2026-08-01 · thundersnap v0.01: an undo button for everything

Transcript

Transcript. Paraphrased; sources in notes.md.

HostWelcome to August 1st, 2026. I am Daniel, and joining me today is our resident systems skeptic, Fred. We have quite a spread on the menu, covering everything from experimental distributed filesystems to the existential dread of modern code review, and wrapping up with the privacy implications of iOS device fingerprinting. Fred, I saw you eyeing the Thundersnap link before we started recording. You looked like you were ready to start debugging it before even reading the documentation.

GuestDaniel, the name alone, Thundersnap, is a red flag. It is essentially a marketing department admitting they are about to perform a controlled demolition on your production data. They explicitly warn that it might lose half your data. I appreciate the honesty, really, but it is the kind of project where the author says it is just a toy, and then three years later, it is running a critical banking backend somewhere. I have seen this movie before.

HostWell, let us lean into that. Thundersnap is an experimental distributed filesystem that uses content-addressed snapshots instead of real-time syncing. It essentially lets you bundle your entire environment, your root, home, and work files, into a container and move it between nodes using Tailscale. Tailscale, for those who haven't used it, is a popular mesh VPN service that simplifies secure connections between devices. The author claims it is designed to be fun, not careful. Over on Lobsters, though, the conversation took a sharp turn. Some users are calling it slop just because of a CLAUDE.md file in the repo.

GuestYes, the classic aesthetic judgment. A user named natkr flagged it as slop because of that file, which is just a set of instructions for an AI to help maintain the code. It is a bit cynical to write off the architecture based on the developer’s choice of text editor assistants, though I get the frustration. If the system is as unstable as the author claims, I suspect the bugs will be far more interesting than the commit history. The real problem is the premise. Content-addressing your entire filesystem sounds elegant until you start dealing with the realities of mutation and orphan collection. But sure, let us all play with it, as long as it stays on your local machine and nowhere near anything that actually matters.

HostMoving from experimental storage to the daily grind of code review. There is an interesting post floating around titled Make Reviews Possible Again With This One Simple Trick, which advocates for using stacked branches to manage large code changes. The author’s claim is that AI-generated code tends to be massive and overwhelming, and breaking it into a stack of smaller, dependent branches makes it easier for humans to digest.

GuestStacking is the industry’s favorite way to hide the fact that we have stopped thinking about architecture. A Lobsters user, singpolyma, pointed out that the issue is not really the workflow but that GitHub’s tooling for reviews has always been fundamentally broken. We are using these branch stacks as a salve for the fact that we are generating too much code, too fast, with too little context. If your diff is so big that you need a complex branching strategy just to get a human to look at it, you have already lost. The solution is not better branch management, it is writing less code.

HostIt is a valid point. The thread actually turned into a bit of a meta-discussion about the site itself, with one commenter, repl, noting that the title felt like classic clickbait and questioning the author’s tendency to post their own work. The author, noon, jumped in to defend the frequency of their contributions. It is a reminder that even in technical circles, the delivery often gets as much attention as the engineering.

GuestThat is the nature of the beast. We want the trick, the hack, the shortcut. But when it comes to code review, there is no shortcut for actually understanding what the computer is supposed to do. If you dump a giant AI-generated blob into a stack, you are still responsible for the integration mess at the end of the chain.

HostLet us close on privacy. There is a project called Loupe, which is an open-source iOS app that exposes how much information your device is leaking to third-party apps through public APIs. It categorizes these signals into categories like passive data, things the app can just see, and permission-based data. It is meant to raise awareness about how these seemingly innocent data points act like a fingerprint to track you.

GuestThis is actually useful, in a terrifying way. Privacy on mobile has become this game of whack-a-mole. You block location, so they track your clipboard usage count or your battery status or your screen resolution. A user named schuelermine in the Lobsters thread noted that the clipboard counter looks useless for gathering real info, but another user, gerow, had the perfect counterpoint. It is not about one signal being unique. It is about the longitudinal data. If you have a specific clipboard paste count, and you match that with a device ID, you have a solid anchor to track that person across sessions. It is the aggregate that kills you.

HostIt is a reminder that the surface area for tracking is much larger than just GPS or contacts. I suppose if you are truly worried about your fingerprint, the only solution is to never pick up the device in the first place.

GuestExactly. Or just accept that you are living in a giant, automated panopticon and enjoy the convenience of the apps anyway. I think I will stick to my terminal-based life for the weekend. I have some code that needs fixing, and luckily, it is not nearly large enough to require a stack of branches.

HostThat sounds like a quiet weekend. Thank you, Fred, for keeping us honest today. All the stories we discussed can be found on Lobsters, and we will be back tomorrow with more. Thanks for listening.