rain 🌦️

Well I guess the LLM slopocalypse is finally starting to descend on nextest

lift yr skinny fists

like antennas

Recent example was, I started pulling a thread and found that the DB was missing a fairly important uniqueness constraint that an upcoming feature would sometimes hit. (iddqd is awesome because you can express uniqueness in the type system)

I'd have 100% missed the defect had I not seen Claude leave a suspicious TODO marker in a spot and start probing it

The two big projects I work on: * the update system at Oxide. If this breaks, our product cannot be updated any more. The entire update process is autonomous, too * cargo-nextest. ~a quarter of professional Rust devs use it. If this breaks, many people have a bad time So yes, I look at the code

There's a few random smaller things that I don't pay as close attention to, often things for my own use. But for things I'm held accountable for, I'm compelled to pay attention. adversarial automated review tends to catch most defects, but there's so much more to building for the long term

Regent run with triple Comet+ and a Genesis last night. Holy shit, annihilated everything

I think this is just the UK being a backwards culture tbh

Switched to a Sennheiser HD 480 Pro -- honestly I can't imagine needing anything beyond this for wired closed-back headphones. They sound so good and are so remarkably comfortable

Reminder that "formally proved program X" is a meaningless statement. You formally prove that some program or a reduced model of it upholds some kind of property or invariant

Where formal verification can also be useful is in exploring arbitrary events and interleavings in a distributed system. Today's experience:

Reminder that "formally proved program X" is a meaningless statement. You formally prove that some program or a reduced model of it upholds some kind of property or invariant

One rule of thumb is that formal verification matters more when assuming the program is arbitrarily adversarial. For iddqd I do assume this, which is why I was happy to be able to formally verify some important properties: oxide.computer/blog/iddqd-u... But in general, you don't have to assume this

Reminder that "formally proved program X" is a meaningless statement. You formally prove that some program or a reduced model of it upholds some kind of property or invariant

Formal verification is powerful but it has quite significant tradeoffs. For example, a proof often just takes a while to run (especially if it's a more lightweight form like model checking which exhaustively enumerates every state). Shrinking counterexamples is annoying

Reminder that "formally proved program X" is a meaningless statement. You formally prove that some program or a reduced model of it upholds some kind of property or invariant

To be fair, nextest does the same :) Not because Rust doesn't have fork, but because fork is slow and messy

A thing I realized at some point is that a lot of the paranoia that has held me back from building certain things was completely justified. I used to admire others who could pull it off, but at some point it clicked that many of them just didn't know what they didn't know

So eMudhra is exiting the webpki business because the scrutiny they got was out of proportion. you affect the entire internet so your practices should be to a global standard. but also, they still have a large domestic digital signing business which might also be irresponsibly run

Real concern is that a lot of businesses just aren't at the caliber and skill level required to do this stuff