00:38:08 oh, hmm, https://illumos.org/man/ isn't up to date with the gate I see - changes from https://github.com/illumos/illumos-gate/commit/bc19178a2736ca6e1e752c880b01f5d86b270c75 don't seem to show there 00:52:29 I think jclulow runs a script to update that 00:53:48 if we ever get everything converted to mandoc, one really nice thing is that it outputs html as an option and can use a common css amongst all of the pages 06:26:14 LLM should be useful for such conversion. 06:37:17 would seem wildly inappropriate for the task (converting a presentational to a semantic markup language) 07:18:22 you can check the results for ls.1: https://code.illumos.org/c/illumos-gate/+/4817 07:18:23 → CODE REVIEW 4817: convert ls.1 to mdoc(7) Assisted-By: Claude Opus 4.8 (1M context) (NEW) 07:27:10 it did spend 14 minutes on it (with confirmation prompts). 09:30:23 jbk we can do that now - mandoc can output html for the older format too, and that's probably what's used for illumos.org/man. The OmniOS pages - https://man.omnios.org/ls for example - include all the extra userland that OmniOS ships too, and has a stylesheet that could probably be improved. 12:14:15 oh, i figured it jsut passed through non-mandoc pages 13:48:41 There is also https://github.com/omniosorg/tomdoc which is what I use to get started with mdoc conversion. It takes care of some of the mechanical bits. You'd have to carefully read through an LLM-assisted output anyway, so I find it useful to do the mechanical bits then read through top to bottom tidying it up and filling in things like the synopsis by hand. 13:58:17 i wrote a c program to assist a while ago... though you have to give it some input on certain things 13:58:33 but it does try to notice certain patterns and translate them into lists (for example) 15:13:32 heh.. why am I not surprised that vmware's nvme emulation is terrible 16:46:51 Yep... I see proof every SmartOS release with test-flake. 19:34:37 proof-reading any bulk conversion of the manual sounds like it would be hell 20:04:45 you'll save power and headaches doing it andyf's way 20:09:44 given the state of our manuals (or at least, the decade of effort alanc has spent improving the same manuals, which we haven't), you're probably going to find/fix bugs too 20:10:49 ptribble has done quite a bit on your side 20:11:27 but yes, there is so much more that could be fixed 20:18:05 I wonder if you asked an LLM to report all the outdated bits in the man pages what it would say - if it would find stuff like the "workman" reference I removed from rmmount recently 20:20:35 Another one is the use of IPv6 site-local addresses in man page examples -- they were deprecated years ago. 20:22:00 alanc: at this point, "whenever a music CD" is not that much less antique 20:22:12 true 20:22:47 we removed most mentions of diskettes when we dropped our driver for floppy disk drives 20:23:44 from the notes I sent Peter comparing my changes to his: 20:23:46 https://github.com/illumos/illumos-gate/blob/master/usr/src/man/man1/rmformat.1#L524-L527 20:23:46 - listing a bunch of media types here is silly, since the media type doesn't 20:23:46 matter, UFS is endian-dependent on any media. 20:23:46 https://github.com/illumos/illumos-gate/blob/master/usr/src/man/man1/rmformat.1#L530-L531 20:23:46 - or, even better, ZFS! 20:27:43 gwr_: do you still need me to test a test-runner thing? 20:53:41 [illumos-gate] 18102 bhyve: Populate the device version from the backend -- ShengYi Hung 21:11:04 to be really honest, I can not care less which way those updates are done as long as they are done. 21:11:56 unfortunately working with anything roff-like doesn't seem to be particularly inviting to people 21:12:38 and markdown is going to suffer from fidelity problems (there's no equivalent markdown syntax for .In or .Ev or .Dv for example) 21:13:08 which is what everyone seems to want to do these days 21:14:01 (also the resulting formartting it produces if you convert is usually crap -- see vmadm on smartos) 21:14:07 I have -- shall we say -- worries about letting a system known for making shit up out of whole cloth, or copying the answers for other operating systems and hoping for the best, loose on our manual 21:14:12 I would hope those worries were shared 21:14:46 if a thorough proof reading is actually easier than doing the conversion, I guess that's someone elses problem to think about 21:17:15 i'm sure if you just tell it 'don't hallucinate and don't copy content from other operating systems' it'll be fine... 21:17:20 (/s if that's not obvious) 21:25:50 why it should copy content from other operating systems when you tell it to convert from one macro language to another? you are not changing the text there. grab the patch from https://code.illumos.org/c/illumos-gate/+/4817 and compare output from two man commands (before and after). I used this prompt to create it: > consult with usr/src/man/man7/mdoc.7 and convert ls.1 to mdoc format 21:25:51 → CODE REVIEW 4817: convert ls.1 to mdoc(7) Assisted-By: Claude Opus 4.8 (1M context) (NEW) 21:29:12 and btw, it did hallucinate about some macros, it did use 'mandoc -Tlint ls.1' as part of the process to detect and fix conversion errors there. 21:38:18 if illumos merges llm stuff I think I would no longer be interested in the project 21:50:52 tsoome: the issue is that the system architecture underlying LLMs does not allow for a clean airtight code vs data separation. 22:13:58 richlowe: that's why I'm leery about using them for code generation 22:15:13 but they can be useful to help figure out what code is doing or for code review (though I would put the onus on the person using it to curate the results before posting them) 22:15:31 I actually do not want to use llm for code generation because I want to understand what I write;) 22:15:32 of course that's all while they're still affordable :P 22:15:50 or should i say insanely subsidized