France has put procurement muscle behind a real attempt at digital sovereignty, macOS power users keep finding clever native workarounds, and NASA’s redundancy-first avionics design reminds us what engineering discipline looks like when failure isn’t an option. Today’s picks balance policy, practical tweaks, and high-stakes systems design.
In Brief
France launches government Linux desktop plan
Why this matters now: France’s interministerial digital agency (DINUM) has launched a cross-government push to migrate government desktops away from Windows toward Linux, and ministries must submit migration plans by autumn.
DINUM’s seminar set explicit procurement and migration steps — covering desktop OS, collaboration tools, antivirus, AI, databases and virtualization — and signaled concrete programs already underway, like moving 80,000 Caisse nationale d'Assurance maladie agents to the state’s sovereign stack, and aiming to migrate the health data platform to a “trusted solution d'ici à fin 2026.” The announcement frames this as procurement-driven digital sovereignty rather than rhetoric: ministries will coordinate with ANSSI and industry coalitions via upcoming “rencontres industrielles.” Read the government briefing at the DINUM press release.
"au profit de postes sous système d'exploitation Linux."
Expect the real debate to be about integration pain: Active Directory, legacy Windows-only apps, specialized drivers, and desktop management workflows. For procurement teams and EU suppliers this is a strong, immediate signal; for IT ops it means planning migrations that touch identity, app compatibility, and device ecosystems.
Native instant space switching on macOS
Why this matters now: A tiny open-source tool, InstantSpaceSwitcher, gives macOS users near-instant desktop switches without disabling System Integrity Protection or installing big tiling managers.
The author documents why the usual advice — toggling “Reduce motion” — is lame and why heavyweight options like yabai carry real security and support costs. Instead, InstantSpaceSwitcher simulates a high-velocity trackpad swipe to trigger native space switching immediately; it’s a menu-bar app with a CLI for left/right or numbered spaces. Try the write-up at the blog post.
"This is always the default answer to this question online, and I’m sick of it!"
For anyone who uses Spaces heavily, this is the pragmatic middle ground: minimal permissions, native feeling, and low surface area. Power users will still debate 120Hz behaviour and edge cases, but this is a neat example of picking a small, local fix that respects platform constraints.
RAM has a design flaw from 1966. I bypassed it.
Why this matters now: An engineering deep-dive and demo shows DRAM refresh creates measurable tail-latencies, and the author ships a library—Tailslayer—that hedges reads across replicated memory channels to dodge refresh stalls.
LaurieWired measures periodic ~100–300ns latency spikes caused by DRAM refresh and implements a pragmatic software workaround: keep hot data replicated across independent channels with uncorrelated refresh schedules and issue hedged reads to whichever replica returns first. The project is on video and explained in detail in the demo; see the video.
"we're still paying for a latency penalty that's existed since the 60s!"
This is brilliant systems detective work. But it’s also expensive: you double memory footprint and bandwidth, increase cache pressure, and rely on software-level hedging because there's no ISA-level hedged_load primitive yet. The work matters mostly to extreme-latency-sensitive systems, and it makes a persuasive case for hardware support if tail-latency matters in your stack.
Deep Dive
How NASA built Artemis II’s fault-tolerant computer
Why this matters now: Artemis II’s onboard avionics use a layered, redundant architecture—eight CPUs running lockstep across independent modules—to ensure that transient faults, radiation, or software divergence won’t give the crew a “wrong answer” at 250,000 miles.
NASA’s team leaned hard into redundancy and determinism. The vehicle runs two Vehicle Management Computers, each containing four Flight Control Modules (FCMs). Each FCM is a self-checking pair of processors, so eight CPUs execute flight software in lockstep with a “fail-silent” policy: a misbehaving module is designed to stop talking rather than output corrupt state. The architecture tolerates multiple module failures in quick succession—the reported claim is you can lose three FCMs in 22 seconds and still continue safely on the remaining FCM.
Determinism is enforced everywhere. The avionics use time-triggered Ethernet, ARINC653 scheduling, triple-modular memory, and three independent network planes so timing and data flow remain predictable. There’s also a separately developed Backup Flight Software (BFS) on different hardware and an alternate OS to reduce common-mode failure risk. The goal isn’t novelty — it’s provable behavior under fault and radiation stress.
"A faulty computer will fail silent, rather than transmit the ‘wrong answer’."
The verification pipeline matches the architecture: full-environment simulation, massive fault-injection runs, Monte Carlo stress tests, and extensive hardware-in-the-loop validation. Those aren’t optional steps; they’re the way the team proves the system will behave under rare but catastrophic conditions. This reads like a masterclass in conservative engineering discipline: minimize assumptions, limit reliance on per-component correctness, and make failure modes observable and safe.
There’s a trade-off the reporting highlights and HN commenters debated: cost and complexity. Hardware redundancy and multiple independent stacks are expensive in mass, power, and engineering hours. Some argue this is “throwing money at the problem,” while others remind us that for crewed deep-space missions the right answer is often to buy safety with redundancy rather than rely on software processes or patch cycles. The broader lesson for systems engineers is practical: when human lives or multi-hundred-million-dollar missions are on the line, deterministic architectures and rigorous verification beat agility and incrementalism.
Closing Thought
Big systems, small hacks, and policy moves all share one trait this morning: choices that trade convenience for predictability. France’s procurement nudges want predictable supply chains and control; InstantSpaceSwitcher picks predictability of UX without breaking platform safety; Artemis II buys predictability through redundancy and verification. When uptime, safety, or sovereignty matters, engineers and policymakers revert to designs that limit surprises.