2025-04-21 — Last Week in Carbon

[josh11b]

Updates on leads issues

• Decisions
  • #5143 Do different integer literals have different types?
    • No, because that has too much overhead: more insts, more types, more specifics, more witness tables.
    • Note that this means that the constant values of expressions are now inputs to typing judgements (in addition to types and expression categories), but we're continuing to avoid dependent typing of runtime values.
    • This is largely capturing the outcome of discussions from late last year.
  • #5253 Should we change `impl fn` to `override fn`?
    • Yes. impl fn creates a parse ambiguity, and the aspiration to model overriding in terms of interface-implementation hasn't borne fruit.
    • This may make the syntax less elegant in the rare corner case of declaring a function that's abstract, but has an implementation (abstract override fn instead of abstract impl fn).
• New issues
  • #5319 extend final impl as
    • Currently we don't allow combining extend and final, but it seems to have valid use cases. Should we allow it?
• Active discussions
  • #4109 how to make a non-virtual call to a virtual function?

Updates on proposals

• Decisions
  • None
• New RFC proposals
  • None
• Active discussions
  • #3802 extend api
    • Should this be extend alias instead?
  • #5168 Forward impl declaration of an incomplete interface
    • No concerns with overall direction. Some discussion of the details.

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.
  • DONE: shared summary of insights from Swift folks in OSS projects with the moderation team to help us with community building
• Last week’s stats across our spaces:
  • In the last 7 days in our repo: 8 authors pushed 27 commits to trunk, we changed 223 files with 3,617 additions and 1,576 deletions, with 27 merged PRs, 9 open PRs, 3 closed issue and 2 new ones

Activity on toolchain

• 04-15 #toolchain #5304 #5323 Making the code less verbose when referring to the error singletons (& other singletons)
  • #toolchain The error instruction is a bit polymorphic, sometimes it is a type, sometimes not.
• 04-15 #toolchain FYI we were considering using a "redeclaration" instruction instead of another declaration instruction in #5310. Ultimately that approach was abandoned.
• 04-15 Work on incrementally building generics is getting close
• 04-15 Figuring out the new approach for impl witness instructions. Resulted in #5321, which is in review.
• 04-17 What part of the impl declaration should be included in the name mangling? Currently we don't detect duplicates, but we will.
• 04-17 Update on lowering specifics patch
• #toolchain Continued talking about importing var patterns
• #toolchain Interaction between Check::GetOrAddInst() and implicit instructions, as a result of recent LocId changes
• #toolchain FYI, destructors will soon use an interface from the prelude, affecting tests
• #toolchain #5322 What should find_if return? Made FindOrNull and FindIfOrNull that returns a pointer to the found element or nullptr, but there are some concerns about UB if the pointer is dereferenced without checking if it is null first.
• #toolchain Determining whether you are in a function scope, and not a (for example class) scope nested within a function scope
• #toolchain Recovering from a function declared with a package.... name in a non-package scope.
• #toolchain first increment of runtimes building PR out: #5338
• PRs:
  • associated constants in #5295, #5303
  • impl witness in #5305
  • calls and no-op calls in #5306 #5302 #5318
  • 🎉 virtual call dispatch in #5308
  • running lldb from the command line in #5324
  • facets in #5325
  • generics in #5326 #5332
  • LLVM and freeing in #5333 #5334
  • others #5297 #5299 #5301 #5312 #5311 #5317 #5316 #5327 #5329 #5330 #5331

Other active discussions

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

• #naming: Should it be spelled "noop"/Noop/noop or "no-op"/NoOp/no_op?
  • "nop" is also common in assembly.
  • Consensus favors "no-op". Applied in #5306.
• #contributing-help: Not all of our GitHub docs follow the 80-character limit.
  • The limit only applies to code that's under source control, not to things like GitHub issues.
  • Weirdly, line breaks in GitHub issues are visible on the rendered web page, so probably better to avoid them when they're not structurally needed.
• #interop: New design doc on interop for primitive types, based on discussion in #5263.
  • Actively seeking feedback.
• #infra: Getting a weird test failure with no backtrace, that doesn't reproduce when run manually.
  • Failed reproduction was because the tests need TEST_TMPDIR to be set.
  • Underlying problem turned out to be a stack overflow.
• #infra: Discussion of configuring the LLDB source mapping.
• #infra: Updated LLVM in #5333, so the first build after syncing may be slow.
• 04-16, 04-17: Discussion of changing the meaning of extend in an interface (prompted by #5168)
  • We probably want to support both meanings, but the question is what syntax to use for which.
  • Right now extend B in interface A requires types that implement A to also (separately) implement B.
  • Alternatively, it could add methods to A with the same names/signatures as the methods of B, so types that implement A have to implement the methods of B as part of the implementation of A.
  • This would implicitly generate a final impl forall [T:! B] T as A, forwarding the methods of A to the methods of B
  • That final impl could conflict with other impls under our current rules, so we're considering changing the rules:
    • Instead of rejecting impls that overlap, we only reject queries that overlap (which could be resolved by manually prioritizing, e.g. with match_first).
    • Not yet clear how we can definition-check in the presence of overlapping impls, but we think we can make progress.
• Discussion topic
  • Any interesting points

RSS Feed for "Last Week in Carbon" here