Hacker News Evening Brief: 2026-06-19
A Friday evening roundup covering Java’s decade-long Valhalla project, a claimed Linear A decipherment, a massive GitHub malware campaign, Norway’s ship tunnel greenlight, and much more across 30 stories.
Tech Tools & Projects
Project Valhalla, Explained: How a Decade of Work Arrives in JDK 28
Summary: JEP 401 (Value Classes and Objects) has been integrated into OpenJDK targeting JDK 28 after years of development. The pull request adds over 197,000 lines across 1,816 files and ships as a preview feature disabled by default. Value classes allow the JVM to store objects inline in arrays without header or pointer overhead. Brian Goetz cautioned that this is only the first part of Valhalla; heap flattening for objects larger than 64 bits is not yet included.
HN Discussion: Commenters debate whether the null-safety semantics of value types are genuinely simpler or just a different mental model. Defenders cite Oracle-era constraints on Java development as context, while critics want the missing heap flattening. Observers find the mailing-list design evolution a compelling case study in major language feature incubation.
Ten years of ClickHouse in open source
Summary: ClickHouse marks a decade as an open-source column-oriented OLAP database, reflecting on growth from an internal Yandex project to an independent cloud service on AWS, GCP, and Azure. The post covers architectural evolution and ecosystem expansion including chDB, the ClickStack observability stack, and the Agentic Data Stack for AI workloads. The team emphasizes balancing C++ core maintenance with user-facing features like a web admin panel.
HN Discussion: Several users share success stories replacing Elasticsearch and TimescaleDB; one reports 5x better storage and QPS in a PoC but was blocked by management wary of Russian-originated software. Users praise ClickHouse for dashboard analytics and logging where Postgres becomes too expensive. The project’s post-sanctions resilience is noted as part of its longevity narrative.
Gribouille 0.3.0: A Grammar of Graphics for Typst
Summary: Gribouille 0.3.0 adds fine-grained guide control for hiding ticks and legends without touching the theme, a themed compose() function, and default area stacking. The library implements a Grammar of Graphics approach for Typst, similar to ggplot2 in R or Observable Plot. New defer() replaces the old plot(defer: true) syntax, and annotate() can let marks overflow the panel. Editor integration includes Tinymist docstrings support.
HN Discussion: Commenters debate whether the API borrows patterns from Observable Plot or ggplot2 more closely, and whether the aes wrapper is necessary. A user finds the shorthand ‘labs’ (for labels) jarring among otherwise full-word parameter names. General enthusiasm about Typst as a TeX alternative with a cleaner macro model.
Show HN: Modeloop – From visual algorithms to microcontroller C code
Summary: Modeloop is a visual model-based design environment generating Python and MISRA-oriented C code from block diagrams and StateCharts. It targets the model-based engineering workflow Simulink occupies but with modern DevOps features and no binary lock-in (models are plain JSON). The tool runs in-browser with a native desktop app in waitlist and is currently in closed beta. It is positioned as 10x faster to production by generating reviewable code, tests, and artifacts from a single visual model.
HN Discussion: HN users found a disappearing mouse cursor bug faster than any QA team, noted as a rough first impression. Commenters question whether the main audience is students, makers, or embedded engineers, noting Simulink has long needed a proper competitor. Some find the website’s AI-generated marketing copy off-putting, with language patterns described as formulaic.
Akse3D – open-source 3D modelling anyone can master
Summary: Akse3D is a browser-based 3D modelling tool designed for kids and teens, built by the Norwegian makerspace Skaperiet. Users combine primitive shapes or draw 2D blueprints and export STL files ready for 3D printing with real-millimetre measurements. No sign-up is required to try it, and self-hosting is available for schools and makerspaces.
HN Discussion: Users appreciate the frictionless onboarding — clicking “Try for free” goes straight to the app with no sign-up wall. A commenter requests a desktop version (e.g. Electron) for offline use, preferring local tools over browser-based ones like Tinkercad. A skeptical view holds that every “easy CAD” tool becomes useless for experts, suggesting layered complexity instead of simplification.
Lift Challenge
Summary: DARPA’s Lift Challenge incentivizes novel drone designs that lift at least 4x their own weight while flying a 5-nautical-mile circuit. The challenge aims to break the mold of distance and speed optimization in favor of heavy vertical-lift capacity for military and infrastructure inspection use cases. Cash prizes encourage non-traditional entrants beyond the usual defense contractors. The competition will be held at the world’s largest military aviation museum.
HN Discussion: A commenter working on the challenge says it breaks the distance/speed mold but is not a crazy hard problem, offering to discuss tuning knobs. Another predicts a hydrogen-filled balloon would win, arguing the 5NM distance is too short to rule out buoyant lift. A third proposes a 50NM course or minimum 100kg payload to prevent edge-case engineering.
Show HN: Metiq: a real time 3D globe for 100 public datasets
Summary: Metiq presents a real-time 3D globe visualization of roughly 100 public datasets, overlaying global data layers on an interactive Earth model in the browser. The project’s minimal landing page does not elaborate on data sources or update frequency. It is positioned as a “global intelligence” tool, likely aggregating environmental, economic, or geospatial data.
HN Discussion: The submission had no HN comments at the time of capture. The project appears visually compelling but the website provides little detail about data sources or methodology.
Ubiquiti: Enterprise NAS, Built on ZFS
Summary: Ubiquiti announced an enterprise NAS product built on ZFS with dual 25GbE SFP28 ports and redundant power supplies. The device marks Ubiquiti’s entry into the network-attached storage market, leveraging ZFS for data integrity and snapshots. Pricing emphasizes no recurring license fees, contrasting with competitors moving to subscription models.
HN Discussion: Commenters question whether spinning drives can saturate dual 25GbE links, citing difficulties achieving even 10Gbps with ZFS on HDDs even with L2ARC. Ubiquiti’s track record on software quality and security comes under scrutiny, with past AWS key leaks and misleading encryption claims mentioned. The no-subscription pricing is praised as a welcome departure from industry MRR trends.
Modos Color Monitor Pushes E-Paper Displays Further
Summary: The two-person startup Modos is fundraising for Modos Flow, a 13.3-inch color e-paper monitor targeting 3200x2400 resolution, touch input, and a 60Hz refresh rate. E-paper technology has traditionally been limited to low refresh rates and monochrome; Modos aims to make it viable as a primary computer monitor with much lower power consumption and sunlight readability.
HN Discussion: A YouTube video from the creator provides a detailed technical walkthrough of the development process. A commenter tracking the e-paper space calls Modos one of the most exciting developments but questions whether higher refresh rates reduce Carta panel longevity. The specs are widely acknowledged as ambitious.
Launch HN: TesterArmy (YC P26) – Agents that test web and mobile apps
Summary: TesterArmy uses AI agents to continuously monitor critical user flows across web and mobile apps, alerting teams when something breaks. No SDK, test scripts, or infrastructure is needed; it connects via staging URL or app binary with Slack, CI/CD, and GitHub integrations. The YC P26-backed product positions itself as replacing traditional E2E testing with always-on agent-based monitoring.
HN Discussion: Commenters question how it differentiates from established players like Rainforest QA and whether token costs of running continuous agents beat traditional E2E tests. The nondeterminism of LLM agents is raised as a concern for stable test results. One asks about compatibility with Cloudflare Turnstile or captchas and whether a human-in-the-loop is involved.
Linux Maintainer Greg Kroah-Hartman Says AI Tools Now Useful, Finding Real Bugs
Summary: Greg Kroah-Hartman states that AI-generated bug reports for the Linux kernel have shifted from junk to genuinely useful in detecting real defects. The tool Sashiko is specifically credited for finding obscure race conditions, stack leaks, and kernel-panic-level bugs. GKH cannot pinpoint the inflection point but confirms the improvement is persistent. The AI tools can now analyze subsystem-specific nuances, such as checking IIO subsystem code against chip datasheets.
HN Discussion: A commenter who sends patches and reviews confirms Sashiko detects very obscure race conditions and kernel bugs that would otherwise be missed. One notes the article is from March 2026 and questions whether the finding remains current. The shift is seen as significant given kernel maintainers’ well-known skepticism toward automation-generated reports.
Security & Privacy
I found 10k GitHub repositories distributing Trojan malware
Summary: A developer discovered 10,000 GitHub repositories distributing Trojan malware through a systematic cloning operation. Attackers clone legitimate repos, push a commit adding a link to a zip archive, then delete and re-push the commit every few hours. The repositories are from different contributors with different names and are not forks, making detection harder. The targeting prioritizes newer repositories over popular ones, likely aiming to infect automated dependency resolution agents.
HN Discussion: Commenters theorize the rapid commit-rotate pattern is designed to appear at the top of “Last Updated” search results to trick users into cloning the infected version. Several maintainers report finding their own names attached to unknown repos, suggesting an impersonation tactic beyond simple cloning. One connects this to a recent WSJ story about a Disney engineer who downloaded an AI tool from GitHub and had his accounts compromised.
History & Science
Amateur may have cracked Linear A, a 120-year-old puzzle
Summary: Self-taught AI engineer Tom Di Mino claims to have deciphered Linear A, a Minoan Bronze-Age script that has resisted decipherment since its discovery in 1900. Di Mino believes Linear A maps to an extinct Semitic language predating biblical Hebrew, whereas prior Semitic hypotheses by Cyrus Gordon in 1957 did not gain acceptance. Over 300 words have been translated so far, and the work is under review by linguists at Rutgers and Cambridge. Claude Code was reportedly instrumental in the pattern-matching work behind the breakthrough.
HN Discussion: Commenters note that many prior Linear A claims came from fringe sources, but the Rutgers and Cambridge peer review lends credibility. Skeptics point out that the Semitic family is broad and proving a full decipherment requires consistent grammatical parsing, not just vocabulary matches. The use of an AI assistant as a research companion in a humanities discovery is itself seen as noteworthy.
Reinventing the Renaissance
Summary: A review of Ada Palmer’s book Inventing the Renaissance: Myths of a Golden Age, which argues the Renaissance as popularly understood is a constructed myth rather than a clean historical period. Palmer initially wrote the book as blog posts before expanding it into a 768-page volume published by the University of Chicago Press. The book was completed under chronic illness constraints with no archive travel, relying on accessible English-language sources.
HN Discussion: A commenter links a Dwarkesh Patel interview with Ada Palmer covering why Leonardo was a saboteur and Gutenberg went broke. One reader draws a parallel to current efforts to frame the AI era as a similarly constructed historical narrative, calling the book highly relevant today.
How Alberta Eradicated Rats
Summary: Deena Mousa documents how Alberta, Canada has remained rat-free for over 70 years through a sustained eradication program involving bait stations, building rat-proofing requirements, and public cooperation. A pest control officer famously ate warfarin-treated rolled oats at public meetings to prove safety. Alberta is a blank spot on global rat distribution maps — by comparison, New York City alone estimates three million rats.
HN Discussion: A commenter corrects the article: the mayor who refused to cooperate opposed the Social Credit Party, not the UCP which formed decades later. An Alberta resident confirms no rats and notes ticks in the region also do not spread Lyme disease. The warfarin-eating demonstration is flagged as a memorable piece of community engagement in public health policy.
Academic & Research
The Productivity J-Curve [pdf] (2018)
Summary: An MIT paper explaining the productivity J-curve pattern that occurs when general-purpose technologies are adopted. Initial productivity dips as firms invest in complementary assets such as retraining and process redesign before gains materialize. The paper uses a formal model with Hamiltonian dynamics to show why measured productivity often drops before rising. The effect is distinct from hype cycles and depends on whether adopters are price-takers in capital markets.
HN Discussion: A commenter notes the model breaks under current AI-investment reality where GPU, memory, and power supply are finite and marginal costs are not constant. Another warns readers not to conflate “GPT” in the paper (general-purpose technology) with OpenAI’s GPT. One thread jokes about the irony of MS Word being used to typeset a paper about productivity.
Zen and the Art of Machine Learning Research
Summary: Jack Morris argues that success in ML research depends more on temperament than talent, drawing parallels to Zen meditation. The core process is a cycle of reading papers and building things; neither alone suffices. Scientific insights come unpredictably, and the key trait is discipline to put in time regardless of daily results. Noam Shazeer is cited on the inherent randomness of which research directions prove fruitful.
HN Discussion: Commenters distinguish Western Zen (equanimity and beginner’s mind from Zen and the Art of Motorcycle Maintenance) from East Asian Zen, which emphasizes aimlessness or non-purpose. One reader notes ML research feels more like alchemy or biology than mathematics, making temperament particularly important. Several engineers share their own transitions between ML and back-end roles, affirming that resilience against failure is the real differentiator.
CS 6120: Advanced Compilers: The Self-Guided Online Course (2020)
Summary: Cornell’s PhD-level compilers course by Adrian Sampson is freely available online with videos, written notes, and open-ended implementation tasks. It covers intermediate representations, data-flow analysis, classic optimizations, parallelization, JIT compilation, and garbage collection. The course uses LLVM plus a custom educational IR for practical hacking. The self-guided version omits deadlines and Zulip discussions but follows the same reading and coding schedule.
HN Discussion: A compiler expert notes the dynamic compilers section focuses heavily on trace compilation, which has been repeatedly abandoned, and suggests type feedback, speculation, and deoptimization are more important. One reader questions what makes the course “advanced” given many topics appear in introductory compilers courses. The course has been previously discussed on HN in 2020, 2023, and 2024.
Hospitals and universities repurposing drugs at lower cost
Summary: King’s College London reports that hospitals and universities are repurposing existing drugs at 90% lower cost than developing new ones. A key example is Bevacizumab (Avastin, $50/dose) used for macular degeneration instead of Lucentis ($2,000/dose), which is the same molecule. The nonprofit Cures Within Reach funds repurposed-drug studies for rare diseases like Huntington’s. Drug repurposing bypasses patent and R&D costs that typically drive pharmaceutical pricing.
HN Discussion: An ophthalmology insider confirms Avastin and Lucentis are molecularly identical; the difference is packaging and sterile injection qualification. Esketamine (Spravato) is given as an example of modifying an off-patent drug for re-patenting despite evidence the modification may be less effective. Healthcare systems in lower-income countries achieve similar savings by stocking generics from China and India instead of branded equivalents.
Web & Infrastructure
There Are No Instances in ATProto
Summary: Dan Abramov argues that the concept of “instances” is a Mastodon-brained category error when applied to the AT Protocol (Bluesky). ATProto separates hosting from aggregation: personal data servers host content while relay and indexing services aggregate across servers. The architecture mirrors RSS and Google Reader where blogs were hosted independently and readers aggregated them, not the ActivityPub model of distinct community servers. Users choose a Personal Data Server and can switch without losing identity or content.
HN Discussion: The post serves as a clear explainer for those confused about ATProto’s architecture compared to Mastodon’s federation model. Commenters find the RSS analogy helpful for understanding the hosting-versus-aggregation separation at the protocol’s core.
Norway greenlights first full-scale ship tunnel
Summary: Norway allocated NOK 8.6 billion (£671m) to build the Stad Ship Tunnel, the world’s first full-size tunnel for ships. The 1.8 km tunnel will let coastal ferries and small passenger vessels bypass the dangerous Stad peninsula in Vestland county, where 33 maritime deaths have occurred since World War II. Storms hit the peninsula roughly 100 days per year. Construction is scheduled to begin next year with an estimated four-year timeline.
HN Discussion: Commenters note the article lacks a map and point to the Wikipedia page for the Stad Ship Tunnel for better context. One questions how this compares to existing canal tunnels like the Rove Tunnel, suggesting the “first full-scale” claim needs qualification. A reader wonders about cost comparisons with alternative canal tunnels to address Panama Canal drought issues. The four-year timeline is met with skepticism, especially from European readers accustomed to infrastructure delays.
So You Want to Define a Well-Known URI
Summary: Mark Nottingham, co-author of the Well-Known URI specification and current Designated Expert for the registry, offers best practices for defining .well-known URIs. These locations work best when a client knows the site and needs to discover something about the whole site efficiently, with robots.txt as the canonical example. The post covers what makes a good candidate for registration versus what should remain in the root namespace.
HN Discussion: A commenter wishes more projects would use .well-known instead of polluting the root namespace, giving llms.txt as a counterexample. Another finds the post lacking substance, with no concrete examples leading to actionable recommendations. One reader questions whether the .well-known/change-password registry actually gets used by bots, suggesting the discovery value is limited.
AI & Tech Policy
W Social, public institutions and the theater of European digital sovereignty
Summary: Elena Rossini investigates W Social, a Bluesky fork branding itself as Europe’s alternative to X with identity verification and EU data hosting. The article alleges discrepancies between its public image and internal reality, including plans to train European AI models on user data. The platform attracted high-profile EU politicians and national news coverage, while an existing ATProto-based European network (Eurosky) receives no press. W Social is described as a for-profit company run by Swedish entrepreneurs.
HN Discussion: Commenters note W Social felt shady since its first HN advertisement; one created six different accounts despite its claims about human verification. A comparison is drawn to TruthSocial with a European accent, suggesting the core user base is EU politicians wanting independence from US-owned platforms. The lack of press coverage for Eurosky, a transparent non-profit alternative, is flagged as suspicious alongside W Social’s heavy media promotion.
System Administration
.gitignore Isn’t the only way to ignore files in Git
Summary: Three levels of Git ignore files exist: .gitignore (repo-wide, tracked), .git/info/exclude (per-repo but untracked), and ~/.config/git/ignore (global, machine-wide). The per-repo exclude file is useful for personal workflow files that should not be added to the project’s .gitignore. The global ignore file covers IDE artifacts, OS files, and AI tool outputs across all repos without modifying each project.
HN Discussion: Commenters highlight .gitattributes for ignoring diffs (e.g., package-lock.json noise) as a missing feature from the article. A user shares a trick using a scratch directory with .gitignore containing just ’*’ for dumping temporary notes without git status clutter. Several advocate for the global exclude as the proper place for IDE, OS, and AI files rather than polluting every project’s .gitignore.
SMTP Relay with Web Dashboard
Summary: SimpleRelay is a self-hosted SMTP relay with a web dashboard, built on FastAPI, Postfix, and Docker. Features include DNS validation, per-sender access control, and a web UI for managing relay rules. It is designed to simplify the setup of a local SMTP relay without the complexity of configuring Postfix or Exim manually, targeting teams that need transactional email sending without relying on external SMTP services.
HN Discussion: A commenter questions what it adds over apt-get installing Postfix or Exim and configuring them directly for LAN relay. Another notes similarities to Mailrise and Apprise, observing that Apprise pushes notifications to many destinations beyond just email. OAuth2 support for upstream SMTP (Gmail, Microsoft 365) is raised as a missing requirement for modern setups where simple credentials no longer work.
Other
Telescope Ranchers
Summary: Starfront Observatories in Rockwood, Texas runs a telescope ranch where astrophotographers ship their gear for remote operation. Owner Bray Falls hosts 550 telescopes on 40 acres under Class 1 Bortle ultra-dark skies with fast internet. Customers control their scopes from a laptop anywhere for as little as $99 per month. The ranch handles maintenance, power, and connectivity so hobbyists avoid transporting expensive optics.
HN Discussion: The story had no comments on HN at the time of capture. The concept appeals as a practical solution for urban astrophotographers who lack dark-sky access or the ability to set up and tear down equipment each session.
The AirPods Effect
Summary: Markham Heid reports on the social impact of ubiquitous earbuds, noting a 28% decline in spoken words per person between 2005 and 2019, with 44% of Americans now using Bluetooth earphones. The trend extends beyond music into podcasts and always-on audio consumption. The author observes cultural differences between the US, where AirPods are everywhere, and Germany, where they are far less common, and raises concerns about the erosion of spontaneous conversation in public spaces.
HN Discussion: Several commenters push back, noting that noisy urban environments are themselves unnatural and earbuds provide acoustic comfort in crowded spaces. The default mode network and lost daydreaming time is raised as a bigger concern than social isolation. Thomas Ptacek offers a contrarian perspective that chatting with strangers was never universally normal or desired.
Show HN: Gerrymandle - Daily puzzle game where you redraw electoral districts
Summary: Gerrymandle is a daily puzzle game where players draw electoral district boundaries to advantage their assigned party. Players group adjacent tiles into equal-sized districts; a party wins by having more houses in a district than others, while ties mean nobody wins. The game prioritizes fun and puzzle design over accurately modeling real gerrymandering, with a hint button for stuck players.
HN Discussion: A commenter sees it as a great tool for high school civics classes but notes the tie rule is unrealistic since ties essentially never happen in real elections. Gameplay is praised for its clear instructions, polished presentation, and thoughtful design tradeoffs favoring fun over accuracy. Links to academic papers on provably fair districting protocols are shared in the thread.
”No Feigning Surprise”
Summary: Julia Evans’s comic references the Recurce Center’s social rule “No Feigning Surprise,” which discourages acting shocked when someone lacks expected knowledge. The comic itself is a writing tip: instead of feigning surprise, say something actually surprising. A linked fact notes that a Seagate 2X18 dual-actuator HDD can write at 528 MiB/s, far above the commonly cited ~100 MB/s HDD speed.
HN Discussion: Commenters note the social rule originates from the Recurse Center’s code of conduct, which effectively maintains a positive learning environment. A reader points out the URL was slightly wrong — linking to /surprise/ instead of /no-feigning-surprise/ — but the content resonated. The HDD speed fact generated its own discussion about modern dual-actuator drives far exceeding common speed assumptions.
Ice water drowning survival of young patient (2025)
Summary: A JACC case report details a child who survived 2.5 hours underwater in ice water, followed by 1.5 hours of CPR before resuscitation. The medical team was instructed not to begin rewarming until the patient could receive comprehensive hospital treatment. At six-month follow-up, the patient could give short commands, stand without support, ride a tricycle, and eat soft foods, illustrating the “you’re not dead until you’re warm and dead” principle in cold-water drowning.
HN Discussion: A parallel case from Norway involved a tourist technically dead for 20 hours in hypothermia before resuscitation with good outcomes. Commenters note that “survival” is not binary; the patient had significant neurological deficits at six months while still improving. The Chris Lemons case — 30+ minutes without oxygen in the North Sea returning to diving weeks later — is cited as another example of cold-water protection.