I am also suspicious. The way the topic shifts seemingly at random is very common with GPT generated text. It has several correct technical details but they don't support a central thesis, another telltale sign of a GAN in use.
I stopped reading when they said its architecture should be more like Photoshop. This made me actually LOL.
Has the author ever tried coding a photoshop plugin? Or read a PSD file from scratch, its a notoriously poor interface, bogged by bad legacy decisions and duct tape and string quick-fixes. The PSD file format is a meme and prime example of garbage programming.
LLVM does use a plugin architecture, that’s what the whole IR system is.
Comparing LLVMs architecture to needing to have plugins “Like Photoshop” is like saying the Starship Enterprise should have more rope because it can’t pull this donkey cart.
I read it, and is not that bad, in fact the points it make are probably as close as non-controversial in how make a fast compiler.
And LLVM is certainly major slow in the compiling pipeline. And very complex at it.
What is unclear is how design "a photoshop-like plugin architecture" that allow to do all the optimizations it have, ie: How much the complexity of LLVM is related to the need to intermix all the optimizations and IF is possible to modularize them to the extend to make a lean core?
(A better way to say the same is that it must be. like the nano pass framework, that it have more pedigree in this case than use of photoshop :;)
I read this twice and I'm not really sure what's being asserted here. I got the impression there's conflation here between LLVM the project (which indeed is quite large) and what LLVM actually does when it generates code.
If someone's going to whip out ancient slang like 'sweathog', at least use it right. It's either an overpaid/underworked caricature of a union worker used in anti-union propaganda, or it's the gang from Welcome Back Kotter. That's pretty much it.
The headline may or may not be true, but this essay is a bizarre mishmash of superficialities and non sequiturs, with a dash of "anyone could rewrite it better" thrown in.
Hmm, I guess that's possible, but it seems a little strange given this website is attached to a GitHub repo for what seems to be a real (albeit also weird) programming language. The langauge pitch page was posted earlier: https://news.ycombinator.com/item?id=27287989
19 comments
[ 2.9 ms ] story [ 54.1 ms ] threadHas the author ever tried coding a photoshop plugin? Or read a PSD file from scratch, its a notoriously poor interface, bogged by bad legacy decisions and duct tape and string quick-fixes. The PSD file format is a meme and prime example of garbage programming.
LLVM does use a plugin architecture, that’s what the whole IR system is.
Comparing LLVMs architecture to needing to have plugins “Like Photoshop” is like saying the Starship Enterprise should have more rope because it can’t pull this donkey cart.
And LLVM is certainly major slow in the compiling pipeline. And very complex at it.
What is unclear is how design "a photoshop-like plugin architecture" that allow to do all the optimizations it have, ie: How much the complexity of LLVM is related to the need to intermix all the optimizations and IF is possible to modularize them to the extend to make a lean core?
(A better way to say the same is that it must be. like the nano pass framework, that it have more pedigree in this case than use of photoshop :;)
Rants about LLVM to then give a completely unrelated story about cache?
Yes, I signed up just for this.
Correction: also a band from 1971.
Actually arranging ideas as an argument and adding fresh thought will be a stickier wicket.