development

[wide reef]

right, we're up, config is generated

# Configuration file for Synapse.
#
# This is a YAML file: see [1] for a quick introduction. Note in particular
# that *indentation is important*: all the elements of a list or dictionary
# should have the same indentation.
#
# [1] https://docs.ansible.com/ansible/latest/reference_appendices/YAMLSyntax.html
#
# For more information on how to configure Synapse, including a complete accounting of
# each option, go to docs/usage/configuration/config_documentation.md or
# https://element-hq.github.io/synapse/latest/usage/configuration/config_documentation.html
server_name: "192.68.50.95"
pid_file: /data/homeserver.pid
listeners:
  - port: 8008
    tls: false
    type: http
    x_forwarded: true
    resources:
      - names: [client, federation]
        compress: false
database:
  name: sqlite3
  args:
    database: /data/homeserver.db
log_config: "/data/192.68.50.95.log.config"
media_store_path: /data/media_store
registration_shared_secret: "SECRET"
report_stats: false
macaroon_secret_key: "SECRET"
form_secret: "SECRET"
signing_key_path: "/data/192.68.50.95.signing.key"
trusted_key_servers:
  - server_name: "matrix.org"

[old arrow]

that sounds about right

just add stuff to disable federation and youre good

[wide reef]

right. is there any ways to check the server status?

grep is saying it's healthy

3ce816648fe2   matrixdotorg/synapse:latest                    "/start.py"              About a minute ago   Up About a minute (healthy)   8008-8009/tcp, 0.0.0.0:8448->8448/tcp, [::]:8448->8448/tcp   dfiles-synapse-1

[old arrow]

i think just try connecting to it with element

[wide reef]

right, let's see

doesn't seem to be working

let me grab the logs

[wide reef]

right, I figured it out but Element says it's unsupported

"Your home server needs to be upgraded to support Matrix Authentication" apparently

[old arrow]

Wha

[wide reef]

There is no matrix authentication service

so element refuses to connect to it plain and simple

[old arrow]

you might wanna ask in the synapse support room on matrix.org, I haven't encountered this before

Its #synapse:matrix.org I believe

[wide reef]

Thank you!

[granite gazelle]

Anyone familiar with Dockge or manageing docker containers? I have a docker compose file like;

services:
  dotnet-gateway:
    build: .
    container_name: dotnet-gateway
    restart: unless-stopped
    volumes:
      - ./bin/configs:/app/configs

And then I have a Dockerfile like;

FROM mcr.microsoft.com/dotnet/aspnet:9.0 AS base
WORKDIR /app

COPY ./bin/ ./

ENTRYPOINT ["dotnet", "Gateway.dll"]

When I try to update it in Dockge, it says that there is nothing to do, even though the binaries have been replaced.

docker compose up -d --build --force-recreate seems to have done the trick

[old arrow]

Yeah it doesn't have any feature to automatically detect when the source files change. You have to manually tell it to build (another way is docker compose build), then doing docker compose up -d will notice the change

but what you put above does those both in at once

[wide reef]

Who would've guessed that 5GHz DFS works terribly in midwar Russian city

certainly not me

[hollow basalt]

because they jam it no?

[austere quest]

i should probably learn to use Figma if i wanna become a developer

[soft fern]

Yes and no, it is good to be familiar with how it is used and how to get CSS or icons and the like out of it, but less important to be highly proficient at ux or visual design unless plan to work alone or primarily on frontend (learn to navigate and extract from it but don't need to necessarily learn to do ux or visual design to be good at coding). Also, think penpot worth a mention as an alternative can self host (also also am a fan of excalidraw can also self host if want to pretty easily with docker or in my case microk8s, is more for rough sketching than CSS based UX development but also useful as a digital whiteboard for brain storming/sketching)

[keen sorrel]

wow, clangd is so much better than intellisense in vscode. intellisense was bugging out with the obj-c++ needed for macos dev so I needed to find an alternative, but I wish I had switched to clangd sooner

[hollow basalt]

ok

[autumn flint]

Hello smart people. I've been developing an app for file syncs to my home NAS/Server. I've only had me and my setup to test and build on. I'm wondering if ya'll could ask some use case questions to help me broaden my development. Maybe some people would be interested in testing? I started development because I wanted an app on mine and my wife's phones that would handle automatic backups to the server like a cloud service. No subscriptions, no ads. Thanks!

[cosmic hatch]

Is it open source

[autumn flint]

No, I've been solo developing for the last 2 years now.

[neon oriole]

so prolly no licence then wich isnt inherently a problem, depending on how you would like testers to engage , unfortunatly ive been trying to ditch my mobile phone in favor of my landline so cant really help you out there ,(and for what its worth ive been using nextcloud for the little i do need to sync between my mobile and my currently dead rasberrypi server at home)

[autumn flint]

Thanks! I didn't even think about licensing.. There's a lot I don't know.

[frozen flame]

My question would be what can it do that syncthing can't

Although you said both sync and backup, which are different

[neon oriole]

pythonistas arn't gonna like the herresey im comitting here 😄 :

:p

and i mispelled mutable but meh

not sure how to solve this one yet but...

[silk eagle]

im something of a pythoner myself

(lambda s:
    (lambda ol:
        (lambda fo:
            print("".join([f"{chr(_ORD)}" for _ORD in fo]))
        )(
            [
                *(lambda t, offsets:
                    [  
                        (
                            (t >> 8) - ord(offsets[0])
                        ),
                        (
                            (t >> 8) + ord(offsets[1])
                        ),
                        (
                            (t >> 7) +- ord(offsets[2])
                        ),
                        (
                            (t >> 7) +- ord(offsets[2])
                        ),
                        (
                            (t >> 7) +- ord(offsets[3])
                        )
                    ]
                )(ol[0], [
                    "\x0b", "\x12", ";", "8"
                ]),
                ord(" "),
                *(lambda w, offsets:
                    [
                        (
                            (w >> 8) - ord(offsets[0])
                        ),
                        (
                            (w >> 8) + ord(offsets[1])
                        ),
                        (
                            (w >> 8) + ord(offsets[2])
                        ),
                        (
                            (w >> 8) - ord(offsets[3])
                        ),
                        (
                            (w >> 8) - ord(offsets[4])
                        ),
                        (
                            (w >> 9) - ord(offsets[5])
                        )
                    ]
                )(ol[1], [
                    "\x17", "\x01", "\x04", "\x02", "\n", "\x16"
                ])
            ]
        )
    )([ord(c) for c in s])
)("台湾")

[neon oriole]

at some point i should also add the ranges for emoji's that combine but :

pode ```py

it really kills 🙂

(you can see its not the python code that is the part that hangs since when ran with wc -c the python code finishes fine, its when the terminal has to display the text that it goes down the drain :D)

[bitter edge]

i am somewhat of a python coder myself

[opaque finch]

Python is just psuedocode

[delicate meteor]

Lmao

[delicate meteor]

Also vs code for java, use intellij community.. its better for java.

[neon oriole]

inline python code that executes inside a bash script,...

and here is the kicker:
it also works when ran with python:

note the difference in argument positions 😉 meaning you get to conditionally change what runs depending on how the script is executed 🙂

(besides the obvious fact that a file with the same code both runs with bash and python ofc)

[bitter edge]

i mean i've tried it, it just runs pretty slow on my system so i use vscode

[hollow basalt]

Skill issue

[soft fern]

Hmm really cause takes more effort to get Java working in vs-code than intellij so seems like not a skill issue. I tend to go vs-code for all the things because with extensions like eclipse before it is swiss army knife of editors. IntelliJ/JetBrains editors in general are more "pre baked" for a given language or working environment so can be better at one specific thing.

[bitter edge]

not really i mean i just installed the java extension and it just worked

[soft fern]

Yeah for basic projects with just one build gradle file or maven pom.xml it'll handle setting up the workspace/class path info for libs and can be a smooth experience but once deviate from any sort of basic project setup there tends to be more config involved

Debugging is generally "okay" as well but can be easier and cleaner view of things within intellij by default since it's geared towards the one language already (can make certain assumptions or work with the one debugger instead of doing generic/catch all type behaviors)

[old arrow]

Just use a build system like gradle, then install whatever the plugin for vscode is
problem solved

[neon oriole]

mm back when i was forced to learn some java, netbeans was "THE-" java ide that one must have 🙂 ... if i ever decide to code in java again , i figure my stenographer may decide what IDE he/she prefers 🦹 😄

[old arrow]

now intellij is that

[cloud knot]

Eclipse you meant

[grizzled steeple]

The JetBrains IDEs are rather expensive for commercial use, but they're genuinely great to use imo

[frozen flame]

They've got community editions for lots of their IDEs and free pro licenses for students so take advantage of those

[bitter edge]

anyone know how to get ptp and mtp data in java without using the uh JMTP library

I've found the results to be sometimes missing or incorrect using JMTP

MTP Device Info:
  Name         : Nintendo Switch
  Model        : Nintendo Switch
  Type         : CAMERA
  Manufacturer :
  Serial       : XTW50231870262
  Power Level  : -1%
  Power Source : BATTERY
  Protocol     : PTP: 1.00
  Sync Partner : N/A
  Storage Info:
    Name          : Album
    Description   : Album
    Type          : FIXED_ROM
    File System   : Generic hierarchical
    Capacity      : 18446744073709551615
    Free Space    : 18446744073709551615
    Size          : N/A
    Serial        :
    ID            : s80001
    Sync ID       : N/A
    Date Created  : null
    Date Modified : null
    Date Authored : null

MTP Device Info:
  Name         : Galaxy A03s
  Model        : SM-S134DL
  Type         : PHONE
  Manufacturer : ???
  Serial       : ADEDF32E23376BCF52AB97B4DFDE5E6A
  Power Level  : 57%
  Power Source : BATTERY
  Protocol     : MTP: 1.00
  Sync Partner : Galaxy A03s

MTP Storage Info:
  Name           : Internal storage
  Description    : Internal storage
  Type           : FIXED_RAM
  File System    : Generic hierarchical
  Format         : 30010000-ae6c-4804-98ba-c57b46965fe7
  Capacity       : 23870816256
  Free Space     : 7861571584
  Size           : N/A
  Serial         :
  ID             : s10001
  Sync ID        : N/A
  Date Created   : null
  Date Modified  : null
  Date Authored  : null
--------------------------------------------------
MTP Storage Info:
  Name           : SD card
  Description    : SD card
  Type           : REMOVABLE_RAM
  Format         : 30010000-ae6c-4804-98ba-c57b46965fe7
  Capacity       : 15926525952
  Free Space     : 15922593792
  Size           : N/A
  ID             : s20001
  Date Created   : null
  Date Modified  : null
  Date Authored  : null

[little oak]

Does this even make sense?

[little oak]

How does the Switch have -1% lol

[sturdy ridge]

Trying to Google this, I'm getting nowhere. I'm trying to make an android app that renames files inside a folder in a random order. For that, I need it to get the MANAGE_EXTERNAL_STORAGE permission. But, even though my code compiles and the apk is built, it just doesn't request the permission. Here are my Java, gradle, and manifest xml files.

[sturdy ridge]

After some chat in another server, I was able to figure out the permissions issue. I'm now able to get on with the real work

[gilded crystal]

No, I've been solo developing for the last 2 years now.

[hollow basalt]

congrats

[bitter edge]

if you would have read the statement above the result you would understand

[little oak]

Yeah, I know it's wrong but how does the code even look at it and think it's -

[old arrow]

because it's wrong

[bitter edge]

it can get like my phones battery accuretely most of the time but like for most information it varies each time i run it

[bitter edge]

for (PortableDevice device : pdm.getDevices()) {
    device.open();
    System.out.println("Device: " + device.getModel());
    System.out.println("Name: " + device.getFriendlyName());
     System.out.println("Type: " + device.getType());
     System.out.println("Manufacturer: " + device.getManufacturer());
     System.out.println("Battey %: " + device.getPowerLevel());

     for (PortableDeviceObject pdo : device.getRootObjects()) {
         if (pdo instanceof PortableDeviceStorageObject psto) {
            System.out.println("Storage Device: " + psto.getName());
            System.out.println("   File System Type: " + psto.getFileSystemType());
            System.out.println("   Free Space: " + psto.getFreeSpace() + " bytes");
                }
            }

            device.close();
        }

    }

 'App' ;be2da3ef-1a2e-4804-b2fe-e17e6f1beda0Device: Nintendo Switch
Name: Nintendo Switch
Type: CAMERA
Manufacturer: ??
Battey %: -1
Storage Device: Album
   File System Type: Generic hierarchical
   Free Space: 18446744073709551615 bytes

[neon oriole]

prolly because it cant read the power because its getting also charged trough the cable u use to connect to the device:)

(aside from that , i made a program that listed all info i could get from a usb device usign usblib1.1 on linux as a side project on something i was working on , and i gave up on the main program since there as no single usb device that had its properties like manufacturor , model, serialnumer , ... filled in consistenly or not most not even correctly, not to mention not even all using the dame character encoding ...

[bitter edge]

it reads my phone just fine even though it charges

[neon oriole]

i think you misunderstood what i meant , since the thing is connected to a powersource , its powerlevel is basicly infinite. how the device reports this depends on the internal circuitry and the implementation of it in software , eg for android an app exists by the name of Ampere , wich tells you how fast you are charging, depending on the device it can actually tell you from just using the device's facilities , or it has to calculate a best guess by analizing how fast the battery level is increasing over time by knowing the total capacity of that battery as the device itself might not have the required circuitry to measure the actual charging rate ...

and my screens were there to show that maybe your phone has filled them all out correctly , but i wouldent trust that to be the case for every usb device out there

[gilded crystal]

HELLO

[neon oriole]

EHLO

|| technially its HELO not HELLO if were doing smtp, 😛 EHLO is not a typo in that case ||

[old arrow]

MAIL FROM:[e-mail]

[neon oriole]

3

[gilded crystal]

FROM:[e-mail]

[gilded crystal]

FROM:[e-mail]

[fading osprey]

why are we posting email adresses here?

[little oak]

She sending SQL queries into the Discord server, just trust the process

[fading osprey]

ok ¯_(ツ)_/¯

[livid jungle]

https://youtube.com/shorts/WY2xTiq0DM4?si=XoifXfOUTU2BNX0s

[weak glen]

Robert') DROP TABLE Students;--

[old arrow]

i have no idea why they are randomly and repeatedly reposting that

[peak acorn]

maybe it's an XMPP bridge thats broken lol

[old arrow]

that's a normal user, is it common for people here to selfbot xmpp bridge or something

[peak acorn]

I dont think its common no

[old arrow]

why do you think that then

[peak acorn]

i just said maybe idk ☹️

[dusky coyote]

hi gamers- made my first python project. gonna add more to it soon! https://github.com/brixoric/python-calc

[tropic haven]

Does anybody over here know how to make a web app from scratch?

I need to make one for my uni mini-project

It's basically an employee management system for a government agency, atleast that was the problem statement allotted to me

[old arrow]

Yes but what specifically do you need help with

Like what part of it, it's best if you start trying yourself first

[tropic haven]

A web page would do I guess, the only crutch is that I have to use Java for it. I have like basically no experience in Java

[tropic haven]

Guess you're right, any resources you have handy?

Damn they're that much different I thought they'd be like C and C++

I only know traditional languages

I'll start from there I gues

Imma have to check if the syllabus even contains javascript, the instructions only were to use the technologies from the course material

[old arrow]

java would be a backend, you would still need HTML+CSS+JS for the frontend. Unless you're targeting stuff from like 15 years ago with Java Applets.

the backend is the server stuff, the frontend is the interface you se

[neon oriole]

java applets still exist 🙂 so ... 🙂

[old arrow]

It's hard to recommend developing a new app only supported by browsers that are at minimum 7 years old

Or I guess very niche browsers that bring back support for it

I am not so up to date on Java applet support. Websites that only work on IE also still exist but that doesn't really help anything

[neon oriole]

well ... dunno how to explain this as its beyond my capabilities but its javaruntime to html5 ish
https://boardspace.net/login.html

also javavm backend but with selectable frontends to it :https://teavm.org/gallery/software3d/index.html

or running minecraft (wich is java i think :)) in the browser :https://browsercraft.cheerpj.com/

[old arrow]

None of those are java applets. It's kinda unfair because technically you still need a bit of HTML/JS to tie it back to the DOM as far as I know

some of those links you show even are just compiled down to js it seems?

[neon oriole]

theye not oldschool java applets however they are technically running a javavm, as the browsercraft kind of shows (since the game is a java game)
to me it looks like more: take an oldschool javaapplet wich is no longer supported , instead of putting it into an html page ising <applet> , the thing you do is wrap it in a different javavm like teavm or cheerp , and put that on the page,... so the thing you made is still just a javaapplet , the way its presented to the user is just no longer the need a local javavm and a browserplugin to run it ...

[old arrow]

it's not a java applet, that's a specific, deprecated thing; that's what i was getting at
i don't know if any of these tools are really meant for running an entire SPA, moreso just for little components as a part of the page

[neon oriole]

if i write a java applet today but instead of including it with the html applet tag , i inlude it with the <script> and a <canvas>tag and a jvm running on javascript,... , im still writing javaapplets right? only thing that changed is the way its included in the webpage... aside from the 2 lines to inlcude the thing,.. im not writing any js, html or css to make my application ... im simply writing java.

if you literally were talking about the 2 tags <applet></applet> as being targetting old technology,... im sorry then yes those 2 tags arent in use anymore,... but if you were talking about the code running hat gets included by those tags , the java code , then no , sill verry much in use , only the way that it gets ran and included changed

[cloud knot]

That is running WebAssembly, which is literally an assembly language for web (https://webassembly.org/). And that one is just Java runtime running in webassembly

https://madewithwebassembly.com/ see examples of what people did

[old arrow]

i am aware of webassembly

i think i mentioned it earlier

[cloud knot]

Java Applets and JNLP are both deprecated, so webassembly is your only option to include anything better than JS within a browser.

[old arrow]

@neon oriole i see what you meant

i thought it was just various unrelated frameworks for making wasm and wasi apps using java

[marble ocean]

any ruby devs here?

[limpid plaza]

Where do you suppose I can ask for help regarding PCB making?
it does say development, not exactly programming, I was hoping if that included electrical/electronics?

[neon oriole]

your prolly gonna wanna look for Altera/Altium communities, or KiCad , or used to be verry popular for designing the schematics : EagleCad but i dont know if its still the case (was opensource when i was in school but since then it got bought up by autodesk (makers of autocad ) so i dont know and didnt follow it tbh
also we were tought Orcad isn school sop thats another aveneu to look for discord communities on 🙂

if its a verry simple question , i might be able to answer it myself , but my info might be more or less 20years out of date so beware 😄

[flat pewter]

PCB?
do you mean Physics , Chemistry , and Biology?

[old arrow]

lmao

[weak brook]

what's the question?

might as well ask it, worst case is someone just doesn't answer

[marble ocean]

i'm so fed up developing AI features 😩😩, been like a year with AI this AI that, and I know they are less reliable and they are not adopted that much, they are only for marketing

[weak brook]

EagleCAD is awful now that they've integrated into into the awfulness that is Fusion360

[neon oriole]

someone should really start archiving the last "ownable" versions of software that dont require any clould or ornline access to work , with LLM's free,able and allowed to break copyright'ed code/ code licences as they please , reverse engeneering them to be opensource (like wine or reactos do for windows code) will be a walk in the park in a year or so if we are to believe the ai guru's (or atleast in a year or 10 but still worth doing imho 🙂 )

[weak brook]

I think the funny part is that I would 100% buy software

I bought software years ago that is just a license that I continue to use

I Just want to be able to use it

[neon oriole]

i dont have a problem buying software eitehr , but i have a problme with :

• not getting anything i can use the way i want after spending money on it ,
  ... so cloud software , unless i get to own the cloud ,...
  ... or rental that requires a theter to their servers , or a connection at all... ,
  I
  some of the things i own i have no regreta about spending money on it:
• Windows PRo licence (from back when i bought win7 pro (not ultimate) got upgraded to 11 pro over the years , but to this day i refuse to log in with my microsoft account or to even connect it to my install ..
• Ableton (8...12 suite), but i have them installed with a crack (reasons...)
• a licence for FuityLoops Producer edition (bought it at version 4 but comes with lifetime free upgrades so were at version 25 now :)) (installed no crack)
• MS Office versions 95 .98,2000,2003,2007,2010,2016,2021(most of them with visio) while i do have valid retail licences, installed with cracks (reasons)
  -old visual studio versions but not new ones , still use the 2005 one if i really need it and the free version of 2022 if i need that one.
  ithings i regret buying include :
  -RDR2 on epic game store , i expected to get a key ,.. .and be able to run it without the connection and launcher ,..
• the game was worth the money, but not with the store / launcher attached to it,... no i cant actually play the game the way i would want to and when i would want to...
• some vst's that use the modern ilok licence manager ... i didnt mind the old hardwaer ilok , that one atleast could be emulated ... now they require a connection and some ring zero software to be runnng 🙁

that is wh i specifically said , (with a typo: 🙁 )

renting softwar that is thetherd to the cloud feels like renting shoes , but you cant leave the city with them and only walk on level flat surfaces since the shoumaker does not want to support his shoes walkng on cobbelstones, grass or gravvel , but you do get a new color of shoelaces ever month for that 50euros you have to pay him ... ow and theres holes in them that maybe get patched someday

[weak brook]

yeah, just let me buy software with a perpetual license

give me the ISO, let me use it.

one software i bought a perpetual license to was like $6500, but i use it all the time and i've had it since 2017 so

[frozen flame]

you bought all the ableton versions??

[neon oriole]

no 8, , and i got 9 for free as beta , then came the switch to 64bit , and i had allot of max for live stuff and vst's that were 32bit , and ableton couldnt handle both 64 and 32bit vsts (unlike flstudio wich could) , so i only recently switched to 12 , partly becasue my inbox was filling up with mails of people that were using one of my max devices that apparenly broke in 12./// so ...

wow and i did 11 lite for a short period... 🙂

[soft fern]

Not sure if you already got an answer but am actively working with kicad for doing some PCBs at home, so far some positive results but about to try some more difficult boards with tighter pin spacing smaller pads etc

Am using dry film photoresist and laminator to stick to the copper clad stock boards and then cure clean and etch, the process doesn't take long once have everything in place, but is a setup and getting transparencies aligned by hand is tedious (trade off in hands on time vs cost of ordering professional boards)

I just ended up ordering some dry film solder mask to use too since the paint on kind is sort of a mess to work with

Software wise I had used eaglecad in the past (like 7yrs ago), there's a learning curve to any sort of CAD but once you have the gist of a given type of CAD tooling (mechanical vs electronic) then it is pretty easy to move around between programs (slower because of hotkey switches maybe but can figure out what to look/what to expect/workflow). Kicad is definitely sufficiently advanced especially if you use a few add-ons, I have free-routing and pcbway and a couple others enabled that make it easier to work with too.

For kicad here too I just pushed repo with my project in it, there's some git integration in the UI but can just use other git GUI or CLI tools for remote copies (git branching/forking/history etc)

[weak brook]

KiCad is actually really good; that's what I use nowadays too.

KiCad I think is the first EDA that I used that was able to export boards in STEP format so I could make 3D enclosures in Solidworks

I think Altium is integrated into Solidworks but it was awfully expensive so

[grim frost]

I’ve been trying to develop a new protocol ssp secure stream protocol works alongside tcp and udp how does it work communicates with a server when traffic is sent its encrypted and decrypted server side performs a handshake with the server and the client then routes the traffic to the requested destination was wondering if anyone wanted to help

[soft fern]

Guess I'm wondering what you would be adding that TLS encrypted connection or ssh tunnel wouldn't already do? Er like what's the use case for the protocol? Generally speaking though I think usually just make a GitHub with whatever initial work or ideas you have and work out details in issues and/or wiki with others as starting point then have git repo ready once ready to start coding some implementations (or just start sharing whatever you have insofar as a current implementation is concerned)

Usually easiest to manage random internet collaborators as part of a GitHub project and/or organization for the project and then if need be have discord for the project too (things like iNav or elrs basically do that)

[grim frost]

So it’s AES encryption and I’m just trying to secure the internet the internet was not made secure by design so I’m just trying to revamp the way we use the internet at the protocol layer

[grim frost]

Good to know thank you for that

[old arrow]

What does tls not do which you are trying to solve

[grim frost]

Basically just trying to live my life I like this stuff you know it’s fun

[old arrow]

Also fun fact, its very impractical to make a new transport layer (layer 4) protocol that actually reaches adoption on the open internet, if that's what you were hoping to do. That's why newer stuff, runs on top of an existing protocol, i.e. QUIC running on top of UDP

[old arrow]

I'm just curious what you mean by the internet not being made secure by design

like it kinda sounds like you're describing TLS. are you just making something similar to tls for fun

Or maybe it's more like a VPN. I don't know if I read it right the first time

[grim frost]

Oh ok so basically 1. Secure Transport Layer Enhancement
• It operates alongside TCP/UDP rather than replacing them.
• Adds encryption, authentication, and handshake control before standard transport begins.
• Useful where legacy applications cannot be modified, but security is required.

2. Traffic Protection
   • Client encrypts packets → sends via TCP/UDP → SSP server decrypts.
   • Server then relays clean traffic to the requested destination.
   • Protects data in transit from passive eavesdropping or tampering.

3. Use Cases
   • Building VPN-like tunnels without needing a full VPN stack.
   • Secure relay/proxy for HTTP, DNS, or custom protocols.
   • IoT or embedded devices that lack built-in TLS/DTLS support.
   • Environments where you want uniform encryption but apps don’t support TLS.

[grim frost]

5. What It Solves
   • Security gap in legacy TCP/UDP traffic.
   • Provides forward secrecy and replay protection.
   • Can be deployed incrementally without rewriting apps.

[old arrow]

It sounds like you're describing exactly whar VPN protocols, i.e. Wireguard which is very lightweight, or proxy protocols like SOCKS5, do

If you wanna make that you could look at their protocols specs for reference since they'd probably be similar

[grim frost]

Yes it’s very similar I appreciate the feedback and support

[grim frost]

Yep so basically just saying the internet was not made secure at the first level the internet was not made secure by design

[old arrow]

I feel like that is an odd definition

[grim frost]

It most certainly is not because if you think about it the internet was made for speed tcp udp was made for speeds in mind not security I’m talking about the standard protocol,s not other protocols we did not make tcp udp with other forms of security in mind just plopped them on top not from the root cause

[grim frost]

I’m not saying it solves everything but am willing to build more

[old arrow]

the internet was made for speed
if you look at TCP at all it's super obvious that is not the case... TCP is famously slow to figure out the right transfer speed, especially if you drop an extra packet and it overreacts

The transport kayer doesn't 'have security in mind' because that's a different layer of concern. The main concern at this layer is figuring out how to split data to be transferred into small units called packets, and that's by design, for simplicity.

You can then just do the security at a different layer like you see with VPN protocols, or encryption protocols like TLS and it's fine. It's still enclosing your sensitive data

[old arrow]

what you're making doesn't seem to solve anything that a VPN doesn't already

[grim frost]

I agree with that

[old arrow]

Take a look at the OSI model for context here, about layers

[grim frost]

TCP is my best friend rn because the way it works is more secure in my opinion I actually learned that when I went to school I have a precision in network fundamentals and other in computer maintenance and repair I also learned cybersecurity

[old arrow]

Maybe you mean more reliable instead of secure? I don't think there's really anything about TCP that makes it more literally secure than UDP as far as I'm aware

like using TCP isn't going to make it any harder to snoop on data than if you used UDP. May make it easier to write since you don't have to worry about handling transmission rate and error correction yourself

[grim frost]

To send data to an application using TCP, you first have to establish a connection. Until the connection is established, packets only get to the OS layer, not the application. Establishing a connection requires that you receive packets back to the initiating end. If you wanted to forge an IP address not on your own network and establish a TCP connection, you'd need to be able to intercept the packets the other side sent out. (you need to be "in between" the endpoint, and where the packets to the forged IP address would normally go, or do some other clever routing tricks.)

[old arrow]

Some would argue that TCP is worse for security because it does expose a decent amount of metadata. I believe something to do with being able to use source port numbers to estimate the volume of traffic you're sending

[hollow basalt]

what

[grim frost]

https://security.stackexchange.com/questions/165626/why-is-tcp-more-secure-than-udp

[old arrow]

it doesn't really mean anything

[grim frost]

lol I know

[grim frost]

But is udp any more better than not leaving meta data

[old arrow]

I guess from the perspective of application development, using UDP makes it so you have to do more to prevent spoofing. But this was completely unrelated to what you were talking about earlier. Earlier was more about the security of data during transport

[old arrow]

this should make it pretty clear

[grim frost]

I’m team tcp lol

[grim frost]

Yes it was

[old arrow]

The argument about TCP being less secure due to more metadata would be more of a fingerprinting one. Obviously that wouldn't affect whether your data in transit is safe

I think it is misleading at best, to say that either TCP or UDP is 'more secure' than the other

especially in the context of the task essentially being running a proxy over TCP or UDP, where you likely could be concerned about fingerprinting or metadata

[grim frost]

Ya I’m not really worried about fingerprinting

[grim frost]

Thanks for talking to me I appreciate it I love talking about this stuff I can actually talk to you to and understand and you’re not coming after me ether I actually feel safe here other severe,s took me as spam because I don’t lisining unless I see clear rong so they took me as spam became I am a very opened person not listening because I’m trying to protect my knowledge I didn’t know that I’ve learned a lot today

[old arrow]

Ah, yeah, I feel like there's a lot of places that will just see your original message and just make fun of it instead of actually thinking about it

[old arrow]

I can't really find what I was thinking of. Not sure if it exists. I think it's still accurate that, if anything at all, TCP would be more fingerprintable than UDP

[old arrow]

I have more questions than answers

[frozen flame]

What the fuck even is this

[silk eagle]

why is notion there

what does notion have to do with any of this and also obsidian

and google drive

google drive is a website it should work on all platforms that run websites 😭

[tall gulch]

Guys I got this crazy idea in a dream that we had profile chat lobbies baked into Twitter

So I made a concept of it

It’s like discord on Twitter profiles 🤤

Or YouTube livestream lobbies

who wants to help me build it for real?

[dusky echo]

Any tips for using a vm for the first time

[soft fern]

Install guest additions packages if applicable in the guest OS can help with working with the VM host in some cases (like qemu-guest-agent or virtualbox guest agent) but without some more info about what VM you setup and how and why (or any issues) can't really say you should change anything from defaults

[dusky echo]

I want to use it for penetration testing

[soft fern]

Gotcha, nothing really special in that case, I guess just don't run VMs with any sort of escalated privileges on the host but shouldn't be an issue really

[soft fern]

Vs code, Arduino IDE, obsidian... You can basically run all of these things on all platforms (g drive and OneDrive maybe exceptions but also trash :D). I also have no idea what's going on in the graphic

[soft fern]

I mean gdrive can use in browser but the native sync service on desktop has limited install targets (same with OneDrive)

[neon oriole]

^^100% true , and then they made http/3 wich trows everything upside down and uses udp, i dont know half of it but its a mess when you look at it 😄

[hollow basalt]

0/10 ragebait

[neon oriole]

i dont even get that image lol 😄

[uneven tiger]

https://security.apple.com/blog/memory-integrity-enforcement/

[hollow basalt]

Ok

[dreamy sandal]

GUYS IT RAN

[weak brook]

What compiler are you using? mingw?

[old arrow]

industry-first
the humble google pixel 8
although to be fair it's not enabled in the pixel OS

[dreamy sandal]

gcc

[dreamy sandal]

i fucking love my google pixel 8a
grapheneos my beloved

[old arrow]

yeah plus that's me being annoying because they probably really mean simply ahead-of-the-curve

which is pretty fair to say about apple security

[keen sorrel]

if only they just rewrite everything in rust, they wouldn't have to worry about this /s

[dreamy sandal]

https://tenor.com/view/taking-a-deep-breath-nicolas-cage-history-of-swear-words-smelling-deep-breath-gif-19883465

[neon oriole]

this is prolly one of the dumbest takes ive heard in a long time, rust does not prevent memory bug, memory corruption , not even useafter free or memory leaks ...

also : verryu much worth it 🙂 https://www.youtube.com/watch?v=kLXXZddWCJU

[keen sorrel]

What does /s mean to you?

[neon oriole]

\s would mean whitespace but thats about it

[keen sorrel]

it means sarcasm... I guess you must be new to the internet. welcome...

[neon oriole]

hahah im on the internet since before javascript

||half the people here , hu was there ever internet without javascript , how is that possible ???? 😄 and yes no css either , sites that had a layout used tables to create that layout lol||

[old arrow]

sure rust doesn't make it impossible to have memory related mistakes, but it's disingenuous to say that it doesn't help at all (which is how that reads, unless you didn't intend that)

[neon oriole]

i really meant that it doesnt prevent them , entirely , i can see the value of rust ofc (for me just not the value it brings over what you have to give up for it , wich is nicely documented in the videoclip) , but certainly not for every codebase , kernels being one of them , (even if redox seems to be proving me wrong there) )

(also enyon notice that youtux has the same terminal as Mental Outlaw ? :d

also specifically on that blogpost that was mentioned id like to point out this :https://www.youtube.com/watch?v=-D1gf3omRnw

ow this is the better one:https://www.youtube.com/watch?v=PCXwpLaQZ7s

or even a better one :https://www.youtube.com/watch?v=uZEBkOrfUzM

the last one specifically talks about the memory tagging , that is also mentioned in the article

[neon oriole]

i tried coding with copilot for the past week ,... conclusion it makes me less productive and really annoyed at it , if it provides completion for code , it might have the current line correct (sometimes) but never the next ones , exept when its like the 3 or 4th item in a pattern , else its just wrong , but i see the current line being right so i comlete , only to then have to remove 3..10 lines below...:( disablingh that shit asap

[weak brook]

The only thing its good for is explaining my code to me like dr Seuss

Particularly for embedded, it's awful.

[cloud knot]

i use ollama + qwen3-coder as 'helper' for stuff like 'how do i do this or that' in chat mode. I certainly not feel like doing more than that for me.

Also today in agent mode i managed to get the AI agent into a loop 😄

(the thing literally repeated the same thing over and over until it ran out of time/context length)

[neon oriole]

it was suggesting me methods that did not exist eg in autocomplete , messing up variablenames, okay granted im not following "convention" but still the whole dam project is consistent in naming... and it was python prolly the language best known by copilot...
(i use *a,**k instead of '*args,**kwargs') and it started sugesting me __a , __k

[cloud knot]

as i said, i only use it as quicker stackoverflow for generic questions where they can't really make mistake (like how to do X in CSS)

[silk eagle]

i use it for specific things that i use once a month but can never remember

mostly jinja2 templating

[cloud knot]

well, vertically centering text in a DIV is exactly such question of "how did i do it last time" 😄

[silk eagle]

oh nah i got good at that one i know all the tricks

but the jinja2 like % include 'nav.html' % weird stuff, i cant always figure out how that works cause sometimes itll import the nav.html template in a really weird way

but ai is pretty decent at figuring it out if i just write a book to describe what i want

i end up having to type more than the LLM does but it works out cause i dont have to google

[neon oriole]

frontpage best ide ever 😄

(i had to study macromedia dreamweaver in school at some point ,... that was time well spent ... :S

[neon oriole]

i actually prefer browsing stackoverflow, usually i take to long because some other related intresting thing caught my eye,... but i considder that not a waste of time , found some gems that way 🙂

tables was ow so mindbreaking hard , since internetexplorer included the border in the celwith and firefox did not or vice versa i forgot 🙁

[soft fern]

Yeah browser compatibility was a nightmare is why I worked in plugin based things in the first place (also having a "canvas" to control things on), also why jQuery was such a thing

I mostly skipped on the jQuery period and was happily oblivious working on top of flash runtime till mobile hit but then soon after html5/css3 started landing angularjs and then typescript and modern wave of web frameworks with "transpilers" and poly fills pretty much smooth things over... Still the occasional client who is dead set on using IE 6 and XP or something

Guess to be fair the modern stuff was initially at least just building on jQuery too but think a lot of that is gone now just "jqlite" or other bits and pieces

[weak brook]

AI in embedded: takes you longer to fix the code it gives you than just writing it yourself

[silk eagle]

i think its a dopamine thing where people dont really get the dopamine hit from writing code, they get it from seeing the code work (which i think is a consequence of coding classes and such; needing your code to work for an assignment)

people who write the code because the process itself is fun are probably more likely to not want AI to spit out a half-baked solution they have to fix because it removes the fun part (like people who crochet wanting to make a hat but they have this magical tool that makes half of the hat for them)

[weak brook]

normally it's the guys with project management breathing down their neck wanting the code done that day, that end up using it and AI spitting out something completely unbuildable, resulting in the poor guy having to fix it.

having been the one to fix that stuff before, it's just not very good with embedded systems

idk if it's any better nowadays.

[cloud knot]

coworker put down estimates on 6 tasks which were assigned to me. He estimated 4 days of work to do those features. After 1.5 day i was done. No AI 🤷‍♂️

[weak brook]

I once had a schematic and pcb made by the time the programmer was done with his program, because he was using AI.

We were supposed to be in paralell, he should have had his code done before the pcb was fabbed lol

[cloud knot]

today's use of local AI tool - "how many items are in XY array?" (where XY array was a list of configuration items, and i needed to know how many are in that array, because we will have to add some properties to most of them 😄 )

so i jus threw the question and file on it, and while it took it 1 minute to do it, still less than it would have taken me to count them manually

[weak brook]

Yeah, and that makes sense; the vibe coding thing is dumb.

Paying someone to babysit a chatbot okok

[cloud knot]

or as i mentioned before, asking it CSS questions which i could surely find on internet, but then i can also tell it to do the same thing using tailwind classes, and boom, time saved on things i do not enjoy (CSS) 😄

like this. Sure, a bing/google AI thingy would probably show the same code, but hey, at least my 4080 Super is not bored during work hours

[weak brook]

I think for stuff like that, it works pretty good. The issue is more that for the small, low-level embedded type stuff it just doesn't work that great

mostly because parts change, packages change, and it assumes a specific model and often won't assign GPIO correctly or stuff like that.

if it's a "write a function that calculates freedom degrees from Celsius" thing, sure, it's just drop in and easy.

or it'll assume you're using HAL, which confuses the newbies

[neon oriole]

ask it go calculate degrees F from degrees K using plank's conststant 😄

[old arrow]

how do you involve planck's constant with converting between fahrenheit and kelvin

[silk eagle]

the aristocrats

[undone pond]

im more worried on why he thinks he should be using planck's constant at all

[neon oriole]

i know its possible 🙂

[neon oriole]

🙂

the right part is for 1 kelvin so you have to m,ltiply it by whatever value kelvin youhave 😄

[neon oriole]

simple theoretically it should be possible to create an anything to anything converter using that (wich would be a really nice thing to have)... so thats why

[old arrow]

What are these variables

[neon oriole]

for energy in joules

kb is boltzman constant

[old arrow]

what's F

Fahrenheit?

[neon oriole]

its basicly the link tho the planc constant, i could pretend im wicked smart but i have to look it up aswell (:))https://en.wikipedia.org/wiki/Boltzmann_constant

[neon oriole]

idd , there is a degree circle in front if you look carefully

ofc its a dumb procedure to do since you can factor out 9/5+12 from that formulle , wich would allow you to directly go F->K

bnut it is possible , and i can keep hoping , at some point someone will have the determination to make that anything to anyting converter program 😄

(there was a video from some guy on that topic but since youtube ,... no way in hell im ever finding that again :()

[weak brook]

that would be a wild thing to make, and takes the fun out of dimensional analysis

do it by hand, you look like a wizard.

I used to write homework problems that were entirely dimensional analysis questions for semiconductor physics because it's the most fun thing to watch people do

[neon oriole]

we had to do fourrier transforms ... on paper, i'll leave it at that , it took me 25years to finally get a decent grasp on how to think about them (learing how to do them was easier then understanding what on earth i was doing lol)

[keen sorrel]

In what way does the use of Boltzmann's constant somehow mean Planck's constant can be used as an "anything to anything" converter. This is all nonsensical

[neon oriole]

since you can write any unit as some formulla containing one of the variations of the plank constant... you can use that as a point for converting one to the other,... theoretically as for some its an insane undertaking to get a formulla with onething on the left side and the other on the right side in a way that fits (not to mention that i think that somehow you have to reduce it to be dimentionless waswell to be correct)

[keen sorrel]

You just repeated your claim without actually proving anything. It remains nonsensical. Planck's constant is a specific physical constant describing a property of the universe. It isn't some magical tool to describe any proportionality

[neon oriole]

a few weeks ago i kind of lost that discussion not because i was wrong but i couldnt find anything that could explain it better than i can, that also had some form of authority that some random guys would accept,.. but in the mean time this was posted to youtube https://www.youtube.com/watch?v=ZR-1Jol_nUM

[weak brook]

that's not that difficult tbh

that's pretty standard in signals and systems and DSP classes.

[neon oriole]

depends on the wave how difficult that is,.. i can think of some pretty impossible ones to do 🙂 or atleast verry hard ones ,... (an impossible one would be a square wave ,... you can only approximate that one)

and i think in dsp classes they would do the fast approximations ...

for the transforms fft ..

[old arrow]

you can calculate a perfect square wave if you just dedicate all your time to it....

[weak brook]

I mean you do both, you do the ones easily doable by hand and use MATLAB for the rest

I used to write fancy word problems about AM radio for practice lol

[neon oriole]

made this a while back 🙂 in the mean time it has gotten a bit more complex than just two but anyway 🙂 was fun project

(its in pygame :?)

(currently in windows ,and i cant seem to read NILFS2 filesystems with wsl so i have to reboot to open the project again ...)

and no , i haddent noticed but someonee else aready pointed out that i stole 3blue1brown's colors , totally by accident actually 😄

(i never got mathlab in school so , we had to do everything by hand , but we actually had a great math teacher 😄 if only i would have realized it at the time ,... i remember one exam 4 hours 3 questions , one of wich was , there are 3 proofs for the Pythagoras Theorem in your book, these : (all 3 there ) prove the theorem with a proof other than these ...

[neon oriole]

well it would still be an approximation ,hehe a verry close one but ...

[old arrow]

maybe if you never die...

[old arrow]

funny you speak of that because last night I was trying to get MATLAB to work on Linux, the add on store doesn't seem to work after using the aur package

and I need to add simulink

[weak brook]

you shouldn't need the AUR package for MATLAB.

i've never used it anyway

[old arrow]

what do you suggest I do? run the installer manually and make a mess of the system? how would that be any better

[weak brook]

wat?

[old arrow]

what would you do

[weak brook]

how would trusting the AUR be any better than the official install?

[old arrow]

the aur is just package scripts, its not the prepackaged thing like repo packages. It downloads from mathworks

[neon oriole]

it seems that the mathlab thing , is not so much an installer but a selfcontained folder you extract and run from within so ... you can dump it in /opt , and create a symlink to /usr/bin or /opt/bin

[old arrow]

which is why it'd be great for a package, maybe their prelicensed installer would be different and I could try that instead of the general one with the product activator

[weak brook]

you need a license to use MATLAB. it's cheap, like $50 if you're a student or have a .edu

if you want an open-source free one, Octave is pretty good.

[old arrow]

yeah I already have a free license from my college which is why I want the real thing to work so badly, instead of a Foss thing

[weak brook]

I've never had issues running MATLAB on linux, always through their official install script

I used simulink very heavily, as well.

communications toolbox, DSP toolbox, etc.

Arch actually isn't officially supported for MATLAB anyway, so you're kind of on your own.

I think it's mpm that works the best, and is their official way too.

[old arrow]

it doesn't practically mean anything for it to "not support arch"
the aur package uses mpm

[weak brook]

I mean, im looking through the stuff in the AUR you posted

[weak brook]

What distro are you using?

yes, arch, but arch base, manjaro, what flavour

I shall create a virtual machine and try this aur package myself, as i've never used the aur one in my life so i'll see if i can recreate your issue.

[old arrow]

real arch

[weak brook]

can i get DE, too?

[old arrow]

manjaro isn't even arch arguably and shouldn't use the aur

[old arrow]

none, just a Wayland compositor

[weak brook]

ok ty

[frozen flame]

Linux users try not to have stupid takes challenge: impossible

[old arrow]

Manjaro has different packages, sometimes on different versions, which the aur isn't really designed for

it's not really close enough to arch for it to be reliable

[weak brook]

I did not have this issue.

MATLAB works, although idk that I trust the aur one is that reliable.

So idk what the issue with your install is. Maybe its because mines a VM idk

[old arrow]

you were abgle to use the addon menu?

[weak brook]

ye.

idk if maybe something went wrong with your install, not sure

I am assuming you're starting MATLAB with sudo?

[old arrow]

WHAT 💀

who told you to do that

[weak brook]

not for daily use.

you may need it if you're trying to install stuff to the directory that's owned by root.

[old arrow]

it's not the problem

[weak brook]

I was asking a question, thanks for your clarification. Has anyone else tried to help you yet?

[old arrow]

this the error i get https://aur.archlinux.org/packages/matlab#comment-1039542

[weak brook]

javaclasspath?

[old arrow]

Unable to resolve the name 'com.mathworks.jmi.ClassLoaderManager.getClassLoaderManager'.

Error in javaclasspath

Error in javaclasspath

[weak brook]

I mean, at that point might just wanna nuke path shadowing.

that is, a folder named com/org/javax on MATLAB's path sometimes causes issues.

[old arrow]

i cannot find such a folder

neither under /opt/MATLAB or in the path output

[weak brook]

I think you might wanna check prefdir In that folder, if there's stuff like: javaclasspath.txt javaclasspath.txt javalibrarypath.txt You may benefit from just adding .bak to the end.

honestly i think either your install messed up or you picked a weird PATH.

[old arrow]

uh

[weak brook]

that looks ok

echo $MATLAB_JAVA?

[old arrow]

no java anything .txt
it's supposed to use system java but i don't know how

[weak brook]

that's a good thing, we don't want java anything .txt

if echo $MATLAB_JAVA returns a path, you got bit by the AUR.

[old arrow]

theres no path

wdym bit by the aur

oh hold on you're not gonna believe this

[weak brook]

it's always something really simple lol

[old arrow]

there's a package it provided which i assumed was gonna be installed but apparently not
it's called java-matlab
now it does this

Warning: Unable to find Java library:
/opt/MATLAB/R2025a/sys/java/jre/glnxa64/jre/lib/server/libjvm.so
Please check if you have a MATLAB_JAVA environment variable.
Note that MATLAB requires Java version 1.8 or higher. 
> In javaclasspath
In javaclasspath 
Disabling Java support

Warning: Disabling Java support 
> In javaclasspath
In javaclasspath 
Cannot initialize Java Virtual Machine.
MATLAB Java services cannot be initialized.

Error in javaclasspath

Error in javaclasspath

[weak brook]

I have never seen this problem and i've used MATLAB since version 2007 lmao

so now if you $MATLAB_JAVA?

oh wouldn't you be able to do the following:

if that file exists, you should be able to launch MATLAB with env vars that force the bundled JRE.

or maybe a compatibility symlink

[old arrow]

when i echo that in matlab it acts like it's trying to read a file named that
there's no /opt/MATLAB/R2025a/sys/jre , the aur package is supposed to remove the bundled java so it uses the system one i believe

do you think the java part is necessary for the addon thing to work

am i supposed to have java_home set. maybe that's the issue

now it's really messed up

[weak brook]

I believe the addon explorer needs the JVM

[weak brook]

don't echo that in matlab, that's shell

in matlab use: getenv('MATLAB_JAVA')

Why would you want MATLAB to use the system java instead of the bundled?

that screams compatibility issues with toolboxes

[old arrow]

it uh crashes when i do the getenv in matlab

[weak brook]

lmao

[weak brook]

ooooooookay, you need to start it without Java to fix it.

terrible formatting

I need to fix that lmao

eh, you can figure it out. you use arch linux

You can't copy and paste from the picture, but it looks way better.

[hollow basalt]

People still use matlab

[weak brook]

Yes

Its extremely common in systems engineering.

Python is great and all, but it is much more work than Matlab sometimes.

[hollow basalt]

Ok

[midnight wind]

Simulink is very popular

[wraith turtle]

so i had a random fun idea that serves no purpose other than entertainment. i created an idle game on an esp32 😂 the main game itself and all the logic and ticking is done on the esp32. you then open a client html page on your phone and can use the browser to connect to the esp32 over BLE and use it as a gui for the game XD. wheter you are connected or not it keeps on ticking as long as it has power. the html is basically just a thin client displaying its data nicely.

[keen sorrel]

TIL there's a syntax alternative to the traditional template variadic recursive unpacking; the fold expression:

template <class ...Printers>
void print_all(const Printers& ...printers) {
    (printers.print(), ...);
    bool succeeded = (printers.success() && ...);
    auto errors[sizeof...(Printers)] = { ((printers.get_error())...) };
}

[willow lance]

Ha I have a doubt I create a website but when I open it from another device the storage is not working properly that cloud computing things

Can someone help

[hollow basalt]

so the cloud computing does not compute on different device?

[silk eagle]

clouds don't compute they're just water vapor in the sky I think

[formal cargo]

hey guys
i made a chess bot pls give any suggestions: https://destined-git.github.io/playchess.github.io/

[soft fern]

You using an LLM as the other player or custom ai or some existing chess engine?

I made a chess game and hooked into LLM a while back if doing that can be kinda fun to ask it to explain it's moves before making them, can see if it has some kind of reasoned response and/or just makes something up that sounds like chess words... At least with gpt 3.5 or 4 it would still make invalid moves at times but entertaining nonetheless

[formal cargo]

yep its actually my first time doing that

i used winsurf for this using the free trial and also bit outside my comfort zone

this is the github repository: https://github.com/destined-git/playchess.github.io

[soft fern]

Heh oh right forgot gh-pages site so source is there will take a look in a bit here, I posted mine on gh-pages too but didn't bother with deploying backend so still need to run that locally

[formal cargo]

ask more in dms if you want

[wraith turtle]

anyone else ever take a look at some of the ios and safari specs and really wonder wtf apple is doing with their engineering and api's along with hardware support?
like i was just looking into web-bluetooth and ble implementation setup in general for mobile to figure out reasonable throughput to optimize for. you kow figure out the smallest bottleneck in the ble communcation on any playform and make sure we stay below it.... oh my god is it a huge difference between ios safari web-bluetooth and android web-bluetooth.... ios/safari just gets trounced. its a bloodbath

the throughput difference is pretty sizable

[cloud knot]

that is because Apple still wants you to make apps for their store, which gives them 30% cut. While they "support" PWAs, they don't want them really. Plus as usual, Safari is the new IE6

[old arrow]

is there a primary source for that other than chatgpt

I am just curious

[wraith turtle]

With more research Safari doesn't support it at all lol

Just Android Chrome and Android brave lol

Though I'm adjusting my approach anyways

[old arrow]

That tracks

[wraith turtle]

Using web ble just for configuration of Wi-Fi and setting up user for the device then once it connects to Wi-Fi it serves an API over http and the static client

[old arrow]

It seems like only chromium is on the curve with supporting these hardware web APIs, like WebUSB for example

[wraith turtle]

Well the choice is web serial or web Bluetooth lol

Honestly might use both tbh

One for flashing the devices and the blue ble to configure

Yes you can flash usb serial devices from a browser lol

[keen sorrel]

a great site for browser features is https://caniuse.com/web-bluetooth

[wraith turtle]

I've been using mdn for browser spec so that's pretty useful, ty

[wraith turtle]

Honestly surprised by how many actually support the Web Bluetooth lol

[old arrow]

It's only google chromium

Just happens than chrome is 80% of users

[wraith turtle]

And opera, uc and Samsung

[old arrow]

those are all based on chromium

it's like reskinning chrome essentially

[wraith turtle]

Didn't think mobile opera and Samsung were chromium based

Though the main though to note is it's honestly just browser on Android, Android the key supporter lol

[old arrow]

Huh

[wraith turtle]

I don't think the desktop browser expose the hardware

At least not on Linux Chrome

Though honestly thinking of just using web serial for all of the flashing and configuring

Has more adaption than web Bluetooth

That and chrome just knows what's up with hardware API lol

[old arrow]

What are you using it for?

[wraith turtle]

Game server running on an esp32. Using web serial to flash the device and web Bluetooth to configure it and connect it to Wi-Fi and then ble For the game updates and status to the game

I want to swap the game updates to http due to the packet size limitation

[old arrow]

Well wouldn't you just do the flashing via WebUSB instead of web serial

[keen sorrel]

https://caniuse.com/web-serial
looks worse wdym, android chromium browsers don't support it

[wraith turtle]

Fair lol

[old arrow]

That's kinda the normal way. Or a separate desktop app like the esphome CLI for example

[wraith turtle]

Yes but most desktop browser do. So it works for the flashing at least? Lol

[old arrow]

I think you should be able to do whatever you need via webusb which has more support than web serial

[wraith turtle]

Well the firmware can be done a billion ways so the way it's done just needs to be accessible, but we need a good way to flash and configure and then swap to http after deployment

So webusb is a solid option

[wraith turtle]

Actually it seems they both have the same support

[old arrow]

Webusb is supported by android chromium browsers

Web serial seems it is not

[wraith turtle]

That makes no sense at all lol. Web serial just just an abstraction of webusb... Just a module for serial specifically

[old arrow]

serial devices don't necessarily have to be usb

[wraith turtle]

Like the underpinning of web serial is webusb lol

No the API itself

[old arrow]

that would make more sense though, because you support the lower level thing but not the higher level one

[wraith turtle]

Like serial is made using webusb

[old arrow]

Yeah. So webUSB doesn't depend on webserial

[wraith turtle]

Other way

[old arrow]

If it was the other way round it would be weird

[old arrow]

you said webserial depends on webusb

[wraith turtle]

Poor wording on my post lol

[wraith turtle]

Yes, web serial is an abstraction, it uses web USB but handles serial situations of USB specifically

[old arrow]

Yeah. Then it wouldn't be impossible to just not have the abstraction since you don't need webserial to have webusb

WebUSB is lower on the pyramid than webserial

[wraith turtle]

Webusb is lower level and works with all usb

[old arrow]

You can remove the top of the pyramid

[wraith turtle]

Correct

Just a layer in the tech sandwich

[old arrow]

So android chromium just has the bottom of the pyramid. They removed the top which is fine because it doesn't collapse

[wraith turtle]

Though honestly having the flashing and config with serial seems like a lot less work lol

Just need to make flash and configure accessible and easy to use from a laptop or phone (one or the other not both)

From there the client and game data can go over http

[wraith turtle]

time for an architecture change lol

ESP32 Boot → Serial Config → WiFi → HTTP Server → Web Client

can just make a separate tool for flashing and configuring over serial later and when flashing during dev just temp hardcode the creds or use serial to set them.

honestly was probably just having fun with ble for a bit

will also make it so that later if i want to add an x86 build target for people to do nodes on x86 the adjustments would be minimal

just have the server check its build target when starting up and if its x86 just skip the network config step and go right to serving the player/node setup step over http

[wraith turtle]

redid the implementation and its working like a charm for now and just using http and websockets

pretty good asteroid game server running on an esp32 😄

performance is even stellar:

=== PERFORMANCE METRICS ===
Uptime: 15000ms
Heap Free: 238320 bytes
Heap Min Free: 179356 bytes

--- GAME LOOP (Core 0) ---
Frames: 938
Avg FPS: 62.5
Avg Frame Time: 0.02ms
Min Frame Time: 0ms
Max Frame Time: 1ms

--- NETWORK LOOP (Core 1) ---
Frames: 455
Avg FPS: 30.3
Avg Frame Time: 1.53ms
Min Frame Time: 1ms
Max Frame Time: 3ms

--- WEBSOCKET PERFORMANCE ---
Sends: 227
Avg Send Time: 1.01ms
Min Send Time: 1ms
Max Send Time: 2ms

--- SYSTEM STATUS ---
Game Running: YES
Asteroids: 40
Projectiles: 0
WebSocket Clients: 1
========================
Heap free: 238320 bytes

hell networking is taking up more time than the game itself lol

never underestimate how much you can do with 512kb of ram 😂

[cloud knot]

i guess you don't want to pay apple the yearly 99 dollars, cause otherwise you could make a normal app, with Cordova/Capacitor (if you want to use web technologies and delegate BLE to native plugins), Flutter etc

[wraith turtle]

i have played a bit with capacitor but honestly i just wanted to aim for platform agnostic

[keen sorrel]

You don't actually need to be in the Apple Developer Program for ios anymore. Though what you can do with just an Apple ID is limited to up to 3 of your own personal devices, for up to 10 apps at a time, and they'll only be valid for up to 1 week.

[wraith turtle]

no having to build apps and distribute them. honestly i have focused heavily on pwa's and the such so the plan was to just use that instead of actual system apps.

since i migrated to using http over ip for the game communication i have no need for the specialized api's that a native app would call for

[frozen flame]

And limited Capabilities

Idk I pay for it for my hobby work, it's not a huge amount compared to a lot of other hobbies lol

[wraith turtle]

i just dont want to touch apples ecosystem with a 20m pole

[frozen flame]

Ask a warhammer fan if $100 a year is a lot

[wraith turtle]

and bam, we now have support and implementation of the use of psram 😄

=== PERFORMANCE METRICS ===
Uptime: 4200ms
Heap Free: 252908 bytes
Heap Min Free: 252204 bytes
PSRAM Total: 8385735 bytes
PSRAM Free: 8369547 bytes
PSRAM Used: 16188 bytes
PSRAM Available: YES
Memory Mode: PSRAM + Internal RAM

--- GAME LOOP (Core 0) ---
Frames: 263
Avg FPS: 62.6
Avg Frame Time: 0.98ms
Min Frame Time: 0ms
Max Frame Time: 1ms

--- NETWORK LOOP (Core 1) ---
Frames: 128
Avg FPS: 30.5
Avg Frame Time: 1.00ms
Min Frame Time: 1ms
Max Frame Time: 1ms

--- SYSTEM STATUS ---
Game Running: YES
Asteroids: 40
Projectiles: 0
WebSocket Clients: 0
========================

though i want to change the user/player structure. right now nodes are tied to individual players. instead i want to make them have their own user managment so each node can handle more than one ship and player owned/maintained by a node.

though will admit getting the networking working well was a pain and to get consistant network times i had to implement a custom binary data object to make the packets super small and super efficent for either side to encode or decode and transmit. json was just too heavy and slowing down the network core

[cloud knot]

without apple id you can't use any features which require you to do configuration on the Apple developer portal. Stuff like push notifications via APNS for example. Or sign in with apple (you can't create a key required for that)

[keen sorrel]

I'm aware of limitations to capabilities. None of the missing capabilities matter to this context

[cloud knot]

https://grpc.io/ + https://github.com/grpc/grpc-web ? Never tried it tho, but it is protobuf based

[wraith turtle]

That's a bit overkill, I just needed to optimize the data format and not use json and instead use a binary data setup so encoding is super efficient and so is decoding and transmission. The packets are down to 244byte contents

Can send and encode packets that small and simple way faster than 30hz on way worse hardware lol

[cloud knot]

the point of using protocol buffers is that you have the specification of definition shared between frontend and backend

(and then you can auto-generate the clients for said definition)

[wraith turtle]

Yeah, but size and optimization matter greatly in this form factor so the less anything is better

We are running on a microcontroller

The less the mcu has to store, send or process the better

So just using a simple binary object is really efficient

[cloud knot]

unrelated: I hate when people archive their repositories with nothing to indicate why they did archive it. At least say "sorry people, i have no time for this project anymore, ok thx bye" in the Readme 😄

[wraith turtle]

Yo fr, you preach the truth

[cloud knot]

rant from cleaning up my bookmarks from last 20 years

[wraith turtle]

It's even worse when another person and repo takes over the project and the original owner makes no comments or anything about it...

Pwnagotchi I'm looking at you specifically

That original one isn't even archived but is long dead with no mention or pointing that another project had taken over or that that repo is dead... And it's the first result on Google... And there is a whole different maintainer and repo now with many more updates and commits and the whole community

[keen sorrel]

flatbuffers could be a good fit imo. it's technically not as network efficient as protobuf, but it's far more minimal and device resource efficient
https://flatbuffers.dev

[wraith turtle]

Honestly might use that one! Ty ty, looks amazing for embedded use

Next to no footprint or overhead

[keen sorrel]

not sure how larger it'll be on an esp, but the last time they updated the benchmark page the entire library was 15kB on x86

super tiny

[wraith turtle]

Compiled or uncompiled?

[keen sorrel]

compiled

if you tree shake it should get even smaller too

[wraith turtle]

Not bad, though the esp internal ram is only 512kb so it would be about its operational footprint

Not just the flash storage it takes up

Though with utilizing psram it should be a non issue

[keen sorrel]

oh huh they say "library source code". not sure why that's even a metric but ig that's the only cross platform way to describe size or smthn. The benchmarks.md that's from might also not have been updated in quite a while. The git history is missing

[wraith turtle]

Though on an xtensa core on an embedded system system with many less instructions that are much smaller (uses risc ideology for the instruction set) it probably compiles smaller

I'll have to implement it and see the footprint

See what we are using of the heap

[keen sorrel]

in any case it's so tiny that it realistically should fit on even incredibly constrainted devices

[wraith turtle]

Was just a factor to at least find the info on lol, even if it didn't turn out to matter lol

I'm so used to json with JS I forgot how freaking big it is so I tried to use it at first and found out the esp was struggling to encode them fast enough to json lol

That's why I just made a custom binary structure and encoding

Though will probably migrate to flatbuff tomorrow as standardization is always nice especially if anyone wants to write a custom game client for it one day lol

Thanks a bunch for that reference and save @keen sorrel

Or well, after I sleeps lol

Though it does say on the main page it doesn't use heap so that is an interesting aspect

Wonder how they pull that off

[keen sorrel]

It just won't use any more memory than the message itself. Since deserializing from the message is done without any intermediate parsing

It's why it's less network/wire efficient than protobuf, since they can't do anything fancy that would require parsing with extra allocations

[wraith turtle]

Yeah, just read through it and it's pretty ingenuous imo

The compiler just formats the read to directly get the data from buffer without any additional code or parser

Basically just shortcuts a step with smart formatting and compiler time optimization

[keen sorrel]

it's a trade-off but worth it. Games was the original motivation but MCUs definitely fit too

[wraith turtle]

It knows the data is x size and x exact binary structuring so it knows that if the it's at this spot in memory it just goes x spaces to the exact right address and knows how to directly read it

[wraith turtle]

Well it's still for a game lol, just adding embedded context on top of it lol

[hollow basalt]

lol

[keen sorrel]

What's so funny?

[wraith turtle]

@keen sorrel just did some tests and if you want to chuckle.... my implementation is actually more efficent for my usecase lol.

1. since im doing direct struct copy which is fater than the serialization of flatbuff
2. our data is actually smaller with my setup
3. my implementation has next to no overhead or footprint
4. since only two things have to understand it (either side) flatbuff is a bit overkill

[cloud knot]

i can already see how this is going to be pain if the SMS doesn't arrive in timely window

[wraith turtle]

though i will be saving that lib for future use for dam sure lol

also would be a lot of work to swap to flatbuff in this situation. would be a lot of work for just easily portable structure and actually a loss in space and slight performance.

should have figured though as i just decided to break it down as simple as i could lol. the thought was if i moved to binary data i couldnt get much more efficent or smaller

[cloud knot]

but again, the point of protobuff and flatbuff is the fact that schema exists, so you can't have a mismatch on either side of communication if you use the same schema. Right now you have to match those manually

[wraith turtle]

well seeing as there is only one server and one client, the need for that schema isnt that important

only one thing of code making it and one thing of code reading it and the best efficiency possible is paramount. as long as i dont botch my own code it shouldnt be an issue. and if i do it will be pretty obvious.

if there were many different systems or different clients connecting in and using that data frame it would matter a bit more

[keen sorrel]

raw structs is always the most efficient. flatbuffers isn't intended to beat that and it's one of the things they compare against in their own benchmark. What flatbuffers provides is a considerably more complex abstraction with tables, optionals, and forward+backward compatability.

[wraith turtle]

thats fair, structured data is always nice

[wraith turtle]

though might add keys so the size can be variable

being able to only send 1 projectile when there is only 1 projectile and only 1 player ship when there is only 1 player ship can net as much savings as reading and building the empty spaces in the packets

[wraith turtle]

actually looking like asteroid now 😂

[wraith turtle]

@keen sorrel ended up going with flatbuffer anyways lol, though built a custom scafolding to auto build and insert the pieces into the project. though the hard part was wiring a ts to js converter that was able to make the code run client side (avoiding import and the such) i just got too tempted to try it 😂

though now that i have done that work.... modifying or changing anything is as simple as adjusting the schema and running the build script lol

honestly automating ts to js and converting to client side was a pain....

[rustic hare]

Hello! Besides learn-php.org , are there any other websites that can help refine my php skills ? 😄

[soft fern]

Why? I mean I used to write PHP back when it was a popular choice and didn't want to bother with Java/jsps or c#/asp, but typically modern solution involves separating API from frontend so usually rest or graphql apis with Json request/response... Not to say you can't do that with PHP, but hard to say why you would really want to. Guess question for you might be what's the scenario you're trying to learn it for?

Personally just used the docs like a dictionary back when I used it more so would just incrementally learn stuff by reading docs or if I couldn't really understand the docs would just look at the source behind it in the interpreter for more details

Oh yeah also getting xdebug or otherwise remote debugging working was pretty clutch for when things went really south too

[rustic hare]

I'm a bit salty that for an open source Ticket System I enjoy using there's nothing for a plugin regarding Active Directory operations because SOMEONE DIDN'T ENABLE audit operations on Active Directory. So I'm trying to "squeeze" the active directory operations in to a php plugin for this system and then also publish it because no one bothered to publish this as open source ...

Ergo why I'm learning php. All I it has to do for now is for me to be able to call some simple powershell scripts. Visual enhancing can be done later

And also, this system apparently can only "read" from AD. So I'm gonna make it to be able to write as well hehe

[soft fern]

Gotcha which ticketing system just curious to take a look? About the php used in the project ideally would use GH issues or see if they have a discord or other chat with devs that is still active at all to discuss the feature, might be something they already have and hide behind a flag or at least some cautionary tales if anyone has tried to implement the integration before.

[rustic hare]

I asked about this once already and instead was guided to "one of their global partners because that sounds like a specific plugin development and they don't do that" ....

So I decided to do it myself and just publish it kekw.

The system is called GLPI .

I'll use some github co-pilot for some fast code writes to get a test version up and running and then I'll feed their "plugin requirements" in to after .

[soft fern]

gotcha, yeah was gonna suggest LLM since they should be able to handle most PHP without issue since it hasn't changed a ton over the years, that said this project does look to be using PHP 8.x which I never even touched (was on the 5.x and some of the 7.x versions at latest) so I'm not really sure how much that will effect the LLM ability to write runnable code but I think it'll be fine, just took a look at the composer.json here for dependencies they are using symfony which is a pretty big lib/framework but I never touched that either mostly worked in Laravel, guess worth looking up the latest matching version of PHP docs and that symfony version would be good to be familiar with high level at least but not sure where else to point ya

https://github.com/glpi-project/glpi/blob/11.0/bugfixes/composer.json

Also GH-copilot is okay, I like roo-code extension better myself, it has different "modes" it switches between that have different sets of tools or "modes of operation" for "orchestration, architecture, code, debug, ask" and if start in "orchestration" it creates a high level plan then hands off sub-tasks to other modes to use tools to get stuff done, it works well for medium/larger tasks than copilot itself from what I've seen (can also still use GH copilot licensed LLMs behind roo-code or Ollama locally or OpenAI/OpenRouter etc etc lots of providers can hook into and optionally switch them for different modes)

If decide to check out roo-code would also say grab qdrant (vector db) and create an "index" in roo-code for faster look-ups too helps reduce back and forth between remote LLMs and cost/speed overall improvement

can run docker container for qdrant it's the vector/db local storage, and then you use an "embedding LLM" to handle generating a vector based embedding for doing semantic look-ups on the code-base.... it sounds complicated and sort of is but also is easy to setup and helps the performance overall for making progress on things when LLM has quick look-ups across the code and doesn't need to read every file while working through a task

[silk eagle]

the worst part of that screenshot is that the text is being selected on a phone

[rustic hare]

I did some googling and apparently the codellama:70b is the best LLM for coding. Gonna try to set that up in my visual studio because screw limitations on ChatGPT xD

[cloud knot]

issue is fees (if you run it on cloud) or initial costs (to buy GPUs to be able to run 70b model)

to run codellama:70b which is a 25-49GB model (based on quantization) you probably want at least two 3090/4090/5090 cards. Or a Mac with 64GB RAM. Or 64/128GB Framework desktop/similar AMD AI platforms

[rustic hare]

God damn...I have a shitty 1660 6 gb ...

[cloud knot]

i run qwen2.5-coder:14b (9GB) on my 4080 Super, maybe it could run 30b (20GB) with some offloading to CPU/RAM, but i didn't test that enough to be confident it runs well enough

[cloud knot]

in the end, it's up to you if you pay for 'renting HW' or you buy your own, at cost of getting parts and possible headaches setting it up

[rustic hare]

Damn...So basically my gpu too smoll for local LLM ? 🙁

[cloud knot]

no, but your local LLM options are very small, and could be very dumb

[rustic hare]

Damn .,..

So I need a new GPU xD........

[cloud knot]

6GB VRAM => you could run qwen2.5-coder:7b, deepseek-coder:6.7B, llama3.1:8b etc

[rustic hare]

But how good are those models in comparison to github co-pilot in VSCode ?

[cloud knot]

those are all sub-6GB models, you can always just try, all you lose is some internet download cap, if you have one

for vscode, i use continue (for chat) and cline (for other tasks), but i haven't really dig that deep. As said, you can always just try it and see how it performs for your use case

having said that, i only dug into ollama so far, which is the most well supported local tool imo. llamacpp and others not sure

[rustic hare]

qwen3-coder:30b is what Ollama shows me ...2.5 ain't there 🙁

[cloud knot]

https://ollama.com/library/qwen2.5-coder

literally just run command ollama run qwen2.5-coder:7b and that will download the model and then start a chat session in command line, but then it will also show up in UI

[cloud knot]

or type the model name fully in UI:

[rustic hare]

Worked ty! 😄

[cloud knot]

2.5-coder:14b did well for me, but i mostly bothered it with CSS/HTML questions. I have a 4080 Super and 2080 Ti tho

[rustic hare]

I need PHP and Powershell 😄

[cloud knot]

PHP is also fine, i was asking it about upgrade process from Silex to current Symfony

(for context, silex was a minimalistic framework made by Symfony guys back in time between Symfony 2 & 4)

[rustic hare]

How good is your experience with setting up qwen2.5 with continue in VSCode ? xD

[cloud knot]

as i said, continue i used only in chat mode, so you might as well use the chat mode in ollama directly; i only used "current file" as context, but that one failed miserable on my request (how many items were in TS object).

[rustic hare]

Because all I'm doing is failing with the guide in front of me ..

[cloud knot]

let me try to redo the config here

[cloud knot]

what part is failing ? logging in to local ollama or configuring models ?

[cloud knot]

1. ollama installed
2. install continue
3. select to use local models, pick ollama

[soft fern]

Yeah definitely hit or miss on which models can be used as agents or have been trained for "agentic use" too even with the proprietary ones, roo-code with gpt-5-codex they just patched some issues to make it work more consistently since that model just hit but gpt-5 (not codex version) can work pretty well, Claude code 3.5-4.0 also work well enough for acting as agents in terms of cloud based ones. For local I use ollama too but have mostly used embeddinggemma for generating embeddings locally (somewhat slow but also usually can run once and keep using same index so long as it doesn't get corrupted)

Roo-code will also let you use ollama as a provider haven't seen/tried "continue"

Open router seems pretty reasonable for pay as you go and access to lots of different models but for the most part am "corporate sponsored" on getting to tinker around with these on projects.. I am using personally on side project too though and has been somewhat expensive but easy to monitor usage too

[rustic hare]

Configuring models ...

I get this dumb error.

[cloud knot]

open your home folder/.continue/config.yaml and set it to:

name: Local Agent
version: 1.0.0
schema: v1
models:
  - name: Autodetect
    provider: ollama
    model: AUTODETECT
    roles:
      - chat
      - edit
      - apply
      - rerank
      - autocomplete

then it willl iterally autofill options with models

after that open the continue sidebar, press settings:

and then you can pick/choose defaults

or you can pick a model you want to use right now under tghe chat window:

https://docs.continue.dev/guides/ollama-guide#method-2%3A-using-autodetect based on this

[rustic hare]

YEY ! It worked 😄 Thank you!

Darn it's so slow xD but free is free I guess

[cloud knot]

it that the big model or one of the smaller ones ?

[soft fern]

Yeah trade off running local vs cloud compute better GPU can definitely help too 1060 vs 3070 here with stable diffusion is like 20-30s for an image vs like 2-3s, a little harder to see the difference with LLMs but is there for sure too

[rustic hare]

The 2.5 coder one

[cloud knot]

yes, but there are many

[rustic hare]

I pulled the latest 7b*

[cloud knot]

so that might be 7B, ok, then i guess it's just slow on your 1660 😄

it should fit in your VRAM, but missing tensor cores are hurting you

[rustic hare]

Yeah. Should bother with getting a new GPU, buuuuuut I guess I'll just run this one in to the ground first kek

I mean, it's been in my PC for the past 4 years ...so it will die sooner or later.

[cloud knot]

as i said, you might want to try https://marketplace.visualstudio.com/items?itemName=saoudrizwan.claude-dev too.

Continue is more like "chat + provide context for tips in chat"
Cline is "ask me to do something and i do those steps on your project/folder as long as you allow me to do those steps" (aka agent)

[soft fern]

I just rent cloud time if need more than my local can handle, eventually will bump gpus but moreso for games than local models (just a bonus)

[cloud knot]

https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif

[rustic hare]

Can I throw things to do at it endlessly or is it limited ?

[cloud knot]

define endlessly. Every LLM has an issue with context window. Longer context window = increased RAM/VRAM requirements and slowing down of response

but if you can separate those tasks in separate jobs/sessions, sure, you can do it "endlessly"

[rustic hare]

I mean it is a bit slower than the usual speed of ChatGPT, but DAYUM I like it's responses ❤️

[cloud knot]

something like 3090 is a middle ground between performance/VRAM/price for LLMs. Used 3090's can be had for ~500 euros, have 24GB VRAM (so you can run a lot of models meant for local use), has tensor cores, so while it is not peak performance anymore, it still does well enough.

[cloud knot]

"usual speed of chatpt" - until it says 'come back at 17:00 cause you ran out of your tokens'

[rustic hare]

I hate that

hate this too :

[cloud knot]

i mean qwen2.5-coder with Continue did bad for me with certain tasks (but maybe with Cline it would do better) and it was stuck in a response loop once, but then again, free is free. Firing a question towards it once in a while for weird HTML/CSS question before going to google is only costing me some time if it responds with BS or unhelpful answer

(unhelpful was for example when i asked it what to do when i got Android SDKs from two suppliers, where both of them in infinite wisdom decided to fully obfuscate every package structure, so we ended with two SDKs both having classes and packages named a.a.a.a.A, which then trips the Android compiler; and it said to ask SDKs supplier to fix their SDK, which while it is something we will have to do in the end, it was not helpful at the moment)

[cloud knot]

to be fair, for me the biggest surprise is how much general knowledge still remains in "coder" models. I asked qwen2.5-coder:14b for it to create an STL model for 3D printing for certain shape... and it started it giving me guides about how to do it in FreeCAD, Blender, Meshmixer, and even via Python script for FreeCAD 😄

[soft fern]

Hah nice, typically they can output some openscad but spatial reasoning is still questionable at best

I use onshape quite a bit too there think can be helpful for "feature scripts" probably but haven't used them much, there's stuff like trellis.ai or meshy.ai that are more geared towards 3d space and meshes

[rustic hare]

OMEGALUL

That was 1 request xD

[soft fern]

Think space-time really the biggest remaining issue with modern wave of AI, lots of people trying to solve with video or embodiment in bots and whatnot but is still a problem where the models don't know something is running in background or that processes may persist or run in foreground when using them as agents, most often I'm telling it to stop or check what's running before it tries to relaunch something already building in the background

[cloud knot]

were you running against cloud or ollama ? Press "Use your own API key', pick Ollam and model

[devout vortex]

anyone in here work with ESP32's and C? Im having a very interesting issue and looking for some assistance. general searches of the web have yeilded no results.

[keen sorrel]

https://dontasktoask.com

I have experience with embedded, but not ESP32 and not really C either. I do C++ on nRF primarily (zephyr rtos)

[devout vortex]

touche. i have an esp32 where the program works just fine when i have the programmer connected but when i disconnect the programmer the program doesnt run. i dont have any references for serial 0 but i am opening serial 1 for reading bytes for further manipulation and comparison.

I can post the code upon request

[keen sorrel]

What procedure are you using to flash it?

[devout vortex]

not sure exactly what youre asking, im using the arduino2 ide with the esp32 boards added in, is that what you are looking for?

[keen sorrel]

ESP supports a handful of different ways of flashing their chips. Are you flashing through the USB debug serial port? Some ESP32 devices have a USB peripheral on the chip itself that can be used for USB DFU or CDC flashing. If you're flashing the USB debug serial port (the normal method for most ESP32 devices), then you may need to hold the boot button down when you click upload so that it goes into flash download mode.

https://docs.espressif.com/projects/esptool/en/latest/esp32/advanced-topics/boot-mode-selection.html

https://docs.espressif.com/projects/arduino-esp32/en/latest/guides/tools_menu.html

[devout vortex]

correct, i have a switch that breaks/makes gpio0 go to ground, i have to flip that every time i want to program the device

[keen sorrel]

What messages are sent on the debug serial port if you hit the en button? The bootloader should report a few things during boot

[devout vortex]

not sure cause anytime i have the serial lines connected it works just fine, its when i disconnect the serial lines that the program doesnt run anymore.

i even went as far as a simple led blink code just to be 100% sure it was actually failing to get past boot

[keen sorrel]

Have you tried an alternate power supply? What board do you have?

There isn't a mechanism for the device to know if serial is connected, so I'm guessing it's a power issue that USB VBUS is resolving when the debug usb is plugged in

[devout vortex]

the programmer is a seperate device connecting to the rxd and txd pins of an esp32 wt eth using the wt32-s1 esp chip. im using a bench power supply through the same pins the programmer uses when plugged in to reduce as many variables as possible. i measured the voltage of the programmer board(3.31) and the bench power supply(3.299) at the input to the board.

[keen sorrel]

Are you saying if you simply plug in the RXD/TXD pins, it'll suddenly boot?

[devout vortex]

well, good news i figured it out. apparently my bench power supply is bad even though its putting out the correct voltage and showing current flowing. i used a different power supply after building a second unit and it worked without the programmer. yes it did start working at one point when i plugged the programmer in after it was externally powered but i think it was cause of the power issue. i apologize for not trying it sooner. i legitimately thought that the bench psu was nigh infallable so thats a lesson learned for me!

[keen sorrel]

@spring lily dont ask to ask in my DMs lol. No, I don't know Unity. I do know c++ and some C# though

[spring lily]

alright, my fault

[gloomy girder]

I’m looking for self development. should I grow this out

[neon oriole]

this is actually genious :https://www.youtube.com/watch?v=4Qj5y0Vtdds

the number of times i wished i had something that could do that ...

(it generates functions from datapoints that are simple and dont use fourrier)

[keen sorrel]

I appreciate the thought, but if you're making a unity game as an indie company you should find someone else with experience in unity

[frozen flame]

sir this is a Wendy's

[woven path]

hii

Imagine a rack server setup powering your projects behind the scenes. These servers fit neatly in a rack, giving you massive performance without taking up extra space. You can handle websites, apps, or heavy workloads all at once, with smooth speed and reliability. With organized cabling and smart cooling, everything stays efficient and hassle-free. It’s like having a mini data center at your fingertips—powerful, professional, and ready to grow with your needs. Know More Visit: https://www.lenovo.com/ch/de/c/servers-storage/servers/racks/

[fading osprey]

I feel like this counts as advertising......

and your bio is written from the third person? Very odd choice

[woven path]

no iam not from any brand , This only my experienece about this product

[fading osprey]

still isn't very on topic for software development. That's a hardware product in a software channel

[hollow basalt]

ok

[wide reef]

https://github.com/1a11/billard designed and built this little thingy to host my little blog

[fading osprey]

That's actually really pleasant to look at, good work!

[wide reef]

Thank you!! I'm mostly a designer/backend person so it was kind of a point. Page-wise it probably isn't very good but hey, there's a base there (I halfassed it in collab with Claude hehe)

[fading osprey]

if it works it works, at the end of the day it only needs to meet the needs you have for it, and if it meets those, who cares how you did it? (unless you used slavery, then we would probably care)

[wide reef]

I try to minimize bird deaths!!

[fading osprey]

silently curses all those damn windmills the dutch have

[wide reef]

FUCK THE WIND TOURBINES

[fading osprey]

please do not, that is how they reproduce

[wide reef]

[mental panther]

<@&750150305383186585>

Multiple channels

Every* channel

[quasi cloak]

holly molly,is coding baaad on Android.So basically ,recently I got a keyboard/case combo for my android tablet ,mostly for school but I got curious and decided to try to creat a small game. Nothing special,just a pong clone in asci running on the terminal.And wow that was a ride.First I decided to do it in java ,mostly to practice it.So I hopt on the google play store and there were actually IDE for Java but it was a mistake .They have ads,a subscription model and they run a completely outdated version of java .I thought to myself and thought ok lets give python a shot.Nope,same problem.

O I should also mention that if you want to install a library you need to pay money.JUST FOR ONE

[cloud knot]

you don't develop on mobile platform for mobile platform. you do it on a PC

[cloud knot]

you download this to your Windows/Linux/mac computer https://developer.android.com/studio install it, install SDK from within the Android Studio, then connect your phone/tablet to it to run the app you created

[quasi cloak]

I know,I did just to try out

I have a laptop I know how to do stuff I just wanted to try it out

[cloud knot]

i would use tablet maybe for vs code running on server

like this https://github.com/coder/code-server 😄

have it run inside a web browser from your server, that is as far as i would go with coding on tablet

[quasi cloak]

oh yea

[old arrow]

that's the bottiest bot that has ever botted

[shrewd valve]

<@&750150305383186585> could you look into 1308742061138382910? second time they wrote something very ad like, both for lenovo products. sounds and smells a bit like the current msi bot recommends on reddit https://www.reddit.com/r/GamingLaptops/comments/1np87ra/this_brand_is_hijacking_our_recommendation_posts/

[wide reef]

Have to fix a bunch of CVEs

It sucks

I hate to roll out a full database and validation engines in the back end. It needs to be a minimal validation page thing where json is not validated properly and page just won’t load

[hoary totem]

anyone familar with basic C here?

[old arrow]

Only if you say what the specific question is

[limpid reef]

Yes, I'm aware of what Basic C is.

[keen sorrel]

I'm afraid I only know Holy C

[last sonnet]

not really sure what channel to post this in, but I was walking by one of those EcoATMs that buy phones, and it was on some sort of developer/support menu. thought it was neat

No way to interact with it from the outside of the kiosk though sadge

[frozen flame]

what a horrendous UI. I absolutely love it

[dreamy sandal]

i made vim

[grizzled steeple]

Wait, those things run on LabView??
I recognize that UI. LabView is hella expensive per license

[keen citrus]

but why?

also is there a git repo?

[midnight wind]

you'll be suprised how many things run on labview

[dreamy sandal]

not yet.
also its literally just an entire operating system

[keen citrus]

all x86 asm or have you jumped to c code yet?

[dreamy sandal]

im using c aswell yeah

[silk eagle]

vitamin C is usually found in acidic foods I think

not strictly but like citrus is vitamin C right

so you probably wouldn't find it in something basic

[keen citrus]

yeah whatchu need

[wide reef]

alas

the engine is finished

https://github.com/1a11/billard

have fun hosting or commiting, it supports docker compose out of the box, is also on the docker hub for easy pulling

[old arrow]

it reminds me of something i planned on making a few years ago

very neat

[frosty hemlock]

Hey guys, i have a dilemma.
I want to buy a GPU for machine learning and inference. It absolutely doesn't have to be a top of the line GPU. Theoretically any cuda & 8GB vram gpu would do. I have two places where I can add a GPU. I have a server with a pcie slot of 1 height (for low profile 1 slot cards e.g A400, A1000, GT1030, T1000).
And I have a PC with a full pcie slot able to fit larger cards.

I currently have my eyes on the following

• RTX A1000
• RTX 5060 8GB
• RTX 5060 16GB

And im constantly checking for discounts or occasions on the 5070

I really have no idea which to buy. I would really love to have a low profile card that fits in my server for 24/7 inference but I like how the 5060 has the latest architecture and the most tensorcore precisions

[cloud knot]

Do you have a budget for $500 or so ? then 3090. Seriously.
I don't think there are any reasonable singleslot low profile cards to be used for LLM.

3090 will give you 24GB VRAM, which is kinda huge

[frosty hemlock]

Thats a little high tbh

[cloud knot]

but it's also much more better than 5060 for LLM

not saying it's impossible to run LLM on smaller GPUs, just that the VRAM jump and thus model compatibility jump between a 5060 16GB and used 3090 is big, while price jump is rather small

i mean 5060 8GB sells for like 300-350 euros in my country, 5060 Ti 8G is 400-460 euros, 5060 Ti 16G is 430-560 euros.

Compared to that, used 3090 24GB sits right there around the 5060 Ti 16G (there is no 5060 16G, so i suspect you meant Ti?)

[frosty hemlock]

I must've mentioned that the 5060 16GB is also already a bit too expensive. I really wish to not go any higher than 400 euros as im not planning to run any large LLM's it would mostly be around 1-3B models and probably quantized too.

And yes I think I meant Ti

[vapid grove]

What about the Intel Arcs?

[frosty hemlock]

They have no cuda right?

[old arrow]

nvidia is the only sane option for GPGPU and ai

even amd isn't the greatest experience

[vapid grove]

No, I believe CUDA is specific to Nvidia (?)

[cloud knot]

then pick literally anything. I had no issues running qwen2.5-coder:14b on a 1080 Ti 11GB (9GB model), which is sub-100 euros these days

(having said that, i only ran ollama so far, haven't tried the other local options)

[frosty hemlock]

I currently have AMD and I have tried several times with several methods to run AI models on both linux and windows but its terrible

[frosty hemlock]

Yeah I prefer to have cuda as that currently still the best supported for ML and inference

[cloud knot]

qwen2.5-coder:3b is like 1.9GB, 7b is 4.7GB, so any 8GB card will run 7b model, any 11-16GB card will run 14b model

[frosty hemlock]

I currently have to use my CPU (ryzen 7 7700x) which is insanely slow for inference so I guess in my case any nvidia would work too

[cloud knot]

here is how qwen2.5-coder:14b performs on a 9950x/4080 Super system. Give me a minute, will show you the same on a 2699v3/1080 Ti system

(that is, if the system boots LOL)

[fading osprey]

yikes that token speed is not the best. Don't get me wrong it is not bad given that it is running on a local system but that's with a 4080...

[frosty hemlock]

Interesting, the token speed is indeed not as high as what I would've thought but its still better than running on a CPU

[cloud knot]

well Windows certainly doesn't help

[frosty hemlock]

Are you sure its not running on CPU btw?

It looks like around 8-9 tokens/sec

[cloud knot]

30% CPU, 30-50% GPU

[frosty hemlock]

Mhm

[cloud knot]

asked a different question 30-40% CPU, 50-80% GPU

let me test with 7b

[frosty hemlock]

The 4080 should do around 60-90 tokens/sec

Ive seen some websites reporting nearly 200 tokens/sec on a 4080 super

[cloud knot]

here it is with qwen2.5-coder:7b, but that is a rather small model

[frosty hemlock]

Thats a whole lot faster

[cloud knot]

one more thing, let me test

nah, context window change didn't speed it up with 14b model

nevermind, it did, but had to go really low

16k -> 32k context is what slows it down

[frosty hemlock]

Im still deciding between the 5060, A1000 and now also the 3060 with 12GB

[cloud knot]

here it is with 16k and 32k context length

[frosty hemlock]

Nice looks good

And on the 1080?

[cloud knot]

still trying to set it up LOL, system has problems

[frosty hemlock]

Ah ok haha

I think I might go with the 5060

I won't be running large models anyway

And if I really need to run a large model, we have an H100 at the office

[cloud knot]

few more minutes

1080 Ti with 16k window feels like the 4080 Super when it has that long context window. 1080 Ti with 8k feels faster than with 16k window

1080 Ti 16k window with 14b model

1080 Ti 8k window with 14b model

both certainly do some work on CPU still, but hey, it's a 1080 Ti from 8 years ago 😄

[frosty hemlock]

Mhm thats pretty good

Thanks for showing me some demo's

Gives me a lot better insight into the speed difference

[cloud knot]

unfortunately got nothing inbetween - my 2080 Ti is ~150km away, and i skipped the 3000 and 5000 generation

[frosty hemlock]

Maybe ill give a demo of the 5060 if I buy one

[cloud knot]

obviously it comes down to what you expect from the LLM. If you only ask one-off questions, the context length doesn't matter. If you ask questions with attached files, with back and forth, then context length might become limiting

[frosty hemlock]

Context length would probably hang around 2-4k

Maybe not even that

I just want to LoRA's

[cloud knot]

https://www.youtube.com/watch?v=VGyKwi9Rfhk

[soft fern]

yep largely comes down to if the model fits into VRAM then higher CUDA core counts and clock rates tend to be the limiting factors

[frosty hemlock]

Im probably gonna go with the 5060

[soft fern]

for anything "non-agentic"/not using reasoning and tool calls or longer context lengths can run locally, but otherwise either paying credits through open-router or other similar providers or paying for rented cloud instance time to run things... been looking at new machine in part for AI things too, I want a 5080 but I also don't want to spend $1k on a card 😛

was building the matrix demo from unreal 5 release with nanite stuff the other day though and was like 4hr endeavor, it runs well enough on my current machine but I want to start messing with some that in VR and then I defintiely don't have the VRAM to do that or encode capabilities etc.

[frosty hemlock]

Same haha

My budget for a card is no more than 400 euro

[cloud knot]

it's really sad you can't stretch that tho. 3090 would be an awesome pick for just 20% more 😄

i mean sure, on average they sit more around 600 euros now, but often there is a deal where someone sells it for 500 euros or less. Right now someone is selling one for 489 euros where i live.

[frosty hemlock]

The 3090 in my country is like 1000+

[shrewd valve]

on the used market ofcourse

[cloud knot]

used, not new

[frosty hemlock]

Im not really a huge fan of buying used GPU's

[cloud knot]

want lot of VRAM for reasonable price ? That means used market. NVIDIA usually just stops selling old cards, they don't drop their prices

[frosty hemlock]

I don't need a lot of vram

[shrewd valve]

for llms you absolutely do. going to system memory/cpu slows everything down by a ton

[cloud knot]

he means he want to play just with small models, like 3b and such

[shrewd valve]

i mean all models even with 16 or 24gb available are relativly small. gpt oss 20b already needs 16gb iirc. also what gpu does he already have atm?

[cloud knot]

he is buying a new one, that is why i said to him he should buy an used 3090 in the first place. His budget is 400 euros, used 3090's show up at 500-600 euros here, so it is not a huge stretch, but that's me

[shrewd valve]

dont dissagree with the 3090 recommendation. just dont see the need for a new gpu in the first place if hes targetting 3b models

[cloud knot]

i don't think he uses any GPU right now, so he is buying something (if i remember correctly)

[shrewd valve]

@frosty hemlock do you have any kind of gpu already?

[frosty hemlock]

I want to use it for very small models, if I actually need to run something large, I have access to an H100

[frosty hemlock]

At home I have AMD RX 6650xt

And at the office H100

[shrewd valve]

ah rip. just out of ROCm support. Would definitely look at used 3060 16gb or 3090's then.

[frosty hemlock]

But I really don't need the high vram

8GB will do

RTX 5060 or an RTX A1000

The A1000 fits in my server and is more usefull for 24/7 inference, the 5060 is more modern with more types of tensorcores, but doesn't fit in my server which means I will run 2 GPU's in my PC

[cloud knot]

for me the worst part about the memory requirements is how much of general knowledge remains in the "coder" LLMs. Like this, why does qwen3-coder knows food recipes ? 😄

(and even in multiple languages)

[cloud knot]

btw kinda surprised how well i5-8500 in mac mini with 64GB RAM does:

total duration:       59.867651145s
load duration:        62.888202ms
prompt eval count:    55 token(s)
prompt eval duration: 79.414861ms
prompt eval rate:     692.57 tokens/s
eval count:           673 token(s)
eval duration:        59.724752084s
eval rate:            11.27 tokens/s

[cloud knot]

same prompt using a 3900X + 128GB DDR4-3600 (as opposed to 64GB DDR-2666 in Mac Mini):

total duration:       32.981286479s
load duration:        61.637866ms
prompt eval count:    55 token(s)
prompt eval duration: 601.280936ms
prompt eval rate:     91.47 tokens/s
eval count:           630 token(s)
eval duration:        32.317951373s
eval rate:            19.49 tokens/s

i mean yes, obviously it is not as fast as GPU only processing, but having somewhat reasonable speed (10t/s is good enough for me), while being able to load bigger models (with 64-128GB RAM) with hardware i already own is awesome

[keen citrus]

is that cpu only or is it using the igpu?

[cloud knot]

cpu only, there is no IGPU on 3900x

ran it with OLLAMA_LLM_LIBRARY="cpu_avx2", also nvidia drivers were requiring a restart after first install. but i can test it again

[cloud knot]

qwen3-coder:30b, now running with /set parameter num_gpu 0 instead of using the OLLAMA_LLM_LIBRARY="cpu_avx2" env variable, pretty much identical result:

total duration:       35.186116036s
load duration:        3.125379831s
prompt eval count:    55 token(s)
prompt eval duration: 675.66878ms
prompt eval rate:     81.40 tokens/s
eval count:           608 token(s)
eval duration:        31.384454875s
eval rate:            19.37 tokens/s

[cloud knot]

and to be 100% sure, here is how it looks on nvtop. so yeah, that's CPU only

[keen citrus]

on the i5 tho, would it be any faster with the igpu?

[cloud knot]

i don't think ollama supports intel igp, it complained that i had no supported GPU, so it had to run CPU only

[keen citrus]

right, yeah

llama.cpp does tho https://github.com/ggml-org/llama.cpp it can use vulkan iirc

[cloud knot]

this is a very old IGP tho, i doubt there is any reasonable performance in pre-Arc IGPUs

[keen citrus]

probably not lol

i have a laptop with a celron n3150 and 4gb of ram, wonder how well llms would run on that lol

[cloud knot]

well, you can always just download ollama and run one, but i doubt you got much space to do anything at all. I mean at least 250MB of RAM goes down the drain for the IGP, 2GB or more for OS, so you are left with 1GB usable memory, maybe ? I don't know if there are any models even that small

[keen citrus]

i think some really low param models are a few hundred mb

like embeddinggemma is 300m param, 650ish mb

[cloud knot]

of course different models might perform much worse on same CPU, but for me something like qwen3-coder:30b is the ideal model and at the speed i see on the CPU, it is more than enough for my occasional use.

[keen citrus]

yeah i think i have qwen2.5-coder 7b setup on my desktop, its like 5 gigs and easily runs on my rx 9060 xt

id do a higher param model, but then space could become an issue, im dual booting arch and windows on the same 500 gig ssd

[cloud knot]

9950x + 64GB DDR5-6000 is some improvement, but not as much as one would expect:

total duration:       28.6506126s
load duration:        5.3788934s
prompt eval count:    55 token(s)
prompt eval duration: 418.1631ms
prompt eval rate:     131.53 tokens/s
eval count:           572 token(s)
eval duration:        22.8500438s
eval rate:            25.03 tokens/s

[keen citrus]

i imagine you might be hitting some form of diminishing returns at some point, either that or your using a larger model?

[cloud knot]

it was all done with qwen3-coder:30b, no change there

[shrewd valve]

how are you doing the testing?

[soft fern]

It'll need that when you ask it to code up an app for your recipe book 😉 I hear ya but also coding requires lots of general world knowledge too not just syntax of languages

[cloud knot]

ollama -> on command line ollama run model_name --verbose. If you run it with the --verbose parameter, it prints out those stats at end of response. If you want to run it on cpu only, in chat you enter /set parameter num_gpu 0 (if i didn't do a typo here). In ollama, i used https://ollama.com/library/qwen3-coder:30b

LM Studio - you can set how much you want to offload to GPU when you are loading the model, you can drop it to 0; or you can disable GPU completely in the settings/hardware. LM Studio prints the token rate at end of every response. In case of LM Studio, i used "BasedBase : Qwen3 Coder 30B A3B Instruct 480B Distill v2 Q4_K_S"

not saying that distill in LM Studio is good, just that was the first one i downloaded for testing 😄. Now downloading the version matching ollama model

qwen/qwen3-coder-30b in lm studio is more or less same, no surprise there.

With 4k context length it also loaded, but it clearly still uses CPU, as the performance jumped up only to 33.09 tok/sec. With 16384 context length, 37/48 layers on GPU, again around 32-33 tok/sec.

For me, that is more than enough, others might consider that slow. 🤷‍♂️

[fading osprey]

I had a couple models running at 80+ tokens/sec but I just don't need anything that fast, it's actually distracting when it runs that fast for me

Mostly I find that qwen3-coder-30b seems to like getting itself stuck in repetitive cycles sadly. Had more than a few instances where it locks onto a word and keeps repeating itself until it gets terminated manually

[frosty hemlock]

Lmao

[cloud knot]

In multiple languages too

[cloud knot]

poor Gemma is much slower even with GPU, also first line it wrote was Okay, this is a substantial project! 😄

[old arrow]

i cannot get systemd-nspawn working for the life of me

• started systemd-nsresourced.socket
• started systemd-mountfsd.service
  now i get failed to clone [path]: Permission denied where [path] is the path i'm trying to mount into the container

[old arrow]

well so I had matlab working for a week but now it seems it's impossible

R2025b just gives error 5201

[old arrow]

downgraded to R2025a and somehow it works now, but java doesn't work again... the java installed by mpm doesn't contain libjvm.so or even java/lib at all

@weak brook when i told you i thought aur matlab was using its own java that was wrong, it just installs the java that mpm puts there

the matlab aur package just uses mpm to install everything, and then puts it into management by package manager

it complains

Warning: Unable to find Java library: /opt/MATLAB/R2025a/sys/java/jre/glnxa64/jre/lib/server/libjvm.so
Please check if you have a MATLAB_JAVA environment variable.
Note that MATLAB requires Java version 1.8 or higher. 
> In matlab.internal.addons/Explorer (line 31)
In matlab.internal.addons.Explorer.getInstance (line 42)
In matlab.internal.addons.launchers.showExplorer (line 180) 
Disabling Java support

i also tried setting MATLAB_JAVA to /opt/MATLAB/R2025a/java which is a real dir and same error (but it has the path i provided)

if I compare this to manually installing using mpm, and requesting both MATLAB and simulink, the /opt/MATLAB/java dir is identical to what mpm put in my test dir

DUDE

[old arrow]

i just realized i was wrong again, it does do --no-jre and i dont know why. if i go to settings and set my system installed openjdk-8, it works. now let's see if i can install anything

[frosty hemlock]

I bought the 5060 TI

I have tested it, and the GPU itself is good, very minimal setup when it comes to machine learning and inference. Its really fast and easily outputted text of 3B models in 1-2 seconds. But the second I start configuring it with my linux dualboot in such a way that I can keep using my AMD for video output and nvidia for ML when I want, is impossible. Somehow the NVIDIA drivers literally pushed away and overwritten anything from AMD and I couldn't run games on my AMD, they just kept going to the 5060.

Than came testing on windows with gaming (because I do like to game sometimes). And instead of even offering a slight performance increase over my AMD 6650xt it just was offering the same or even worse. In BeamNG for example, on the vulcan API (With max traffic, and max settings) I got around 40-50fps on the NVIDIA and I was pushing 60fps on the AMD. And yes, BeamNG might not be the best example as its pretty CPU heavy but than still does it have GPU processes it was pushing 100% utilisation in both scenarios. For ETS2 it performed the same.

Im returning it and buying a 9060xt (This time 16GB)

[wide reef]

Updating the blog engine! Now supports bookshelf and book pages with reviews all in the same convenient format and with the same pretty interface

[frosty hemlock]

What's that?

[wide reef]

I’ve got a blogging engine I halfassed for my home lab. That’s a part of my future library system

[frosty hemlock]

Nice nice

Git repo?

[wide reef]

https://github.com/1a11/billard

It’s also on docker hub if you prefer it that way

[frosty hemlock]

Dude thats awesome

I love how you use json that way

Really efficient and small

[wide reef]

I dislike the idea of having a database so it’s a crossbreed between a static page generator and something manageable and structured

[frosty hemlock]

Yeah yeah

[frosty hemlock]

And what about SQLite?

[wide reef]

Oh well uhm at least until you get to the books part, I have to figure out the compression, because right now they’re just base64 encoded. It technically supports normal URLs too so ig you could hook up an S3

[wide reef]

It’s very nice, but I don’t have a feature that needs to rely on it right now

Well, I do but I manage to use a token instead of having a proper auth

[frosty hemlock]

I like it

[wide reef]

Have fun selfhosting, it’s minimal, there’s a client for page uploads and it kinda partially supports logging. If something goes awry make an issue and I’ll try to resolve it asap.

Books are coming in the next week or so if I’d have time

[old arrow]

well somehow i have matlab working on linux at the moment. this is interesting

[frosty hemlock]

Idk if im gonna use it but Im always interested into projects like these

How they work and what tech they use

[wide reef]

Well hey, contributions are always welcome, give it a poke, maybe you’ll find a mistake

[frosty hemlock]

Ill see

[weak brook]

was it not working for like a month? lol

[old arrow]

yeah i gave up for 2-3 weeks whatever it was

i think 2 weeks because i had used it the week before 2 meetings ago

[weak brook]

are you able to install stuff?

[old arrow]

havent tested it but i can open the addon menu

idk what i should test, trying to find if we have code that uses simulink or anything

[weak brook]

make an FFT plot in Simulink

you'll need some toolboxes.

it's audio-adjacent. Impulse response test your room!

[old arrow]

i havent done the simulink training yet so idk how to use any of this
can i like series some ac sources at different frequencies or something and send it through that

hold on just let me look at a getting started guidew

oh

you know i wonder if it would work if i tried just adapting the fixes from the aur package to a manual installation so it's not in a privileged location
i will chown it for now

how long is it supposed to take to install an addon

my dad just saw my screen and was surprised matlab still exists today

https://aur.archlinux.org/packages/matlab#comment-1041965

[old arrow]

well i dont need simulink right now so i'll see what the maintainer responds about that

[wide reef]

Matlab is horible on Linux

just get a VM, you'd save your nerves

[old arrow]

vm seems like an even harder path because I can't seem to get NAT working on qemu
tried the recommended fix of switching it to use iptables (which i have iptables-nat installed), but that doesn't work either
I can't just make a bridge network like you normally would because I need to use it on a laptop, connected to the network via wifi

I also tried wine but the installer just doesn't even start

[frosty hemlock]

Idk if you even wanted it, but I clone the repo and added color functionality

[wide reef]

Let’s see!

[frosty hemlock]

Supporting hex, rgb and verbal colors

Its really simple

And its non required, so you could just remove the color field and it will default to black

[wide reef]

Hey, why not! I’m all for classic tumblresq personalizations lol

[frosty hemlock]

Cool

[wide reef]

Bring it in lol

[frosty hemlock]

I will 😀

[signal ginkgo]

helloworld("print")

[frosty hemlock]

🥀

[wide reef]

Books are now available in the dev branch

[frosty hemlock]

Yeah I saw you just made that branch haha

You want me to make a PR to main or a different branch?

[wide reef]

dev would be cool, I'll see what's up

[frosty hemlock]

Cool

I think github is having issues

And its back

[frosty hemlock]

Made a PR

[wide reef]

Yep. Do review the Copilot suggestions. There's also an issue with a file you've created. I only ship hello, world as an example file, so I won't be able to add your example directly. Consider appending the base one?

[frosty hemlock]

Yes will do

[wide reef]

Might or might not accidentally pushed the dev branch to docker hub but whatever

[frosty hemlock]

Haha

I solved the issues

[frosty hemlock]

@wide reef would this be good?

[wide reef]

Sure, let's keep it that way for now

[frosty hemlock]

Pushed the changes and also added the suggestions

[wide reef]

Merged into dev, now available generally

[frosty hemlock]

Cool!