freenode

← digests

Swift cells, LLVM AI rules, and Git cleanup

Languages & Toolchains2026-11-17

Swift Evolution is reviewing new cell types while LLVM argues over how to limit AI-generated submissions that are wearing out reviewers. Git, Julia, and Haskell also moved: merge-detection patches, Apple Neural Engine access, and a published ecosystem vision.

Swift reviews Cell, ConstCell, and Volatile

Swift Evolution is reviewing SE-0556, which proposes Cell, ConstCell, and Volatile. Discussion has concentrated on aliasing, Sendable, and C and C++ interop. The outcome matters for anyone sharing mutable state across Swift and C-family code, where those rules decide what the compiler may assume.

LLVM RFC would ban AI-generated project text

An LLVM RFC proposes forbidding AI-generated text in pull requests and RFCs, aiming to stop verbose, low-quality submissions from consuming reviewer time. The thread drew 130 messages from 65 participants and is among the day's most contested toolchain governance debates.

LLVM weighs pull-request limits for non-committers

A separate LLVM RFC would limit pull requests from non-committers, again citing maintainer burnout from AI-generated low-quality submissions. The proposal would change how newcomers land patches, so the debate is about contribution policy as well as review load.

Julia package targets the Apple Neural Engine

AppleNeuralEngine.jl was announced as a Julia package for direct Apple Neural Engine access via tracing, with cross-Mac benchmarks against CPU BLAS. Developers who want on-device inference from Julia now have a concrete entry point and a performance comparison rather than only higher-level frameworks.

Git patch would find squash-merged branches

A patch series adds --delete-integrated, extending --delete-merged, so Git can detect squash and rebase merges by file content rather than commit reachability. That closes a gap for teams that squash on merge and then cannot clean local branches with the existing flag.

LLVM debates AI use on good first issues

An LLVM governance thread is clarifying the boundaries of AI use on good first issues and confronts one contributor over high-volume AI-assisted pull requests across projects. The result will shape whether beginner-labeled work stays a reviewable on-ramp or becomes another source of unreviewed volume.

Haskell Foundation publishes a technical vision

The Haskell Foundation technical leadership committee says it is operational and has published a shared technical vision, inviting proposals and funding. The document is the clearest recent signal of where the foundation wants ecosystem work to go.

Git discusses a log --follow regression

The Git mailing list is discussing a performance regression in git log --follow for non-linear histories. The author is iterating patches after review, so users who follow renames through complex history may see a fix once the series lands.