Hacker News Reader

Comment by mdaniel | Hacker News Reader

I suspect this was prematurely published to HN and was in fact just someone's weekend project

https://github.com/Foreseerr/TScale/blob/aa2638c53c74dd33280...

and I struggle to think of what would lead one to the urge to implement a key=value config file parser in 2025 https://github.com/Foreseerr/TScale/blob/aa2638c53c74dd33280...

On top of that, people who do $(git add . && git commit -myolo) drive me crazy https://github.com/Foreseerr/TScale/blob/main/logs/125m_1T_f...

View on HN

> and I struggle to think of what would lead one to the urge to implement a key=value config file parser in 2025

C/C++ culture never changes.

As many new build tools and package managers as people come up with, the ‘default’ environment is still one where adding dependencies is hard, so people roll their own utilities instead.

CMake makes it a lot easier. couple that with conda and it's pretty good.

I'm coming from a Java/python background originally and compared to that it's more finicky but not bad at all.

I can only speak for myself but I don't think you've got the cause and effect right. Dependences tend to have their own dependencies (which have ...). It's not so much the difficulty as it is the awareness of it that leads me to minimize my dependencies to the bare minimum.

All my dependencies are locally cloned. I build them from source in a network isolated environment. And yeah, that makes it more expensive to bring new ones in so I tend to shy away from it. I see that as a good thing.

That said, if you're willing to give cmake access to the network things largely just work as long as you don't attempt anything too exotic compared to the original authors. For that matter boost already has a decent solution for pretty much anything and is available from your distros repos. Rolling your own is very much a cultural past time as opposed to a technical necessity.

> and I struggle to think of what would lead one to the urge to implement a key=value config file parser in 2025

That could be a symptom of LLM coding. I have found at times they will go down a rabbit hole of coding up a complicated solution to something when I know that a library already exists it could’ve used. I’m sure part of the problem is that it isn’t able to search for libraries to solve problems, so if its training data didn’t use a particular library, it will not be able to use it.

Reply on news.ycombinator.com