2025-08-04 — Last Week in Carbon #5902

geoffromer · 2025-08-04T18:48:57Z

geoffromer
Aug 4, 2025
Maintainer

Decisions
- #5884 Should type-checking inside a facet type constraint have access to rewrite constraints?
  - Both asked and answered in the last week.
  - Discussed in #generics-and-templates and 07-31.
  - Decisions:
    - Rewrites are still applied eagerly so that they can be used for subsequent name lookup.
    - Going to use implied constraints to allow constraint information propagation within a declaration.
    - Reaffirmed we don't want implied constraints to propagate across declarations
New issues
- #5896 C++ interop: Using C++ constructors syntax
  - What should be the basic syntax for calling C++ constructors from Carbon?
Active discussions
- (none)

Decisions
- #5689 Semantic Identity and Order-Dependent Resolution for Rewrite Constraints
  - Two rewrites of the same associated constant are allowed as long as they result in the same constant value semantically (rather than a syntactic test).
  - Approved
- #5434 ref parameters, arguments, returns and val returns
  - A ref parameter takes and gives a reference expression. The caller must use ref to mark the argument.
  - Can return a reference or value expression by marking the return with ref or val keywords.
  - Approved
New RFC proposals
- (none)
Active discussions
- #5545 Expression form basics
  - Update based on recent discussions in progress

Moderation and conduct teams are still the same size, available in theory from UTC-7 until UTC-5.5: 10 trained members altogether + community lead (hands-off trainer, coach, guide)
- WIP: Outreach for new interview round of community leaders of other programming language communities about their community journey
- DONE: Conversation to reassure team members who felt they were not contributing enough to stay on the team and to emphasize that being a trained moderator and making oneself available to help is also a way to contribute to the Carbon Language Project. Sometimes just being there is enough. 💜
Activity in the past week:
- One new contributor!
- In the last 7 days in our repo: 11 authors pushed 37 commits to trunk, we changed 125 files with 10,145 additions and 2,665 deletions, we merged 37 PRs, opened 7 PRs, we closed 2 issues and opened 2 new ones.

#toolchain #5871 Import symbolic generics
#toolchain Tried non-null pointer annotations in the explorer codebase and the consensus was that we didn't like the verbosity, and the C++ support had gaps.
#toolchain Re-use processing of the prelude
#toolchain Exposing compiler functionality such as determining the set of imports for use by language servers, build, and other tools
#toolchain Caching during build
PRs:
- #toolchain lldb debugging in #5824
- #toolchain ErrorOr in #5857
- C++ import in #5855 #5856 #5860 #5858 #5866 #5868 #5865
- vtables for generics in #5853
- rewrite constraints in #5617 #5673 #5864 #5872 #5885 #5892
- docs in #5869 #5846 #5886 #5861
- golden tests in #5870 #5877 #5878 #5897
- diagnostics in #5874 #5876 #5887 #5894
- invalid impl in #5875 #5881
- iterate for array types in #5895
- characters in #5893
- others in #5862 #5882
New contributors:
- @kazutakahirata made their first contribution in #5867

#interop 07-29 Decided to defer struct literal / initializer lists to avoid complex overload resolution.
- Create a lead question to decide how this should be handled.
07-29 To allow checking if a parameter is int32 or int64, we could add these types when we map the builtin types so we know they’re complete.

#typesystem How to model what initializing expressions evaluate to.
#infra Is it OK to post the memory safety strategy doc now that the summit is over?
- We'd prefer to turn it into a proposal first (hopefully soon).
#strategy Should we cancel the toolchain sync meeting due to lack of activity?
#pattern-matching Does pattern matching invoke explicit conversions?
- No.
#typesystem Are we OK with allowing code like this?

fn F() ->? form((var i32, var i32));
var a: array(i32, 2) = F();

Yes.
Side observation: we probably don't want to allow initializing a class from a struct value.
#summit Summit participants, please review the draft summit report so it can be published.
- Aiming to publish the week of 08-04
#strategy Carbon is being discussed on Hacker News again.
#typesystem Clarifying exact struct->class conversions
#typesystem What should the form of an explicit tuple/struct conversion be?
- Shouldn't introduce tuple/struct forms that weren't already there.
- Side consequence: As can't be the extension point that defines tuple/struct pattern matching.
#generics-and-templates Substituting into an ImplWitnessAccess sometimes fails because the result depends on the result of another substitution that hasn't happened yet.
- Seems to resolve if we substitute into the operands, rebuild, and then repeat substitution if rebuilding changed the instruction.
#generics-and-templates When should "constant" be replaced with "template constant" in #5890?
#generics-and-templates Certain cases have trouble seeing through .Self.
- Implied constraints should help here
#generics-and-templates Missing information on the node stack when dealing with RequirementEqual.
- Probably need a separate inst kind for rewritten impl accesses.