[00:25] *** MasterDuke joined [00:26] *** librasteve_ left [00:57] *** human-blip left [01:36] *** MasterDuke left [01:59] *** kylese left [02:02] *** kylese joined [02:15] *** kylese left [02:15] *** kylese joined [02:40] *** annamalai left [02:59] *** ShimmerFairy left [03:00] *** ShimmerFairy joined [03:30] *** dustinm` left [03:43] *** dustinm` joined [03:58] *** ntv joined [03:58] *** apogee_ntv left [03:59] *** ntv is now known as apogee_ntv [04:19] *** annamalai joined [04:27] *** Aedil joined [04:47] timo: yeah . I don’t know , maybe I had to clarify , sorry . One of the reason I post my LLM related stuff here is it that Sparrow is written on Raku and it uses Raku regexps as a part of DSL and I find it quite interesting that code generated by some LLMs for this regexp rules is quite accurate, but I general I am agree with you that none Raku relate led stuff should be somewhere else . I don’t mind post this in other place , it’s just I [04:47] don’t have any other audience rather then here ) sorry [04:50] From practical point of view anyone could your code from this snippets and adapt it and run it for their infrastructure compliance checks using sparrow, again I don’t post all the LLM stuff here but the one I find of good quality and maybe the one showing that some LLM are quite good in generating Raku regexps [04:54] Another thought … u guess AI in general and reasonable and wise way to leverage it could be a really a good a vehicle to popularize Raku among others none Raku folks . And please don’t get me wrong I am not a supporter of vibe coding and LLM for the sake of it. But imho this a future we need to embrace or better to say to look forward ))) [04:58] *** dg left [05:04] *** dg joined [06:17] topnep: cool module, but some things may not work reliably - `get-terminal-width` is attempting to be defensive and have timeouts for `tput cols` in a way that will spawn a process and forget about it [06:18] consider simpler `qx/tput cols/.Int` [06:18] Yes, it will hang if tput cols hangs for some reason [06:19] but spawning promises and then killing the process when it's not started will definitely not work [06:19] it might if `tput cols` hangs, but if process doesn't start because of high load in 2 seconds, it eventually might start [06:20] and since it's for ffmpeg, high load is not unlikely [06:21] I'm also not sure how I feel about FFmpegFlags and reimplementing ffmpeg command line parsing, it feels icky [06:31] well, maybe not icky, but brittle, if there's edge case (and there are), iirc old ffmpeg used to accept `-crf:20` for example [06:34] you can't possibly make it handle all ffmpeg versions reliably, because there's been changes [06:38] if you need to add some flag to every stream, consider more lightweight aproach, like asking users to add it [06:55] *** kylese left [06:56] *** kylese joined [07:35] *** kylese left [07:35] fwiw I think there is a latent question “will my LLM be as expert and as efficient in raku as it is in .any” so I encourage the sharing of comparison evidence that they do a good job and of prompts to foster that [07:36] *** kylese joined [07:38] there is (at least for me) the worry that any niche language will suffer from lack of training set “out there” eg the number of SO answers, or general GH repos is relatively low [07:39] bigger threat is GH repos that are llm-made [07:40] so if we have good results let’s post them … but in a side channel so as not to flood the #raku main channel [07:40] yeah an infinite regression to slop beckons [07:41] *** kylese left [07:42] *** kylese joined [07:42] other than that, I don't, in general, believe in llm-generated code [07:43] it's often bug-ridden and simply stupid – the other day I generated a bunch of scripts that literally did `my %opts = :thing(Nil); %opts:delete;` [07:44] oddly (?) I find I am a late adopter of tooling … happy to let others do the trial and error part (odd? because I am an early adopter of Raku) [07:45] I'm hoping that AI in general (not just llm) will keep improving so at some point the code not only works, but doesn't feel painful to look at [07:45] s/works/work/ [07:48] I was late to syntax highlighting, IDEs and now LLMs … but now I am giving it a fair shot (Claude) and it is undeniably a time saver for me (keep a close eye, TDD, and NOT in my important code) [07:49] the worst pain is probably sycophantic behavior – mistakes in prompt integrate falsehoods into code :( [07:49] amen [07:49] either way, llms are good at generating tests, fortunately [07:50] Ok. The fact is I am surprised how good for example deep seek in generating good quality code for sparrow framework which is 1) written on Raku 2) has Raku regexps based API [07:50] so you can test their code with their own tests and see bugs :P [07:51] It’s that good that could be used by devops / sre / mfrs people straight way with little adaption to their custom needs [07:51] yeah … test generation solves a pain and often LLM test covers something I have not thought of (wood for trees) [07:51] I've seen the validators for configs generated by deepseek, they seemed fairly complete [07:51] And if it gets some traction will result in promoting Raku in general [07:52] (the openssh iirc) [07:52] Think about Ansible - Python , sort of example [07:53] mfrs -> infra [07:54] imo python is go-to language for people who don't do programming, kinda like improved basic [07:54] Also posting anything Raku related ( maybe not directly ) in channel rather then #raku honestly does not make a sense for me. Too few people we even have here in this channel so to fracture stuff into other channels [07:55] I don't think raku can compete with python on that field [07:55] The example of Ansible Python is a it or people first t knew Ansible and then Python through that [07:58] My belief is that none Raku ( ops, devops ) people will starry using Sparrow ( with almost minimal Raku impact at first ) and then will discover Raku through the tool ( which has SDK for many languages including of course Raku ) [07:59] It can . I don’t see why. This exactly what’s Sparrow framework for, specially targeted for none dev people [08:00] I don’t see why -> I don’t see why it’s not [08:00] eh, most of ansible code is playbooks, you can do quite a lot without python [08:00] I think that DSLs (like Sparrow) are a secret weapon for LLM effectiveness [08:01] Exactly [08:02] since they are human readable, domain centric and constrain the training set and the LLM output [08:02] This was my eye opener , cause it’s so much easier to generate DSL via LLM [08:02] DSLs are also quite limited (not for general programming) which is also a great thing [08:02] Yep [08:03] And sparrow task check dsl by itself is very simple [08:05] @melezhik. let’s get an example sparrow on https://raku.org front page … can you take a look at the other examples and note the typical length and then provide me your best proposal (code and words) via a gist? [08:07] a PR https://github.com/Raku/raku.org here would be even better [08:13] @librasteve good idea , will take a look , I think I can adjust some of my LLM generated examples, and btw you can try to generate something by the prompt example I use [08:13] deep seek is the best in that way [08:14] Of course the final code still need to review but it’s pretty accurate [08:14] thanks! &afk for a bit [08:16] Sure [08:17] For anyone who cares - this is a prompt I use https://gist.github.com/melezhik/d57132c9d3ba0ae9cee30a7c04c98399 [08:18] You can feed it into deep seek expert level , you can alter prompt to generate generator: and code: blocks on Raku instead of Python , it should work [08:19] https://gist.github.com/melezhik/d57132c9d3ba0ae9cee30a7c04c98399#file-gistfile1-txt-L54 [08:19] line 54 [08:20] You can also just replace !python by !raku in the prompt [08:20] I use deep seek free chat bot in browser [08:20] It’s impressive [08:22] Line https://gist.github.com/melezhik/d57132c9d3ba0ae9cee30a7c04c98399#file-gistfile1-txt-L51 and https://gist.github.com/melezhik/d57132c9d3ba0ae9cee30a7c04c98399#file-gistfile1-txt-L56 need also to be changed to Raku - it’s easy to guess how [08:23] In https://gist.github.com/melezhik/d57132c9d3ba0ae9cee30a7c04c98399#file-gistfile1-txt-L7 replace sshd by any word / tool you like and check it out ! )) [08:23] It’s fun 😊 [08:37] Here is Raku adapted version of Sparrow prompt for deep seek https://gist.github.com/melezhik/79fa653739458729013e0b9b8fa7f570 [08:42] *** apogee_ntv left [08:43] *** apogee_ntv joined [08:48] *** Sgeo left [09:09] *** itaipu left [11:43] *** shareable6 left [11:44] *** shareable6 joined