Full Width 
Shortcuts 
Sign Up 
Log In 
Articles 
Reading List More from Tyler Cipriani: blog
.title {text-wrap:balance;} #content > p:first-child {text-wrap:balance;} .gallery { display:flex; justify-content:center; } figure a:link, figure a:visited, figure a:hover, figure a:active {color:inherit; text-decoration: underline;} Netbooks are dead, but the Chuwi Minibook X scratches the same itch. The Minibook X is a 10.5″ x86_64 sub-ultrabook with 16GB RAM, a 512GB NVMe drive, and only one majorly annyoing Linux quirk. I needed a knock-around laptop, so I bought myself a Minibook for my birthday last year. The more I tote it around, the more fun I’m having with this ridiculous little computer. Chuwi Minibook X Quick specs Much like the netbooks of yore, the Minibook is a budget machine. But it’s 2026, so even budget machines pack more oomph than I need from a utility laptop. CPU 4-core/4-thread 3.6GHz Intel N150 Twin Lake 16 GB RAM – LPDDR5-6400 – soldered 😿 512GB NVMe – upgradable 10.51” IPS 2K 16:10 screen 28.88Wh Li-Ion battery Weight: 911g Ports: 2×USB-C (1×PD charging) Cost: $350 Chuwi Minibook Guts One oddity is that the Minibook comes bundled with a 12V/2A USB-C charger. I chucked the charger; I worried I’d fry some 5V SoC someday. The Minibook works fine with a PD charger. Minibook X using a PD Charger at 20V I’d assume the 12V charger was a cost-saving choice, but it also creates some weird possibilities for DC/off-grid setups. Linux and weirdness: sideways panels and kernel parameters Charlie Stross, a favorite SciFi author, talks up the Chuwi Minibook X on Mastodon The fediverse told me that Minibook runs Linux “boringly well,” which was almost true. I tried Debian, then jumped to NixOS for kicks. What works: Camera/Microphone/Speakers Touchscreen Sleep/Suspend Hibernate Keyboard backlight USB-C HDMI Bluetooth (non-free blobs – Intel) Wi-Fi 6 (non-free blobs – Intel) But on first boot, the screen orientation is 270° clockwise: Linux setup screen rotated 270°. The Chuwi’s screen is a panel from a cheap tablet; the screen rotation issue is a hardware problem (the screen is mounted sideways). To fix the screen’s rotation, I had to tweak screen orientation at every software layer. Fixing this problem was a journey: Bootloader – Switched from systemd-boot to grub, carrying some unmerged GRUB rotation patches on top. Initrd – Tell the Intel display driver about the panel orientation via a kernel parameter, and force the Intel driver to load in the initramfs. On NixOS: boot.kernelParams = ["video=DSI-1:panel_orientation=right_side_up"]; and boot.initrd.kernelModules = ["i915"]; (see Kernel docs for modedb default video mode support) Desktop environment – For X11, good ole xrandr --output DSI-1 --rotate right. Wayland picked this up from the DRM connector. This one was easy. Framebuffer – Ensure all TTYs have the proper orientation by adding fbcon=rotate:1 to kernel parameters boot.kernelParams = ["fbcon=rotate:1"]; (see Kernel docs for framebuffer console boot options) Behold, the final result in all its glory: Non-rotated system boot. Zero Cool's bootscreen courtesy of mainframed/Hackers-Plymouth Size, weight, and build This computer is mind-bogglingly small. The build is sturdy and totable; it’ll hold up to a backpack jostling. Chuwi Minibook X with "banana" for scale The laptop’s case is MacBook-esque: aluminum and good-looking. The MacBook Air’s dimensions dwarf the Chuwi’s, but the two laptops are about the same thickness. Chuwi Minibook X alongside the Macbook Air Chuwi Minibook atop the Macbook Air A notebook that weighs more than a kilo is simply not a good thing – Linus Torvalds The Minibook weighs in just shy of a kilo at 912 grams. My Minibook X weighs 912g Perf, thermals, and power tl;dr: you get what you pay for. But battery life and cooling are better than I’d have guessed. The Minibook X was never going to compile the Linux kernel in record time. But the performance matches the specs, it stays cool, and it has enough battery life to run a movie marathon. Numbers: Geekbench6 (a fun side-quest to get running on NixOS), better than I expected. Single-core: 1295 Multi-core: 3332 Wi-Fi 6 speed: 424 Mbps, more than enough to stream a 4K movie. Power Idle: 3.8W During benchmark: ~15W Battery: When I left the 1995 classic film “Hackers” looping in VLC, the battery lasted about 6 hours. Heat: Running stress-ng for 10 minutes, the hottest part of the laptop chassis remained below 90°F (32°C): Thermal camera view of Chuwi Minibook X running stress-ng What I dislike There’s so much to dislike about this laptop: Screen is terrible – 2K? 50Hz refresh rate? Why!? Keyboard is terrible – it only registers keystrokes when you hit the exact center of each key. Touchpad is terrible – It’s a diving board-style, without physical buttons. Sound is meh – I can hear the tinny laptop speaker fine, but it’s underwhelming. I’ve never tried tweaking it in Pipewire, though; it’s possible it could be better. But “terrible” is in comparison to the nicest modern laptops in existence. Everything I listed here works fine. I’m honestly blown away when I tune my expectations to the sub-$400 laptop range. Verdict In The Death and Life of Great American Cities, Jane Jacobs wrote, “new ideas require old buildings”: cheap spaces let people try risky ideas. The Chuwi Minibook X is an old building. I can brick the Minibook and have a normal Monday on my serious work laptop. Nothing has to work, which makes it perfect to try out new Linux desktop stuff: NixOS – I’ve been using Debian for 15 years+, figured I’d try joining the NixOS cult for a while. RiverWM – I’m on a quest to find the Wayland version of XMonad; River is pretty close. KDE Plasma – I’ve used a tiling window manager for over a decade. What’s it like to use a desktop that Just Works™? Steam – Never been much into games, but I decided to give Steam a try since, well, why not? Cheap, weird computers like the Chuwi make it safe to play. And playing with computers is still fun. Playing Melatonin on Steam on the Chuwi
.title {text-wrap:balance;} #content > p:first-child {text-wrap:balance;} Hackers are pwning packages at an exhausting clip. But the hacks are hackneyed. What’s new is the doom cycle: Code that steals keys to publish code to steal more keys. A zombie army of infected code. And AI is making it worse. GitHub Actions are a trap Trivy is an open-source security scanner. But if you used Trivy in late March, you had a bad time. On March 19th, hackers pushed a version of Trivy that tried to smuggle secrets from anywhere it ran. Trivy cited a “misconfiguration” in their continuous integration (CI) system, GitHub Actions. But the exploit was less a misconfiguration and more a GitHub Actions trap. Admiral Ackbar warning about the trap in GitHub Actions Here’s a simplified version of how Trivy got pwnd1: # INSECURE. DO NOT USE. on: pull_request_target jobs: check: steps: - uses: action/checkout@deadbeefdeadbeefdeadbeefdeadbeefdeadbeef with: ref: refs/pull/${{ github.event.pull_request.number }}/merge - uses: ./.github/actions/setup-go - uses: some/go-static-analysis@c0ffeec0ffeec0ffeec0ffeec0ffeec0ffeec0ff At first glance, this code looks fine: No secrets referenced. Third-party actions pinned to an immutable hash. Check out a pull request. Perform some static analysis. But this code is a verbatim antipattern from a 2021 GitHub blog post titled “preventing pwn requests”: if the pull_request_target workflow only […] runs untrusted code but doesn’t reference any secrets, is it still vulnerable? Yes it is – GitHub Security Lab The problem is pull_request_target: pull_request_target – plunks a nice, juicy GITHUB_TOKEN into the environment. actions/checkout – takes an optional parameter persist-credentials, which removes secrets if set to false. But the default for the parameter is true. Setting the persist-credentials parameter to false has been an open issue in GitHub Actions since 2021. Your $HOME is a crime scene Once hackers had Trivy’s keys, they published a new version of Trivy to steal more keys. LiteLLM used Trivy in their CI. The same CI they used to publish code to PyPI, the Python software registry. When LiteLLM’s CI ran the compromised Trivy, hackers nabbed their publishing key. And on March 24th, when Callum McMahon fired up his IDE, his MacBook froze. And that’s how he discovered the LiteLLM hijack. McMahon’s MacBook was flailing at bad code that hackers snuck into LiteLLM. And the bad code trying to steal credentials: ~/.netrc ~/.aws/credentials ~/.config/gcloud ~/.config/gh ~/.azure ~/.docker/config.json ~/.npmrc ~/.git-credentials ~/.kube/ Files that are typically strewn around $HOME directories, full of tokens and keys, often unencrypted. AI and the supply chain doom spiral We’ve dealt with problems like unencrypted credentials, unpinned dependencies, and CI footguns forever. But AI has accelerated everything, including repeating security mistakes. On the day of the Trivy compromise, I asked Claude, “how do I scan docker registry images for security vulnerabilities?” The reply, in part: CI/CD Integration Example (GitHub Actions with Trivy) - name: Scan image for vulnerabilities uses: aquasecurity/trivy-action@master Broken in two ways: Unpinned references – master is a reference that changes all the time. If hackers zombify the repo, I’d be the first victim. Active vulnerability – No mention whatsoever of the CVE posted that day. I never asked, so Claude never checked. Meanwhile, Vercel’s CEO has attributed his company’s recent data breach to a hacker that was “accelerated by AI.” And Anthropic’s latest hype tour includes briefing the US Federal Reserve Chair about vulnerabilities unearthed by their frontier model. Bad guys with LLMs get superpowers. Good guys with LLMs fall prey to mid-2010’s CI problems. And the same tool that can root out 27-year-old security problems in OpenBSD, will still tell you to pin your GitHub actions to @master. My GitHub Actions example is a simpler verison of the action removed in aquasecurity/trivy #10259.↩︎
.title {text-wrap:balance;} #content > p:first-child {text-wrap:balance;} If Git had a nemesis, it’d be large files. Large files bloat Git’s storage, slow down git clone, and wreak havoc on Git forges. In 2015, GitHub released Git LFS—a Git extension that hacked around problems with large files. But Git LFS added new complications and storage costs. Meanwhile, the Git project has been quietly working on large files. And while LFS ain’t dead yet, the latest Git release shows the path towards a future where LFS is, finally, obsolete. What you can do today: replace Git LFS with Git partial clone Git LFS works by storing large files outside your repo. When you clone a project via LFS, you get the repo’s history and small files, but skip large files. Instead, Git LFS downloads only the large files you need for your working copy. In 2017, the Git project introduced partial clones that provide the same benefits as Git LFS: Partial clone allows us to avoid downloading [large binary assets] in advance during clone and fetch operations and thereby reduce download times and disk usage. – Partial Clone Design Notes, git-scm.com Git’s partial clone and LFS both make for: Small checkouts – On clone, you get the latest copy of big files instead of every copy. Fast clones – Because you avoid downloading large files, each clone is fast. Quick setup – Unlike shallow clones, you get the entire history of the project—you can get to work right away. What is a partial clone? A Git partial clone is a clone with a --filter. For example, to avoid downloading files bigger than 100KB, you’d use: git clone --filter='blobs:size=100k' <repo> Later, Git will lazily download any files over 100KB you need for your checkout. By default, if I git clone a repo with many revisions of a noisome 25 MB PNG file, then cloning is slow and the checkout is obnoxiously large: $ time git clone https://github.com/thcipriani/noise-over-git Cloning into '/tmp/noise-over-git'... ... Receiving objects: 100% (153/153), 1.19 GiB real 3m49.052s Almost four minutes to check out a single 25MB file! $ du --max-depth=0 --human-readable noise-over-git/. 1.3G noise-over-git/. $ ^ 🤬 And 50 revisions of that single 25MB file eat 1.3GB of space. But a partial clone side-steps these problems: $ git config --global alias.pclone 'clone --filter=blob:limit=100k' $ time git pclone https://github.com/thcipriani/noise-over-git Cloning into '/tmp/noise-over-git'... ... Receiving objects: 100% (1/1), 24.03 MiB real 0m6.132s $ du --max-depth=0 --human-readable noise-over-git/. 49M noise-over-git/ $ ^ 😻 (the same size as a git lfs checkout) My filter made cloning 97% faster (3m 49s → 6s), and it reduced my checkout size by 96% (1.3GB → 49M)! But there are still some caveats here. If you run a command that needs data you filtered out, Git will need to make a trip to the server to get it. So, commands like git diff, git blame, and git checkout will require a trip to your Git host to run. But, for large files, this is the same behavior as Git LFS. Plus, I can’t remember the last time I ran git blame on a PNG 🙃. Why go to the trouble? What’s wrong with Git LFS? Git LFS foists Git’s problems with large files onto users. And the problems are significant: 🖕 High vendor lock-in – When GitHub wrote Git LFS, the other large file systems—Git Fat, Git Annex, and Git Media—were agnostic about the server-side. But GitHub locked users to their proprietary server implementation and charged folks to use it.1 💸 Costly – GitHub won because it let users host repositories for free. But Git LFS started as a paid product. Nowadays, there’s a free tier, but you’re dependent on the whims of GitHub to set pricing. Today, a 50GB repo on GitHub will cost $40/year for storage. In contrast, storing 50GB on Amazon’s S3 standard storage is $13/year. 😰 Hard to undo – Once you’ve moved to Git LFS, it’s impossible to undo the move without rewriting history. 🌀 Ongoing set-up costs – All your collaborators need to install Git LFS. Without Git LFS installed, your collaborators will get confusing, metadata-filled text files instead of the large files they expect. The future: Git large object promisors Large files create problems for Git forges, too. GitHub and GitLab put limits on file size2 because big files cost more money to host. Git LFS keeps server-side costs low by offloading large files to CDNs. But the Git project has a new solution. Earlier this year, Git merged a new feature: large object promisers. Large object promisors aim to provide the same server-side benefits as LFS, minus the hassle to users. This effort aims to especially improve things on the server side, and especially for large blobs that are already compressed in a binary format. This effort aims to provide an alternative to Git LFS – Large Object Promisors, git-scm.com What is a large object promisor? Large object promisors are special Git remotes that only house large files. In the bright, shiny future, large object promisors will work like this: You push a large file to your Git host. In the background, your Git host offloads that large file to a large object promisor. When you clone, the Git host tells your Git client about the promisor. Your client will clone from the Git host, and automagically nab large files from the promisor remote. But we’re still a ways off from that bright, shiny future. Git large object promisors are still a work in progress. Pieces of large object promisors merged to Git in March of 2025. But there’s more to do and open questions yet to answer. And so, for today, you’re stuck with Git LFS for giant files. But once large object promisors see broad adoption, maybe GitHub will let you push files bigger than 100MB. The future of large files in Git is Git. The Git project is thinking hard about large files, so you don’t have to. Today, we’re stuck with Git LFS. But soon, the only obstacle for large files in Git will be your half-remembered, ominous hunch that it’s a bad idea to stow your MP3 library in Git. Edited by Refactoring English Later, other Git forges made their own LFS servers. Today, you can push to multiple Git forges or use an LFS transfer agent, but all this makes set up harder for contributors. You’re pretty much locked-in unless you put in extra effort to get unlocked.↩︎ File size limits: 100MB for GitHub, 100MB for GitLab.com↩︎
Any code of your own that you haven’t looked at for six or more months might as well have been written by someone else. – Eagleson’s Law After scouring git history, I found the correct config file, but someone removed it. Their full commit message read: Remove config. Don't bring it back. Very. helpful. But I get it; it’s hard to care about commit messages when you’re making a quick change. Git commit templates can help. Commit templates provide a scaffold for your commit messages, reminding you to answer questions like: What problem are you solving? Why is this the solution? What alternatives did you consider? Where can I read more? What is a git commit template? When you type git commit, git pops open your text editor1. Git can pre-fill your editor with a commit template—a form that reminds you of everything it’s easy to forget when writing a commit. Creating a commit template is simple. Create a plaintext file – mine lives at ~/.config/git/message.txt Tell git to use it: git config --global \ commit.template '~/.config/git/message.txt' My template packs everything I know about writing a commit. Project-specific templates Large projects, such as Linux kernel, git, and MediaWiki, have their own commit guidelines. Git templates can remind you about these per-project requirements if you add a commit template to a project’s .git/config file. Another way to do this is git’s includeIf configuration setting. includeIf lets you override git config settings when you’re working under directories you define. For example, all my Wikimedia work lives in ~/Projects/Wikimedia and at the bottom of my ~/.config/git/config I have: [includeIf "gitdir:~/Projects/Wikimedia/**"] path = ~/.config/git/config.wikimedia In config.wikimedia, I point to my Wikimedia-specific commit template (along with other necessary git settings: my user.email, core.hooksPath, and a pushInsteadOf url to push to ssh even when I clone via https). Forge-specific templates Personal git commit templates lead to better commits, which make for a better history. The forge-specific pull-request templates are a band-aid, the cheap kind that falls off in the shower. There’s no incentive for GitHub to make git history better: the worse your commit history, the more you rely on GitHub. Still, all the major pull-request-style forges let you foist a pull-request template on your contributors. As a contributor, I dislike filling those out—they add unnecessary friction. Commit message contents Your commit template allows you do the hard thinking upfront. Then, when you make a commit, you simply follow the template. My template asks questions I answer with my commit message: 72ch. wide -------------------------------------------------------- BODY # | # - Why should this change be made? | # - What problem are you solving? | # - Why this solution? | # - What's wrong with the current code? | # - Are there other ways to do it? | # - How can the reviewer confirm it works? | # | # ---------------------------------------------------------------- /BODY But other clever folks cooked up conventions you could incorporate: Conventional commits – how do your commits relate to semantic versioning? This makes it easier for SRE and downstream users. Problem/Solution format – first pioneered by ZeroMQ2, this format anticipates the questions of future developers and reviewers. Gitmoji – developed for the GitHub crowd, this format defines an emoji shorthand that makes it easy to spot changes of a particular type. Commit message formatting How you format text affects how people read it. My template also deals with text formatting rules3: Subject – 50 characters or less, capitalized, no end punctuation. Body – Wrap at 72 characters with a blank line separating it from the subject. Trailers – Standard formats with a blank line separating them from the body. People will read your commit in different contexts: git log, git shortlog, and git rebase. But git’s pager has no line wrapping by default. I hard wrap at 72 characters because that makes text easier to read in wide terminals.4 Finally, my template addresses trailers, reminding me about standard trailers supported in the projects I’m working on. Git can interpret trailers, which can be useful later. For example, if I wanted a tab-separated list of commits and their related tasks I could find that with git log: $ TAB=%x09 $ BUG_TRAILER='%(trailers:key=Bug,valueonly=true,separator=%x2C )' $ SHORT_HASH=%h $ SUBJ=%s $ FORMAT="${SHORT_HASH}${TAB}${BUG_TRAILER}${TAB}${GIT_SUBJ}" $ git log --topo-order --no-merges \ --format="$FORMAT" d2b09deb12f T359762 Rewrite Kurdish (ku) Latin to Arabic converter 28123a6a262 T332865 tests: Remove non-static fallback in HookRunnerTestBase 4e919a307a4 T328919 tests: Remove unused argument from data provider in PageUpdaterTest bedd0f685f9 objectcache: Improve `RESTBagOStuff::handleError()` 2182a0c4490 T393219 tests: Remove two data provider in RestStructureTest Git commit templates free your brain from remembering what you should write, allowing you to focus on the story you should tell. Your future self will thank you for the effort. Starting with core.editor in your git config, $VISUAL or $EDITOR in your shell, finally falling back to vi.↩︎ I think…↩︎ All cribbed from Tim Pope↩︎ Another story I’ve heard: a standard terminal allows 80 characters per line. git log indents commit messages with 4 spaces. A 72-character-per-line commit centers text on an 80-character-per-line terminal. To me, readability in modern terminals is a better reason to wrap than kowtowing to antiquated terminals.↩︎
[The] Linux kernel uses GPLv2, and if you distribute GPLv2 code, you have to provide a copy of the source (and modifications) once someone asks for it. And now I’m asking nicely for you to do so 🙂 – Joga, bbs.onyx-international.com Boox in split screen, typewriter mode In January, I bought a Boox Go 10.3—a 10.3-inch, 300-ppi, e-ink Android tablet. After two months, I use the Boox daily—it’s replaced my planner, notebook, countless PDF print-offs, and the good parts of my phone. But Boox’s parent company, Onyx, is sketchy. I’m conflicted. The Boox Go is a beautiful, capable tablet that I use every day, but I recommend avoiding as long as Onyx continues to disregard the rights of its users. How I’m using my Boox My e-ink floor desk Each morning, I plop down in front of my MagicHold laptop stand and journal on my Boox with Obsidian. I use Syncthing to back up my planner and sync my Zotero library between my Boox and laptop. In the evening, I review my PDF planner and plot for tomorrow. I use these apps: Obsidian – a markdown editor that syncs between all my devices with no fuss for $8/mo. Syncthing – I love Syncthing—it’s an encrypted, continuous file sync-er without a centralized server. Meditation apps1 – Guided meditation away from the blue light glow of my phone or computer is better. Before buying the Boox, I considered a reMarkable. The reMarkable Paper Pro has a beautiful color screen with a frontlight, a nice pen, and a “type folio,” plus it’s certified by the Calm Tech Institute. But the reMarkable is a distraction-free e-ink tablet. Meanwhile, I need distraction-lite. What I like Calm(ish) technology – The Boox is an intentional device. Browsing the internet, reading emails, and watching videos is hard, but that’s good. Apps – Google Play works out of the box. I can install F-Droid and change my launcher without difficulty. Split screen – The built-in launcher has a split screen feature. I use it to open a PDF side-by-side with a notes doc. Reading – The screen is a 300ppi Carta 1200, making text crisp and clear. What I dislike I filmed myself typing at 240fps, each frame is 4.17ms. Boox’s typing latency is between 150ms and 275ms at the fastest refresh rate inside Obsidian. Typing – Typing latency is noticeable. At Boox’s highest refresh rate, after hitting a key, text takes between 150ms to 275ms to appear. I can still type, though it’s distracting at times. The horror of the default pen Accessories Pen – The default pen looks like a child’s whiteboard marker and feels cheap. I replaced it with the Kindle Scribe Premium pen, and the writing experience is vastly improved. Cover – It’s impossible to find a nice cover. I’m using a $15 cover that I’m encasing in stickers. Tool switching – Swapping between apps is slow and clunky. I blame Android and the current limitations of e-ink more than Boox. No frontlight – The Boox’s lack of frontlight prevents me from reading more with it. I knew this when I bought my Boox, but devices with frontlights seem to make other compromises. Onyx The Chinese company behind Boox, Onyx International, Inc., runs the servers where Boox shuttles tracking information. I block this traffic with Pi-Hole2. pihole-ing whatever telemetry Boox collects I inspected this traffic via Mitm proxy—most traffic was benign, though I never opted into sending any telemetry (nor am I logged in to a Boox account). But it’s also an Android device, so it’s feeding telemetry into Google’s gaping maw, too. Worse, Onyx is flouting the terms of the GNU Public License, declining to release Linux kernel modifications to users. This is anathema to me—GPL violations are tantamount to theft. Onyx’s disregard for user rights makes me regret buying the Boox. Verdict I’ll continue to use the Boox and feel bad about it. I hope my digging in this post will help the next person. Unfortunately, the e-ink tablet market is too niche to support the kind of solarpunk future I’d always imagined. But there’s an opportunity for an open, Linux-based tablet to dominate e-ink. Linux is playing catch-up on phones with PostmarketOS. Meanwhile, the best e-ink tablets have to offer are old, unupdateable versions of Android, like the OS on the Boox. In the future, I’d love to pay a license- and privacy-respecting company for beautiful, calm technology and recommend their product to everyone. But today is not the future. I go back and forth between “Waking Up” and “Calm”↩︎ Using github.com/JordanEJ/Onyx-Boox-Blocklist↩︎
More in programming
My now-traditional “summary” of the conference along with my own commentary and thoughts.
Every year, browser vendors collaborate via the Interop project to synchronize their efforts on new HTML/CSS/JS features. This year, there’s a ton of super exciting features being considered! In this blog post, we’ll explore the CSS of the future, to get a glimpse of what we’ll soon be able to do.
Imprint’s average engineer is merging 4x more PRs than 2 years ago. Yay. Aggregated behind the average is something even more interesting: we have two teams that are merging 10x their PRs over that period. One engineer is merging 25x the PRs per week of our average engineer two years ago, and a number are merging 10x. That’s pretty neat, and the slope of the line suggests our average in a year is going to look much closer to today’s p90 than today’s p50. Even more importantly, we aren’t just shipping more, our quality has improved: incidents are down 42% half-over-half, and critical incidents are down 78% half-over-half. I think we’ve always known the quality/velocity tradeoff wasn’t a particularly useful construct, but that’s undeniable today. As we’ve walked this path, we–like everyone else progressing down it–have run into the problem of pull request review. The p50 pull request review took about 33 minutes to complete 2 years ago, but now takes us about 77 minutes. The slope here correlates tightly with the number of pull requests getting merged. Tightly enough that addressing it became one of my priorities for the week. We solved it! Just kidding, we’ve absolutely not solved it, but I think our current approach is an interesting one, which we call the small-change. The work discussed here was shaped and done by many folks at Imprint including Dan, Joy, Daniel, Lenny, Taylor, Ted, and Annie, among many others. Why review pull requests? There are four reasons we review pull requests: Knowledge sharing between author and reviewer – software engineering is generally learned via apprenticeship, and reviews are a primary mechanism of experienced engineers teaching others what good software looks like. We believe the current volume means that this knowledge sharing will have to move towards operating on clusters of pull requests, rather than individual pull requests Preventing errors – review has always been an expensive time to catch errors, and we believe that effective CICD along with harness verification techniques like poteto’s pstack are more effective mechanisms for preventing errors than review Preventing malicious behavior – my experience is that even careful code review will generally miss competent malicious behavior, although fortunately most malicious behavior is incompetent. We believe both sorts of behavior are best identified using a combination of fixed rules (modifying these specific objects/classes/files requires review), anti-abuse mechanisms that we’ve refined in our LLM-based member support experiences, auditing, and audit log inspection Compliance obligations in our documented controls – these can be changed, but today they require that an independent human reviews changes in a subset of our repositories In the modern development paradigm, we think there is a better solution for each of these problems. But, we also want to make these changes very cautiously. Small steps with small-change We believe we move fast by finding narrow, contained ways to validate new practices, so to iterate our way into an effective solution, we’ve developed the concept of a small-change. Here’s how it works: When a pull request is marked ready, CICD determines whether it is eligible to be a small-change. This is represented as a passing Github check. There is also a label for visibility, but it is the check itself, managed by the build and not human modifiable, that determines eligibility. (The label is used for initial filtering, but the check is still independently verified.) Something is a small-change if it meets all of these criteria: It is less than 1,000 lines of code There are no p1 or p2 findings from our ai-code-review automated review bot The author has “experience” in the service or application, defined as 10+ prior merged pull requests into that service or application’s directories. For changes working across multiple services or applications, the author must have experience in all of them to qualify Impacts fewer than N distinct services or applications. N is 1 on our frontend monorepo, and N is 3 in our backend monorepo, where typical changes will modify both a gRPC service and a common HTTP gateway Shared libraries and utilities files are not eligible for small-change Each repository has a .small_change file which allows per-path customization while we continue to tune the behaviors. For example, a service dealing with money movement can be excluded Our Slack automation tags every reference to a pull request with reacji labeling the state of that pull request, for example if it has failing tests, if it has been approved but tests are still running, or if it is ready to merge. We added a small-change reacji that indicates which of those pull requests meet the small-change criteria In our internal skills repository, we created two skills that have been distributed across harnesses: review-small-change which takes a list of repositories and reviews all open, non-draft, small-changes, including providing approval if they meet the review’s criteria (all checks pass including small-change check, and not touching on sensitive topics such as payments and security) review-pr-cluster which reviews batches of pull requests to identify common themes that came up, and suggest opportunities for discussion and adding verifications We set the social expectation that any error caused by a small-change is a failure in our system, and has to be solved at the system level: improvements to ai-code-review, to our verification mechanisms, to our CICD, to our harness, etc. This took about two days of infrequent attention to implement and rollout, although I imagine we will be tuning the knobs here for a long time. Maybe forever. fleet: centrally orchestrated harness I’ll briefly mention our other approach to reducing code review volume, which anchors on our centrally-orchestrated harness, fleet. (Which is similar to Stripe’s Minions although not following the one-shot approach in particular). We rely on fleet to implement and verify software in a highly controlled harness running in the cloud, with independent evaluation mechanisms to identify the correct number of independent reviewers for any given change. That number can be as low as one (e.g. fleet writes, one human reviews) for changes evaluated as safe. Results / next steps / reception I’ll be honest, it’s too early to give a useful result here. From early results, about 10% of changes qualify as a small-change. Those are now eligible for very fast approval, and generally get it unless they trip a flag, generally due to touching a security or money adjacent topic. While 10% is a small percentage, the core concept here is working, and ultimately will succeed or fail based on our ability to expand that percentage over the next few months in tandem with expanding our verification mechanisms to be increasingly comprehensive. This is ultimately going to fail or succeed depending on our ability to improve our verifications, which does require more difficult sorts of techniques to accomplish in low scale-complexity, high domain-complexity software than in the high scale-complexity, low domain-complexity products where many techniques like incremental rollouts are exceptionally effective. (Don’t get me wrong, they still work, they just provide less coverage when most users have different configuration and behavior.) Folks have not been uniformly excited about this, it’s a substantial change from how we’ve learned to be good engineers, and it’s uncomfortable to reevaluate whether a long-held best practice has become merely dogma. The most energizing part of the current cycle is that we get to reevaluate these practices we grew up on every week; the most painful part of the current cycle is that we’re likely to fall behind if we ever stop reevaluating them.
The park near my house has lots of chestnuts and my kids have been collecting them like crazy (it’s actually their favorite post-school activity). We accumulated quite a hoard. I thought it would be fun to make a launcher so we had something to do with our new chestnut collection. I wanted to go with a very simple crossbow-style design (like the famous D2 item) using latex tubing and a leather pouch for holding the chestnuts. This is what I came up with. It launches chestnuts about 220 feet, and it’s simple enough to build in an afternoon. The frame is a 42 inch 2x2 and a spare IKEA bed slat for the crossbeam, glued and screwed together. A 3/8 inch eye bolt goes through each end of the crossbeam, and the latex tubing is tied to those and to the pouch from a leather shepherd’s sling. Tying the tubing was the fiddliest part. It has to be stretched while you wrap it with butcher’s twine, so I clamped a couple of scrap blocks to my workbench to hold it. I like this design because the tubing can be wrapped around the eye bolts to easily add more tension and power. The Trigger My favorite part of the design was that I decided to use an archery release like this one for the trigger. It provides a safe and reliable way to hold the tubing back, and it’s made to let go cleanly. I 3D printed a little housing that mounts the release securely to the back of the stock, and the release clips onto a loop of archery D-loop cord on the back of the pouch. Then I added a dowel for a rear grip and an old handle I had previously removed from my shop vac. How Well Does It Work? We tried it at the park and it works surprisingly well. Aimed at about 35 to 40 degrees, it sends our chestnuts (on the small side, about 10 to 15 grams) about 220 feet. Thanks to the archery release, we haven’t had a single misfire. I got curious and did some rough math. To go that far with air drag, the chestnuts are leaving at somewhere around 85 to 100 mph, which works out to about the same energy as a paintball. So safety glasses are important and you never point it at anyone. You can find the 3D print on Printables and a full build guide on Instructables. Make sure to always use it with adult supervision :)
I have a lot of digital cross-stitch patterns and until recently they were a mess. My “collection” was a series of several hundred PDFs, disorganised and spread across my disk. Recently, I was looking for a pattern and it took me several minutes to find it. I had a brief moment of panic, thinking that I’d lost it, and the website this pattern came from is long since gone. Fortunately it did turn up, but it was clear I had to organise my pattern collection. Some people would use apps like Pattern Keeper or Flosshaven, but I just created another static website on my media drive. This is what it looks like: The patterns are arranged in a masonry layout with a preview thumbnail for each pattern, which is normally the first page of the PDF. If a pattern has multiple files – for example, different files for colour and black-and-white printing – there’s a list of links below the thumbnail. I can also add a source link to where I bought the pattern, if I want to find more patterns by the same author. Internally, it’s structured like my other static websites. The metadata for the patterns is stored in a JavaScript file, and then I have an HTML page which renders that metadata in this visual design. My pattern collection doesn’t change very often, so I’m adding new files by hand. There are no sorting or filtering options, because my collection is small enough that they’re unnecessary. I can scroll through the entire list in less than a minute and find whatever I need. I’d like to start making my own clothes soon, and I expect to copy this design to store clothing patterns. This collection now sits alongside my saved bookmarks, podcasts, TV shows, and more. I can keep all my patterns organised, and I won’t lose them in the depths of my hard drive. Technical tricks There are a couple of technical pieces I’m quite pleased with in this process: Masonry layout. I love the cascading grid style of a masonry layout, and I think they’re great when you have variable-sized content – like my pattern cards. I’ve been using them for over a decade; screenshots of my early web apps show my experiments with David Desandro’s Masonry.js library. Masonry layouts are now a native feature of CSS called “Grid Lanes”, which is simpler and more performant than the JavaScript approach. I’m excited to use these layouts in even more places. Thumbnails of PDFs. Most digital postage patterns are shipped as PDFs, and I’m creating the thumbnail images with my create_thumbnail tool. I had to extend my tool to support PDFs, which turned out to be straightforward with the pdf_oxide crate. I’d not come across Oxide before, but it’s a cross-language library for working with PDFs, and I can imagine using it in future. A coloured stripe for the palette. In the metadata for each project, I have the complete list of DMC thread colours and the pattern dimensions. I have a lookup table that converts DMC thread colours to hex codes, and then I’m able to create a linear gradient in CSS to render a coloured stripe of the pattern colours. Here’s an example from a pattern with four colours: border-bottom: 12px solid transparent; border-image: linear-gradient( to right, #7a451f 0%, #7a451f 25%, #985e33 25%, #985e33 50%, #e4bb8e 50%, #e4bb8e 75%, #361f0e 75%, #361f0e 100% ) 1; Building the HTML, CSS, and JavaScript was very quick – I have notes on how I build my media sites, and I’ve done it plenty of times before. I threw together the initial prototype in less than 15 minutes. The longer part was finding all the PDFs scattered across my disk – but now that’s done, and I know exactly where to find my patterns. [If the formatting of this post looks odd in your feed reader, visit the original article]
