🪅-progaming

[supple whale]

big difference

[royal nymph]

not what i asked

[supple whale]

no it actually matters

because threading is "fuck it, dont care throw this task on another thread, w/e, jsut dont fuck my main thread"

multithreading is "fuck i need this complex task computed as fast as possible across all cores"

[royal nymph]

it's kinda the first but i also don't want it to take for ages so multiple workers on demand

[supple whale]

so im making sure this aint an A/B problem

[royal nymph]

i wish i was using go rn

[supple whale]

what are you trying to do?

[royal nymph]

go doThing()

[supple whale]

because i know few things that scale well with multi-threading in JS

[royal nymph]

just image manip

it's not about scaling, it's just for @elder yarrow

just not to block the main thread (and also be able to do multiple requests at once)

rn if u generate a gif with venbot it just blocks the entire bot lmao

[supple whale]

yeah workerpool is useless for that then, you want to spawn a separate worker for each manip, as the requests come in

since you cant do image manip across multiple threads at once

[royal nymph]

how is it useless? isn't using a pool that can reuse workers way more efficient than always spawning a new one

[royal nymph]

why not?

[supple whale]

because it simply doesnt scale with threads

so you'd be wasting compute

[royal nymph]

how does it not scale wdym

this isn't gonna be editing 1000 images at once

a few at once at max

those should be done parallely

[supple whale]

yeah, and you'll be performing likely synchronous merge operations on a single data array, this isnt C, and you wont be micro optimising mutexes

node threads are green threads, so you can simply spawn 10k of them, they are very low cost to start

but sorry, to answer your question, i've never used workerpool, since as a rule of thumb, all worker pool implementations in JS are dogshit, because its JS, when I need that kind of performance I usually opt for NAPI and C multi threading, which performs x1000 better

lemme have a gander at the codebase and i'll give u my opinion

its dogshit

its a very generic worker pool implementation, but it converts all functions you pass to it to a string first

which it then executes in the worler

you could write something better yourself in 2 minutes

[royal nymph]

yeah i wouldnt use that lol you can also create a worker pool from a file

2 minutes is a bit of an overstatement

[supple whale]

no

really

i'm not even joking

this is rudementary at best

it horrendously overcomplicates this lol

[supple whale]

implementing a promise task queue is like what, a minute? then simply manage the state of the task que via message channel between a list of workers

anyways, i'd still probs just spawn a worker per action, since i'm lazy as shit, and worry about perf if that problem shows up then, but it shouldnt

[crude star]

it is

[supple whale]

no worker pools have very good uses, its just.... they arent very common

[crude star]

launching a worker obviously has its overhead

[supple whale]

V's use case is decent enough

just the language itself isnt quite suited to it

[crude star]

and idle workers are idle so its not wasting anything

[supple whale]

yeah, but realistically its no more than creating an array

its just an object with its own scope

its a green thread

anyways @royal nymph https://github.com/piscinajs/piscina

this seems like a less... insane version of what u need

i dont think it has serialization doe, which might be a problem

[crude star]

if you launch 1 per request and dont limit it you'll selfddos

[supple whale]

nope

[crude star]

yope

[shrewd canopy]

Why not just find a proper async version of library for image manipulation

[crude star]

what's there to async

[supple whale]

because they dont exist

[crude star]

js is single threaded

[supple whale]

node-canvas is the only thing

and its sync

[royal nymph]

this example from their readme is insane

'use strict';

function add({ a, b }) { return a + b; }

function multiply({ a, b }) { return a * b; }

add.add = add;
add.multiply = multiply;

module.exports = add;

why are they abusing the add function as an object instead of just creating a new object lmao

module.exports = { add, multiply } 😭

[supple whale]

hm?

seems quite sane no?

[royal nymph]

it's below

[royal nymph]

I'm talking about this example specifically

[supple whale]

yeah thats for multiple functions per worker

i dont think you'd care abt that tbf

[royal nymph]

did you even read what I said

[supple whale]

yeh

[royal nymph]

the cursed part is that they're doing ```js
add.add = add;
add.multiply = multiply;
export add

[supple whale]

well i skimmed over it because i'm looking for that one benchmark that shows that nodejs can easily handle like 20k workers at once

for rini

[shrewd canopy]

Probably so const add = require("mymodule"); console.log(add(1, 2)); and const { add } = require("mymodule"); console.log(add(1, 2)); both work

Eris does same afaik

[crude star]

add({ a: 1, b: 1 })

so sane

[supple whale]

https://pkolaczk.github.io/memory-consumption-of-async/ ah yeah it was this

[royal nymph]

is that actually real everyone online says you should only create n cores where n = cpu threads lmao

[crude star]

i dont doubt that the selfdos is one million image generations at once

[supple whale]

yep, its green threads

aka "fake in software threads"

[royal nymph]

I mean nothing wrong with that

[supple whale]

like you can easily run 1 mil threads in nodejs

just like go

[royal nymph]

go threading is really efficient

[supple whale]

yeah, its GC isnt tho :////

i wish they'd improve it, go would be unbeatable if they did

[crude star]

never had problems with it

[royal nymph]

go has the best multithreading I've ever used

[supple whale]

oh yeah no 100% agreed

[royal nymph]

and improve the error handling while they're at it (mainly reduce boilerplate via stuff like trust's ?🫨

go would be best language then

[supple whale]

the only thing stopping nodejs from having go-like threading is the language syntax, because the featureset allows it

which suuuucks major dick

we need inline threading so much its unfunny

[royal nymph]

honestly i might just use a single worker with a queue

[supple whale]

like just let us do

function goCommitDie() {}

new Worker(goCommitDie)

[royal nymph]

people can wait a bit for their request it's not like 200 people will be using it at once

[supple whale]

yeah pretty much

[supple whale]

still probs more work than just using this lib

since it alr implements a q

but yeah

[royal nymph]

is there any point even sharing small buffers (like few hundred kb) or is it cheaper to just copy them

[supple whale]

always share them

sorry

always transfer them

[crude star]

you should add an image generation currency

[supple whale]

always use transferred buffers

its 0 cost

[royal nymph]

yeah that's what I mean

transferring

[supple whale]

and always faster

[royal nymph]

didn't know the exact word

[supple whale]

it can cause errors if you dont handle it correctly yes

but its 0 cost, and always faster

[royal nymph]

is it really 0 cost tho

[supple whale]

yes

its the same process, it operates on the same vm, in the same memory

[royal nymph]

In go it's usually faster to copy function parameters than pass a reference

[supple whale]

Unlike child_process or cluster, worker_threads can share memory.

[crude star]

for small arguments this makes sense

[supple whale]

it depends on ur throughput, if the data is like >50MB/s it REALLY helps

if its not then fuck cares tbf

[royal nymph]

well duh

how often are u passing 50mb to a func tho

[supple whale]

no sir, 50MB/s

not 50MB/call

it could be 1k calls of 50kb

objects are also automagically transferred

so no need to worry there

you only need to specify transferable arguments manually for potentially problematic things like direct memory aka arraybuffers, GPU bitmaps, and images

since those are often controlled

[royal nymph]

nah small data is actually cheaper to copy due to cache

[supple whale]

in go maybe?

im fairly sure not in v8

again, its pretty much the same process, so there's no copying even happening

[royal nymph]

if you pass a reference it has to deref it from memory which is slower

[supple whale]

yes but its pretty much always faster than structure cloning it

unless the data is like 8 bytes

[royal nymph]

and more efficient allocation

it can put the value on the stack

[royal nymph]

go really optimised struct cloning

Also gc shenanigans

[supple whale]

yeah for go its defo true

but i'm 99% sure thats not true for v8

[royal nymph]

if it can put it on the stack it doesn't have to gc

if you use a pointer it likely has to put it on the heap and then it has to gc it later

[royal nymph]

:P

[supple whale]

the smallest data benchmark i can find is for 32MB

which is still a nuts difference

[royal nymph]

this post seems fire https://medium.com/eureka-engineering/understanding-allocations-in-go-stack-heap-memory-9a2631b5035d

[royal nymph]

yeah id always transfer larger data like this

I was mainly interested in smaller data

[supple whale]

32MB isnt large XD

[royal nymph]

well

compared to 500kb it is

[supple whale]

i fucking do 3GB/s for my subtitle renderer

.

[royal nymph]

I always shit myself when I see huge data structures until I realise it's actually not as much as it seems

like casual array with 1 million elements

it's not actually that much

[supple whale]

lmao yeah

i recently got my torrent client to run at like 2Gbit/s @ 1k connections on a single fucking thread

and its so horrendously simple to do high perf at scale in JS its crazy

just dont .slice() and dont copy

its always that simple

[royal nymph]

holy fuck js array limit is 4 billion

time to make a 1 billion element array

[royal nymph]

yeah, but it depends a lot on how you write code

[supple whale]

arraybuffer is bigger afaik, but you need to manually increase the max V8 VM memory size

[supple whale]

eventemitter

[royal nymph]

a good while ago I was playing around with js performance a bit

I wrote a simple password brute force algorithm

my first implementation was so slow cause I used string manipulation & comparison

then I rewrote it with nodejs Buffer and it was better but still slow

then I rewrite it with raw Uint8Array and it was incredibly fast

[supple whale]

anyways, always use trasnferables, its a good habit to have, and its good to get used to the edge cases

[supple whale]

yeah nodejs buffer is doggy

[royal nymph]

lesson of the story Buffer sucks

[supple whale]

its ancient

so it suffers a lot from backwards compatibility problems

it sucks because it needs to support its first versions, which sucked

and readables suck for the same reason do

[royal nymph]

I compared my final implementation to a similar C implementation and it was surprisingly close

only like 20% slower iirc

[supple whale]

i always replace all readable usages with streamx, because its API is 95% the same, and its x10 faster, and interoperable with streams

[royal nymph]

what's streamx

[supple whale]

you could get this down to even less im sure

but not bad for a fucking GC language eh???

v8 is NUTS

[supple whale]

its like nodejs streams

just, not dogshit

3rd party lib

https://github.com/mafintosh/streamx

im not joking when i say ALWAYS use streamx instead of nodejs streams if you need to

but its always best if you use async iterators

but u cant always, so just use streamx

its SO much faster, because it doesnt touch buffer

nodejs streams always do a Buffer.from() on incoming data

streamx doesnt care, it just passes whatever data is given to it

[royal nymph]

I never use streams much cause I don't usually have a need for them

[supple whale]

and it also uses microtasks instead of timers

so the perf diff its nuts

[supple whale]

nice, keep it that way!

and if you DO need to just use:

for await (const buffer of fs.createReadStream('/path/to/data')){
    //stuff
}

and

async function data * () {
   yield 42
}

Readable.from(data()).pipe(fs.createWriteStream('/path/to/data'))

its unfunny how much faster it is than streams, even with the promise overhead

[royal nymph]

I think the only time I used streams is when uploading / downloading files with fetch

res.body.pipe(createWriteStream())

[supple whale]

i legit did this for the torrent client lib i work on, made it 80% faster, reduced bundle size down to 400kb from 3MB, and dropped memory usage by over half

its nuts XD

hashing in nodejs still sucks tho

[royal nymph]

I hate how in nodejs there's both web and node streams and different apis use either one so it's a pain to make them compatible and you have to constantly Readable.fromWeb

like i get the desire to be spec compliant but man does it suck

[supple whale]

the web crypto api in nodejs is fucking slow, and the node api for it is synchronous and blocking, but x4 faster :/

[supple whale]

oh fuck both of them

i strictly operate only on async iterators

seriously u should adopt this ideology

just async iterators everywhere

fuck stream bullshit

[royal nymph]

how do you do this with async iterators :P

[supple whale]

here u dont need to since u dont douch the data

[royal nymph]

res.body is response from nodejs global fetch

[supple whale]

so what u have works fine

[royal nymph]

this is what I mean btw https://codeberg.org/vee/bot/src/branch/main/src/util/fetch.ts#L39-L40

[pearl stag]

fetch.ts: Lines 39-40

const body = Readable.fromWeb(res.body);
await finished(body.pipe(createWriteStream(path)));

[supple whale]

i'd use file handles, ut yeah

yeah that works perfectly fine

[royal nymph]

the fact that you gotta convert it to node from web first annoys me

[supple whale]

its not that bad

node fs is stream only for the most part

ofc i have my own implementations which are faster, using handles and data flushing

for random access file

but u dont need that perf, so that is perfectly usable

[royal nymph]

yeah it almost never matters for the things I'm doing

[supple whale]

import 'fast-readable-async-iterator'
import { open } from 'node:fs/promises'

const res = await fetch('gay')
const fd = await open('sample.txt')

for await (const data of res.body) {
  await fd.writeFile(data)
}
await fd.close()

i think this should work

i dont remember i didnt use file handles for quite some time now

i dont remember if writeFile appended or overwrote

[royal nymph]

completely unrelated but I wish web had something like android's recyclerview

its really more difficult than it should be to make performant lists in html

like rendering thousands of items efficiently

android's recyclerview is so nice to work with for stuff like this

[supple whale]

you mean dynamic lists?

idk what recyclerview is

yeah dynamic lists

[royal nymph]

take a lot of items and render them efficiently in complex layouts with smooth scroll, filtering, etc

[supple whale]

.parent {
  will-change: scroll-position;
  contain: strict;
}

thank you for coming to my ted talk

[royal nymph]

well it's not that simple lmao

[supple whale]

it actually is

[royal nymph]

if you just render all elements at once it's gonna be uber laggy

[supple whale]

skia is actually that good

[supple whale]

nope

tested this on lists with like 40k items

works like a charm

its like 80% of the perf of a dynamic list

for how easy it is, might as well

there's a CSS property for this

if you have fixed sizes

one sec

contain-intrinsic-height

pretty much does dynamic lists like that

with content-visibility: auto

that is

here

https://web.dev/articles/content-visibility

this is simply why web is better than android or switft, you dont need to bother with implementing dogshit like that for rendering UI, the web does it for you

and its craaaaaaaaaazy discord uses none of this

i had to write a fuckload of css just to unfuck the discord UI perf

[royal nymph]

extremely simple example: there's noticeable lag here when you click generate and only will get worse the more complex your layout becomes & styles are added

this just rendered everything and if you had 1000 notifications it would lag for like 2 seconds no joke

i migrated it to Discord's lazy scroller implementation

[supple whale]

this uses none of the things i mentioned doe

[royal nymph]

okay so make it speeeedddy

show

also real word example is here https://badges.vencord.dev/explorer, the performance is really shitty

although the lag is kinda expected cause it's loading like 400mb worth of images (and this site is incredibly low effort)

[supple whale]

ez.

but yeah 99% of the perf here is the images, not the list

so its not even the same problem

ez

[royal nymph]

is that with the images

[supple whale]

actually

[supple whale]

without the changes i cant get a single frame to drop on this

i added 2 0's to the list

and no problems

ah it finally died at 10000000

[royal nymph]

i added two zeroes and got this

and with one added zero theres incredible lag

this is 100000 elements

[supple whale]

yeah i did a

for (let i = 0; i < elements.length; i += 100000) {
    const batch = elements.slice(i, i + 100000)
    container.append(...batch)
}

[royal nymph]

i see

thats also laggy for me tho

[royal nymph]

like here

[supple whale]

yeah i gor the repro

yeah

i see your skill issue

you're thinking that the JS lagging is the UI lagging

the UI isnt frozen, its the JS chugging iterating the array

so this is a bad repro

[supple whale]

and this is a bad repro, because 95% of the render goes towards the images

[royal nymph]

there's still noticeable lag between the append calls finishing and the ui appearing

also those js measurements seem really off so the js thread seems to be freezing a lot

because it's rendering?

[supple whale]

no, because its iterating tru the array

<style>
    #container {
        height: 100vh;
        overflow: auto;
        will-change: scroll-position;
        contain: strict;
        content-visibility: auto;
        contain-intrinsic-height: 34px;
    }
</style>
<div id="container"></div>
<button id="generate">Generate</button>

<script>

    generate.onclick = () => {
        const data = Array.from({ length: 1000000 }, () => Math.random())

        const elements = data.map(n => {
            const p = document.createElement("p")
            p.textContent = n
            return p
        })

        for (let i = 0; i < elements.length; i += 100000) {
            const batch = elements.slice(i, i + 100000)
            container.append(...batch)
        }
    }
</script>

this will get the render down to 0

if u enable rendering -> frame renderign stats

and if u go on badges, and enable paint flashing

you'll see why that's lagging

you're also not using decoding="async" on the image

[royal nymph]

this still has really laggy scrolling

try scrolling fast you will get a lot of white screen

[supple whale]

not for me?

oh yeah

fucking duh

thats ur GPU running out of memory

any virtual list will have 1:1 the same problem

scratch that, no virual list will catch up to those speeds

but i'm scrolling at my display's vsync

aka 165fps

[crude star]

virtual lists are supposed to fix that as well 😭

[supple whale]

no sir, virtual lists do not manage GPU memory

they cant tell skia to evict a bitmap

they can remove an element, which will schedule it for eviction

but that's still slower than simply scrolling it out of view

[crude star]

why would it not deallocate immediately

[supple whale]

because that's expensive

its always done in batches

thats how GC works sir

[crude star]

alright ma'am

[royal nymph]

no?

[supple whale]

sir skia limits itself to <1.4GB of GPU bitmap memory

[royal nymph]

why

[supple whale]

worst part, its shared across all active chromium processes

so if you have, discord, steam open, its reduced even more

[royal nymph]

wait what 😭

[supple whale]

lord i wish i knew, i've been trying to change this behavior for 2 years now

[royal nymph]

i have so many chromiums open is that really why

[royal nymph]

[supple whale]

especially if its shit like steam's or discord's animated image frames, which are gifs, which allocate a single bitmap per image frame yes

because gifs are horrendously inefficient when it comes to repaints

[royal nymph]

oh i know

[supple whale]

again, open ur badges thingy and enable paint flashing

[royal nymph]

ive had my share of pain with gifs

[royal nymph]

this is like genuinely so dumb it should just be configurable

i dont mind it using up most of my resources for a better experience

that's what i bought it for

[supple whale]

fucking

mood

as

fuck

ive been fighting this for ages

and i still dont know the exact in and outs of it

but its horrendous

worst part? your shit will start blinking if ur fps is too high

and u have other views

https://cdn.discordapp.com/attachments/1390481081626525786/1390481092083056660/IMG_9062.mov?ex=689d25ee&is=689bd46e&hm=6fd46f89fe0b835b5011876b056db61c78b6d0b6bddb34be7f9d19af7829916b&

its REALLY bad

[royal nymph]

also i love how this kept stressing my pc until i closed the page just now even minutes after i stopped

[royal nymph]

how do i make that site faster

i get that it's always gonna be laggy cause of rendering so many fucking images but man

why does it even take this long to load from disk cache

[supple whale]

decoding='async'

on the img tag

and not using gifs

also if ur framework supports it, enhanced images

https://svelte.dev/docs/kit/images

[royal nymph]

"my framework" bro it's just a single html file 😭

[royal nymph]

yeah this seems to have made it better but now the ui is kinda glitchy

like this white empty bar at the top for a bit while it's loading

[supple whale]

LMAO

[supple whale]

its legit this XD

[royal nymph]

why does this happen

shit is already loaded in and should just show empty black border

[supple whale]

reload

and it probs wont look like that

[royal nymph]

i threw this crap together super quick

[supple whale]

it for the most part works for me

scorllign cant keep up

[supple whale]

but u could do this to fix htat

[royal nymph]

nah look

it wasnt like this without async decoding (instead of empty white bar it would show empty black borders)

[supple whale]

disable gpu acceleration and that will go away

:))))))))

[royal nymph]

really

[supple whale]

yes!

[royal nymph]

dude it got way worse lmao

the white boxes

[supple whale]

lmao srs?

[royal nymph]

this is with --disable-gpu

[supple whale]

X D?

[gilded surge]

at least vencord badge explorer uses less ram than nixos config options page

[supple whale]

enjoy linux i guess?

[royal nymph]

what does this mean

[supple whale]

that you manually override rgb profiles for ur chromium instance most likely?

most likely?

[royal nymph]

i def dont

[supple whale]

simple linux enjoyment then

[royal nymph]

well i dont have those in chrome fwiw

[supple whale]

thats what i get

but yeah simply too many animated images at once

[royal nymph]

nah wtf just happened 😭

[supple whale]

and shit shits itself

[royal nymph]

the entire ui is fuckded up

[supple whale]

yes, contain: strict isnt easy to work with

and intrinsic size needs to be sthe size of the element and its padding

[royal nymph]

oh my god zooming out was a mistake

[supple whale]

you're legit working with the worst possible case scenario here

aka a fuckload of animated gifs

and massively oversized images

[royal nymph]

i should just merge them into one big spritesheet

[royal nymph]

lol yeah a lot of the old badges are insane

all the new ones are sane (mostly webp and all 64x64)

should probably write a script at some point that optimises all the old ones but no point really

[supple whale]

just throw it into sveltekit real quick and use enhance:img

XD

[royal nymph]

i mean it's just taken directly from https://badges.vencord.dev/badges.json

[supple whale]

the UI isnt

[royal nymph]

ya

is there any way to have fast scrolling like this not look terrible

or do we just have to accept that

[supple whale]

uuuh

nuke all ur chromium processes and start ur chromium process with --force-gpu-mem-available-mb=9000 then use d3d9 ANGLE

will improve it a bit

not by much tho

[royal nymph]

crazy man https://badges.vencord.dev/badges/1162297670254215198/1-a42510fe5b56b575c703cde128cb54cdcee60e4c.gif

[supple whale]

also disabling raw-draw might help

[royal nymph]

only 500x500
only 75 frames
still 15mb

god i love the gif file format

dude i converted it to webp with no other changes and it's just 1mb 😭

[supple whale]

convert it to a webm

legit will probs cost less to render

XD

[royal nymph]

nah bro this shit is put in a img tag

[supple whale]

worst part? still asaggy as the original gif

as it causes a paint update each frame

KEKW

[royal nymph]

anyway the explorer isnt that important anyway i just made it for fun

this is just for displaying badges in discord @steady cloak

[royal nymph]

wouldn't it be a lot better cause it's way smaller?

40kb vs 15mb

if it really is less efficient than video why don't browsers make it more efficient

gif i get - gif is just inherently atrocious

but an optimised webp/avif?

[supple whale]

a bit yeah

but not much

the repaints are the biggest problem

[royal nymph]

i will simply merge all images into one single sprite sheet

[supple whale]

I honestly don't know what's the best solution here

[royal nymph]

idea for you: make the gif picker in discord less terrible

[supple whale]

most people just use videos for this since they are less laggy

[supple whale]

height:100%

fixed

legit no cap

[supple whale]

is it finally a PR you'd merge in <3 months?

[royal nymph]

didnt feel any different (also had to use vh not %)

i have like 500+ favourite gifs

when you scroll it's just blue loading for eternity

a while ago i thought about making a plugin that generates thumbnails for each gif and displays those while stuff is loading

[woeful perch]

Hiya !

[jade stone]

what is the rust equivalent of make -C someDir

[tired vigil]

For me it's like 1.5x times faster than before, I think the media proxy now returns webps/webms by default

[lavish frigate]

cargo build ––manifest-path someDir? Not sure what make -C exactly does (I do not use make)

[supple whale]

oh yeah @royal nymph add loading='lazy' to the images too

[royal nymph]

yeah i did already ages ago

[supple whale]

hm android wack

[austere idol]

make -C:

1. move to the directory specified after -C
2. call Makefile in that directory
3. return back

[crude star]

-C <DIRECTORY> Change to DIRECTORY before doing anything (nightly-only)

why is it nightly rust is so fucking dumb

[solemn ravine]

@woven mesa trying to add something that presses cmd v to paste into a field but seems like it doesnt work idk how to fix

[woven mesa]

i havent used these apis yet

unfortunately i cant test it and figure out what to do correctly for you

sorry oomf

[solemn ravine]

sad because the apis are super straight forward but idk what’s wrong with it

makes a source and disables keyboard events for the time being

then press keys

also maccy making an entire class just for AXIsProcessTrustedWithOptions is kinda funny

might as well just use AXIsProcessTrustedWithOptions only, it’s just a bool

[woven mesa]

lmao

[ivory heath]

Why isn’t this default?

[ivory heath]

Animated webp support is complicated because it uses vp8 in RIFE in a non standard way

Most people and Linux distributions ship outdated versions of av1 encoders

[supple whale]

riiiiiiight?

[supple whale]

fuck cares, it has software fallback

[ivory heath]

I remember when Debian was shipping libaom 1.1 or whatever when 3.0 was about to release which was orders of magnitude faster

[ivory heath]

Huh?

I said encoders

[supple whale]

ah

[ivory heath]

Avif isn’t common cause to most people they’re expensive to make

And cheaper to just use more bandwidth

Because server hardware is typically old as shit

[supple whale]

no, avif isnt used because it doesnt have full browser support and because its slower to render as it causes more repaints

how difficult it is to make has NOTHING to do with it

for web def what matters is first contentful paint, and browser share support, legitimately nothing else matters, because those 2 things are DIRECTLY tied to revenue

4 cents of extra cpu time to encode an image is not the determining factor, trust me

[ivory heath]

Like, you’re wrong? Lol?

I’ve talked to actual discord staff about this

[supple whale]

its also why progressive loading formats have been completly abandoned, becuase they imploded first contentful paint times

[supple whale]

are you...

that dense?

or just pretending?

discord isnt a website

its an electron desktop app

its constraints are different

discord has a pre-determined featureset and feature support since its always running the same chromium version

[ivory heath]

Discord electron has supported av1/avif for years

[supple whale]

great, cool

[ivory heath]

We used a bug to embed av1 for a long time

But yes discord is different, but most CDNs also have strict transcode deadlines as well

Because of the amount they deal with

[supple whale]

yep, is still is no way related to global adoption of AVIF

[ivory heath]

Then why don’t they if 96% of all web users have browsers with support?

HTML supports built in fall backs

[supple whale]

thats the cool thing

the versions that support fallback support avif

the ones that dont support fallbacks dont support avif

its TRULLY that simple

[ivory heath]

So we’re stuck using jpeg forever by that logic

[supple whale]

yes!

exactly!

its why webp failed

and one of the reasons why JXL failed

tho thats a MUCH longer list

google pretty much said that JXL and AVIF is effectively unusable because they implode first contentful paint times, and implode CPU on lower end phones, which are horrendously common

i could probably find the exact comment

its somewhere on web.dev

[ivory heath]

Webp wasn’t a failure

[supple whale]

it is

[ivory heath]

All of Google uses webp, plenty of popular websites use webp

[supple whale]

yeah this is simply wrong

they still use jpg

[ivory heath]

Probably cause the images being served are too big

[ivory heath]

All of Google images is transcoded to webp, their own assets use webp, all YouTube tumbnails are webp

[supple whale]

why are you making shit up lol?

[ivory heath]

I’m not?

[supple whale]

legit none of google images is webp, none of youtube is webp

they have PARTIAL support for those formats, but 80% of the time they fall back to jpg because they dont have most thumbnails in webp

on google images out of 230 loaded images only one was a webp

on yyoutube not a single one was, and half is avif, since only half are actually in that format

since youtube simply doesnt do avif for most video thumbnails

they also fall back to jpeg based on user agent

[ivory heath]

Correction. The thumbnail for the latest Mr beast video is avif

[supple whale]

cool

go back a year and not a single will be

[ivory heath]

Google mislabels the extensions for images

They’re all typically webp or avif

[supple whale]

yes im going off the actual media type

not extension

im not stupid

[ivory heath]

Idk you’re saying I make shit up when I’m right about YouTube

[supple whale]

and they miss-label the extension of the image, because the source image is still just a jpeg, they just trancode it on the fly based on user agent

[ivory heath]

Maybe a year ago it was different. Does not matter

[supple whale]

ie: https://i.ytimg.com/vi/bTkocnwq45U/hqdefault.jpg?sqp=-oaymwEnCOADEI4CSFryq4qpAxkIARUAAIhCGAHYAQHiAQoIGBACGAY4AUAB&rs=AOn4CLCxWbA0ue94Y4aURK7A-I1Vog2HyQ
vs https://i.ytimg.com/vi/bTkocnwq45U/hqdefault.jpg

source image is simply a jpg

[ivory heath]

Yet it’s avif

[supple whale]

motherfucker are you reading what im saying

[ivory heath]

Google uses their own image formats it’s amazing

[supple whale]

thats the sqp based converted image

[ivory heath]

Yes.

[supple whale]

and the source image is a jpg

they convert it on the fly

based on browser's support

[ivory heath]

You said YouTube does not use avif/webp

They do use it

[supple whale]

yeah the source images are still jpg mate

[ivory heath]

Why does that matter?

[alpine pecan]

@alpine pecan, to be frank you are an annoying, rash, and intolerable person to talk to.
I'm far more invested into the project you have been seeing me complain about in #🪅-progaming than you ever will be and i know more than you ever will.
I'm not sure how the hell a person like you who is so insufferable can exist.

[supple whale]

...?

are you insane or insane?

they cant use avif, because browsers dont support is simply

they always need to use jpeg

they simply detect my browser supports it

the source data is still jpg, it still operates on jpg

because they cant abandon it

because avif isnt good enough to replace jpg

[ivory heath]

Yes?

This is how internet works

Deleting the original jpg also means they add generation loss when AV2 releases in the next 1-2 years

They STILL use it for users who can see it

[supple whale]

yes

but we're still gonna be using jpgs in 2 years

because webp and avif will never be able to replace it

thats why those formats failed

[pearl dawn]

you are both stupid

[ivory heath]

We’re always going to have to use JPEG but as a fallback.

[pearl dawn]

you should be using protobuf instead

[ivory heath]

We can use anything better for 90% or all users

FALLBACKS are a good thing

I don’t think they should go away

[supple whale]

youtube is giga specific because it doesnt run on mobile

which is likely why google decided to use avif only on youtube

and nowhere else

[ivory heath]

Even on mobile they serve avif

[supple whale]

yes but the mobile userbase is 0.1% of what google image has as the mobile userbase

[pearl dawn]

avif was developed at google lmao of course they're going to be using it

also youtube does use webp for things like those animated thumbnail previews

[ivory heath]

Jop

[supple whale]

same for webp and av1

and it still takes ages to adopt it

[pearl dawn]

avif is the newer format so it's the one they predominantly use

[ivory heath]

You’re saying adoption is the total destruction of jpeg and it’s not

[pearl dawn]

googles image proxy encodes as avif by default iirc

[ivory heath]

Discord even on mobile uses webp for thumbnails

[pearl dawn]

jpeg is never dying out lol

[ivory heath]

Yeah

[pearl dawn]

every consumer camera I know outputs jpeg

[supple whale]

google photos does webp for 20px previews, and jpeg for everything else

because webp is too computationally expensive to allow on mobile

[pearl dawn]

every phone automatically converts to jpeg when you try to use pictures taken in anything but the built in gallery

[ivory heath]

They’re moving to avif on pixels

It even has causes some peoples libraries to become corrupted

[supple whale]

i think they are changing that on android with the whole HDR picture/screenshot thing, dont quote me on that, afaik its still WIP

[deep mulch]

yop

[supple whale]

it really is, but w/e

[nimble minnow]

but it really is

[nimble minnow]

my consumer camera outputs avif sorry to say

[deep mulch]

my camera has option to save in HEIF

[supple whale]

thats when you use their "compress your pictures and we'll give you infinite image storage" thing

[pearl dawn]

My friends are super into photography and their cameras output raws and jpeg by default

because jpeg "just works" on everything

[ivory heath]

Hopefully eventually JpegXL not jpeg

[deep mulch]

I had to use this to save my storage
14.8 gb down to 4 gb

[ivory heath]

Do they even use 10 bit jpegs

[supple whale]

google said they will never allow JXL in chromium

[pearl dawn]

jpeg xl isn't ever happening unless chrome adopts it

[ivory heath]

Wrong. They made a compromise to wait for a rust rewrite

Look at the mailing list

Libjxl is too big and.. buggy

[supple whale]

not even remotely close to its biggest problems

[ivory heath]

Memory consumption for encoding yes that’s the biggest one /j

[supple whale]

nope

[deep mulch]

consume my memory

[ivory heath]

What is it then oh great oracle

[supple whale]

decode times, cpu times for decoding, repaints caused during decoding, long large contentful paint times

keep in mind they are targetting android 8 phones from india with e-waste for CPUs with 3G internet

[ivory heath]

Avif is only slightly faster than jpegxl and that’s only 8bit avifs

Enter the domain of HDR and JpegXL starts to win or anything > 8bit really

JpegXL internally represents everything as fp32

[supple whale]

hm i see the problem

you're back in your "lets debate the best format in the world" mode where u ignore all other parts of the conversation, where im saying "whats the most viale format for websites"

JXL is good, but its not a good fit for the web

[deep mulch]

Gif /j

[ivory heath]

It’s good for artists and why safari has native support for it

[pearl dawn]

safari

https://cdn.discordapp.com/emojis/1165720260968710236.png?quality=lossless&name=lol&size=48

[supple whale]

JXL decode times are so bad, this exists as an option for google's image encoder

[ivory heath]

It’s a library feature in libjxl and svt av1 has a similar feature called fast decode

The main problem why avif is faster is chroma sub sampling and unfortunately 444 isn’t supported in the main profile

[supple whale]

man is still talking to himself about the 2 slowest decoding formats

nuts

[ivory heath]

I’m sorry you can’t fathom anything outside of your world

[supple whale]

yes

because my world is web development

your world is "my best media format in the world"

[ivory heath]

It’s not. If it was then I’d be talking about VVC

Both avif and JpegXL have their place. Everyone on the chromium team agrees as well as almost every other person in web dev that actually knows a thing or two about image formats

[supple whale]

yeah they do have their hyper specific use cases

[nimble minnow]

Every person in web dev that knows their stuff agrees that png is the best image format sorry to say

never seen anyone actually use avif

[ivory heath]

So true

[supple whale]

but they simply wont superseed the most commonyl used formats

it wont happen

[ivory heath]

Nothing will replace jpeg not even PNG

it will see usage tho

[supple whale]

png does replace jpeg, its pretty much a back and forth

the both have the same support

[deep mulch]

jpeg isn't gonna go away for a long time probably never

[supple whale]

jpeg is better for compression, png is better for high quality and well... alpha

[supple whale]

yep, which is simply why google's attempt att making webp the default image format failed

15 years and only like 3 years ago did all browsers fully support it

[deep mulch]

oh I thought webp was newer than that

[supple whale]

nah its 2010

I think?

yeah 2010

i remembered right

ff added support after 9 yeasr XD

safari after 12

nuts.

[ivory heath]

Cause Mozjpeg is or was better than webp

It never had a reason to exist

But unfortunately most people don’t use mozjpeg

[nimble minnow]

Breaking: Google announces JPEG support ending in Chrome in 2026

[supple whale]

its still jpeg

doesnt change anything

[nimble minnow]

saw the article yesterday but i forgot the link

[supple whale]

it still has all the limitations of jpeg

[ivory heath]

Webp has more limitations than JPEG

[supple whale]

what encoder u use doesnt matter

[ivory heath]

No 422 no 444

[supple whale]

it could lick my balls for all i care

[supple whale]

that really doesnt matter for web

like trully doesnt matter

it has alpha, animations and better compression ratios

so fuck cares

web isnt about lossless quality

[ivory heath]

For a 8bit 420 image mozjpeg is better than webp at sane quality levels. It’s not a format limitation.

[supple whale]

its about loading content quickly

[supple whale]

there are heavy asterisks around that

sometimes yes sometimes no

more often than not, yes

[ivory heath]

When webp first release jpeg was still the winner

15 years later webp wins by 10%

In MOS rankings

Not 30-50 like google claimed

Fucking 10

[ivory heath]

This still holds true depending on who is looking at the images

Or what format it’s in

I don’t think 10% is worth adding webp support

[ionic lake]

webpiss

[ivory heath]

True.

I will keep using the browser on my Nintendo DS so the unknown can never adopt anything better than jpeg

Or IE6 in a vm works too

[frosty skiff]

nice banner

[spark tiger]

stole*

[frosty skiff]

thanks I made it myself

no need to thank me

[deep mulch]

@young flicker

[valid jetty]

@hoary sluice @deep mulch @royal nymph

[deep mulch]

where is elle

[valid jetty]

obviously

[deep mulch]

I love

@valid jetty I have funny idea im gonna make

@valid jetty IR data transmitter and receiver for transmitting files and whatever

[supple whale]

add rust as a block somewhere on the side

just like off to the side

XD

[deep mulch]

will be fun i think to experiment with different protocol ideas

@valid jetty

[supple whale]

isnt this gonna take like... hours?

like 4-10 kbps

with a 50% error rate

idk bro

[deep mulch]

nop

i estimate I can get maybe at least 50 kbps

with error checking and possibly correction

[supple whale]

i estimate at best 10kbps

either way 10 or 50, its not usable speeds XD

do wifi

[deep mulch]

nop

too complex

probably even 1 mbps is possible with the right driving circuitry

[hoary sluice]

most true statement made this year

[winged mantle]

TIL you can do this in c

typedef struct {
    Color color;
    bool bold;
} SuperPrintOptions;

#define super_print(text, ...) _super_print(text, (SuperPrintOptions){ __VA_ARGS__ })
void _super_print(const char *text, SuperPrintOptions opt);

// somewhere in code...
super_print("hello world!", .bold = true, .color = Color_Green);

just like python

[ivory heath]

Yeah the C preprocessor is very powerful.

[winged mantle]

i never thought of this though...

or used varargs

i'm not a huge fan of macro magic

but this is cool

it's enough to confuse vscode though..

autocompletion doesn't work

[valid jetty]

one trick i found useful was this

#include <stdio.h>
#include <stdbool.h>
#include <stdarg.h>

typedef struct {
    int value;
    bool __last;
} Foo;

void _print_Foos(int _, ...) {
    va_list args;
    va_start(args, _);

    for (;;) {
        Foo foo = va_arg(args, Foo);
        if (foo.__last) break;
        printf("Foo { value = %d }\n", foo.value);
    }

    va_end(args);
}

#define print_Foos(...) _print_Foos(0, __VA_ARGS__, (Foo){ .value = 0, .__last = true })

int main() {
    print_Foos((Foo){ .value = 3 }, (Foo){ .value = 5 });
    print_Foos((Foo){ .value = -42 }, (Foo){ .value = 32 }, (Foo){ .value = 39 });

    return 0;
}

only works for single typed structs

but pretty awesome

[crude star]

insane

use (Foo[]) {} and sizeof

[valid jetty]

i mean yeah you can do that too

[nimble bone]

@valid jetty add // @ts-ignore to Elle

[deep mulch]

@valid jetty make coal

[winged mantle]

does the abs function in c give you abs

[crude star]

no need for vaargs...

[valid jetty]

thats boring..

[deep mulch]

@valid jetty add fast math to elle

use avx512

[supple whale]

fast math doesnt use avx512

it just lowers precision

no?

[deep mulch]

fasterr

elle needs the fastest implementation

[supple whale]

you dare imply that elle should not support wasm

you dare sugggest it to have worse compatibility?

[deep mulch]

yes

anyways won't matter when elle works for java applets and runs on the jvm

@valid jetty Elle JVM NOW

[ivory heath]

Yeah this is why I love that languages like golang implement generics within their own LSP. Rust does as well

And native things like reflection

[jade stone]

what is the proper way to store callbacks in rust

example of what i mean in c++

static std::vector<std::function<void()> callbacks;

void add_callback(std::function<void()> callback) {
  callbacks.push_back(callback);
}

int start_thread() {
  new std::thread([] {
    while (true) {
      // wait for something
      for (const auto& cb : callbacks) {
        cb();
      }
    }
  });
}

[frosty obsidian]

fast math (only integer arithmetic)

[jade stone]

rust statics seem to require unsafe to be mutable and i'd like to avoid that if possible

(if this can't be done in safe code then rust is insane)

[valid jetty]

why does it have to be static

just pass it around

or use https://doc.rust-lang.org/std/cell/struct.LazyCell.html

[jade stone]

bad example

i have one background thread

a function from javascript adds a callback for when something changes

[valid jetty]

@jade stone ```rs
use std::sync::{LazyLock, Mutex};

static CALLBACKS: LazyLock<Mutex<Vec<Box<dyn Fn() + Send + Sync>>>> =
LazyLock::new(|| Mutex::new(Vec::new()));

fn do_x() {
let callbacks = CALLBACKS.lock().unwrap();
for callback in callbacks.iter() {
callback();
}
}

fn main() {
CALLBACKS.lock().unwrap().push(Box::new(|| println!("hello world!")));
CALLBACKS.lock().unwrap().push(Box::new(|| println!("hello sadan!")));
do_x();
}

its a LazyLock not LazyCell

but yeah

this is so fucking terrible to use because rust needs to make sure the things are thread safe

[jade stone]

tysm

[valid jetty]

@jade stone you can also do this if youre sure your functions wont capture anything

use std::sync::{LazyLock, Mutex};

static CALLBACKS: LazyLock<Mutex<Vec<fn()>>> = LazyLock::new(|| Mutex::new(Vec::new()));

fn do_x() {
    let callbacks = CALLBACKS.lock().unwrap();
    for callback in callbacks.iter() {
        callback();
    }
}

fn main() {
    CALLBACKS.lock().unwrap().push(|| println!("hello world!"));
    CALLBACKS.lock().unwrap().push(|| println!("hello sadan!"));

    do_x();
}

[jade stone]

they're not rust functions, they're napi threadsafe_function s

so

ill try it

[valid jetty]

fn() is just a function pointer, Box<dyn Fn()> is a capturing lambda trait item

[valid jetty]

then you can probably just have function pointers

[jade stone]

ah so fn() is like void (*)() and Box<dyn Fn()> is like std::function<void()>

[valid jetty]

i guess so yeah

yeah thats exactly correct

Box<T> is what means "this heap allocates"

[jade stone]

yeah i knew that

[valid jetty]

rust closures dont usually heap allocate, however since theyre being stored, their captures must live long enough and be sharable between threads sooo

[jade stone]

and i assume Mutex is just an raii mutex wrapper

[valid jetty]

in normal sane code youd just see Fn(i32) -> i64 as a type

[valid jetty]

yeah

[jade stone]

but i've never heard of lazylock

[valid jetty]

basically all you have to know lol

[jade stone]

is a lazycell just a lazy value?

[valid jetty]

yeah

[jade stone]

i'll do this after i finish this level of one shot

[valid jetty]

before rust had this, i did it like this

#[macro_export]
macro_rules! global {
    ($name:ident : $type:ty = $value:expr, $getter:ident) => {
        pub static mut $name: Option<$type> = Some($value);

        macro_rules! $getter {
            () => {
                unsafe { $name.unwrap() }
            };
        }

        #[allow(unused)]
        pub(crate) use $getter;
    };
}

so just using unsafe lool

[jade stone]

[valid jetty]

well i technically still do it like this

Kapp

i didnt update yet

thats gonna be part of my migration to rust 2024

[royal nymph]

I did literally the same lol

[valid jetty]

i think i stole it from you ❤️

[royal nymph]

[deep mulch]

@valid jetty

@valid jetty rosinga

[ornate quiver]

ermmm this is actually unsafe and you shouldn't use static mutable globals

[jade stone]

is it possible to get the value of the input to a match clause in the catch all branch without using a variable

something like this

status is a massive enum

[crude star]

error_code =>

else is zig

[jade stone]

that's why i said something like ...

i didn't know the exact syntax

[crude star]

what then

[jade stone]

huh

[crude star]

communication

[twin matrix]

python is cool

[jade stone]

python is evil

give me back my braces

[twin matrix]

ok

[jade stone]

AAAAAAA rust docs are insane

[deep mulch]

@jade stone you love rust

[jade stone]

eh

a bit less than c++

[deep mulch]

@jade stone help

my server is about to explode i think

notihng responsive

[jade stone]

uhhh ssh in

and run journalctl -xef

[deep mulch]

ssh froze

[jade stone]

time to pull out the jetkvm

[deep mulch]

dont have

[deep mulch]

buy for me

bru

i ssh'd another terminal and it froze too

[jade stone]

go reboot

[deep mulch]

guhhh

its in the middle of an upgrade

oh my god

literally every ssh session freezes after like 20 seconds

this is so annoying

[jade stone]

@deep mulch

it's 1 line in c

[deep mulch]

@jade stone Warning: Repository 'http://atl.mirrors.clouvider.net/debian stable InRelease' changed its 'Version' value from '12.11' to '13.0'
Error: Repository 'http://atl.mirrors.clouvider.net/debian stable InRelease' changed its 'Codename' value from 'bookworm' to 'trixie'

[jade stone]

XInternAtom(display, atom_name, False);

[deep mulch]

so much simpler

[jade stone]

yop

and blazingly fast

[deep mulch]

STOP TIMING OUT OMG

????

i dont get it

literally every time i ssh it freezes several seconds later

then im just stuck in this useless frozen ssh session

@jade stone why is it preferring the ipv6 for lan

?????

[fleet cedar]

Which part is one line, ignoring the errors?

Antway it is one line in rust too conn.intern_atom()?.reply()?.atom

[jade stone]

yop

if this errors you have serious issues

rust

[hoary sluice]

what's that font

[jade stone]

https://github.com/ryanoasis/nerd-fonts/tree/master/patched-fonts/ComicShannsMono

[hoary sluice]

do you use this unironically

[crimson sparrow]

I do

It’s great

[jade stone]

Yes

I think I'd pay like $50 if someone made ligatures for it.

[alpine pecan]

dumb

[jade stone]

[alpine pecan]

dumb

[jade stone]

That's not what I mean

It should retain the typedef in docs.

Type aliases often convey a lot of meaning in their names

[frosty obsidian]

typealiases are a very good way to self document code

[fleet cedar]

.map(_).flatten() is better written.and_then()

As clippy points out

Why are you not running clippy

[crude star]

rust api suffers from perl syndrome 💔

[hearty lintel]

whats perl syndrome

[crude star]

"There's more than one way to do it"

that's perl's slogan

[valid jetty]

is flat_map .flatten().map()?

[fleet cedar]

No, .map().flatten()

[valid jetty]

so and_then == flat_map???

[fleet cedar]

It's called and_then on Result/Option/Future and flat_map on iterators

[valid jetty]

ohhh i see

[fleet cedar]

Both are traditionally called >>=

[valid jetty]

that makes sense

[valid jetty]

you use this kind of font so you can get icons in fastfetch and whatever

[fleet cedar]

And in nvim devicons, and powerline-style prompts

[crude star]

rosie dumb

he's referring to the fact it's comic sans

[clear bison]

is it possible to modify and load plugins into discord web on firefox

like saving/loading plugins on local files and running themes in real-time

[limpid mica]

i need someone to talk about x86 instrinsics with

mostly optimizing codesize

rather than performance

[placid cape]

https://una.im/5-css-functions/#

you can create functions in css

[winged mantle]

is it weird that i'm wishing rust had templates

[lavish frigate]

??

[winged mantle]

like with duck typing like c++

[fleet cedar]

That's what macros are for

I hate c++'s duck typed templates

[winged mantle]

a library i'm using has two types which happen to have the same methods but it's not shared with a trait

rust needs templates

template <typename T>
fn quack(duck: T) {
    duck.quack();
}

[fleet cedar]

You could also try writing code that doesn't suck

[winged mantle]

it's the libraries fault though

or maybe not

library limitation

[crude star]

https://ziglang.org/

[lavish frigate]

fn quack<T: Duck>(duck: &T) {
    duck.quack();
}

?

[winged mantle]

that function won't automatically support any type that has a method called quack though

[fleet cedar]

Yep, because function names mean nothing

What would you do if someone passes you a Doctor, which also is a quack?

[winged mantle]

solution: make function names as descriptive as possible

put the whole function body as the name

[fleet cedar]

I think Duck::quack is a very descriptive name

[winged mantle]

with illegal chars replaced

[lavish frigate]

are you a java dev

[winged mantle]

wait i guess rust allows you to implement triats for foreign types

a feature i'm not too fond of but oh well...

[fleet cedar]

That's one of traits' biggest advantages over interfaces

[shrewd canopy]

I like C++ templates more

[lavish frigate]

it does but it also does in a way where it shouldnt break stuff (needing to have the trait in scope to access the functions)

[shrewd canopy]

Like idk how would you express this in rust

[winged mantle]

yeah it's a good implementation

[lavish frigate]

are templates not basically just generics

[fleet cedar]

With traits

[winged mantle]

i just... basically everything about go aligns with my opinions and go doesn't have this xd

go is always right

[fleet cedar]

impl Deserialize for BaseChannel
impl<K, V> Deserialize for QMap<K, V>
``` etc

[crude star]

if it quacks like a duck...

[fleet cedar]

But there is nothing in that code that asks for quacks like a duck

It only asks for quacks

[shrewd canopy]

Unfortunately it turns into spam of trait Deserialize for X for everything and where

[crude star]

runtime error

no biggie

[shrewd canopy]

Glad i can have magic like complete math at compile-time

[fleet cedar]

Better than a giant elseif

[crude star]

real answer: the return type is important too

let quack: Quack = doctor.quack()

quacks like a duck

[lavish frigate]

if quack returns something im blowing something else up i expect that to directly print to stdout if it does anything else someone will die

[lavish frigate]

what do you think about destructors

[winged mantle]

ok i guess they're cool...

[valid jetty]

I LOVE DUCK TYPNG

look at this

fn Iterator::flat_map<T, U, AnyEnded>(Iterator<T, AnyEnded> self, fn(T) -> U cb) {
    return self
        .map(cb)
        .map(fn(x) x.__iter__())
        .reduce(fn(acc, it) acc.chain(it))
        .unwrap_or_else(fn() Iterator::empty());
}

if it wasnt duck typed, x.__iter__ would NEED to return Iterator<U, ???>

however if youre returning something like a hashmap, the result is Iterator<(T, U), ???>

its helpful to have duck typing in some cases