Replying to
I appreciate both @Codeberg@social.anoxinon.de's policy decision and the diversity in policies that @codefloe@social.tchncs.de and others provide.
Remote
chrysn
@chrysn@chaos.social
mastodon 4.6.9🏳️🌈 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.)
0 Followers
0 Following
50 Posts
Joined January 21, 2019
Website:
pronouns:
he/him
Other metadata for machine consumption:
searchable (eg. through tootfinder), https://justmytoots.com/@chrysn@chaos.social
Also contributes posts to:
I've known that #RustLang's Cargo has been taught to not update dependencies that it can't use (because they require a newer Rust version), but just learned that it can be told to respect the own crate's MSRV for that:
CARGO_RESOLVER_INCOMPATIBLE_RUST_VERSIONS=fallback cargo update
allows me to utilize this even though my MSRV is so old that Cargo didn't have that feature back then.
Thanks to @epage@hachyderm.io for pointing this out in the handling of @epage@hachyderm.io@chaos.social.
Replying to
@fluepke Wir hätten je nach Inhalt der Kanäle noch die untätigen Korruptionsbehörden (wenn nicht-dienstliche und damit nicht an Aufzeichnungspflichten gebundene Kommunikationskanäle für politische Entscheidungen genutzt werden) oder die IT-Abteilungen, die Signal für Dienstkommunikation freigegeben haben.
Aber das wär nicht so lustig.
Preparing a demo for the @ariel@floss.social booth at #CLT. If things goes to plan, you can control different #embedded systems running the same application from your browser.
Replying to
@Crell@phpc.social @dereuromark@phpc.social Any of them is preferable over you-never-know-the-dialect markdown with HTML mixed in.
I'd root for ReST, but doing links with abbreviations with titles shows its rough edges.
With the speed around #JpegXL picking up (now in #Firefox and #Chrome nightlies), I'm wondering what good use of an #HTML srcset would be there.
As JXL's progressive compression is good, we don't need different files in the srcset (so there's no need to load the first few KiB again when zooming in), but how can or should the HTML somehow tell the browser that for (say) up to 800px width it can request the Range: bytes=0-12345, and the next usable zoom level of 1600px needs the first 42KiB?
Replying to
@alexglow@chaos.social It has a pronoun you can use for explicitness: is/ea/id. But in this case, as @Legit_Spaghetti@mastodo.neoliber.al, there is also the group hic(/haec/hoc), iste and ille for "this one here / myself", "that one next to me" (often negative, as in "he did it!") and "that one over there", of which the ring would probably be a "hoc".
CC'ing @baeroline@mastodon.social who I hope will correct me as she knows better.
Replying to
@Doomed_Daniel @dpk Windows and MacOS builders are the reason why a single project of mine is staying at GitHub: In a sense, they pay for my project's presence with the capability of building binary wheels for all platforms.
Kind of an OK deal as long as they don't raise my pain level too far, especially as currently I don't know any platform to which I could go for that.
Replying to
@rysiek I'm all for migrating off GitHub, but this looks like an odd item to peg it on: From your description, it only affect usage that is *already* using Copilot, and frankly, if I were using anyone's AI tools, I'd be surprised if it was *not* using its users for data collection.
Replying to
@lowk3y@infosec.exchange I don't think it's "most people". It's a particular class of power users who know enough to do some things on their own, but then eg. apply patterns they learned in IPv4, or the available tools don't cater for v6 in that corner case. (Similarly, some Austrian ISPs can do static IPv4 because the regulator demands this must be an option, but then that user becomes v4-only).
Replying to
@admin@floss.social Can I vote d) Require it to be declared? (eg. through hashtags or TBD metadata)
Replying to
@tschaefer In a sense it's great: It produces an implicit list of media publishers without any competence in contextualizing information.
Unangekündigte Medienkompetenzübung.
Replying to
@l17r @Eugen_pfister Na super, da les ich endlich mal ne Frage auf fedi, bei der ich denk, ich könnte helfen, und dann seh ich, wer's schrieb und boostete, und dann hat sich das wieder erübrigt.
Replying to
@oli@hachyderm.io Awesome, thanks for pushing this. I was hoping for ranges stabilizing, but seeing that pattern types are the more powerful variant and apparently more feasible to handle in language development, I'll now root for them on the long path through an RFC towards stability.
Replying to
@diondokter@fosstodon.org Thinking of this again while implementing an embassy_net_driver_channel made me realize that there can be a fn f<'t>(data: &'t [u8]) -> impl Future + use<> { … }, and the data buffer can be freed before the future completes. 🤯
Replying to
RE: https://toot.yosh.is/@yosh/116190471042101677
@yosh@toot.yosh.is I was trying to contribute to the question of language that bans (or at least requires transparency about use of) LLMs but not compilers.
n the analogy of https://chaos.social/@yosh@toot.yosh.is/116190472066903497 you can sure fix a bug in source code generated by an LLM, but when comparing the LLM to a compiler, that would be fixing the product of the compiler (sure possible but tedious), whereas fixing a bug in the compiler ensures that henceforth, all instances of that bug are fixed.
Replying to
@yosh@toot.yosh.is @soph@grrl.me Something I've been working with in that distinction is "it produces bugs, can you or someone else go find what went wrong and fix it?"
This clearly tells apart compilers, linters and dictionary based spellcheckers from AI tools. (It may also place non-FLOSS, especially online, tools in the AI category accidentally, but that's not too bad as you can't disprove that they are).
Replying to
@RichiH@chaos.social https://www.scanprofi.at/ hat unsere Dias und Negative super digitalisiert.
Replying to
@diondokter Could an async fn / block waive the "does not do anything until polled" guarantee? (Like #[allow_eager_parts]. Ideally, this might even become default in a future edition).
The only reason I see for this guarantee to be there is so that all the inner parts can be pinned, but then if anything relevant is not Unpin, the compiler could still go into the old mode.
Replying to
@sovtechfund How focused do you want the applications w/rt single drafts? Of my dozen-or-so active IETF drafts, about half of them should qualify for public interest relevance (the rest are more niche topics), and no single of them would take a year of 10h/week to complete. Is improving an ecosystem on several fronts something you'd like to support, or do you need a single use case linking them all?
Replying to
@MicroSFF Seems like this and Opossum in the Underworld are set in the same universe. https://www.nickmaskell.com/comicsandstories/opossum-in-the-underworld-1
Replying to
@diondokter Yeah, reading on, I kept thinking: The compiler would hoist send_response out of the match normally, why doesn't it here?
Looking forward to it!
Replying to
@alexglow@chaos.social @Legit_Spaghetti@mastodo.neoliber.al I think the word order to justify the "hoc" (it is an emphasis, otherwise it'd be moot to say anything at all) would be to have it first. And I'd probably not repeat it; all is spoken in a hurry.
Hoc absconde. Protege.
Replying to
@vampirdaddy @alios Indeed.
But a great benefit of running no_std/no_alloc is that many kinds of resource exhaustion (full disk, RAM exhausted, file descriptors) can't happen or are handled explicitly, and stack overflow is the one exception.
Replying to
@vampirdaddy @alios Also recursing on the stack for storage.
But these are relatively few antipatterns to keep an eye out for.
A pattern that avoids trouble from fixed size as well as preallocation is having (type-)generic pluggable back-end. Then there are two simple impls provided, one for alloc and one for a const generic fixed size.
Replying to
@pivotal@sunny.garden @phaedr0s@masto.hackers.town … and at the end of that exercise, you find the comment that explains it all and that, once reaching this point, you are invited to increment the counter in that comment.
Replying to
@yosh@toot.yosh.is I may not myself fix the bug, but there is a clear process around how it is done, and generally there are people whom I could hire to fix this. When there is nobody I could find to ensure that the issue is understood and resolved, then the I'd call the tool an "AI tool", and (in the context of FLOSS projects I'm contributing to) ask that the use of the tools be declared.
Replying to
@nils I did have some reservations about using @ratatui_rs for embedded devices (mainly that they're natively pixel based, so why go through a typewriter interface when there's no legacy) -- but seeing how reasonable that looks in code already at this quick prototype makes me reconsider.
Don't hold your breath for a text based #CoAP web browser, but I'd love to build one.
Replying to
@KIT_Karlsruhe@social.kit.edu Cool, dass ihr das macht.
Koordiniert ihr euch da mit anderen Unis, oder gibt's irgendwo eine Liste, in welchen Regionen sowas noch gemacht wird? (Hagel hält ja selten an (Bundes)landesgrenzen…)
Replying to
@jens@social.finkhaeuser.de The one thing that keeps me from even trying OpenBSD is their refusal to allow IPv4 on IPv6 sockets (V6ONLY=0). But maybe that doesn't matter any more when used in v6only networks where something outside provides NAT64.
Replying to
Concretely, I'm looking forward to using this when wrapping C libraries that return errors in negative isize. When ranges or pattern types are exploitable in stable, a Rust wrapper could return `Result, NegativeError>` for idiomatic error handling while being no-op (the best form of zero-cost).
Replying to
@mjg59@nondeterministic.computer More software should support minimal distributed config. Like, I (would like to soon) run my web server using git pages, and if two friends just say "yeah I mirror, tell me which domains" and I put them in DNS, there could be sudden redundancy.
Replying to
@tschaefer@ipv6.social @herrbett@ridesbik.es Apple hat doch sicher eine Rechtsabteilung für sowas...
Replying to
@cd0@mastodon.social Which is enough to go back and forth all across a maximally sized PDF (a few hundred km), but would problematic when filling such a PDF with delay lines.
Replying to
@davidrevoy@framapiaf.org In my ideal web, this would be a single image with progressive loading (JXL?), whose metadata indicates that it is composed of different panels (each of which has image descriptions of their own) that can be zoomed to sequentially; clients that don't understand that show the full image, those that do show it as sequence depending on the user's preference.
Until that works, I find 4-pictures more practical.
Replying to
@hjvt@hachyderm.io @nyx@lgbtqia.space @shana@mastodon.gamedev.place Great, now I am nerdsniped into pondering when a compiler can justifiably terminate what is by spec an inftinite loop into an analytical result.
Replying to
@hjvt@hachyderm.io @nyx@lgbtqia.space @shana@mastodon.gamedev.place That's the UB I was thinking of. For the compiler to make that jump, I think the loop still needs a break condition, like "if next == last", otherwise, the compiler would just mark the whole branch as unreachable. With ffastmath or equivalent, a compiler could trust that the sequence converges, and apply analytics such that 1.0+2.0+3.0+... = -1/12 or 1.0+0.9+0.81+... = 10.0, even when due to rounding it might not actually converge.
Replying to
@diondokter Now got the sad confirmation from trying on the playground: Indeed what I thought would be a half-way step towards a more idiomatic loading function looks no less complex than what the async fn does. (That'd be the load_semi function in )
Replying to
Referring to whose author has 'been so growingly averse to [dependencies], preferring to use LLMs to "yoink" functionality when it's simple enough and possible.'
Replying to
@tweedegolf @diondokter Ah, think twice, post once: It's the guarantee that async code only runs when polled, not when constructed. So, (how) can we waive that without too much fuss?
Replying to
Having runners ("agents") both with amd64 and with arm64 is another big plus – not that I'd have wished for it, but it conveniently showed me that there are warnings popping up there. Nothing that needs fixing in this case (just need to silence some specific PyPy warnings), but all of a sudden it has become way much more useful for me that we recently started publishing builds for arm64 in the lakers project.
Replying to
@bunnie Cool design and post, many thanks!
How do quantums/events work out when protocols mix event sources, eg. in Manchester codes where we jump between "read after fixed time" and "wait for edge to restart timer", or when waiting for either an edge or a time (eg. timouts)? Would those always use two BIO blocks at faster rates (like 5Mbd)?
If so, could future baochips have more of them, like, 8?
For example, IIUC, mixing two S/PDIF inputs into one would need at least 5 blocks.
Replying to
@lobingera @mirabilos @leyrer @stoeps Hey, das ist das web, wir haben content format negotiation! Scireibt doch, wie euch der Schnabel gewachsen ist, und in der spec steht dann schon, wie die zu konvertieren sind. (Grad YAML und TOML sind ohnehin ein superset von JSON).
Replying to
@miyuru@ipv6.social It's fine that everyone can publish drafts. Those come with a clear disclaimer:
This document is an Internet-Draft (I-D). Anyone may submit an I-D to the IETF. This I-D is not endorsed by the IETF and has no formal standing in the IETF standards process.
Trouble is people misrepsenting them as IETF plans. If that were a news article, I'd suggest having a with the "journalist". But it's Facebook posts, so what'd you expect...