2025-12-01— Last Week in Carbon #6450

josh11b · 2025-12-01T23:25:29Z

josh11b
Dec 1, 2025
Maintainer

Decisions
- (none)
New issues
- (none)
Active discussions
- #2152 What should we call integer types with mod-2^N arithmetic?
  - Further discussion about what semantics these types should have

Decisions
- #6395 Type completeness in extend
  - Approved
- #6331 Add import Cpp; for C++ built-in entities
  - Closed; superseded by #6358.
New RFC proposals
- #6188 Principle: Flow checking
  - Carbon can have flow-sensitive checking, but it's separate from type checking (and comes after it)
Active discussions
- #5545 Expression form basics
  - 11-24 Could we say that all ephemeral references refer to complete objects?
    - No: if x is an ephemeral reference and not trivially destructible, x.f is an ephemeral reference but we can't implicitly destructure to make it complete.

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)
- DONE: Shared overview of updated moderation team goals
- DONE: Shared input about how to keep it as simple as possible with the moderation team
- WIP: Prepare more resources to help close our skill gaps as identified by the team itself
Activity in the past week:
- In the last 7 days in our repo: 8 authors pushed 18 commits to trunk, we changed 295 files with 15,972 additions and 9,346 deletions, which means we merged 18 PRs, opened 9 PRs, we neither closed nor opened any issue.

#toolchain Question about symbolic bindings that came up in #5982.
#toolchain Imported generics omitted from SemIR dumps
#toolchain Building libc++ with clang modules is slower
#toolchain Question about using GetLocalTypeInstId.
11-25 Should we have end-to-end/integration testing for Carbon?
- This would be costly, because MacOS test runners take a substantial fraction of a second to start a new process.
- Clang doesn't, because build host may not be a valid build target, but it gets some of that "for free" because of bootstrapping.
  - Those considerations may not apply to Carbon.
  - Zig takes an interesting approach: bootstraps using a snapshot of a web-assembly build, which is highly portable.
11-25 More trouble with building runtimes: some tests are 5x-10x slower under ASAN, and libc++ build is slow enough to cause other tests to time out.
- Could address this by making it a build artifact instead of a runtime test action, but it would be tricky.
- Trying to build with modules to see if that helps.
11-25 Unnamed lambda entities could all use a single special name id that means "unnamed lambda/function"
PRs:
- C++ macros in #6418 #6419 #6426 #6432 #6431 #6440
- import in #6415 #6428
- C++ operator overload resolution in #6416
- final impl in #6236
- benchmark test iterations in #6433 #6436
- Carbon Optional(T*) -> C++ T* parameter in #6422
- debug info in #6410
- C++ integer types in #6386 #6442
- libc++ and libc++abi runtimes in #6424
- update dep in #6423

#safety Safety Unit No. 23: On interaction of Types and Flow Checking
11-24 Should we use different types for invalid pointers like we do with unformed state?
11-24 An erased type, like an erased call type, may need to be associated with an alias set to represent captures by the type that was erased.
- Issue was investigated in Safety Unit No. 11: type erasure
11-24 Unlike for interfaces used as constraints, we are considering having defaults and writing negative effects when the default is wrong for conciseness.

#strategy U.S. holidays on 11-27, 12-25, and 01-01 are all on Thursday this year. Open discussions on those days are cancelled because we won't have quorum.
#generics-and-templates Running into bugs when importing the facet type of a require decl.
#generics-and-templates Code that works in a single file has a facet conversion failure when split into multiple files.
- The split arguably violates the orphan rule.
- That implies things like impl forall [T:! MyComparable] T as Core.Comparable can't satisfy the orphan rule. It would be nice to make that less strict.
#last-week-in-carbon Should the last person to contribute updates be responsible for publishing?
#typesystem how to convert a field to a MaybeUnformed reference?
- self.field as MaybeUnformed(Field).
- Casting to compatible types preserves category, and self.field is a reference expression (assuming the method is mutable).
- If the destination isn't const, should this require unsafe somewhere, because it allows you to change the field to an unformed state.
  - But is that different from ordinary write access?
#getting-started Does Carbon's backend use Sea-of-Nodes?
- No, it uses LLVM.
#generics-and-templates Got pretty far with making LookupImplWitness take a TypeInstId for its self, but ran into problems.
- Tracking the facet value as the self may be a better approach.
#infra FizzBot now has a /wake command that regenerates your alerts, and only yours.