IRC channel logs

2023-05-25.log

back to list of logs

<rekado>there’s a chance for a Guix seminar at https://events.hifis.net/event/808/
<rekado>I have no details yet, but would any of you be willing to do a workshop?
<civodul>Berlin end of Sept?
<civodul>hmm
<civodul>there's the possibility of a Guix-HPC event in Montpellier (South of France) in Oct.
<civodul>yet to be confirmed, but that would have to be taken into account in one's agenda
<rekado>yes, at the MDC in Berlin-Buch (northern tip of Berlin)
<PurpleSym>Hm, the “building package cache...” phase of Guix pull is consuming an absurd amount of memory (~80GB) during guix pull when adding guix-cran 😕
<civodul>rekado: are you planning to be there?
<civodul>PurpleSym: ouch
<civodul>can you send a bug report, including the size of the resulting .go file?
<PurpleSym>I can try if it actually finishes. How can I find the resulting .go file, civodul ?
<civodul>3.0.9 improved on assembler/linker memory usage, but we still find ourselve with one or two copies of the ELF file in memory :-/
<civodul>the resulting file is ~/.config/guix/current/lib/guix/package.cache (or similar)
<rekado>civodul: yes, I was approached by one of the organizers. They are interested in a workshop or talk.
<civodul>rekado: excellent, i didn't know they'd started contacted people
<civodul>i hope you can make it 🤞
<civodul>and PurpleSym too!
<PurpleSym>civodul: It finished after about 30 min. Which file’s size do you need in the bug report?
<civodul>30mn, 80G, ...
<civodul>PurpleSym: the size of the package.cache file
<PurpleSym>100G, actually.
<civodul>that not RSS though?
<civodul>i pulled guix-cran on my laptop a couple of months ago, and it has "only" 16G of RAM
<PurpleSym>Well, RSS cannot be more than system memory, which is just 64G. But the system was paging like hell.
<PurpleSym>Actually, it “finished” through the hands of the OOM killer. And no package.cache was ever written to disk. So I’m guessing guix search will be quite slow without that.
<civodul>fun :-)
<civodul>we could implement an proactive OOM strategy (i.e., leave out the cache when there are too many packages)
<civodul>'guix search' doesn't benefit from the cache
<civodul>'guix package -A' does, likewise for lookups by name
<zimoun>hi!
<zimoun>any issue with the manual https://guix.gnu.org/manual/devel/en
<zimoun>?
<civodul>no, it's just bold
<civodul>:-)
<zimoun>that’s an issue for my eyes. ;-)
<rekado>I think it’s because we provide fira sans, but only in the bold variant
<rekado>it’s two problems, really
<rekado>the second is the rule for “.settitle, .top, .chapter, .section, .subsection, .subsubsection”
<rekado>this affects not just the headings but all sections
<civodul>so first we should add a second @fontface for the regular weight, right?
<rekado>I think so, yes
<civodul>done
<civodul>(website is being rebuilt)
<rekado>thank you!
<zimoun>civodul: how many time is the website requiring for being rebuilt?
<civodul>how much time? ~8s says mcron.log
<civodul>it's rebuilt now
<civodul>several times maybe :-)
<civodul>that doesn't make any difference though
<zimoun>fast, nice. I was asking because I do not see any difference after refreshing. Propagation to some cache?
<rekado>the “.settitle, .top, .chapter, .section, .subsection, .subsubsection” rule should be disabled
<rekado>or at least the “font-weight: bold” line.
<rekado>because this applies to all elements in those sections, not just the headings
<civodul>yeah, should be fixed now
<civodul>rebuilding...
<civodul>still i wonder why it suddenly went wrong
<rekado>it’s a difference in generated HTML
<rekado>see http://web.archive.org/web/20220124030451/https://guix.gnu.org/manual/devel/en/html_node/Audio-Services.html
<rekado>the new HTML wraps sections in divs
<rekado>and those divs have the .subsection or .section classes
<rekado>previously the HTML was not as structured; just tag after tag
<rekado>and the only things with .section or .subsection classes were headings
<rekado>so for the new HTML the rule is too broad
<rekado>it was fine for the old HTML
<rekado>oh dear, just found a special snowflake bioinfo package: needs Python 2.7 … and Rust.
<rekado>and node
<rekado>and also go
<civodul>rekado: i see, thanks for checking
<civodul>Python 2.7 and Rust, that's quite an unusual mixture of vintage and cutting-edge software
<civodul>fashionable, even
<zimoun>thanks rekado and civodul. My eyes are now happy with the manual. :-)
<zimoun>the font changed, right?