2026-01-26— Last Week in Carbon #6658

josh11b · 2026-01-26T19:53:29Z

josh11b
Jan 26, 2026
Maintainer

Decisions
- #6333 CLI and separate compilation
  - Accepted
New RFC proposals
- (none)
Active discussions
- #5389 Generic across forms
  - Updated proposal based on #interop discussion:
    - Operator interfaces no longer forbid ref parameters.
    - If an operator impl has a ref parameter not required by the interface, it must be marked with ref at the callsite.

01-20 Implementing heterogeneous operations for long/long long not working if one of the operands is a `let` or `var` variable. Error: non-constant call to compile-time-only function.
- Use int.convert instead of int.convert_checked; int.convert_checked is for conversions that are checked at compile time and require a compile-time-constant value.
01-20 Passing structs and tuples to C++ functions.
- Initial PR for mapping tuples and structs to initializer lists in C++ overload resolution is out. Next: supporting conversions from Carbon expressions (incl. structs and tuples) to C++ types. Discussed ideas and open questions that need further investigation (e.g. aggregate initialization, brace elision, std::initializer_list etc.)
01-20 Reverse interop not yet started, but to begin soon.
#interop Do we represent each overload of a C++ operator as a separate impl, or combine them into a single Carbon entity?

01-20 Like "const West" syntax approach
01-20 01-22 ^(A, B) for union of ^A and ^B follows a general approach of ^ being followed by an ordinary expression, and ^A has a corresponding A in the current name scope
01-20 Implicit declaration of alias sets
- Generally feel that Carbon can be more abbreviated, implicit with safety annotations
#safety Safety Unit No. 31: const and mutable still in progress, but far enough along for others to take a look.
#safety Concerns about removing |...| syntax for alias sets.

#generics-and-templates Continued issues with "inner Self."
#llvm Annoying clangd bug: "find all references" doesn't find accesses via structured bindings.
#generics-and-templates How do we enable Core.Range to deduce its return type from its argument value?
#typesystem If we select user-defined conversions using only the destination type, not the destination category, how do we model C++ conversions, which can overload on the reference-ness of the destination?
#pattern-matching Should we consider match on the critical path for 0.1?
- We should at least have parity with switch, including an interop story for enums.
- Don't need to support mixing expression and non-expression patterns in 0.1.
01-23 Discussion of generics and inner Self.
#syntax Can you pass a struct type literal as a function argument? How do we distinguish struct type literals from struct value literals?
- Yes, you can pass struct type literals as arguments.
- We use lookahead for : vs = to distinguish the two syntaxes.
- Some people have readability concerns around the struct syntaxes.