2026-02-09— Last Week in Carbon

[josh11b]

News

• Interop with a C++ std::vector has improved

Updates on leads issues

• Decisions
  • #6456 Order of operations when initializing from a compound literal
    • Prefer depth-first and as element-wise as possible.
  • #6160 How should array initializers be constrained?
    • Implicit conversions don't take destination form as input.
    • Arrays can be initialized from tuples of compatible types, regardless of form.
• New issues
  • (none)
• Active discussions
  • (none)

Updates on proposals

• Decisions
  • #6668 C++ interop type mapping for integer and floating-point literals
    • Approved.
• New RFC proposals
  • #6699 Diagnostic sorting
    • Primary order is last-token-processed, but diagnostics can opt into a secondary sort by start position
• Active discussions
  • #5389 Generic across forms
    • #typesystem The category modifier in form literals probably shouldn't be optional.
  • #5545 Expression form basics
    • 02-05 Clarified leads guidance on how to explain "initializing results".
  • #6641 Clang IRGen in Carbon
    • 02-03 How much detail belongs in a toolchain proposal?

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

• Conduct team update:
  • Transparency report for Q4 2025 in the making
• Moderation interventions last week:
  • none
• Activity in the past week:
  • 2 new contributors! 🎉
  • 7 authors have pushed 17 commits to trunk.
  • On trunk, 261 files have changed and there have been 8,217 additions and 7,992 deletions
  • We merged 16 PRs, opened 6 additional PRs, closed 2 issues, and opened 3 new issues.

Activity on toolchain

• #toolchain 02-03 How to sort toolchain diagnostics.
• 02-05 When should we call ResolveSpecificDefinition?
• #toolchain Ordering tuple_access insts
• #toolchain Order of operations from conversions changed by pending proposal #5545 reflecting decision in issue #6456
• #toolchain SemIR generated as a side effect of conversion
• #toolchain CHECKs helping debug problems
• #toolchain Producing SemIR that doesn't satisfy invariants from invalid code, producing CHECK failures
• #toolchain Working around bug #6677
• #toolchain Args handling in golden file tests
  • probably best to support a VLOG control directly since that is the most common use case
• #toolchain Parse recovery is documented
• #toolchain Rename types().GetInstId to GetTypeInstId?
• PRs:
  • guaranteed-in-place initializing expressions in #6623
  • diagnostic sorting in #6687
  • forms in #6695
  • build in #6681
  • docs in #6685 #6689
  • infra in #6693 #6700
  • others in #6684 #6686 #6688 #6692 #6690 #6696 #6702 #6704
• New contributors:
  • @justinhorvitz made their first contribution in #6681
  • @nicholasbishop made their first contribution in #6685

Activity on memory safety

• 02-03 02-04 02-05 Always const, preserve const, and mutable syntax
  • Came to a resolution
• 02-05 Follow up to discussion on 01-30
  • Rust example from 01-30 can't be done in currently planned Carbon safety model
  • Will be important to be able to describe the differences between the models
  • Will need some way to bridge to Rust since Rust has the memory safe library ecosystem we'd like to use.

Other active discussions

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

• 02-02a Syntax for const, mutability, and alias sets.
  • Part of the problem is that different contexts want different defaults.
• 02-02b
  • str has an alias set parameter, but std::string_view doesn't. How do we deal with that?
  • Syntax of constexpr/consteval and compile-time/generic parameters more generally
• #syntax Where do we want self to go, particularly given that we probably want to make : Self optional?
  • Seems tied to what semantic model we want for bound methods.
  • May also be tied to the question of who decides whether a function can be called as x.F(y) or F(x, y).
• #interop The Tuesday interop meetings are being folded into the Tuesday toolchain meetings.
• #getting-started How are things going on Carbon? Haven't heard much lately.
  • Added some recent and upcoming talks to README.md.
  • For updates you can follow Last Week in Carbon and the Carbon Copy newsletter here.
• #general-design Can you bind a name to a method like let method: auto = c.Print?
  • Yes, but it's not implemented yet.
• #getting-started Hitting link errors when running the toolchain.
  • Need to use -- to separate Bazel args from Carbon args.
  • Carbon currently doesn't expand ~ in paths, but we should fix that.
• #getting-started How to print a string in Carbon?
  • I/O library is here, but note that right now you have to link the prelude explicitly.
• #typesystem Can we relax the restrictions on what :! can bind to?
  • At least for class parameters, we probably want to allow any phase, but the phase of C(n) is the phase of n.
• #generics-and-templates Interface-with-self is working.
• #infra Is there a good way to index into SmallVector?
  • Possibly in gdb, or lldb on MacOS.
• #syntax Making good progress on safety syntax. See here.
• #general-design VoidBase should probably be moved out of Core.CppCompat since it's the base of all types.
  • And possibly rename it to something like Any?
• #infra Should we use BUILD.bazel instead of BUILD for consistency with LLVM?
• #generics-and-templates What does virtual eval fn mean?
  • Same as virtual constexpr in C++.
• #generics-and-templates A require impls decl needs to have its own generic, with no additional bindings, in order to separate monomorphization errors from the interface containing it.
  • But then noted this currently breaks down for extend require impls.
• #general-design Why does array take a signed size parameter?
  • We only use unsigned when we specifically want modular arithmetic.
• 02-05
  • Continuing compile-time syntax discussion from 02-02b
    • Considering dropping :! and using prefix modifiers to choose the phase (like we already do with template).
    • We probably want the default to be contextual.
  • Variance of alias set parameters on types
• #contributing-help Clarifying int.convert vs int.convert_checked builtins.
• #naming What do we call the non-owning array type (what C++ calls a "span" and Rust calls a "slice")?
  • The Rust [T] syntax may cause ambiguities with indexing.
• #naming Could we use the terms "class method" and "instance method" to avoid C++'s overloading of "static"?
  • We already use "class function" and "method" for this distinction.
• #proposals-prs-and-process Are we wasting the ' token on character literals when we could just use single-character " literals?
  • Character literals seem to be surprisingly common in C++ code.
  • Using ' in an unbalanced way confuses naive syntax highlighting.
• #getting-started Has anyone gotten Zed working with Carbon?

RSS Feed for "Last Week in Carbon" here