My (very) fast zero-allocation webserver using OxCaml
54 points by asb
54 points by asb
I had only had a quick look for now. The "Claude skills" look interesting but I don't care about Claude, especially on advanced topics such as the ones in OxCaml. They look interesting for humans to read but it's a bit sad that they're written for ingeestion by an AI system more than understanding by humans. I seems like only a tiny bit of additional prose, explanations and links between sections would make it a great resource for learning and practicing.
I'm not keen on doing tutorials intended for humans from AI generated code; it's got a different emphasis. The handwritten OxCaml tutorial is an example of something for human consumption, and then Jon Ludlam compiled the oxcaml compiler to Javascript so you can interactively learn: https://jon.ludl.am/experiments/oxcaml-tutorial/
That reminds me, I must update that to the newest tutorial - and then upstream the patches to x-ocaml that make it all work. I think there's a lot of potential there to make a really nice interactive tutorial system.
I already have some knowledge of OxCaml and these tutorials are good but they're learning material rather than day-to-day references or manuals. What I meant (but I feel like this wasn't completely clear) is that the claude skills look like they are fairly close to what would be good day-to-day references. For that, they only require a few sentences here and there.
The feeling behind my comment was that you had written something great that could benefit many people but had meant it for machines instead. I'm not worried it won't happen and I'm not blaming either. It's just a new/weird feeling that an author favored teaching machines. And again, you did it, you published it you and linked to it, while having no obligation to do any of that. That's more than what I have done for my experiments in less unhealthy cookie-like pastries[1]!
PS: somewhat unrelated: it's a bit sad that slipshow's output has no relation to the browser scrolling features: if it extended the viewport as the slipshow runs, it would be possible to quickly scrollback to previous views and use the browser's search.
[1] btw, you can't have healthy cookies: the moist texture we like so much comes from insane amounts of sugar and large amounts of fat
I'm curious what your opinions are about the inter-version changelog from OxCaml minus23 to 25 (they do drops every month or so). I use these to figure out when to migrate code or find a new feature.
Regarding Slipshow @pangled is very receptive to feedback at https://github.com/panglesd/slipshow -- I'm just a happy user!
Funny, I actually found the LLM documentation of modes more readable than the one for humans (though the LLM docs seem to be missing some of the modes like "visibility" and "future").
It probably has to do with being familiar with the concepts involved (which makes the docs for humans too verbose for me), but still..
Oh dear, both @osa1 and @adrien have convinced me to have a go at an LLM-driven OxCaml tutorial next :-) One interesting property of those Claude skills is that the agent tests the correct compilation of those for every compiler release. With @jonludlam's Javascript output, it's quite a low latency way to push out tutorial updates as long as it's clearly flagged as being AI generated.