chrysn

🏳️‍🌈 Technology enthusiast with a focus on Free Software and embedded systems. Science fiction promised us general purpose electronics, let's build them! (And get their security properties right.)

Public Key

npub1fwh05jl2ya7xf0wdur38h7zyts272jnfyn6f22an9622wcxe86esqqnhyf

NIP-05 Address

chrysn@chaos-social.mostr.pub

Profile Code

nprofile1qqsyhth6f04zwlryhhx7pcnmlpz9c909ff5jfay49weja998vrvnavcpz4mhxue69uhhyetvv9ujuerfw36x7tnsw43q85rytk

Publishing to

relay.ditto.pub

Author Public Key

npub1fwh05jl2ya7xf0wdur38h7zyts272jnfyn6f22an9622wcxe86esqqnhyf

Show more details

Last Notes

2026-04-09 08:19:14 UTC

- reply

@nprofile…3eu7 Vielleicht durch wenige Datenpunkte, und eine Punkt-zu-Form-Konvertierung, die mehr Genauigkeit vorspielt, als da ist.

2026-03-25 12:54:13 UTC

Funny how people, in light of the LiteLLM compromise, jump to the conclusion that the solution is to make your supply chain even more intransparent by vendoring in dependencies through an LLM's processing, rather than just using pinned and vetted dependencies by default over tools' defaults "yolo there has been an update and it claims to be semver compatible" attitude.

2026-03-03 11:45:02 UTC

- reply

@nprofile…y9gw Maybe they learned from Bayrische Grenzpolizei.
https://assets.chaos.social/media_attachments/files/116/165/062/394/570/990/original/577c41e00f8b469b.jpg

2026-02-15 18:51:13 UTC

Web archiving is troubled: The Internet Archive is under attack constantly, and archive.is/archive.today seems to be on the way towards turning itself and its users into personae non gratae on the web.
Is it now time to turn web archiving into a distributed affair where obtaining the snapshot happens while browsing, and produces a witness statement a la "at this point in time, I observed these responses to those requests", and storing/accessing those becomes decoupled from obtaining them?

2026-02-11 10:22:35 UTC

- reply

@nprofile…jydu I recently opened https://github.com/typst/svg2pdf/issues/99 on that, which got closed as fix-it-upstream, so I'd "just" follow that path.

2026-02-11 07:37:09 UTC

- reply

@nprofile…jydu Are you sure on 3? Inkscape does a good job pruning data from outside the MediaBox these days.

2026-02-09 06:56:26 UTC

- reply

@nprofile…v673 I have a similar less sophisticated hook. Back when using calypso (a now discontinued fork of Radicale), that had really good commit messages (like, date for calendar entries, and a User-Agent entry to see which client broke things). I wish Radicale had this in scope and didn't just WontFix this with "everyone can write their own hook".

2026-01-28 07:09:29 UTC

- reply

@nprofile…mcwl I don't think you can get rid of that big core switch, but would be delighted to be shown wrong.

2026-01-27 21:24:34 UTC

- reply

@nprofile…mcwl I think in general you can't, you'll need methods like
fn as_service(&self) {
match self {
MyThing::A(a) => Service::APrime(a.into()),
...
}
}
etc.
If you don't want to create them on foot, macros make this reasonably easy.

2026-01-16 17:08:54 UTC

- reply

@nprofile…mudw @nprofile…pafg @nprofile…4esm Genaugenommen bedingt clone das nicht -- ARandR könnte auch die Beamerauflösung nativ einstellen, und scaling für's 2. Display verwenden -- gerade bei Photovorträgen kann das cool sein. Hab ich in ARandR aber leider noch nicht implementiert, und ist für die konkrete Zielgruppe vielleicht auch nicht so essenziell.

2025-11-21 06:09:17 UTC

- reply

@nprofile…mfuv I don't know of any way to count the threads, let alone to turn thread spawning off (esp. as that may not just happen through std::thread, but also through any C library).

2025-10-05 19:27:07 UTC

- reply

@nprofile…00z7 @nprofile…uy6d Zivilgesellschaftliche Organisationen kenn ich keine (nur technische), aber die Kollegys in Tschechien iaben da was vorgelegt: https://konecipv4.cz/en/

2025-10-05 07:30:33 UTC

I share the general concern from DJB in https://blog.cr.yp.to/20251004-weakened.html -- with much of the collective #IETF payroll being subject to the whims of US agencies, there is a risk of policy based strongarming-in of weak #cryptography.
For now, the concern may be premature: WGA is mainly about whether there is drive to do the work.
Until this is done, there are more important checks, especially before Recommended=Y: Tougher chair review, but also the shepherd report pointing out dissent, and the IESG itself.

2025-08-10 18:00:40 UTC

- reply

@nprofile…kxtw @nprofile…j8hy Would they be comfy as seat sized bean bags?

2025-07-25 12:29:17 UTC

- reply

@nprofile…nhea While you are joking: @nprofile…lrh8, what'd it take for an MPTCP host to advertise a v6 option and then unadvertise all prior v4 options?

2025-07-19 06:35:19 UTC

- reply

@nprofile…ss2n @nprofile…88lj Ah! I'm more of a wood person in that department.
https://assets.chaos.social/media_attachments/files/114/878/500/455/967/465/original/9c6eab0dc165a90c.jpg

2025-07-18 13:28:52 UTC

- reply

@nprofile…ss2n A whistle? (Not by design, but from the aspiration of using anything as a musical instrument).

2025-07-16 14:21:34 UTC

- reply

@nprofile…u7qg I don't know anything about UniFi; the Thread routers should (on the long run) follow the SNAC spec as far as I understand.
Do I see this right that it doesn't work for you because your multiple HomePod devices (those are the Thread border routers IIUC) don't see each other on the infrastructure network (where the UniFi router sits) because they are not on the same link (ie. because the UniFi router routes between them)?

2025-07-16 13:58:43 UTC

- reply

@nprofile…u7qg But maybe I misread https://www.ietf.org/archive/id/draft-ietf-snac-simple-07.html -- or they don't deploy it that way yet.

2025-07-16 13:34:57 UTC

- reply

@nprofile…u7qg That's odd, because I vaguely remember that at least on a spec l evel, they talk about advertising an IPv6 ULA on the WiFi, which should even work in regions where due to monopolies, ISPs get away with not offering v6.

2025-07-16 08:10:55 UTC

- reply

@nprofile…528n Do those connections go up high enough to have an official French fediverse instance that could host, say, @academie@fr, and beat the pope to it? (cf. https://chaos.social/@chrysn/114477389021336262)

2025-07-15 15:16:26 UTC

We're launching the embedded-cal project: Providing access to hardware accelerated and formally proven cryptographic algorithms on #embedded systems in #RustLang. For this, I'm teaming up with @nprofile…u6wh Paris and @nprofile…pkv8, supported by the #EU funded @nprofile…zzgr.
Right now we're going through requirements. If you want to add hardware accelerated security to your #IoT project, chime in at our wishlist issue: <https://github.com/lake-rs/embedded-cal/issues/1>
#EmbeddedCAL #NGI0Commons

2025-06-24 07:28:40 UTC

- reply

@nprofile…ss2n @nprofile…8j6a For the symlink trouble I like exposing the repo over WebDAV (eg. loopback nginx) -- helps my TV, and helped back when "copy some photos onto USB" was an everyday operation that'd fail due to Nautilus/Thunar blindly defaulting to `cp -a`.

2025-06-24 06:19:39 UTC

- reply

@nprofile…8j6a @nprofile…ss2n Having a remote sounds better because it is tracked; also, that'd enable getting data from there by other clients.
(E.g. I sometimes load files onto my laptop from a relative's NAS rather than my server just b/c their uplink has more bandwidth.)

2025-06-22 14:18:35 UTC

- reply

@nprofile…uy6d http://test-ipv6.com/ sollte das zeigen.

2025-06-22 13:46:03 UTC

- reply

@nprofile…uy6d Schaffen die Path MTU? Deren Disconter bob vergeigt da voll.

2025-06-21 11:27:01 UTC

- reply

@nprofile…ndcz Hate to break it to you, but last time I saw that was on our 2012'ish server, and while far be it for me to call the constructors "grown up", they've all been approaching 30 back then.

2025-06-17 07:09:05 UTC

- reply

@nprofile…8j6a Well, if it just says it's locking, then I think C already has a copy, it's just not locked, and locking is supposed to be fast so no need to elaborate by default. (There could probably be a different trust model where there is no LOCKCONTENT required; maybe if the remotes were trusted, that'd already suffice).

2025-06-17 06:33:56 UTC

- reply

@nprofile…8j6a What is your numcopies? Maybe it needs more copies than the new one to satisfy that.
And which remotes are trusted, semitrusted and untrusted?

2025-05-18 11:05:38 UTC

- reply

@nprofile…7093 Are the recipients really in a better position to judge what the network state is? The negotiation for block sizes is already ... I wouldn't say brittle, but def'ly not well tested. Adding an extra option and protocol would best be done if there is a clear and benchmarked problem (which there may well be and I don't see it yet)

2025-05-18 10:32:13 UTC

- reply

@nprofile…7093 Block size negotiation is mainly there to account for different supported MTUs in heterogenous 802.15.4 networks -- doesn't AX.25 provide that information already?

2025-05-17 17:19:41 UTC

- reply

@nprofile…8j6a @nprofile…ss2n You won't get around running some script per file, but you don't have to have all at the same time, but also don't have to have a commit per tagging:
Just get "some" (like, however many fit on your system, possibly in smaller tranches to parallelize with the tag processing), assign the tags, and drop them -- but run everything git-annex related with `git -c annex.alwayscommit=false`. That way, your changes only go to the index, and a final `git annex merge` to commit.

2025-05-17 16:45:49 UTC

- reply

@nprofile…ss2n @nprofile…8j6a I've seen there's a new feature to have `compute` as a remote, this might be viable too -- declare that each of those files is a compute result of the other. (pristine-tar says hi).
Still, I'd prefer an ecosystem where we don't have to go through those tricks and just occasionally nudge audio player authors to support sidecar files.

2025-05-17 16:40:04 UTC

- reply

@nprofile…ss2n @nprofile…8j6a `drop --unused` and then force drop them from the remotes where they used to be and then forget them. Or is there some automation for that process by now?

2025-05-17 16:38:31 UTC

- reply

@nprofile…ss2n In @nprofile…8j6a 's case, the files are already in git-annex with bad metadata. A smudge/clean filter that acts on the annexed data would probably be viable, but then still needs to place the actual data (or whiteouts for to-be-deleted tags) in some sidecar file or the git-annex metadata.
(I'm a big fan of git-annex, but am troubled by the metadata feature: It's one more format that is only understood by a single tool.)

2025-05-17 16:01:18 UTC

- reply

@nprofile…8j6a Yes and no... they usually read tags from files, but as that's not a well usable data structure, they all build their media library in separate files anyway. Some gladly read external metadata, but sadly they don't even agree on formats there (some using m-tags, some extended m3u).

2025-05-17 15:45:15 UTC

- reply

@nprofile…8j6a Ok then I phrased it badly.
I never alter the binaries, I only add data in extra files that tell the next user how to use them.
E.g. for photos, that means having raw files checked into git-annex; darktable then stores cropping, exposure and white balance corrections in an extra file next to it (which I add into git).
I have not done this for music yet, but do plan to make my music player (mpd) look at the same kind of extra files (and/or @nprofile…8hqs ) to create its database.

2025-05-17 15:08:33 UTC

- reply

@nprofile…8j6a The perspective I take there is that when binary files are checked in, they're some kind of third party artifact that's immutable (b/c had I made it, it'd be versioned notes with playing instructions precise enough for reproducible builds, amirite?); own statements on how to best use them then go to versioned sidecar files.
I've done this for photos, but sadly don't have a workflow to share for music yet. (But minimally for films, there it looks like this: <https://gitlab.com/chrysn/films-metadata>)

2025-04-24 19:05:14 UTC

Don't trust security check websites.
I just went through 10 online #TLS checkers from <https://techarry.com/top-ssl-tls-testing-tools-open-source-online-scanners/>, and only 2 of them (<https://testtls.com/> and <https://www.ssllabs.com/ssltest/>) even managed to scan the #IPv6-only site I was scanning. For all the others: How do you expect to run a comprehensive test when you can't even resolve half the addresses?

2025-04-24 05:52:18 UTC

- reply

Die @nprofile…xpk0 sind übrigens auch hier.

2025-04-23 09:25:13 UTC

Spring at the stable:
plants out of muzzles' reach
grow undisturbed.
#plant #photography #haiku
https://assets.chaos.social/media_attachments/files/114/386/497/151/666/180/original/ee61324293a51976.jpg
https://assets.chaos.social/media_attachments/files/114/386/497/765/015/057/original/801ea628ad4c2556.jpg
https://assets.chaos.social/media_attachments/files/114/386/498/465/290/637/original/80ae35332db588c2.jpg
https://assets.chaos.social/media_attachments/files/114/386/499/115/208/275/original/05560a664029312c.jpg

2025-04-18 08:03:58 UTC

- reply

@nprofile…qfh3 This is not specific to Python: it's the same in Rust and general mathematics. all(empty) is true and any(empty) is false.
This aligns well both with implementations (a la `for item in iterator: if item: return false` and eventually return true) and with the definition (there is not a single element in there that is false, so the result must be true).

2025-04-03 21:57:39 UTC

Nice new milestone for #embedded #Rust development: defmt 1.0 has been released. The crate enables logging for microcontrollers without text or number formatting on the device – instead, the device sends numbers of interned strings to the host, which composes log messages.
1.0 is not only a stable release, but also enables logging #CBOR data into diagnostic notation on the host.
#RustLang

2025-03-25 06:18:43 UTC

- reply

@nprofile…ndcz Weird recommendation, but OK. Convinced me to watch it.

2025-03-22 07:38:52 UTC

- reply

@nprofile…cwcm Viel Spaß! Ich bin diesmal nicht dabei, aber die Freunde und Kollegen von @nprofile…drvc und @nprofile…86am halten die Stellung für embedded-Systeme.

2025-02-26 11:06:50 UTC

- reply

@nprofile…zfuh Ooh, turtle, my favorite RDF serialization 😻
Technically, RFCXML supports internal links with arbitrary text; maybe I should just give it a try and spread as many links as I'd usually do in hypertext, and see if anyone complains.

2025-02-26 10:48:18 UTC

- reply

@nprofile…zfuh IIUC, the rationale is faithful rendition (the HTML form was not the "official" form).
But the effect applies to the HTMLized versions of older documents. Newer documents (eg. https://www.rfc-editor.org/rfc/rfc9528.html) have a first-class HTML form that does not suffer from this, and in those documents there are no page breaks any more even in the text version either.

2025-02-16 14:04:19 UTC

- reply

@nprofile…8j6a Kind of yes, but what would that mean for my vim modelines?

2025-02-15 11:38:16 UTC

- reply

@nprofile…wwe7 Just spitballing:
trait TypedDaemon<T> {
fn input(&mut self, id: usize, data: T);
// etc, with some fns not carrying T
}
and there is
struct MessageSerializer<D> {
daemon: D,
active_typed: HashMap<usize, std::any::TypeId>,
}
impl<D, T: Ser+De> TypedDaemon<T> for MessageSerializer<D> {
fn connect(&mut self, id: usize) -> {
self.active_typed.insert(id).expect_none();
}
}

2025-02-07 10:36:29 UTC

A good 8-minute summary of the #SIGINT situation in #Vienna, based on @nprofile…x7ar's investigation, has been published by @nprofile…l8jc (a journal affiliated with the Austrian NEOS party): https://www.youtube.com/watch?v=CddUToKXZJ8

2023-08-24 12:05:09 UTC

- reply

@npub1nlt…t44t @npub1ynl…873s @npub1adl…7nuk There was a good session during #IETF 113 on this, that showed one agreement across the technical community:
Guardian controlled devices are the only way to exclude kids from content the guardians don't want them to see.
Video at https://youtu.be/tHbIaQG_UFs?feature=shared&t=867, minutes & slides on https://datatracker.ietf.org/meeting/113/session/secdispatch/ (presentation named "Encrypted Client Hello Deployment Considerations")

2023-08-10 18:12:02 UTC

The new match rules of #OpenSSH 9.4 look like they'll finally let me use jump hosts only on demand: A `!localnetwork=2000::/3` condition on the ProxyJump directive for an #IPv6-only host should make direct connections used as long as I'm on a real network, and send data through a gateway on demand.
I'll try it out as soon as it's available on my system.

2023-08-09 11:51:45 UTC

- reply

@npub1evu…uf5t @npub1llz…08cq I think my #IETF work could be eligible, but the scope is unclear to me -- on the one hand the type-of-contributions description reads like the "boring" everyday stuff (WG chairing, reviewing documents) is part of the plan, but the application form looks very project centric, as do the times (the "long term" parts are 6mo; that'd have worked for most of RFC9290, but that was the fastest doc I ever contributed to).
Can you help me reconcile those?