2025-09-01 — Last Week in Carbon

[josh11b]

Updates on leads issues

• Decisions
  • (none)
• New issues
  • #5979 Should we provide a syntactic shorthand for Core.Optional?
    • Possibly using ? or a keyword like opt
• Active discussions
  • #5903 Built-in character type questions
  • #5922 Built-in character operators
  • #5934 What kinds of type deduction do we want to support?
  • #5962 should we disallow all Unicode control characters in character literals?
    • Expanded list of characters to disallow

Updates on proposals

• Decisions
  • #5776 Replace impl fn with override fn
    • Appears to be abandoned by the author?
• New RFC proposals
  • #5990 Principle: The signature is the contract
    • Declarations are authoritative, we generally don't infer aspects of the usage contract from the definition
    • #proposals-prs-and-process The name of the principle may not effectively communicate the substance.
• Active discussions
  • #5913 Reworking unformed state
    • additional feedback
  • #5914 Updating Carbon's safety strategy
    • post-merge feedback

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

• 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 C# community leaders about their community journey
  • WIP: Moderation skill self-assessment poll to plan for internal training or coaching sessions.
• Activity in the past week:
  • Put 2 contributors in contact who were working on the same issue in different PRs.
  • One new contributor?!
  • In the last 7 days in our repo: 5 authors pushed 15 commits to trunk, we changed 93 files with 4,969 additions and 830 deletions, we merged 15 PRs, opened 9 PRs, we did not close any issue and opened 2 new ones

Activity on toolchain

• 08-26 What representation to use for optional pointers?
  • Since we are going to use an invalid pointer value for "not present", should use MaybeUnformed(T*).
  • There are multiple invalid pointer values, all of which are considered unformed, but some values are distinguished: all zero bits (probably the cheapest), high bit set, whatever the representation of C++ nullptr is. Which is used for the default unformed state? Which is used for "not present"? Which is used for "hardened"?
  • Can we assume the default unformed state is the same as C++ nullptr?
• 08-28 Caching of runtime artifacts as part of the build
• #toolchain Strategies for narrowing the failure down when a test using the full prelude fails
• #toolchain Making golden tests hermetic
• #toolchain Need an instrumented libc++ to be able to enable msan
• #toolchain Looking for a good starting task
• PRs:
  • C++ interop calling convention in #5980 #5972 #5976 #5977 #5987
  • conversions in #5984
  • C++ ABI compatibility in #5986
  • importing C++ enum in #5978
  • C++ std::string_view -> Carbon str in #5985
  • unsafe as in #5993
  • C++ char interop in #5988
  • others in #5981 #5998 #6001 #5994
• New contributors:
  • @pascal754 made their first contribution in #5994

Activity on C++ interop

• 08-26 C++ to Carbon character type mappings: `char` -> `char`, `signed char` -> `i8`, `unsigned char` -> `u8`. Others TBD.
• 08-26 C++ standard library support is being worked on.
• 08-26 Next features: Overload resolution for constructors and operators support. Other important features are blocked on vtable and optional support in Carbon, which is being worked on.

Activity on memory safety

• 08-26 RustConf talk on 09-03
  • Have some examples that show issues arising in real open source C/C++ source code
  • Proposing that Carbon can take on the complexity needed to get memory safety for C++ migration use cases, so Rust doesn't need to
• 08-26 Terminology discussion about what to name compile-time parameters that convey lifetime and aliasing
• 08-26 Considering options for immutability of value bindings
  • Not okay acquiring a lock that might block as an invisible part of value binding
  • Option to project to just the immutable parts of the object in a value binding is viable, but would still have complexity such as having two kinds of accessors and more states in the safety model
  • Other option is types with interior mutability may change even when bound to a value
    • As if the interior mutability was actually externally accessed through a pointer
• #safety Suggestion to use "capture sets", a transitive generalization of "alias sets".
• #safety Group Borrowing: Zero-Cost Memory Safety with Fewer Restrictions discusses an in-progress safety model for Mojo that supports aliased mutability
  • The author of the post and the designer of the feature are both happy to answer questions on our Discord.
  • This approach allows you to invalidate iterators without having to discard other references to the container itself.
  • It does add some annotation burden, but there are some tricks for reducing it.
  • Also trying to allow aliased pointers to be transferred between threads, and only disallowing access across multiple threads.

Other active discussions

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

• #proposals-prs-and-process #5980 seems stuck. Needs review from someone knowledgeable about the "constant" terminology.
  • Not sure what to do about changing "concrete" to "template". That matches the latest adopted proposal, but not the direction we want to be going.
• #pattern-matching Partially Matching Zig Enums talks about how to handle common code between match cases.
• #infra pre-commit sometimes fails after a Python upgrade.
  • Workaround seems to be to uninstall and reinstall pre-commit.
• #general-design 08-28 Looking for examples where C++ makes incorrect assumptions when it sees a forward declaration.
• #getting-started Can I learn Carbon if I only know C?
  • Yes, but we'd definitely recommend learning C++ first.
• #ints-and-floats Do we expect Core.Char to be implicitly converted from integer literals?
  • Probably not. Does that mean it doesn't really have a signedness?
  • No, that's still visible from comparisons.
  • Not entirely clear what the difference of two Chars should be, but i32 is plausible.
• #strategy Carbon mentioned in A Review of Nim2.
  • Author seems to think Carbon will transpile to C++.

RSS Feed for "Last Week in Carbon" here