I don’t think an LLM can’t actually do this if there’s not tons of training data out there using this bytecode. I think the best you could do is have it define some documents on how to use this bytecode, have it spit that into a document, and then feed that document back into the LLM anytime you want it to read or write this bytecode. But this would also be in human-readable language, giving us a way to understand this bytecode.
Besides, if only the LLM understands it, what would execute it?
Ha, brutal but true.
How would you implement the backend for API like this with std.Io and without using io.async (which would waste a thread/coroutine that JS runtimes do not need).
You are forgetting that JS has longer history than async/await.
This discussion on its own is ample demonstration of the reality that some projects, maybe a lot of projects, were going to be lost in the transition to std.Io. It’s a big price for the ecosystem to pay, and it won’t be clear until some time from now whether it was worthwhile.
Not sure I agree with that as, here in the UK, the newish **bank** I am with were doing all that with no more or less VC cash and in a much more regulatory challenging environment
I’m not convinced that a port from 0.14->0.16 is more difficult than a port of 0.14->Rust/another language. It’s just advertising and a move to a language more compatible with “their” “coding” “style”.
In order not to get lost, I was forced to rewrite the POSIX-based network layer to use usockets
Io may be the right direction, but many projects still need a low-level, OS-independent layer.
Io, as a good citizen, should also be a client of this layer
It was a solo endeavor back when it was a private beta (and you had to type ‘I want bun’ in a dedicated Discord channel to get access to the repo).
Right after Bun was first published, Jarred got VC money and founded a startup. He then acted as CEO and hired a bunch of people to work on Bun.
Actually this post and this one raised “Io and low-level programming” issue - wider than Bun
I tried in the past - without success
May be someone more brave or moderators will open new thread?
I agree that what Bun is doing is an unusual “nuclear option” that depends on their clanker-friendliness and effectively limitless resources.
I was thinking more of projects that might just be abandoned, or stay behind on an earlier version of Zig, or use Zig as a language and compiler while doing without the standard library.
Ah I stand corrected, thanks for the clarification.
Company gets bought by “THE” PAP (Prompt and Pray) company. So they put their code base through the slop machine and adopt the pap model. One of the things I like about zig is its weak in LLM so you have to understand what your doing . We have add to “you will own nothing and be happy” to include “you will know nothing and be grateful “ lol
I don’t use Claude. 
Technically that’s not 100% true, sometimes when finishing major work I will use a claude review skill from Codex to get another set of “eyes” on the diff. Typically something like “$claude-review and tell me your thoughts”. Gotta use use those $20/mo claude tokens somehow…
Please participate in new thread New Io and low-level programming
I reached this conclusion about 2 years ago with regards to how Bun worked vs what was marketed/demoed. But same can be said of most VC funded teams or hype-driven orgs.
Hey ! Nice to see you here Peter, small world indeed.
Good work on that Odin / Tina lib and Datastar SDK too.
I thought I recognized the name. The Tina repo is really cool; I spent a couple hours playing with it last month.
Thank you!
How far did you go with? I’d appreciate feedback but I’d be happy to help explain or point you to the right API if you decide to explore more with it.
It ended up doubling as an “explore Odin” side quest for me, so I can’t say that I got very far. The examples directory was helpful. It would also be cool to see other real FOSS projects that are using tina.