David M. Lloyd has another piece on modern reflection, this time on VarHandle - and as usual, it's highly informative. It replaces a lot of Unsafe usage, and demonstrates quite a bit of atomic access fundamentals as well, and it opens up offheap access, too.
AI code is not the enemy. Ideas used to be valued against the weeks we'd spend building them, and most got shelved before we ever found out if they were any good. Now we get to find out in an afternoon.
JavaScript has no standard library, so reading a file looks different in Node, Deno, and Bun. rcompat is a thin abstraction that gives you one API and uses runtime keys to dispatch to each runtime's native implementation underneath, with branded identity checks and structured errors thrown in.
Rasmus Ros wrote "From Stringly to Strongly Typed," an article describing a Kotlin library that solves one half of typed schema access: how to read the data. But data isn't just accessible; it's related to other data, and that half he's deferring to a future post. The deferred half is well-trodden ground, and the reasons we don't tread it are worth knowing.
Alexander Hanff has documented two AI-vendor stealth installs in two weeks: Claude Desktop registering Native Messaging bridges across seven Chromium browsers, and Chrome writing 4GB of Gemini Nano weights to disk on every eligible device.
Simon Willison wrote up a summary of Zig's policy on LLM contribution this morning, suggesting that Zig can afford it because it has PR oversupply and downstream users wealthy enough to fork the project at need. Yet: most communities adopting the same policy don't share those conditions.
Jordan Lord lays out three constraints he applies before building anything: a one-page description that survives without padding, a piece of core tech that can outlive the product, and a single defining constraint that shapes the user experience. Scope, leverage, and identity - one filter each. Great stuff.
Quarkdown, a documentation system using "markdown with superpowers", has been released. Changes include quite a few useful documentation elements (key bindings, path specifications, other useful presentation elements) but the core focus is on the project and its scope.
Dragonfly is a multithreaded ground-up reimplementation of the Redis wire protocol, described as being designed to scale vertically where Redis can't. The interesting thing is that Dragonfly is selling itself as "in-memory infrastructure for caching, feature stores, and job queues," which is the IMDG shape from the cache side of the river and a quiet admission that the cache framing was always too narrow.