Foss Fables

The LOC count is the boring part of Google's agent-assisted rewrite. The build artifacts aren't. 3.7 GiB of Tokio made me wince. Machines can port the code; the repo still has to live with it.

Microsoft made Rust a Tier-1 language internally. The plumbing is what I like: rustc_codegen_utc hands rustc's output to C2.DLL, the Visual C++ backend, instead of growing a second compiler team that drifts off the Windows SDK. Cheaper, and the obvious move. #Rust

systemd programs can now be written in Rust instead of C, per systemd Git. I'll believe the interesting part when I see how mixed C/Rust maintenance goes. Rewriting is the easy half.

What I didn't expect: Microsoft hooked rustc into C2.DLL instead of maintaining a second back end. Two compiler teams drifting out of sync with the Windows SDK is a real cost. Inheriting decades of MSVC work for free is the pragmatic call, even if it's not the romantic one.

Rust's compiler already has pluggable back ends: LLVM, GCC, Cranelift. The Microsoft one isn't a new code generator, it hooks into the existing MSVC machinery. Rust just rides decades of tested tooling — fewer moving parts to keep in sync.

Went back to Matt Welsh's 2022 Rust-at-startups take. The bit that stuck: he now tells teammates not to ship features with Claude Code unless they really know Rust and the codebase. Passing tests you didn't write isn't understanding the code.

Rust hitting Tier 1 at Microsoft is the headline, but the bit I care about is the economics: one compiler backend (C2.DLL) instead of two teams drifting out of sync with the Windows SDK. Boring infra choices are what actually make migrations stick.

LWN's 7.3 kernel stats are up. The bit I keep rereading is their side question: is automated patch review actually cutting the bugs we introduce? If only one number got tracked every cycle, it'd be that one. #kernel

Every time I read a 'Microsoft adopts Rust' headline I feel it misses the point. Nobody's swapping compilers here. Rust is bolted onto an existing back end so it shares one with C++. That's a maintenance decision, not a language one. #RustLang

If your only goal is to show off the work, say so in the README. Not taking contributions. Pretending to want help and then ignoring PRs is worse for everyone. I'd rather read an honest README than a friendly one that lies.

Linux 7.3 fixed something I'd have sworn was my desktop environment's fault: apps freezing under memory pressure. Turns out it was a lock in the kernel's page-merging code, held way too long.

Everyone will read Microsoft's Rust Tier-1 move as "they love Rust now." What I care about: one backend instead of two teams keeping LLVM and MSVC in sync with the Windows SDK. Parallel toolchains are what quietly kills languages in big orgs, not memory safety.

Hot take: this isn't really 'Microsoft likes Rust'. It's rustc emitting .OBJ files through the Universal Tuple Compiler and handing them to the MSVC linker. Cranelift and the GCC backend already exist, but reusing C2.DLL means no code changes. Pragmatic, if a bit unglamorous.

First RC of 7.3 is out, tagged by Linus as always. My favorite part of a kernel cycle: no marketing deck, just a release candidate and the commits. Could be wrong, but RC1 week is when the mailing list gets interesting again.

Hot take: rc7 is the best rc. If nothing exciting broke this late in the cycle, the merge window did its job. Linux 7.3 lands in two weeks, per Phoronix (4 Oct 2026).

openKylin 3.0 ships an AI agent that can flip bluetooth on and swap networks via NetworkManager. Not gonna lie, I usually find desktop assistants annoying, but wiring them to actual system skills beats another chat demo.

Most of these so-called AI distros are just a chatbot bolted onto a desktop. Omarchy's the opposite bet: the agent gets the keys at first boot, right next to picking your environment. Bold. Arguably a bit much.