LodeHQSubscribe →

Bun shreds 13.8GB file in 9 seconds, Zig rethinks I/O

Dev · 2026-06-21

Languages & Frameworks
Inside Signals: How Push-Pull Reactivity Powers Modern UI Frameworks13 MIN

The article dissects the push-pull algorithm that drives Signals, the reactive primitive now baked into Solid, Svelte, Angular and others. It shows how the dependency graph schedules updates, turning simple values into spreadsheet-like cells that auto-propagate changes. Understanding this loop helps developers debug and extend reactivity in their apps.

Bun parses 1 billion‑row 13.8 GB file in under 10 seconds15 MIN

A developer tackled the 1 Billion Row Challenge using Bun/TypeScript, parsing a 13.8 GB temperature file in under 10 seconds. By opening the file with low‑level descriptors and zero‑copy chunk reads, he avoided Bun’s 4 GB buffer cap. The result proves Bun can handle massive streaming workloads far faster than typical Node.js approaches.

APL drives a real‑time 3D voxel engine1 MIN

A full‑screen voxel engine runs entirely in Dyalog APL, proving an array language can handle real‑time terrain generation and rendering. It compiles to Vulkan, DirectX12 or Metal via SDL3, offering a proof‑of‑concept for game development in a language usually confined to data crunching.

Zig's Io Interface Lets You Plug Any I/O Backend and Parallelize Tasks12 MIN

Zig now requires the caller to supply an Io implementation, letting you inject custom storage, network stacks, or mock objects. The interface also exposes async primitives, so code that declares concurrency can be run in parallel automatically. This opens Zig to finer‑grained testing, sandboxing, and high‑performance event loops.

Tools & Platforms
Your ATProto identity is controlled by your PDS operator, not you4 MIN

A Personal Data Server (PDS) holds the signing key that authenticates every ATProto action. That means the PDS operator can post, like, or delete as you across Bluesky, Tangled, Grain, and any future ATProto app, and can even kill your identity entirely. The promise of decentralization is therefore fragile, hinging on trusting a single operator.

LFM2 vision model runs client‑side in the browser with WebGPU1 MIN

Liquid AI now lets its LFM2 vision-language model run entirely in the browser via WebGPU. No server round‑trips are needed, so inference is local, fast, and privacy‑preserving. This opens up on‑device multimodal AI for web apps without heavyweight back‑ends.

Chrome’s MV3 bug lets adblockers resurrect webRequestBlocking5 MIN

A Chrome developer uncovered a 2023 bug that restores the webRequestBlocking permission in Manifest V3, letting extensions block network requests despite Google’s planned removal. The flaw sidesteps Google’s ad‑tech crackdown, meaning adblockers can keep working on Chrome without reverting to legacy MV2. It highlights how hidden JavaScript bindings can re‑introduce legacy capabilities.

Google’s traffic finally hits 50% IPv6, a tipping point for the internet8 MIN

Google reports half of its user traffic now reaches its services over IPv6, confirming the protocol’s maturity and global roll‑out. The milestone, backed by APNIC’s independent measurements, signals that IPv6 is no longer a niche upgrade but the new default for internet connectivity.

Engineering Practice
Why Vintage Software Ran Faster, Modern Bloat Is a Choice7 MIN

The essay shows that early software was lean because hardware constraints forced efficiency, while today we waste those gains on oversized containers, default settings, and added complexity. Recognizing this choice can shrink resource use and restore performance without sacrificing essential modern features.

Why io_uring Beats epoll for Scalable Linux I/O5 MIN

A hands‑on benchmark shows io_uring cuts syscall overhead and doubles throughput compared to epoll in a reverse‑proxy workload. Use io_uring for latency‑critical, high‑connection servers; stick with epoll only for simple apps or when kernel support is missing.

SMPTE Frees All Standards to Accelerate Media Interoperability1 MIN

The Society of Motion Picture and Television Engineers has lifted the paywall on its entire standards catalog, making every standard, practice and guideline freely available. By removing cost barriers, SMPTE hopes to speed adoption, boost interoperability and fuel the next wave of media‑technology innovation.

Postgres LISTEN/NOTIFY locks the whole DB and kills scale9 MIN

Recall.ai’s high‑write workload crashed when thousands of bots issued NOTIFY inside transactions. The feature acquires a global lock during commit, serializing all writes and causing severe bottlenecks. The fix landed in PostgreSQL core, but the incident warns against using LISTEN/NOTIFY for massive concurrent writers.

Careers & Open Source
AI‑generated résumés are wrecking hiring pipelines, here’s how firms can reclaim signal1 MIN

Generative AI lets candidates inflate résumés and ace scripted interviews, eroding the trust recruiters place in traditional signals. HBR outlines concrete fixes: adopt AI‑augmented verification, redesign interview formats, and shift to performance‑based assessments to restore hiring integrity.

Get Dev in your inbox, every issue.
Subscribe free
Privacy · Terms · About · Contact
© 2026 LodeHQ