Hacker Newsnew | past | comments | ask | show | jobs | submit | ra1231963's commentslogin

Based on all the negative comments, and being a non-expert, I had an LLM generate an article for that same title. I suspect it was more approachable and more technically accurate than this. In fact I asked it to compare its article with this one and the negative comments.

The author of any technical blog post should be using an LLM to vet things before posting to find gaps or inconsistencies.


I believe with a little (okay a lot of) motivation and a solid LLM, these can all be rewritten faster than everyone says. Especially if it’s gradual.

Mozilla and Dropbox did it. LLMs are good at translating between languages, and writing unit tests to make sure things still work the same.


How much Rust is actually in Firefox these days? The last numbers I see from 3 years ago is 10%.


> https://4e6.github.io/firefox-lang-stats/

12%. Assume the progress is linear (not logarithmic like most cases), we just need 60 more years to migrate those c/c++ code.


That was my point — with LLMs the progress would not be at the same slope as with people only.


Have there been any successful attempts yet of translating 'idiomatic' C++ to 'idiomatic' Rust for a large codebase that has been developed over 30 years? What does the output look like? Does the code look maintainable (because mechanical solutions to translate from other languages into Rust exist, the result is just not what a human would write or ever want to work on). Are the prompts to guide the LLM shorter than the actual code base to be translated? Etc etc... eg the idea to use LLMs for migration is quite 'obvious', but the devil must be in the details, otherwise everybody would be doing it by now.


Here are a few

https://www.phoronix.com/news/Google-Linux-Binder-In-Rust

https://arxiv.org/abs/2503.23791v1

https://www.darpa.mil/research/programs/translating-all-c-to...

https://link.springer.com/content/pdf/10.1007/s10664-024-105...

> Everybody would be doing it by now

Models and agents have progressed significantly in the last few months. Migrating projects to rust can definitely be a thing in the coming years if there is sufficient motivation. But oftentimes c/c++ devs have aversions to the rust language itself, so the biggest challenge can be an issue of motivation in general.


Idiomatic C++ allows too much "freestyle" and duck-typed code. Rust language basically doesn't support half of the things that C++ allows because there is no type or memory safe ways to achieve them. When translating things into Rust, borrow checker forces you to invert the logic or completely redo the architecture. Oftentimes it requires trying multiple things and evaluating the performance cost, generated machine code quality and interface flexibility. It is quite difficult without an actual person to translate things.


Yeah that's why I'm very sceptical about any 'LLM will magically fix all the things' claims. The mechanical translations basically work by compiling the source language down to LLVM bitcode and then translating that back to another language, it works but is basically assembly code written in a highlevel language, it loses all 'semantics' and is entirely unmaintainable.


> starting with (now Google-dead) "Material Design"

I thought it was still actively developed. Are you saying this has been killed by Google?

https://m3.material.io/


No Material 3 (Material You) is the design that stock android uses for all its UI components, and more and more third party apps start using it.

Though I haven't seen Material Design Website in a while.


At least the web components are in maintenance mode[0]. Google seems to focus on a new framework called Wiz now.

[0]: https://github.com/material-components/material-web/discussi...


Unrelated to Wiz, the cloud security company Google tried to merge with. [0]

https://news.ycombinator.com/item?id=41042034


I seem to recall an announcement a while back about it having been sunset.-


> and preferably general LSP support so I can code in my own programming languages with it

I’m curious, do you really use your own languages for day to day development?


About 70% of the programming I'm doing right now is in my own language. But like, my full time work is programming language research so I accept that's an uncommon use-case.


It doesn’t follow that a language server written with JavaScript and run via node will bloat the binary by hundreds of MB. Are they bundling a node runtime too?

Maybe if they are embedding dozens of language servers and runtimes it could bloat the binary, but I assumed the extensions and language servers would be downloaded on demand.

But a rust binary by itself shouldn’t be that large. LSP is just a simple json protocol, so parsing it doesn’t require hundreds of MB.


When I opened a Rust project I think the status bar said it was downloading Rust support, which supports your hypothesis. (Also it doesn't make much sense to bundle these things when they are all making new releases at different schedules.)


Hey, Zed developer here. Indeed we do not bundle the LSP binaries into the final binary for the reasons you've stated; and I do agree that the binary is kind of big, though at present .dmg compression gets us a long way (as the .dmg itself is ~115Mb). Right now we ship an universal binary, so half of that size is essentially unused:

  size /Applications/Zed.app/Contents/MacOS/zed
  __TEXT __DATA __OBJC others dec hex
  120979456 475136 0 4336828416 4458283008 109bc0000 zed (for 
  architecture x86_64)
  117587968 458752 0 4336680960 4454727680 10985c000 zed (for 
  architecture arm64)
Then, each of these binaries includes about 40MB of assets. I've actually had a PR up (https://github.com/zed-industries/zed/pull/3997) that reduced their size quite significantly, though that did not end up reducing the size of a .dmg itself, so I've scraped that. On top of that, we ship with debug symbols for symbolication of crashes (https://github.com/zed-industries/zed/blob/main/Cargo.toml#L...).


It seems inevitable we’ll also be able to run LLMs locally, which would make this type of feature more appealing.


Pdfml is a great name because everytime I work with pdfs, I think fml


This come off so… un-human. Like a machine trying to preserve a human rather than a human trying to maximize real experiences with them.


Nah, you just have a bias along the lines of "if it weren't done before, it is unhuman".


So it’s competing with vercel?


Never learned awk or committed esoteric cli incantations to memory. Don’t get me wrong, I can get around on the cli, but sed, awk, etc just didn’t seem like a good cost/benefit investment. I’m also not a sysadmin.

Thankfully I waited long enough and LLMs can write them for me better than I ever could.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: