Should we add --output-last-input-only to the Compiler Explorer default flags?
• #evolution Blog post about using source-level inlining to clean up deprecated APIs.
  • There's a similar tool for C++.
• #infra Was closing #6892 and the issues linked from it the right response?
• #contributing-help When using jj, having trouble with bazel-execroot showing as deleted in a commit.
• #pattern-matching Feedback and corrections on pattern-matching design, including some concerns about expression patterns.
• #control-flow Is there a design for mutable iteration?
  • Core.Iterate needs to be made form-aware. With that change it should mostly just work.
  • Should we require a separate syntax, e.g. in ref vec?
• #generics-and-templates Getting errors related to circularity while changing pattern-matching.
  • Looks unrelated to the kinds of circularity we're dealing with in generics.
• #build-help Do we have a way of locally simulating the CI that runs on GitHub?
  • No, but bazel test //toolchain/... catches most issues, although some only show up with -c dbg or -c opt.
• #infra FizzBot broke. Needed to restart the host machine.
  • Currently only Googlers can do that, because it runs on Google infrastructure.
• #style Should we rely on ADL in our C++ code?
  • No. Among other things, this follows from trying to pre-adopt Carbon style in C++ code.
• #naming Safety unit 42 proposes renaming "alias set" to "place set".
• 03-19 How to implement a dynamic check that a node kind is compatible with an inst kind?
• 03-19 How to invoke placement new in generated C++?
• #syntax Interesting survey of semicolon omission techniques.
• #infra Trying to remove Clang headers from Carbon headers, but encountering various challenges.
• #editor-integrations Third-party project is planning to implement support for Carbon in IJ.

RSS Feed for "Last Week in Carbon" here