2025-04-07 — Last Week in Carbon

[josh11b]

Introduction

Welcome to the new Last Week in Carbon posts!

This is a new format we're using to summarize activity across the Carbon project in the last week. You can find all the detail in the proposal that introduced our new structure: #5233.

Updates on leads issues

• Decisions
  • (none)
• New issues
  • #5253: Should we change impl fn to override fn?
  • #5251 impl declarations in a generic class context
  • #5250: How do var patterns interact with binding patterns?
    • A question from proposal #5164: Updates to pattern matching for objects
    • Discussion in #pattern-matching
      • The central question is whether var (x: X, y: Y) declares two separate objects, or references to two members of a single object.
      • If they reference members of a single object, they can't be moved.
      • Either option involves some degree of context-sensitivity, which we try to avoid.
      • There are non-context-sensitive options, but they're verbose or surprising.
      • If we have reference bindings at all, we probably want this syntax rather than (or in addition to) an explicit opt-in.
      • Regardless of the choice here, we'll probably want a separate expression category for things that can be moved.
  • #5241: var of type FacetType
• Active discussions
  • #4109: how to make a non-virtual call to a virtual function?
  • #4711: Syntax for implementing a single-function interfaces
  • #5143: Do different integer literals have different types?

Updates on proposals

• Decisions
  • #5233: Towards more async "sync"-ing
• New RFC proposals
  • (none)
• Active discussions
  • #5168: Forward impl declaration of an incomplete interface
    • Partially implemented in toolchain PR #5219

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

• 10 trained moderation & conduct team members + community lead,
  in theory available from UTC-7 to UTC+5.5.
  • WIP: reaching out to other communities we’d like to learn from in terms of journey and interview some of their leaders, and started summarizing first input (currently: Swift folks in OSS projects)
  • Pending: Transparency report for Q1 of 2025
• Last week’s stats across our spaces:
  • In the last 7 days in our repo: 10 authors pushed 26 commits to all branches, 25 of which to trunk, we changed 387 files with 9,066 additions and 4,680 deletions, with 25 merged PRs, 6 open PRs, 1 closed issue and 5 new ones

Activity on toolchain

• #generics-and-templates 03-31 04-01 04-03 changing generic evaluation to be incremental so we can support name lookup into a generic entity while it is being defined
  • Concerns about increased allocation of specifics and code complexity, but we don't see another path forward and we have been finding fixes for the problems
• #generics-and-templates Progress on using associated constants from final impls
• #generics-and-templates 03-31 04-01 04-03 solving the problems with using absolute inst blocks in impl witnesses
  • Fixing limitation that associated constants can only be given concrete values
• #generics-and-templates Would it help to have another person on generics, or have we already saturated the available parallel work?
  • Currently most major work is blocked by two big tasks currently in progress, but there may be more available parallelism soon.
• #toolchain Fuzzer found a crash
• #toolchain Shared doc about linkage for discussion
• #toolchain How to write a test that two type expressions are equal
• #toolchain Idea: encoding the expected instruction kind of the constant value in the type of an InstId.
• #toolchain WIP code has virtual dispatch working. Example:

$ cat a.carbon
import Core library "io";
base class Base {
  virtual fn F[self: Self]() {
    Core.Print(1);
  }
}
class Derived {
  extend base: Base;
  impl fn F[self: Self]() {
    Core.Print(2);
  }
}

fn Do(d: Base) {
  d.F();
}

fn Run() {
  var d: Derived = {.base = {}};
  Do(d);
}
$ ./bazel-bin/toolchain/install/prefix_root/lib/carbon/carbon-busybox compile a.carbon
$ ./bazel-bin/toolchain/install/prefix_root/lib/carbon/carbon-busybox link a.o --output=a.out
$ ./a.out
2

• #build-help: Tests failing on GitHub but not locally, because test runners have an old version of LLVM.
• 04-03 When are two specifics for the same generic going to produce the same lowering?
• 04-03 Constant values of instructions produced by "subst"
• 04-03 Do we want to annotate SemIR with stringified versions of constant values in general?
• PRs:
  • diagnostic improvements in #5221 #5234 #5236 #5235
  • to Carbon's C++ style in #5207 #5229 #5230 #5228 #5227
  • generics and specifics in #5222 #5215 #5232
  • facets and facet types in #5220 #5219 #5226 #5242 #5252
  • lowering class initialization in #5199
  • vtables in #5231 #5244
  • import in #5217
  • lowering generic functions in #5090
  • refactor in #5240 #5249
  • fix crashes in #5243 #5237
  • tidy in #5239 #5248
• New contributors:
  • @tkoeppe made their first contribution in #5248

Other active discussions

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

• #generics-and-templates 03-31 Discussion about recent posts about dynamic ("dyn") types and their interaction with generics in Rust (1, 2)
  • Covers implicit bounds, and issues with the design of Dyn
  • Carbon can learn from that experience, but will take a slightly more explicit approach than Rust has to match its philosophy
• 04-02 Decided we should not do deduction through facet types.
• #explorer: What should we do with the Explorer codebase?
  • p3532 stopped active development of Explorer.
  • Keeping it in our main repo is producing ongoing maintenance costs and user confusion.
  • It's still sometimes useful to have as a reference.
  • Moving it to a separate repo/branch/tag could minimize the costs while preserving most of the benefits.
  • If we ever did restart active development, we'd have to do more work to bring it up to date.
• #strategy: Top comment on a HN discussion mentions Carbon's compiler architecture as a positive example
• #language-questions: How will bidirectional interop work?
• #variadics: Discussion of how to refer to all but the first N parameters of a function with positional parameters, for "at least N" type use cases.
• #general-design: It would be nice to have my_tuple.last.
  • Or maybe my_tuple.from_end(0)?
• #infra: We're getting weird messages in our repository due to questionable enterprise settings. We're following up to try to address.
• #syntax: Draft proposals for compile-time entities using ‽ and ¡syntaxes, and for keyword abbreviations.
  • Note that this occurred on April 1.

RSS Feed for "Last Week in Carbon" here