2025-05-05 — Last Week in Carbon

[josh11b]

Updates on leads issues

• Decisions
  • #4579 When are interface requirements enforced for an impl?
  • #4672 Declaration and definition of impls and their associated functions
  • #4566 Implementing multiple interfaces with a single impl definition
    • All three resolved by proposal #5168
• New issues
  • #5413 Does struct field order matter for impl lookup and type structure?
    • #generics-and-templates Are {.a: X, .b: Y} and {.b: Y, .a: X} considered the same?
• Active discussions
  • #4711 Syntax for implementing a single-function interfaces
    • A bunch of questions asked, discussed 04-30, and answered
  • #5261 We should add ref bindings to to Carbon, paralleling reference expressions
    • 05-01 Rationale for change
    • 05-01 Initial experiment will be with noalias and nocapture LLVM semantics
    • 05-01 ref and value returns using bound parameters
    • 05-01 Concerns with interaction with effects like optional, errors, ...
  • #5367 impl in class redeclaration syntax with parameterization
    • Second option more consistent with previous discussion on 04-21 and #3763

Updates on proposals

• Decisions
  • #5168 Forward impl declaration of an incomplete interface
    • Approved after one last edit about generic and compile-time functions
• New RFC proposals
  • #5366 The name of an impl in class scope
    • Implements decision in leads issue #5367
    • Initial support for accepting
• Active discussions
  • #5337 Interface extension and final impl update
    • Another round of review
    • #generics-and-templates New concern about being able to concatenate files without significant changes leading to a reconsideration of how to prioritize between overlapping final impls

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 some suggestions for the moderation team to prepare for growing our community in the near future
• Last week’s stats across our spaces:
  • In the last 7 days in our repo: 9 authors pushed 34 commits to trunk, we changed 713 files with 23,222 additions and 14,531 deletions, with 35 merged PRs, 14 open PRs, 4 closed issues and 6 new ones.
  • One new contributor!

Activity on toolchain

• 04-28 Poisoning impl lookups that got a concrete result is resulting in extra SemIR instructions
• 04-28 Trying to get good source locations from impl witnesses for diagnostics
• 04-28 Next tasks related to impl lookup and final
• 04-28 Would like to have a helper function like ResolveAs that works for the cases where ResolveAs doesn't
• 04-29 Working on thunk support for impls, seems to be working.
  • Easy rather than good diagnostics for now
  • Possible rule about return types: Thunk always has an explicit return type, and it always returns a call to the impl function.
• 04-29 Impl overlap analysis
• 04-29 Diagnostics should include the query for poisoning PR
• 04-29 GitHub giving errors on some download requests, causing trouble for our builds
• #toolchain (continued): Discussion of changing poisoned names to point to an inst instead of a location.
  • Some concerns about how that will affect diagnostics
• #toolchain: Reminder to remove the corresponding NOLINT comments when disabling a clang-tidy check.
• #toolchain: Discussion about thresholds for when to factor out a separate file.
  • BUILD visibility could be a useful tool when we want to factor out a component to keep files small but don't want to support it as a public API.
• #toolchain: autoupdate sometimes puts note diagnostics in the wrong place.
  • Right now autoupdate just looks at line numbers. We could maybe extend it to be smarter about clustering.
• #toolchain: What do we want templates that are constrained on inst category to look like?
  • The style guide says we should use requires rather than embedding the constraint in the parameter type.
• #toolchain: Should Vtable be an EntityWithParamsBase?
  • EntityWithParamsBase is for things that are explicitly declared, with names.
• #toolchain: Missing test coverage for the three-phase case of TryResolveTypedInst for AssociatedConstantDecl.
• #toolchain: GetSubtreeTokenRange is an O(N) walk of the node's children, and we're doing it per-instruction for debug info, leading to O(M*N) behavior.
  • We wanted debug info to be cheap enough for it to be on by default.
  • Not clear how we can fix this without substantial memory overhead.
• #toolchain: Discussion of whether to sort typed_insts.h (see #5401).
• #toolchain: Fuzzer found a crash in local generic let bindings (not implemented yet, but still shouldn't crash)
  • Deduction shouldn't be looking at types that have already been substituted away.
• #toolchain: Subtle bug in the interaction between LocId, importing, and thunks. Fix in #5418
• #generics-and-templates: What goes in the constants block, and interaction with imports
• #generics-and-templates: What is included in type structure.
• PRs:
  • documentation in #5300, #5369 #5382
  • clang-tidy in #5368 #5370
  • patterns in #5360 #5361 #5385 #5410
  • LocId in #5355 #5371 #5376
  • SemIR formatter in #5372 #5375
  • dump-sem-ir ranges in #5378 #5379 #5394 #5399 #5408
  • impl lookup poison in #5373
  • impl witnesses in #5374 #5409
  • incremental generic eval blocks in #5313
  • impl function thunks in #5390 #5404
  • build infra in #5386 #5407 #5415
  • others in #5364 #5387 #5388 #5395 #5396 #5391 #5414
• New contributors:
  • @IvanDuran02 made their first contribution in #5300: Document lambdas

Other active discussions

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

• #infra: Seeing issues with downloads failing in GitHub actions.
  • #5386 is a mitigation. GitHub seems to have fixed the underlying issue.
• #infra: file_test failures being attributed to the wrong test. #5387 is the fix.
• #strategy: Any public Carbon talks coming up?
  • None announced, but some submissions are pending.
• #variadics: Should we support tuple patterns in parameter lists?
  • It seems possible that we can always avoid them by doing the expansion in the function body instead.
  • The ergonomic cost of that is unclear, so it would be nice to keep the option open, but supporting it in the Call interface might be difficult.
• #language-questions: What's the Carbon equivalent of thread_local?
  • No design yet; async work is deferred to a future milestone.
• 04-29 Getting rid of forall in impl declarations
  • Previously suggested 03-27 in #generics-and-templates on Discord and #3763
  • #generics-and-templates concern about overlap with lambda syntax, though the compiler can tell the cases apart
• Discussion topic
  • Any interesting points

RSS Feed for "Last Week in Carbon" here