|
01:11
hulk joined
01:12
kylese left
01:18
japhb left
01:22
japhb joined
|
|||
| apogee_ntv | timo: I will try to repro on Linux. | 01:50 | |
| I have a Xeon box that I mostly use for running local LLMs, shouldnt be an issue to test Selkie stuff on it. | 01:51 | ||
| The problem is its a bit of a heisenbug, I've hit it twice in 3 weeks. | |||
| And I was doing different things both times | 01:52 | ||
|
02:14
kst joined
02:15
hulk left
02:16
kylese joined
|
|||
| ugexe | huh, i guess i have another class of rakudo ast test failures: running the test with RAKUDO_RAKUAST=1 with a rakudo that was also built with RAKUDO_RAKUAST=1 (instead of just the former) | 02:19 | |
| for example a rakudo built with rakudo_rakuast=1 doesn't have this !INIT_VALUE symbol and thus fails this (and the other corekeys) test github.com/ugexe/rakudo/blob/4fb9c...ys-6c.t#L9 | 02:23 | ||
|
03:13
annamalai left
03:14
annamalai joined
03:26
vasko4535586 left
03:47
vasko4535586 joined
05:14
Aedil left
05:45
Aedil joined
06:15
kurahaupo left
06:28
kurahaupo joined
06:33
kylese left
06:37
kylese joined
06:44
sacomo left
06:58
Sgeo left
08:04
DarthGandalf left
08:06
DarthGandalf joined
08:37
Aedil left
|
|||
| disbot2 | <melezhik.> o/ | 08:46 | |
| <librasteve> hi | 08:55 | ||
|
09:00
cpli left
09:13
cpli joined
09:32
human-blip left
10:09
human-blip joined
|
|||
| disbot2 | <librasteve> just noticed this from the perl community Have folks been reading this? red.anthropic.com/2026/mythos-preview/ It seems like we may be seeing a big wave of new security reports soon for all software, and not just open source. I’m wondering how well CPAN will be able to react. As shown by them finding 27 year old bugs, it’s no longer true that long-lived, heavily used and reviewed code (most of CPAN) can be considered in a safe in a | 10:32 | |
| stable state. | |||
| apogee_ntv | Probably badly, I imagine a lot of CPAN authors haven't written perl for years. | 10:38 | |
| Though tbh Mythos's capabilities are massively overstated. | |||
| The bug in openbsd for example is completely unexploitable in practice unless you turn all the layered security off. | 10:39 | ||
| disbot2 | <melezhik.> Sparrow6 logrotate configuration parser created by Qwen - gist.github.com/melezhik/ebf795b57...7004b3f726 | ||
| <melezhik.> It’s pretty impressive | |||
| ab5tract | weekly: lobste.rs/c/zkpqfl | 11:29 | |
| notable6 | ab5tract, Noted! (weekly) | ||
| ab5tract | ugexe: I think some of that is because RakuAST is much more intentional about what actually should be a core key | 11:32 | |
| But it does seem a bit surprising that it only surfaces on a rakudo compiled by RakuAST code paths | |||
| disbot2 | <librasteve> Yeah, I am just wary that some potential in house Raku project falls foul of a poor Mythos report - hoping that our eco-system is still manageable enough to come out faily clean wrt eg cpan | 11:36 | |
|
13:05
vrurg_ joined,
vrurg left
|
|||
| apogee_ntv | librasteve: If we want to be sure we should point Kimi K2.6 at the major ecosystem projects but both it and Mythos will pick up false positives. | 13:39 | |
| disbot2 | <librasteve> makes sense to me - I would like to see what others think | 13:40 | |
| <antononcube> How about starting with a simple human-programmed in Raku audit? 1. Get latests version of all packages. 2. How many packages have READMEs? 3. How many packages have complete META6.json files? 4. How many packages have tests (bot "./t" and "./xt")? 5. What are the coverages of those tests? | 13:50 | ||
| <antononcube> Implementing the answer-finding for those questions should be fairly easy in Raku. | 13:51 | ||
| timo | how about making it opt-in to receive automated LLM agent based reviews of modules in the ecosystem | 13:53 | |
| disbot2 | <antononcube> I will add/implement a prompt for that in "LLM::Prompts" and corresponding agent "LLM::Resources". | 13:56 | |
| <antononcube> "LLM::Resoruces" has a comprehensive text summarization agent (with a corresponding CLI) the architecture of which is very close to the package auditor agent. | 13:59 | ||
| timo | do remember that we cannot have automated ecosystem review by LLM agents until we have a low latency connection to github APIs, as well as an automatically scaling GPU compute kubernetes cluster and a portion of our ops team to monitor it 24/7 | 14:01 | |
| disbot2 | <librasteve> we have an ops team? | 14:03 | |
| <antononcube> Well, there can be a service for that agent, but people should be able to apply locally. The Zef ecosystem can run that automation, also, right? | |||
| timo | you don't recall the plan that an expert laid out for us for implementing such a thing? | ||
| [Coke] | raku does have an ops team, yes. | 14:06 | |
| disbot2 | <antononcube> TBH, I am more interested in the deterministic overall statistics over the latest versions of the ecosystem packages. | ||
| timo | github.com/Raku/problem-solving/is...3015267225 | ||
| disbot2 | <aruniecrisps> Just wanted to ask anyone if they've used github.com/raku-community-modules/Badger | 14:14 | |
| <aruniecrisps> I'm thinking of pivoting from Red to this cause I want to go all in on functions | |||
| <aruniecrisps> And reduce ORM usage | |||
| <aruniecrisps> Basically to mimic F#, Clojures or Gleams approach to storing values in a database | 14:15 | ||
| SmokeMachine | :( | ||
| ugexe | i'm a heavy ai user and the automated gemini scans we get on PRs at work are almost always noise | 14:16 | |
| disbot2 | <aruniecrisps> Haha I'm sorry @SmokeMachine, Red is a fantastic project, I just feel more comfortable dealing with SQL directly | 14:17 | |
| <aruniecrisps> I've been doing migrations and everything in SQL itself | |||
| SmokeMachine | :'( | ||
| disbot2 | <librasteve> i prefer Red … timtowdi | ||
| ugexe | automated scans are almost always using fast non state of the art models. nor do they take a holisitic view of the code, attempt to poke and debug, etc | ||
| disbot2 | <aruniecrisps> @librasteve I love Red don't get me wrong | 14:18 | |
| ugexe | for example when i pre review code at work using claude, claude almost always dismisses all the comment gemini left on the PR | ||
| disbot2 | <librasteve> i “love” SQL … but i prefer Red | ||
| <librasteve> but I would not use an ORM everywhere … | 14:19 | ||
| <aruniecrisps> Yea, I think for my use case I don't necessarily need an ORM, I think I need tighter control over the SQL that I'm writing | |||
| ugexe | i suppose they can be sort of useful for enforcing certain standards / policy, e.g. we prefer to use pattern X or tool Y | 14:20 | |
|
14:20
sibl joined
14:21
avuserow left
|
|||
| disbot2 | <aruniecrisps> Let's see, if I feel like Badger doesn't fit my use case I'll stick with Red | 14:22 | |
| <librasteve> =b | |||
| SmokeMachine | Please, I'm just kidding! You should use whatever you feel comfortable with! sorry for the joke | 14:23 | |
|
14:58
sibl left
|
|||
| disbot2 | <librasteve> SmokeMachine: just a nudge - do you still have adding Cromponent route authentication on your backlog? Now that stubs is done, I am hoping to go all in on CRUD quite soon. | 14:59 | |
| SmokeMachine | librasteve_: yes, that's still on my backlog, but I'm having so much fun playing with Selkie::UI that I'm not sure how long it will take to stop playing with that... | 15:01 | |
|
15:03
annamalai left
15:04
annamalai joined
15:13
avuserow joined
|
|||
| librasteve_ | ok cool … don’t forget! | 15:26 | |
|
16:16
Aedil joined
16:30
annamalai left
16:34
swaggboi joined
16:35
annamalai joined,
annamalai left,
annamalai joined
|
|||
| disbot2 | <antononcube> Can this be solved with agentic workflows? Probably there is some research on that... | 16:48 | |
|
17:11
kylese left,
kylese joined
|
|||