JPEG XL

Idk if newer browsers are still susceptible or not but it was pretty effective in the past

yes, that is an issue with implementation, but most sites that do this now just pretend to do so

You used to have to force-quit the browser process

Some of them were very effective at putting the browser in a useless state

But it's just as annoying to me, to interrupt me with a bunch of obnoxious banners and automatic video and sound playback when I'm in the middle of reading.

I agree thats annoying, but for me, im fine with just closing the window, if I cant close the window, thats an issue with the browser

"Interactive" in this context shouldn't mean "preemptively acts on its own"

If I'm reading it should not suddenly start doing something else all of a sudden. That's no different than hijacking the browser

> As you may have heard Microsoft Edge no longer uses Chakra. Microsoft will continue to provide security updates for ChakraCore 1.11 until 9th March 2021 but do not intend to support it after that. ah, that’s what it was

I knew the name sounded familiar

it was the only part of EdgeHTML that got open-sourced afaik

does anyone remember KHTML

WebKit is based on it right?

yeah

but it also continued as its own engine for a while

I think I used it until… I want to say 2014?

> It originated as the engine of the Konqueror browser in the late 1990s, but active development ceased in 2016.[1][4] It was officially discontinued in 2023.[3]

khtml was nice, sad it died

I liked that `<q>` used the right quotation marks for the language declared by the page

whereas webkit would just unconditionally use “” from what I recall

(why do I remember this detail, >10 years later)

it seems blink and gecko correctly use «» for French nowadays, but without the appropriate spacing

ideally, should be « like this »

with thin non-breakable spaces inside

(non-thin is accepted by some, but I see no reason not to make them thin in a web engine)

Along with the usual WebP awnsers, there's actually quite a few complaining about AVIF now instead. Interesting to see how the trends are changing https://x.com/wobbuuu/status/1967301378071802097

soo much webp and webm kek

gonna bet that if jxl gets adopted seriously it's going to start appearing on those too 😛

That's how we know we've made it in the world

one difference is that with jxl, it looks like browsers are the _last_ ones to adopt it while with webp and avif they were the _first_ ones to adopt it 🙂

Hmm, good point. Might be the first time people see a new extension and it just works™

webp [lossy mode] deserve the hate

https://www.ctrl.blog/entry/webp-avif-comparison.html

jpegli my beloved

and the full strength of Jpegli hasn't been widely adopted yet

jpegli jpeg2png packJPG JPEG XT WebP lossless These good things don't break compatibility

packJPG <:ugly:805106754668068868>

who tf uses that

Repacks

JXL beats packJPG

afaik

depends on the image

the webp hate in a nutshell

in 2025, some people are still claiming that no app supports it

You can't expect everyone having common sense

Based on my work from a year ago, I had to do the following: My own image interpolation (downscaling) algorithm uses nearest neighbor matching to search for the best pixel based on the original pixels and a standard downscaling image. This means that when downscaling, all pixels within the area where they could be located are searched after the image is downscaling. Therefore, the algorithm takes a long time to run. This results in the clearest, most noise-free image possible, containing only the original pixels without any artifacts. Images: —Original —Regular LANCZOS downscaling —Regular nearest neighbor downscaling —My clever nearest neighbor algorithm

anyone knowledgeable on android? my phone (poco f7 pro running hyperos) is currently basically unusable due to some weird issue with the permission system. most apps just crash immediately when i start them, and i can't go into permission settings for any app. sometimes when an app crashes i get a popup saying the permission manager keeps crashing

i was able to get these stack traces from that popup

from a quick google search, `RANGING` seems to be an android 16 thing? but this phone is running android 15

i assume/hope i can fix this by downgrading some app or doing some adb fuckery instead of having to factory reset

i didn't do anything weird btw, i have the bootloader unlocked but didn't flash anything

seems like i can get at least individual apps to work again by just granting all permissions via adb, so that's something (i did reset them earlier in an attempt to fix only some apps being broken)

i was kinda able to track it down in aosp:

somehow the checks for these flags which only exist on 16+ are returning true on my phone...

running the command `device_config list` seems to list the types of flags this checks for, but these specific ones aren't listed

not sure what else i can do now without root

report the issue to hyperOS would be a good start

did some more thinking and research, i *think* what happened is that i booted an (official) android 16 gsi via dsu once, and that updated stuff in `/metadata/aconfig` which is now mismatched with my actual os. no idea why i can't find a single similar report though

Likely because the amount of people playing around with Roms is disappearingly small nowadays

fair, but DSU is an official feature in dev tools and doesn't even require a bootloader unlock on some devices (like pixels i think)

the sad part is that this probably wouldn't result in that many safety improvements in JavaScript handling because most vulnerabilities are still in or caused by JIT compiled JavaScript

this is true for Chromium as well, which is why Edge decided to disable the JIT compiler for most websites except those that you frequently visit

I've disabled the JS JIT tiers in Firefox (only having the C++ interpreter and baseline interpreter tiers enabled) and I barely notice any performance differences in day to day browsing, except some specific things like hashing a file in VirusTotal takes like 10 times longer. I still have the Wasm JITs enabled since Wasm is usually necessarily expected to be fast, and since Wasm has structured control flow and strict typing it's presumably a lot harder to exploit its JIT components

Hello to everyone! Wondering if there are servers about data compression on Discord. Honestly this is the only server I found about compression.

👋

Learning to draw

What?

Did a perspective study

another day, another piece

Oh lovely

oh they doin an internexit

Zero planning. Zero pencil sketching. Pen exploded in the middle of working on it. Used the butt end of a broken pen like it was a brush to paint the shirt.

Pen exploded?

A dip pen where the nib was made of two pieces. I pressed down a bit too hard and the pieces flew apart.

Drew for a local expo

still full AI though ahahah

Yeah by a dumb AI that took about 36 hours

The first time I block someone in this server

What an insult by a shit

AI = Amazing Individual it seems

Really cool artwork

I even haven't posted so-called "full AI" images for months

about Interop 2026, firefox team made a page to gather feedback from web developpers about wich proposals are particularly notable https://mastodon.social/@FirefoxDevTools/115345278669277934

if you do web development, look through the list and rank up those you find most useful with an open mind (not just jxl)

during some (or at least one) of previous interops I upvoted some other cool stuff besides JXL although none of them got accepted as well

upvoted on github, or?

yeah

yeah that doesn't do anything. this is at least the devs of a particular browser asking the community for rankings.

I was thinking I shouldn't, then I realized I recently started helping a non-profit with their website, which means I technically speaking am a web dev now.

man I just want jxl or webusb

both are probably gonna be declined this year. The point of interop seems to be to make sure things that all browsers support or agree on supporting work consistently across all of said browsers

Mozilla does not want WebUSB and Chrome does not want JPEG XL

I wonder how many people have gotten confused over the years seeing Google devs saying they are willing and ready to provide JPEG XL support for browsers and then Chrome says they don't want JPEG XL

most people just think Google is a hivemind from what I've seen

Mozilla not wanting webusb will always be such a massive meme to me

I didn't know that anyone wants WebUSB

yeah, I’m not too sure what the use case is

browser-based desktop apps like discord websites are still only showcases

USB keyboards, mice, webcams and microphones (maybe also MIDI keyboards?) already have dedicated interfaces that most people would probably rather use

also storage media to some extent

and gamepads (https://hardwaretester.com/gamepad)

ah, yes, MIDI too https://hardwaretester.com/midi

I guess maybe manufacturers of various devices could offer firmware updates that way?

plug your device, go to the manufacturer’s website and update the firmware right there and then without having to download an app

but that sounds a bit niche

That's the only thing I've seen it used for

(but only hobby stuff, nothing professional)

https://flash.android.com/

and similarly https://grapheneos.org/install/web

other than that idk

the slight irony here is that installing an additional driver is anyway required on Windows

webUSB is amazing

a lot of manufactures use it for firmware updates and configuration, I use it to control my keychron keyboard for instance

(keychron uses a modified qmk, qmk also supports web usb with via)

sure one could write a dedicated application for each OS, but webusb apps will support any OS that has a webusb capable browser. BSD, Linux, OSX, Windows, Android. Redox Haiku etc could all be supported too via a webusb capable browser

Amazing security nightmare

WebUEFI next

WebBootloader

WebKernel, WebSwap, WebHiberfil.sys

I can't wait until people need a government ID and seatbelts and airbags to use the internet

After all, your computer could crash

Well we've already got the first one here

WebUSB can be fine security wise, but it *needs* strict restrictions on stuff

as-is on Chrome(ium), you can yank some devices to do horrible shit (cypress released some USB-FPGA bridges that essentially have an equivalent of unremovable backdoor - allowing those to be used with WebUSB is just asking for trouble)

but with some actual thinking about how to go with all of this securely, it can certainly be done - with more limitations than just "libusb bindings for web", but with still a lot of abilities nonetheless

as long as it asks permission its all OK for me

(just like Android should at most ask for permission to sideload, nothing more <:KekDog:805390049033191445> )

FWIW asking for permission only from the user isn't enough? the cypress bridges I mentioned can be re-programmed to act as whatever USB device - so in essence, you could just turn whatever device into a rubber ducky and takeover the inputs for the device. Users can't really be assumed to know this, and stuff like "hey, this neat online tool can help you mod your cool device or something" is believable enough that users will fall for it

IMO the real solution should involve the devices themselves saying "yes, I'm safe to WebUSB with" at the minimum

if someone has a rubber ducky in my system, I have literally zero concern over webusb lol

you can turn devices that aren't ruber ducky into ruber ducky with webusb tho

that's the issue

the premise relies on someone having actual hard accsess to the system already, at least enough to send arbitrary inputs. With that level of threat to the system, webusb doesn't change anything.

I don't think you understand?

the threat model here is me having some device that uses said cypress bridge for a totally normal purpose, using some web page that promises to use WebUSB for interacting with said device, and then, *via WebUSB*, re-programming the cypress bridge on the device to act as a rubber ducky

essentially, it's privilege escalation chaining through WebUSB and an external device with a specific USB chip

idk, maybe this github comment will explain it better https://github.com/mozilla/standards-positions/pull/193#issuecomment-1872606016

thats a threat I dont care about at all

if I give permission, thats my own damn fault, this would be like removing support for running apps with root because you could run a malicious app

perhaps I could see the value in *optionally* supporting a feature for arbitrarily limiting to "safe peripherals" but it would need to be an option

I don't think it's a danger that can reasonably be expected or adequately explained to a common user, and that's the problem with it

" [ ] allow all devices Warning: This is akin to running apps as admin/root " would be fine

or just "[ ] allow unsafe devices" would work

like, bob living on the other side of the street won't have a clue that by clicking allow he might just get completely pwned

and warning in that sort of way doesn't really work? Like, there is a reason why windows moved to requiring *entering the admin credentials* for such things

I mean, if you click "unsafe" and get pwned, I really don't care about your safety as a user

I don't think that's a reasonable stance tbh

because people click "unsafe", essentially ignoring it, because they just want the thing

thats their fault

im all for removing footguns, but not at the expense of useful features

All of modern cybersecurity is operating on the principle of "it's never the users fault", but i guess go off, I don't have much more to discuss if you don't agree with that

the idea of "it's never the users fault" is fundamentally flawed and how we get to android blocking all apps that aren't signed and registered unless it's installed over ADB. Either at some point we have to accept that it is the users fault, or lock down devices so much that they become unmodifable bricks. not to mention "it's never the users fault" directly contradicts that fact that users get in trouble for doing things like leaving workstations unlocked, forgetting 2fa keys, etc.

so all modern cybersecurity is in fact, NOT operating on said principal

> lock down devices so much that they become unmodifiable bricks <:Apple:806136610659237888>

ok, let me clarify - all of *good* modern cybersecurity is operating on that principle, stupid shit from corporations ignored

like, if a user clicks a phishing link and gets pwned - there's a massive chain of issues that should have been caught, and the fact that they didn't, is a problem

if all of your security relies on the user not making mistakes, there's no real security

sure, you can't rely on just the user not making mistakes, but in the end, personal responsibility is a thing. A check like this is perfectly standard in most computing ecosystems

People do not understand the security implications of what they are granting permission for.

"Do you agree to give this website hardware-level access to your PC and connected devices? No take-backs."

People don't usually expect a website to be able to permanently modify their hardware

The problem is that the standards don't really have, or respect, any notion of boundaries or scope or obtaining consent. Every website operates on the assumption that the user implicitly consents to you being allowed to collect a ton of unnecessary fingerprint info and run arbitrary programs with access to your GPU...

I don't trust these idiots to actually make a good standard that isn't designed to screw us over like they usually do. The people writing and publishing the standards are the same people who get paid to spy on you for profit.

If you want it to be useful enough to allow your browser to update device firmware, then it's also useful enough for everyone else that wants to abuse that to screw you over.

Personally I don't think the browser is the appropriate place or context or trust model.

But I do agree with you that devices are way too locked down and security is being used against us rather than for us. But that doesn't mean a web browser is the appropriate place to put all these things.

idk, I'd browsers currently have the best trust model out of any widely-available system (though that says more about everyone else than about browsers)

and stuff like apple and google locking down phones to me is more of a monopolistic practice that's being excused as a security practice than a real security practice

doesn’t Flatpak supposedly have a nice permission model?

if they did things with security in mind from bottom to top, they wouldn't need locking stuff down to anywhere near that extent

it's ok-ish? But browsers and culture around browsers is way more locked-down

e.g. the way WASM doesn't have any abilities besides those that are explicitly given comes from the way browser things are designed with principle of least privilege

Not if you just want to read an article without the page constantly changing itself while you're trying to scroll and read. Running scripts and autoplaying videos at random moments while you're just trying to read what's on the page you already downloaded. You never explicitly consented to any of that, it just assumes you want that.

There's no API for getting permission to do all that obnoxious crap

I think you're more angry that the browser is not just a document viewer rather than it's permission system is bad?

which is like, valid, but a different question

And the security of running arbitrary programs on a VM from every web page you visit without asking, and giving them GPU access, is also a house of cards.

No, I am angry that it's permission system is bad. It doesn't have protocols for obtaining consent for anything risky or invasive it does

Except for very specific things like camera access

I think you have somewhat different definitions of risky or invasive?

like, it's applications - of course it's going to run arbitrary code

It SHOULD act like a static document viewer for untrusted websites that haven't obtained any explicit permissions

well, that ship sailed around 30 years ago

so, take it or leave it

If you want to run an app like discord, then it should ask "do you want to give discord permission to run on this computer?" or something like that

it is very much an application platform, designed for a low-friction experience for the users - adding explicit asks for "hi, do you want to run this" will end up with users automatically clicking the button without reading anything

very important point - more asking doesn't give more power to the users

just look at how things turned out with GDPR consent banners - everyone just added a bunch of popups, users got trained to just click yes without thinking

The browser includes exact version information and platform/cpu details in the user agent string for some unknowable reason. Instead of a very generic string like "Browser" or "Mobile"

Browsers are extremely easy to fingerprint by design and the standards and conventions encourage that, to make it easier to profile you

Well that's a terrible example because it makes no sense for websites to ask for permission when the browser is what has the control there

all of this shit depends on your threat model

Every website has their own stupid banner

if your threat model is that paranoid - go use tor browser, it doesn't do any of the things you're annoyed about

All this shit is on purpose because the goal was never YOUR best interest.

but for wast majority of ppl, a constantly looked after VM for executing arbitrary code safely is a small price to pay for the convinience of not thinking on whether to run a script on every website or not

It's the best interest of the people who are in the business of taking advantage of people and spying on them for profit. Who just happen to be the same people in charge of publishing the standards and writing impossibly-complex and unmaintainable web browsers

I'm not going to argue anymore about whether browser permission model is good because you seem to have a very extreme view on that, but I will say that nothing really surpasses it as of right now

its actually horrid

overly broad, and plenty of things are needed to hack around it

I agree, and there are reasonable steps to mitigate it, but out right not having nice features is not the way to go about it

Well, it would be nice if there could be a transition of less and less unreasonable expectations... Like if more and more websites work with Javascript disabled OR a much more restrictive feature set until a user whitelists or explicitly trusts a website with their consent

And it doesn't need to spam the user. It should only ask for permission when visiting a new website for the first time, and only if that website actually needs more invasive features, which the majority of websites should not need.

I don't think the majority of websites need such deep and invasive access to your PC to function correctly.

I think most people would agree with that and not call that extreme. What is extreme is how far we have allowed this to go without saying no.

Now Google Search require JS <:KekDog:805390049033191445>

in the future we could use only something like FrogFind! or Wiby

https://wiby.me/?q=jpeg+xl

I use Searx still

I don't think interpreting an extremely sandboxed script in a dynamic language is a "deep and invasive access"

And this is the part that I'm calling extreme

https://animetv-jp.net/news/anime-tourism-isnt-the-problem-elitism-is/ Some people will not even read and understand this article and will get pissed off anyway.

It's very invasive if you consider all of the different things a random website has access to without obtaining consent. Both websites you visit, and third party scripts from places you aren't normally even informed about. It is NOT normal or acceptable in any other context to give everyone such powerful and implicit permissions on your PC. Everyone who knows anything about security knows that the trust model of browsers is broken by design

What powerful permissions? Interpreting code? Arguably, many compression algorithms also end up doing that, I find this totally nonsensical

font rendering is turing complete, having something explicitly turing complete doesn't change things in the slightest

I guess the sensitive part is the capability to exfiltrate all those data using network requests

https://www.phoronix.com/news/Cache-Aware-Scheduling-Go This after Kevork's comments. Lol.

What comments?

`"We need to find a balance where we use that as an advantage to Intel and not let everyone else take it and run with it."` https://www.theregister.com/2025/10/09/intel_open_source_commitment/

Lol

thankfully iirc this dude is only responsible for data center stuff

intel is one of, if not the biggest players in the "open source ecosystem" so seeing them die loose that entirely would be so sad

https://tenor.com/view/harvey-dent-die-a-hero-or-live-long-enough-to-see-yourself-become-the-villian-gif-21468252

https://www.animenewsnetwork.com/news/2025-10-12/1st-demon-slayer-infinity-castle-film-breaks-25-year-record-as-no.1-non-english-international-film-/.229882

https://videocardz.com/newz/amd-sound-wave-apu-with-32x27mm-package-appears-in-shipping-manifests

<@179701849576833024> Have you seen the Typlite tool? It's in the Tinymist repo, and uses Typst as a library to generate HTML to use as an IR to convert Typst docs to LaTeX and DocX. Still very WIP, but quite neat.

Had not seen it

Don't expect to need it in the near future though 😜

My school requires all reports to be in docx. They even say you have to use the desktop version of Microsoft Word, not the web version and not Libre Office.

So I have an incentive to tinker with it to make it good enough to fool them 😁

yeah it's been a while since I had problems of that shape 😛

reports in docx 💀

all I had to submit in my univ was PDF, nothing else and almost all of the time we used LaTeX to build those PDFs, until last year when I discovered Typst (and wrote my master's thesis with it)

> and wrote my master's thesis with it nice

I know many journals want your TeX sources, not your rendered PDF. The TeX output of Typlite could be useful in that case.

> <https://staging.typst.app/docs/changelog/0.14.0/> > * Typst now produces accessible PDFs out of the box, with opt-in support for stricter checks and conformance to PDF/UA-1 > * Typst now supports all PDF/A standards > * PDFs can now be used as images (thanks to @LaurenzV) > * Added support for character-level justification (can significantly improve the appearance of justified text) > * Added support for many more built-in elements in HTML export > * Added typed HTML API (e.g. html.div) with individually typed attributes I find all of these additions very exciting.

given that requirement, I would probably try pandoc first – does it produce anything acceptable?

seems decent at first glance, even rendered by LibreOffice (although I am not a fan of that bold font)

not sure whether the red “¿” instead of the `\underbrace{marginal probability}` is pandoc’s or LibreOffice’s fault

what is definitely LibreOffice’s fault, however, is offering to replace “post-humous” with “post-hummus”

just buy microsoft

I've found Pandoc's support for styling to be troublesome. Currently I write in Typst and use the HTML feature to get a doc to feed into Pandoc, and then I fix any issues in the docx manually.

The only thing missing in Typlite's docx output (for me) right now is figures. So I figure I can just add that and then skip Pandoc.

I, in fact, have a legal copy of Windows 11 Pro and of Microsoft Word. I just severely dislike them.

just buy microsoft company

I mean if you wanna give me a few trillion

i think that this money could be used more effectively, or even cheaper to create a new OS

Localization was quite literally better, at least for the german community: - Used to have openings sung in german, sometimes even custom ones and the majority of them were actually good. - When fansubbers were around, every little piece of japanese got translated **in a font similar to the japanese original** plus there were translator notes explaing some japanese terms; that ex-piracy site Crunchyroll that took over just seems to shit on this entirely and it's even that worse that one anime with a deaf char has minutes of phone communication untranslated in episode 1 (and no, there is nothing else happening at that time). Besides that openings and endings often don't even have subtitles anymore – it was not unusual for fansubbers to not only translate them but to add **animated** romaji there so that it was kind of easy to sing along. And besides that, there is obviously some more influence from the west nowadays in general.

Did I say anything wrong? https://www.reddit.com/r/firefox/comments/1nu8okb/comment/ngzijln/

lmao nope xD

maybe they wish you had said “accurate” instead of “precise”

"helps you get shit" can be taken 2 ways

ways to interpolate video from 30 to 60 fps? minterpolate (no multithreading, no native 8 bit), rife ai (gpu used)

Apparently Nvidia GPUs since like the 20 series have had a hardware optical flow accelerator

Yeah, for DLSS

Oh yea I guess dlss does still need to work with transparency and other cases where the engine can't supply motion vectors I hadn't thought about that but it makes sense

Gilbert tessellation

https://arstechnica.com/gaming/2025/10/civil-war-gzdoom-fan-developers-split-off-over-use-of-chatgpt-generated-code/

https://pissandshittium.org/ 🤣

so normal chrome lmao

https://servo.org/blog/2025/10/20/servo-0.0.1-release/

neat to see, not much in the way of importance, but there is some meaning in doing this.

"There is some meaning" aka a fart in the wind?

Related to Huawei

Lmao

Why this guy mad

literally me 🥹

I learned how to compile code because there were no binaries available lol

I get him to compile code because the binaries are slow

https://www.npmjs.com/package/@ui5/webcomponents-ai a button, but for AI

(it looks like this: https://ui5.github.io/webcomponents/components/ai/Button/#basic-sample)

Why Rust is hated/loved? I only know it is low-level programming language without garbage collector.

why it is hated: try writing a doubly-linked list in it why it is loved: try writing large-scale software in C or C++ without memory bugs

I read the rust linked lists book and it seemd pretty fun

Assembly 👀

I mostly just like the way the compiler tells me how to fix my bugs and that cross compiling with it means no other make systems tbh

need to compile for android `cargo build --target <arch>-unknown-linux-musl`

Are you sure it is musl?

can buy quantum computer on ebay?

yes? musl compiles entirely static, so this only becomes an issue with x11/wayland

it becomes an issue if you need to dynamically link with anything - so graphics in general at minimum

kinda, wayland can run static since how it works, but building it on rust rn is actually painful because a bunch of nested tangled sys deps

winit is also sucks here since it force uses system libwayland

oh no

quick someone port winit to Smithay

[omegalul~1](https://cdn.discordapp.com/emojis/885026577618980904.webp?size=48&name=omegalul%7E1)

https://github.com/Smithay/wayland-rs/tree/master/wayland-backend `wayland-backend = { version = "0.3.10", default-features = false, features = ["client_system"] }`

unfortunately untangling this is not so simple

ofc the easy solution is to just have a glibc prefix or a musl prefix, termux actually provides a glibc prefix for packages btw

just use the rs backend ? does it not work or something?

only if you don't care about GPU acceleration

which is fine in a lot of cases

see above, no gpu, so they dont want it

jxl-supporting Affinity is no longer paid and relaunched as freemium, with the apps equivalent to photoshop, illustrator and indesign fused into 'affinity studio'

Nope it's just simply Affinity

<@238552565619359744> had one of your friends say hello to me on your behalf – hello back :D

Ah good, so they did find you haha. Unfortunately it was too short notice for me to attend your talk, I only got into London just as it started. I'm down by Bank at the moment meeting some friends

ah nice, hoping to meet you at some point!

musl has a dynamic linker also. It can do both dynamic linking and static linking better than glibc can.

since it was made from scratch it's a lot better at both. The only weakness of musl is the builtin malloc implementation. It's not so good at multi-threaded malloc performance. But it can be easily replaced with a better malloc like mimalloc.

The only other weakness, which isn't really a weakness of musl itself, is that it's not bug-for-bug compatible with glibc and some software relies on glibc-specific quirks.

I look forward to the day when glibc is no longer included in most linux distros as the standard provider of the c library

yeah but you have to choose when compiling whether you're binary is static musl, dynamic musl, or dynamic glibc

so you must be compiling against the right libc in order to use stuff like graphics drivers

static glibc supremacy

relibc when plox

static glibc isn't a thing

it just statically links a stub that will try to load the system glibc at launch

what? It very much is a thing, or else you wouldnt be able to run a static glibc binary on musl or bionic

sure it kinda sucks to work with, but it's still a thing

static glibc still tries to pull in some sub-libraries dynamically

it could be better described as half-static

If you use any locale-aware function, like stdio stuff, it will try to dlopen iconv

basically: https://stackoverflow.com/questions/57476533/why-is-statically-linking-glibc-discouraged

yeah, it's a pain in the ass, thanks dlopen, but it can be worked around

Plus the whole weird plugin system glibc uses for hostname resolution is pretty brain damaged too

Compared to, say, just running a daemon to do that

minterpolate ffmpeg recommendations? better cpu-related alternatives?

We're people, not a chatbot or a search engine... Most questions you ask can just be typed into a browser

you are right, but even gpt said me wrong info about it, i found h264 motion estimation thread on forum (but it has no minterpolate info), so you are last stand

https://cdn.discordapp.com/attachments/873300833775788113/1434523277925028044/572263168_1432394222227502_6579086510108264778_n.jpg?ex=69094c29&is=6907faa9&hm=10c602616a0c50f2843861e3b663188773c0a1530fdac35ee7e5c38f57ce7743

binary data wasn't supposed to have binary parts, huh?

*puts all my data in fixed point*

floats are great for simulations and stuff, but are overused because languages just don't have any tools for dealing with fixedpoint/fractions so you either use floats or end up with some extremely unergonomic solutions

like, realistically, a 8 bit signal most of the time is most accurately described to be in a range between 0 and 1 (inclusive) which each consecutive value being 1/255 larger than last one, but there's zero ways of marking that in a language

related'ish to the current topic: https://posithub.org/

posits are arguably worse than floats and completely misunderstand *why* floats are designed like they are

hm, I don't think I've seen that in any langauge except Julia. Usually 0.8 fixed point would go from 0/256 to 255/256 The fixed point lib for Julia (used for sample values in image data, for example) does have a "normed" fixed point type that uses 2^N-1, e.g. 0/255 to 255/255, but my gut feeling is that that would be a lot hairier to implement than regular fixed point (since regular fixed point is just regular integer math with a little bit of scaling)

(ofc, if you reserve one bit to use as the integer part, e.g. 1.7, then you can easily represent 1)

It's quite simple: fixed point is good if you want uniform precision across the representable range, floating point is good if you want the available precision to be proportional to the amplitude of the number, i.e. it's closer to a constant percentage precision rather than a fixed constant. Both can be useful/desired, it just depends on the application. They both have the problem that they're finite so they cannot represent all countably infinite rationals let alone all uncountably infinite reals, but that's a fundamental trade-off you cannot avoid if you want efficient operations (in particular, constant-time basic operations).

the problem is that floats are used where they don't make sense because they're ergonomically built-in to most languages, unlike fixed point

e.g. there's zero reason for why minecraft should spend half of the total available state space for a single block, but because floats are easy to use, it does

eh?

https://minecraft.wiki/w/Java_Edition_distance_effects

fixed point is kind of also built-in in all languages if you just use ints with an implicit denominator...

implicit denominator sucks, and multiplication/division becomes annoying

float also has the advantage of a way bigger range

because the exponent is itself binary encoded

Using floats for rendering and physics seems quite reasonable to me.

it is actually better than using fixed point because you want the accuracy to not depend on the scale of the number

physics should work in relative precision

if you multiply a number by 2, it shouldnt change the precision

which is only the case for float and not fixed point

This is more confusing for me

think for example with alpha channel - when multiplying another channel by it, you want max alpha to not change the original value, and min alpha to set it to 0

if you map 8 bit integers to be 1/255 * int_value, normal multiplication will result in that

while a 0.8 fixed point with 1/256 * int_value won't

I don't need syntax sugar when low-level programming

Especially when analyze the overflow or underflow

I don't think this can really be called "low-level programming", and it's not syntax sugar

To do the multiplication properly with these semantics without language support you have to do a whole bunch of conversions

https://www.theverge.com/news/812848/chatgpt-legal-medical-advice-rumor

how many messages do I need to send before the bot gives me the core dev role <:KekDog:805390049033191445>

I'm slowly catching up haha

Vote Booster: Vote now for a 10% boost. <https://arcane.bot/vote>

Vote Booster: Vote now for a 10% boost. <https://arcane.bot/vote>

still a long way before 53

Vote Booster: Vote now for a 10% boost. <https://arcane.bot/vote>

Vote Booster: Vote now for a 10% boost. <https://arcane.bot/vote>

my level is too high, can I tell the bot that from now on I want my level to decrement when the bar is full?

I would be a lot less high if fragmented posts didn't count lol, it needs like a 1 min timeout

Vote Booster: Vote now for a 10% boost. <https://arcane.bot/vote>

https://www.threads.com/@thetrickybuddha/post/DQnIBtGDKq3 interesting effect

they vibing

mathematical hot take, consider: instead of writing `f^n(x)` for repeated function application, we could write `exp(log(f)*n)(x)` ||i don't care if it doesn't make sense||

people have written worse things

and, I mean, there are things such as (exp(d)f)(x) I believe

I've seen this in a youtube video about operational calculus: the difference operator is equivalent to exp(d/dx) - 1

https://shnatsel.medium.com/the-state-of-simd-in-rust-in-2025-32c263e5f53d

Not sure how much has changed since y'all looked at Rust SIMD solutions for jxl-rs, but Shnatsel but published an overview.

I follow the space fairly closely 😛

*nod*

How's Pulp, in your opinion? Shnatsel told me a few days ago he didn't understand why jxl-rs has its own SIMD abstraction instead of using Pulp

> Fortunately, this problem only exists on x86. > > ARM made its NEON mandatory in all 64-bit CPUs and then didn’t bother expanding the width beyond 128 bits. (Technically SVE exists, but in 2025 it is still mostly on paper, and Rust support for it is still in progress). yeah not quite, NEON-fp16 is still optional

I think it's OK, but it tries to support everything (including, in some cases, stuff that happens via inline asm) and I have no intention to safety review all of that code 😄 also I am hoping that a custom simd abstraction can more easily be ported to "effective target features" or whatever once I do the work to actually implement & stabilize that

that reminds me, will jxl-rs be able to target wasm?

I like the idea of having only JXL images on my website and serving a wasm-based decoder for browsers that don't support it yet

eventually, sure

adding an instruction set should not be *too* hard

ah, interesting. mind if I forward your comments to him, since he's not in this Discord?

>simdeez lol

another thing is that pulp doesn't do sse4 - understandable, but possibly not the tradeoff I want to make

the real trick/question is no-std support

sure, feel free to do so and/or to create a group DM 😛

also "eventually" 😉 I think I will try to keep the usage of mutex & friends to a minimum, and just stick to atomics, which should make everything work (or be easily adaptable) in no-std as long as you don't need threads, but definitely not a priority

can't wait to decode jxl in a shader xD

I'm not sure how big of a blocker that would be. thinks like heap allocated types and such from the std will work fine in wasm, it's just that I/O functions expect to be running in an OS and things like that. but so long as there are available function exports for decoding images and all the other necessities, that the wasm module exposes, and another wasm module can hook into, it should be possible to use it in combination with other scripts to decode images on websites

there are specialized toolchains for compiling rust to wasm that deal with a lot of it, but someone does need to explicitly add support for those. being as no-std as possible definitely helps to prevent possible incompatibilities though yeah

Debian graphical installer not supports WPA3?

https://user-images.githubusercontent.com/430322/146363876-cffc7936-c79b-407f-a057-51e4abff71dc.png

??

(from https://gist.github.com/zingaburga/805669eb891c820bd220418ee3f0d6bd )

https://www.youtube.com/watch?v=uFI5WpK2sgg

it's two and a half hours, I'm not sure if I'm going to be able to watch this in one sitting but it's bound to be good

Vote Booster: Vote now for a 10% boost. <https://arcane.bot/vote>

Got a job interview in a few days with intoPIX, which created (or co-created) JPEG XS <:KekDog:805390049033191445>

exciting! good luck :D

[⠀](https://cdn.discordapp.com/emojis/674256399412363284.webp?size=48&name=pepelove)

Who at intopix are you talking to? I know some of those people from JPEG meetings.

https://gstreamer.freedesktop.org/documentation/svtjpegxs/svtjpegxsdec.html?gi-language=c#svtjpegxsdec Speaking of JPEG XS, is this Motion JPEG rehashed for the 3rd time?

JPEG XS is not at all like motion JPEG. It's based on a low complexity wavelet transform, and it's meant for ultra low latency applications like wire protocols for video production workflows. It has an inherent latency that can be as low as 4 rows of a frame.

Charles

and apparently they're only 6 people in that "department" for C/C++ software development, which is what I'm searching for

I would love a job where I can develop Shell/Bash scripts as well, we'll see 😄

Haven't met Charles yet. If you see Tim somewhere, send him my regards 🙂

Sure thing!

how does JPEG XS compare to VESA DSC? Seems like they're kinda in the similar ballpark?

Wow... No wonder some people don't care about progressive loading https://x.com/mtlushan/status/1985901533116989839

many servers are limited per client anyway, user speed doesn't mean the same speed will be to every server

But at least the bottleneck will never be your internet connection

And Discord being Discord... https://fixupx.com/itsfolf/status/1985814954990072130

is this why I have recently been seeing people say their CPU usage spikes up when typing in Discord?

actually yeah that's what this has to be lol

there's even a Vencord plugin to disable typing animation <:KekDog:805390049033191445>

weird thing is I have only *recently* heard of this being a problem so they must of somehow made it so much worse that now everybody is noticing lol

https://unit42.paloaltonetworks.com/landfall-is-new-commercial-grade-android-spyware/

after seeing someone mention typst again here I remade my resume in it

I took a template and started editing it with no experience aside from briefly checking typst out before. it was pretty easy. I got a beautiful document out <:BlobYay:806132268186861619>

Yes!

It's great

<:Hypers:808826266060193874>

feature request for discord: don't highlight channels as if they have new messages if the only new messages are from a spammer and they have meanwhile already been deleted.

So that's why I saw those highlights without any new message

I got a flashlight which advertised stepless dimming. there is a dial which appears stepless (just from how it feels), but the flashlight only has six brightness levels. however, it fades between them, so it appears stepless while you're adjusting the dial. so... it has the hardware for being stepless, because it's already showing intermediary values... why didn't they just make it stepless?

very weird

maybe it's because they'd have to use more complicated circuitry to remember the last brightness state or something

aside from that, the flashlight is really nice :3

a peak intensity of 1500 lumen for the main light, which interesting optical design which makes for a smooth wide beam (the single diffused and two smaller emitters) and a focused beam (the single larger emitter). it also has a side light consisting of a bunch of warmer LEDs that go bright enough to light up a room at night. and both the main light and side light support """"stepless"""" dimming

on this one you can better see the optics

The humble potentiometer:

right??

actually it's probably a rotary encoder because it's not limited in its rotation

It has a mug mode?

warmer *(adjective)* LEDs <:KekDog:805390049033191445>

they do not warm things

or that's not their main feature

the main light is actually bright enough to burn my hand if I keep it there for.. hold on

four seconds

Important science going on here tonight

yes yes

I think I have flashlight autism

<@321486891079696385> they are speaking your language

This is normal. All that light energy is getting converted to heat.

it's a crazy experience, so much energy coming just from the light of LEDs

Clearly you never had your friend draw on your hand and then use their phone's flash at school

man I miss crank hand lights with old school bulbs, not the most efficient but they are nice.

I should make a good one

https://openai.com/index/fighting-nyt-user-privacy-invasion/

Lmao, that's rich

It's real https://store.steampowered.com/sale/steamframe

but more importantly https://store.steampowered.com/sale/steamcontroller

<https://store.steampowered.com/sale/steamframe>

https://youtu.be/OmKrKTwtukE

That's gonna annoy me, because the 'new feature' has been in Steam Link for years

the rumous of Valve sponsoring FEX development appear to have been true

Proton will now have FEX to run x86 games on Arm

https://youtu.be/Y1azUz2JVn0

I have been listening to this pretty much on repeat for the past three months (roughly since the upload)

I like how it picks up around 9:30 or so

it’s even kind of visible on the moodbar 😄

wish they would have done box86 instead, it works way better

I just wanna know what they are doing for android support

FEX has a much more solid foundation

Better long term bet

how so?

box86 is very well developed and even supports multiple archs

the basic thing I've heard from people involved with this sort of stuff (like Asahi's support for running Windows x86 games on Linux Arm), Box64 gets its higher perf by being less correct in how it emulates x86_64, while FEX has focused on correctness from the start. So while FEX is harder to set up (part of that correctness compes from having a while x86(_64) base system rather than doing library forwarding), it has a better baseline for compatibility.

(And of course, FEX has been getting a lot faster than it was before, especially since Valve started contributing)

anyway who's excited for HL3?

I'm excited to see what people do modding wise as it will be a non-VR singeplayer first party Source 2 game

I wonder if the fact that Box86 was AArch32-only back when Valve started working on this influenced the decision.

yi

I haven't been able to find people online comparing the performance of fex vs box86, except one very old post before fex got a ton of improvements

box64 been around quite a while tho so I doubt it

android emulation community does quite a bit

But until not long ago, Box64 could only do x86_64

that's not really an issue for steam

Box32 (x86_32 on AArch64) is a relatively recent addition

Arm is trying to phase out AArch32 from their mainstream application processors, there will be no 32-bit support in phones and anything using phone hardware, soon.

So if Valve started working on this all a few years ago, they might not have known whether they'd end up with a SoC with 32-bit support.

maybe eventually, not soon, especially not with the hardware valve is using

Android didn't even stop requiring 32-bit until super recently.

It's the previous generation compared to the one that's in flagship phones now.

Almost all of the new SOCs are going to support 32-bit for quite a while.

anyway that was just speculation over one thing that might've swayed them

if it's not that it must've been something else

Maybe they want to bring it macOS 🤣 (which is AArch64-only)

lel

that would be pretty neat

I'm just happy that steam on android isn't going to need to emulate steam now

not that it actually changes that end of things that mhch

If nothing else I guess maybe CodeWeavers can take advantage of this once Apple phases out Rosetta2.

but yeah, I wondet if they will support x86 apks lol