clice v1.0 Released: Welcome to Modern C++

22 points by jeaye a day ago on lobsters | 11 comments

The part about using all the modern C++ and being a standard for best practices reads to me like: “we asked AI to use all newer features” - kinda like we can ask AI to build a compiler that implements any number of research ideas and type systems.

This is a turn off for me. I simply can’t trust a project like this on the long run — id wait for it to mature for a few years…

[OP] jeaye | 5 hours ago

I agree to wait for more maturity, but I don't want to bash greenfield projects for trying to start with the most modern features. Adding C++20 modules to an existing project, especially as it has become quite large, isn't particularly easy. The same goes for turning a synchronous project asynchronous. I think starting with C++ coroutines is an important design decision which affects overall responsiveness. I'm happy to see projects pushing these features forward, since all of us other devs stuck in our older, larger code bases are going to take a lot longer to get there. So who else would use them?

It’s mostly a matter of being able to handle the flood of new software. Of being able to trust the longevity of vibe coded projects.

BenjaminRi | 9 hours ago

To me, the statement "we use the newest C++ features available" translates to "we made our code maximally incompatible with all the existing systems". As an embedded developer, the toolchains I have to use are often years or even decades old. I maximize compatibility by selecting the lowest possible C++ standard (greater than or equal C++11) that allows me to write robust and elegant code to solve the task at hand.

jonpalmisc | a day ago

Commit history & messages suggest heavy use of LLMs...

jo3_l | 23 hours ago

This is true and the author indicates as such in the post. Nevertheless the post itself seems mostly human-written and contains some interesting technical content + what appears to be a pretty genuine reflection at the end. I also think there's a real gap that this project can potentially fill: clangd really does not handle modules well last I checked.

All that to say: I think it's worth reading the post despite the code mostly being LLM-authored.

[OP] jeaye | 23 hours ago

This is how I felt reading the post as well. It's clearly hand-written and when the author gets to the point of talking about agentic coding, the prose becomes melancholic and boderline nihilistic in a way which I think is very human.

On a technical level, we all know that clangd has some serious usability issues. Exploring a different way of handling LSP for C++ is intriguing.

clangd really does not handle modules well last I checked.

I've been out of the loop with C++ for a while, but the last I looked, modules seemed like they weren't handled well in general; it felt a bit like export template in that sense.

[OP] jeaye | 5 hours ago

Yeah, a bit like that, but with more support than export template. Modules solve a real problem in how C++ compiles (and compiles and compiles), but the path to modularizing code is not easy. The most exciting thing, for me, about clice is actually its automated modularizing. I haven't tried it yet, but that's exactly the sort of tooling we should want in the C++ space. If compiler tools can automatically migrate us to the latest features, we should be able to easily check if those features buy us anything. That's a huge win.

BenjaminRi | 14 hours ago

The architectural decisions outlined in the blog do not look like LLM slop. I don't mind if someone who knows what they're doing designs the architecture and lets AI do the grunt work. I haven't studied the project in depth though.

codekobold | 12 hours ago

There is absolutely a way to do good software engineering while delegating a lot of the coding to agents. It's new, we don't necessarily have widespread best practices around it and it's hard to distinguish between AI slop and well-thought AI-aided software engineering: they look very similar superficially.

Especially given the insight into the developer's thinking from this blog post, however, I would not jump to shouting "slop!" just because code is not handcrafted.

kneeawn | a day ago

Comment removed by its author