Why is it that MacOS detects my keyboard as a JIS layout one, but refuses to apply the layout so all the punctuation is in the wrong places, and the only solution is to install a `.keylayout` file? Even Windows lets me select my keyboard layout manually. msyk.net/macos/winkey...

How do you fuck up AF_UNIX to the point where you can kernel panic from userland, when you steal your OS code from 30+ years of prior work (Not a Rust Bug, but an Apple bug). github.com/rust-lang/ru...

In a further sign of Japan's descent into the grim dark cyberpunk future, the Tokyo Govt is now running an AI powered matchmaking service/dating app. Signing up requires photo ID, proof of income, govt. issued proof of being single, and money. www.futari-story.metro.tokyo.lg.jp/ai-matching/

And people say that panaceas are a myth.

Real life has been "interesting" for the last month in a not amazing way, apologies in advance if responses from me are going to be somewhat slower than usual.

If the TV loicense people are hurting so much for money as the news says, how can they afford to send me junk mail asking me to get a TV loicense. I do not own a TV or any device capable of receiving broadcast TV. NHK can Fuck Off.

Switching Off your Device Does Not Protect Against Fault Attacks (Paul Grandamme, Pierre-Antoine Tissot, Lilian Bossuet, Jean-Max Dutertre, Brice Colombier, Vincent Grosso) ia.cr/2024/1123

It's kind of impressive that the only package I need to implement Dual_EC_DRBG in Go "securely" is @filippo.abyssdomain.expert 's bigmod package, since the `crypto/ecdh` package exposes just enough for the rest. nb: I do use `crypto/elliptic` for the scalar-field order, but that can be removed.

I just spent my afternoon trying to figure out how to use HTTP over AF_UNIX in Rust, and ran into a giant tire fire. My brain lacks wrinkles for Rust and it's ecosystem. ps: Thanks to comments in a still-open bug in reqwest from 2017, I think I figured it out at least.

Time to find out how much MS uses Github repos for training Copilot, and how many people blindly copy-paste code for things that they shouldn't.

Neat.

Sending me a profanity laced tirade about how your software broke because a downstream dependency of something I maintain moved it's repo, is not an effective way to motivate me to help you.

www.bbc.com/news/article... 10k -> 13k SHU is enough to prompt a Govt. initiated recall? WTF.

Well damn, the era of compiler-introduced timing side-channels is here. In the Kyber reference implementation, Clang notices a bitmask is just selecting between zero and a constant and turns it into an if. Very happy the Go compiler is not that smart right now. groups.google.com/a/list.nist....

An old friend noted that there is a distinct feeling of "I just lost a SAN point". Not only is that true, but lately I'm experiencing that a lot more than I should.

As an addendum, as far as personal (non-employer) projects go, Rust has been the safest language I have used because I rage-quit and delete everything due to friction before my code gets far enough to have bugs.

I feel like writing Go full-time for a decade has made me less capable of using difficult languages, but I like it this way - I can focus on what I'm trying to achieve instead.

After using Go for things that require concurrency and networking, dealing with async, tokio and hyper makes me seriously question my life decisions, and want to consume various mind altering substances in large quantities.

A ha. forums.developer.apple.com/forums/threa...

If I ever design a programming language, my benchmark for success will be "pervasive enough that my terrible design decisions lead to dedicated silicon workarounds for performance reasons". developer.arm.com/documentatio...

I like how trying to use dtruss has a reasonable chance of making my system unusable till I power-cycle the thing.

> My investigator's motto is not "99% Mythos Lore, 1% Sanity- don't push me" Has become an increasingly accurate description of my professional career over the years.

Comparing performance of RSA-2048 and secp256kr1 when: - Both are implemented in Mathematica - secp256k1: affine coordinate add/double - secp256k1: add then double scalar multiply Only shows that ECC is really fast, despite the best efforts to the contrary.

One of these years, I seriously should just give up on hosting my own e-mail. I just don't want to deal with the headache of migrating my mailserver state. :/

A Note on Quantum Algorithms for Lattice Problems (Omri Shmueli) ia.cr/2024/583

The Bandai HG ATM-09-ST kit was pretty fun to build, and VOTMS (and the spinoffs) are in my top 3 giant robot animes of all time.

re: eprint.iacr.org/2024/555.pdf, a quick tl;dr: non-fancy crypto fine (Kyber, Dilithium); (efficient) fancy crypto (besides ZKP's) may be boned ~forever (if this attack holds) (based on lattices / LWE specifically, when your q vs n has a sufficient gap)

Despite the automatic 2 month extension and using the "throw money at the problem and have an accountant do it" approach, having to file taxes twice every year fills me with immeasurable rage and hatred.

Share with Care: Breaking E2EE in Nextcloud (Martin R. Albrecht, Matilda Backendal, Daniele Coppola, Kenneth G. Paterson) ia.cr/2024/546

I have a soft spot for hobbyist projects. github.com/SerenityOS/s...

大草原

"Apologies, I cannot answer your maintenance inquiries for international security reasons."

Reading the timeline of the pressure campaign to convince the xz maintainer to hand over control is… awful. Merciless guilt-tripping. One lesson I’m taking from this is to be even more ruthless with blocks. Whining about maintenance? Blocked. Zero chances. research.swtch.com/xz-timeline

@durumcrustulum.com I read through the latest FROST IETF draft, and the bs255 Schnorr scheme isn't different in meaningful ways from Appendix C. The main difference is the choice of H2 and the inputs, but it would be easy to define FROST bs255.

Is there a way to make implementing all the Web PKI stuff not an absolute nightmare? Being doomed to dealing with P-384. P-521, RSA, ASN.1 and X.509 is depressing.

On the GoFetch note. It's depressing that the only way to get microcode updates for consumer AMD CPUs is via BIOS update. The linux-firmware/amd-ucode etc packages do absolutely nothing, because those only include updates for EPYC/Threadripper.

This is the draft bs255 specification. I will likely add to it at some point, and I still need to double check that the reference implementation matches the spec, and generate test vectors. Hopefully it's at least readable. gitlab.com/yawning/bs25...

Been using OSX on the side for about a week, and I don't totally hate it. - The scroll-wheel was inverted by default. - Choosing between "my tinfoil hat crinkles less" and "dtruss works" sucks, as does the 2x reboots to switch. - Super instead of Ctrl really screws with muscle memory.

How do you manage to fuck up your programming language's ecosystem to the point where there is a package manager for package managers?

Design question: How to handle cryptographically insignificant probability internal failures (Eg: deriving a nonce via wide-reduction, it produced 0)? Rejection sampling or abort? I currently do the former (read more from the XOF) with `MAY choose to abort instead`. > Abort, Retry, Fail?

bs255 progress: - Tweaked the synthetic nonce generation to include the per-signature domain separator (Fix for the deterministic signature case). - Added test cases so all reasonable cases/failure modes are covered. - Misc code cleanups. Next: - Write a specification document.

While batch verification is out of scope atm for bs255, 4-way parallel keccak (AVX2) fits in nicely when re-deriving the `e` values. If ds/msg in a batch are the same, `e = H(ds, msg, R, P) mod n` would allow more optimization for large ds and msg, but large is like > 89-ish bytes combined...

As promised: gitlab.com/yawning/bs255 This still is a work in progress and probably has stupid bugs, and needs: - More test cases. - A specification document. But, it was relaxing to write, and I think illustrates "it is possible to do better".

I went and wrote a Schnorr signature implementation on top of ristretto25519, because "Ed25519 also caused me a lot of pain". Given first-class support for domain separation, do I need explicit separate support for pre-hashed messages? ie: Why does Ed25519ph exist when Ed25519ctx exists?

It's been MORE THAN FIFTEEN YEARS and we are still grappling with the clusterfuck of Ed25519 underspecified validation criteria. I think "safecurves" might go down as my all-time pet peeve. A curve with a cofactor is not safe. ESPECIALLY if you don't specify how to handle it and provide no tests.

Am I the only one that finds that it takes more time documenting something than doing a functional/correct implementation?

Peak Shitcoins is the former Safemoon CEO being free on a 3M USD bond, to go live with his parents, and using a Public Defender because the law firm he originally retained withdrew representation for lack of payment.

Whowa. I find myself needing an OSX test-target, but even the low-end Mac mini quickly gets kind of expensive. Maybe I need to be looking at the used/refurb market.

Having used it for about a week, the Thonkpad X13 Gen 4 AMD is indeed, quite nice.