IRC channel logs

2021-02-02.log

back to list of logs

<rekado>true
*civodul runs "git pull" in guix-hpc-web just in case
<zimoun>hi!
<rekado_>zimoun: hi!
<rekado_>I checked MRAN
<zimoun>cool
<rekado_>they don’t have a real archive; they have date-indexed snapshots of the whole CRAN.
<rekado_>so we can’t easily fall back to their archive
<rekado_>we can only fall back to a specific snapshot in their archive
<rekado_>so it seems to me that this is not a good generic solution for CRAN packages in Guix
<zimoun>I do not know what you mean by generic. The generic is to have git-fetch instead of tarballs. Otherwise it is always clunky at point or another, IMHO. Disarchive is nice but just a fix for the short/medium term, not for the long term because it requires to maintain 2 databases: the content and the bridge to lookup for the content; another story. MRAN seems a pragmatic solution for the short term.
<rekado_>for individual packages I agree that MRAN is fine
<rekado_>but we can’t just add an MRAN URL to those tried by cran-uri
<zimoun>I agree.
<rekado_>because a single URL isn’t going to necessarily be better than the CRAN archive or the current CRAN URI
<rekado_>by “generic” I mean adding something as a fallback to cran-uri
<rekado_>generally, we’re doing fine with CRAN, I think. Bioconductor is the bigger problem, but for that we could use git-fetch as you suggest.
<rekado_>AFAIK Bioconductor has not a single repository for packages; is there a “derivable” git repository name for Bioconductor packages or can the name be arbitrary?
<zimoun>about MRAN, I do not know. I trust your wise opinion when you say it is a bad idea to add a fallback to MRAN.
<rekado_>I’d like to switch to git-fetch for Bioconductor soon, so that we no longer have to worry about it :)
<zimoun>Maybe it is worth for indivivual package to write an extension and a cookbook recipe, more or less clean a bit the answer I did by email.
<rekado_>BTW: sorry for dragging my feet on GUIX_EXTENSION_PATH; my brain is mush these days due to terribly short nights
<rekado_>(if my brain is mush, does this make my skull “mushroom”?)
<zimoun>:-) Hope baby is going well
<zimoun>About Bioconductor, I agree. It is an easy move and fix issues about time-machine. :-)
<rekado_>I read that all this is normal. Brain development as they learn to speak can result in terrible sleep. Some are luckier than others.
<rekado_>zimoun: did you have a bug/issue for Bioconductor?
<zimoun>about the switch url-fetch to git-fetch? Let me check
<zimoun> http://issues.guix.gnu.org/issue/39885 is one thing, maybe closed by the switch
<zimoun>related: http://issues.guix.gnu.org/issue/36805
<civodul>does CRAN provide tarballs?
<civodul>currently SWH integration works great for Git snapshots, occasionally with tarballs, and the rest (svn, ...) is not handled at all
<zimoun>civodul: currently Guix uses the CRAN tarballs. I have never checked if Git is provided.
<civodul>so i guess we need Disarchive for these
<zimoun>Well, I have on my TODO to be able to fallback with SVN… and time flies… :-)
<civodul>ISTR there was something that made svn support tricky, but i'm not sure why
<civodul>*what
<zimoun>the SWH API is less nice than for Git. And svn works with revision, so…
<zimoun>civodul: trying to see if a similar trick can be used for Singularity and the --export-<stuff>, I am a bit fighting with Singularity… Is someone used to play with Singularity? v2 or v3
<civodul>zimoun: i've used it just to ensure our images work, both v2 and v3
<civodul>but we don't have v3 in Guix, so i tested on a cluster that does have v3
<zimoun>civodul: I am not able to use v2 from Guix, and I use v3 compiled from sources. Well, I am going to write a blog post (or cookbook recipe) only for Docker and let Singularity as an exercise for the reader :-)
<civodul>zimoun: sounds good!
<civodul>we have a system test that uses v2
***ChanServ sets mode: +o civodul
***civodul changes topic to 'GNU Guix in high-performance computing | this channel, it is logged: http://logs.guix.gnu.org/guix-hpc | FOSDEM: https://guix.gnu.org/en/blog/2021/meet-guix-at-fosdem-2021/ | https://hpc.guix.info'
<zimoun>civodul: the Guix hash is the same on the 2 machines now ie nkvlqwzvxdlhzlc7vhfcngxc19x2ay2f-docker-pack.tar.gz but md5sum is different. And sha256 in the file config.json is different and layer.tar too. The checksum of the tar are different but the content seems the same. Maybe a metadata issue?
<civodul>zimoun: could you run diffoscope on these two tarballs?
<zimoun>not yet, I have untar them and compare all the checksums with “find bla -type f -exec md5sum {} \;”
<zimoun>civodul: https://paste.debian.net/1183728/
<zimoun>what does it mean ’h’ in +hr-x-r-xr-x?
<civodul>"h" in "tar tvf" output must be for hard links (tar supports that)
<zimoun>thanks. Well, that’s the difference. Why?
<civodul>dunno!
<biotim><rekado_ "AFAIK Bioconductor has not a sin"> rekado_: For Bioconductor repository names, the format should be https://git.bioconductor.org/packages/PackageName, with branch names like RELEASE_3_12 for each Bioconductor release