2026-04-28 — Last Week in Carbon

[geoffromer]

Updates on leads issues

• Decisions
  • #6026: Why do we use .Self instead of the binding name?
    • We have addressed the cycles in the toolchain implementation, but we could still make this change if desired.
• New issues
  • (none)
• Active discussions
  • (none)

Updates on proposals

• Decisions
  • (none)
• New RFC proposals
  • (none)
• Active discussions
  • #7016 Updating self syntax and adding static member variables
    • Iterating in review
    • 04-23 Restricting to "static" being about storage, not per-instance

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

• Conduct team updates:
  • DONE: fixed transparent report list display issue on GitHub
• Activity in the past week:
  • DONE: shared another self-learning quiz for the moderation team, to help foster a more diverse and welcoming community
  • WIP: discussing if/how we want such quizzes to be shared with the rest of the community
  • WIP: interviewing TypeScript community leaders to learn from their community journey, with 2nd interview done; pending scheduling more interviews
  • 8 authors have pushed 38 commits to trunk and 40 commits to all branches.
  • On trunk, 319 files have changed and there have been 22,456 additions and 11,850 deletions
  • We merged 38 PRs, opened 8 additional PRs, closed 1 issue, and opened 1 new issue.

Activity on toolchain

• 04-21 Replacing type type -> the empty facet type
• #toolchain How to represent form generics in textual SemIR
• #toolchain Should performing an action be considered a kind of desugaring?
• 04-21 #toolchain errors and diagnostics
• #toolchain Can we improve file_test performance in debug mode?
• #interop What do we want to migrate C++ output streaming to?
• #toolchain How to understand a bug related to attachment and getting an InstId from a TypeId?
  • Only insts in a generic region can have attachment.
  • You can't navigate to such an inst from a TypeId or ConstantId (including via their canonical insts).
  • You also can't navigate to such an inst from an unattached inst, except via an AbsoluteInstId.
  • In principle we could express these differences in the type system, but that may be impractical.
  • It's probably feasible to decouple attachment from ConstantId, at least.
• #toolchain Restructuring ValueStore to improve build performance seems to work, and reduces object file size by 40%.
• #interop How should C++ interop thunks handle by-value C++ params with movable types?
  • For now, provide separate by-move and by-copy overloads.
• #toolchain We need to record the Carbon class -> clang decl mapping somehow.
  • Maybe we should model clang_decls as a bidirectional map.
• PRs:
  • build and bootstrapping in #7084 #7083 #7088 #7095 #7087 #7102 #7080 #7086
  • require constraints in #7081
  • interop/cpp tests in #7085
  • C++ calling Carbon in #7078 #7107
  • exported thunks in #7092
  • return declaration in #7076
  • imported C++ types in #7106 #7094 #7112
  • locations in imported C++ code in #7093
  • named constraints and where in #7104
  • .Self in #7097 #7105
  • update deps in #6951 #7103 #7109 #7110 #7111
  • dependent PRs in #7096 #7101
  • others in #7079 #7089 #7090 #7099 #7091 #7118 #7098 #7108

Activity on memory safety

• 04-20 Want a brief version of the thread safety story
• 04-20 Mutex re-entry
• 04-20 Syntax for safety effects
• #safety Can you modify a ref in a function argument?
• #safety Updated Unit 43 based on feedback and discussion.
• 04-21 TryAcquireLock forces us to use effects for locking & unlocking, if we are going to support a C++-like annotation model
  • #safety This means we need conditional returns. How do those relate to sum types?
• 04-21 Invalidation for arguments to types
  • Can have two objects of the same type but different validity
• #safety How do we model the PT_GUARDED_BY thread-safety attribute?

Other active discussions

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

• 04-20 04-23 Newsletter examples, including mythical crosses and reverse interop
• #generics-and-templates How to deal with attached constants in substitution?
  • Don't think of attachment as part of the constant, it's just extra info that's automatically packed into the ConstantId.
  • Conceptually, each InstId has a ConstantId, and if it's symbolic it also has a GenericId, GenericInstIndex, and ConstantDependence.
• #infra Releases were broken. Fixed in #7080.
• #infra What to do when your auto-assigned reviewer is on vacation?
• #infra Accidentally merged a dependent PR before its dependency.
  • We should have automation to prevent that.
  • Implemented in #7101
• #generics-and-templates Hit a crash in impl lookup. Fuzzers didn't find it.
  • The fuzzer corpus is years old. We should refresh it.
• #syntax Is the documentation for destructors still accurate? It doesn't match the implementation.
  • It's out of date, but the new design isn't fully implemented either.
• #infra False-positive -Wunused-function warning from clangd (and clangd-tidy)
  • clangd is inaccurate for "unused" warnings. We disable others; we should disable this one too.
  • We catch these errors at build time instead.
• #infra Merging #7103 reverted #7106 due to a temporary GitHub bug (!!)
• #infra resolved fields in package-lock.json were reverted #7109.
  • That was intentional, to work around local URLs not matching them.
• #syntax Why can I have a field named a1 or f0 but not f1?
  • That's a floating-point type literal. We should have better diagnostics for that (including suggesting r#f1 instead).
  • We should probably also reserve f0 to avoid confusion.
  • Filed as issue #7120.
• #infra Do we intend to eventually ship with CMake, or just be compatible with it?

RSS Feed for "Last Week in Carbon" here