Aurelle
285 posts

Aurelle
@aurelienb42
SWE | Building an open source alternative to Raycast for Linux: https://t.co/Fp6nNn4WBQ
Paris Joined Mayıs 2021
94 Following133 Followers

@ZPostFacto it started for mostly json. The big highlight of this one is speed and it also supports reflection.
English

@aurelienb42 I suppose it you wanted to support every known file format, that library would be useful!
English

Parsing JSON in C++ has been a solved problem for a long time, and yet I couldnt find a library that I totally liked. So I'm throwing my hat into the ring with vjson!
See the README for why I reinvented this wheel, and why you might like it.
github.com/zpostfacto/vjs…
English

@ryanrhughes yeah those are beasts. Right now I'm not really leveraging the lanes, I have a single 7900xtx on this rig.
English

@aurelienb42 🤤 I need this in my life. I've been eyeing a 9975WX for my AI rig. Only has 1xRTX6000 today but if I toss a few more in there, I'll need more lanes to work with.
English

In a chase for performance I forgot why I picked rust for making fff.
If you write safe idiomatic rust you can really not think about races, deadlocks, and unsafe concurrency. Once you start juggling pointers everywhere segfaults are there. And if not yours, someone else's AI will find them.
English

@leiserfg @badlogicgames aren't those just notifications? no id is legal if it's only one way
English

@badlogicgames I had to implement an mcp server, and found out that anthropic sends some jsonrpc calls without id, worst part is their schemas for mcp have the field as marked as required.
English

@e907700898427e @0xWARP7 @MMatt14 well the enshittification is real, but what makes me lose my mind is stuff that doesn't work or turns my computer into an airplane more than something eating a bit too much memory
English

@aurelienb42 @0xWARP7 @MMatt14 I'm just generally noticing a trend in software nowadays where optimization isn't a big priority, in games for example which kinda makes me sad
English

low iq linux users don't understand how good this design is
an actually functional package system doesn't need 10 billion different package managers.
ThePrimeagen@ThePrimeagen
Wait... It's 2026 and y'all are still dragging icons to install on Mac?
English

@HSVSphere I think that same guy was also bragging about vibe coding his own DE not so long ago so the distro may not be the only issue
English

That's on you, you chose a mutable skid distro, in this case Omarchy.
Use Bazzite KDE, a distro that won't explode on you. Then stop ranting about how Linux is unreliable just because you chose the ticking time bomb.
Greg Schier 👨🏼💻🇨🇦@GregorySchier
Ah, I'm remembering why I switched to Mac ~10 years ago. I just rebooted my Framework/Omarchy setup and now the screen is just... black. I'm sure I could debug/fix it but I have work to do.
English

@e907700898427e @0xWARP7 @MMatt14 no direct correlation I mean. of course lack of optimization can yield to unnecessary RAM usage, it's just not the ultimate proof of it
English

@e907700898427e @0xWARP7 @MMatt14 you are right but there is also no correlation between high RAM usage and poorly optimized software. Modern toolkits will cache a LOT (QT/QML as an example) if they think they can. I'm just so tired of linux people looking at btop to claim whether something is "bloated" or not
English

@e907700898427e @0xWARP7 @MMatt14 optimized software uses RAM where it can to avoid using too much CPU :) Many programs consuming low RAM are in fact poorly optimized
English

@neogoose_btw yeah of course they make change every chip generation but overall they can make a lot more optimizations as the range of supported hardware is narrower than your typical linux/windows computer
English

@aurelienb42 that's not true - example is aarch64 dotprod and some specific instructions that only available starting from M3 version, when inlined those doesn't require an indirect function call in a hot loop
English

You probably have no idea what your CPU is capable of. Here is the same x86 2 core CPU on my vps running fff-demo. One compiled with -march=native, the other one compiled with simple cargo build.
Difference is 83x because Zen 3 architecture is very specific and by utilizing better instruction & order you actual cut of a giant portion of work.


English

@neogoose_btw to be fair since macos targets very specific hardware you probably get all these benefits if using a mac already (I mean for the stuff coming with the OS)
also I will remain a proud gentoo user, no nix for me :D
English

@aurelienb42 yeah seems like I am slowly getting the point of nix everywhere
e.g. the other thing no one compiles but the one that improves the performance dramatically is ffmpeg & libx265
when compiled with -march=native you can hit up to 4x of encoding speed out of nothing
English








