<nckx>raghavgururajan: Bisection only makes sense for reproducible failures (the result of bisecting transient failures is simply random). But if failure/success are reproducible on your machine, great!
<nckx>jackhill: We already have that package, it's called Chromium.
<nckx>I actually did think of a thing that's annoying (and not due to simple operator error): synchronising cached nars & narinfos.
<Aurora_v_kosmose>I also haven't managed to figure out how to get it to convert LAN http requests to query upstream via HTTPS
<nckx>I don't think it's possible to do it right with a generic HTTP proxy.
<nckx>Well, this is all very Squid-specific, I can't help you with that I'm afraid.
<nckx>Anyway, that could be solved by a Guix-aware proxy that would arguably be better for being a separate piece of software from guix publish. Not that I drafted it out ☺
<nckx>Making guix publish do this without making trust transitive seems *hard*.
<nckx>But maybe you want that. Anyway, yay, my build finished, back to work.
<Aurora_v_kosmose>nckx: Basically the idea is that I trust the cache to fetch from Guix's servers & CI properly, but I don't necessarily trust the nodes querying the cache, so I giving them build offloading ability to effectively achieve the same effect is not an option, because there seems to be no way to restrict what you allow building if you allow offloads.
<KarlJoad>lilyp: Using your suggestion, I narrowed my issue down to 2 things. 1) When using straight, guix-mode cannot find guix-scheme-directory. 2) Prepending my configuration path to the load-path list causes everything to break.
<vagrantc>it is pretty variable, though ... also depends on how many dependencies are being built in order to build the package itself
<vagrantc>you can have a package be built because it's source or packaging changed, or any package that it depends on recursively ... so you could have 0 packages that need to be built first, or hundreds ...
<ns12>vagrantc: Okay. I have now successfully installed youtube-dl by using the substitutes, whereas I had to build from source yesterday. Apparently, youtube-dl became available as a substitute sometime in the last ~10 hours.
<opalvaults>Otherwise, it's a fairly similar experience. It's probably not as polished as NixOS but it's certainly imo more fun. Aside from hardware, GUIX has tons of software packaged for it. Also, check the core-updates-frozen repo on the guix git repo if you find that a version of software you're using is a bit older (Gnome for example).
<KarlJoad>With all the weird hardware I run, that support would be my biggest concern.
<opalvaults>you can do a guix pull --branch=core-updates-frozen --allow-downgrade (or something like that) and grab software that's still not in stable.
<KarlJoad>I would like to stay away from GNOME if I can. I am not a fan of it or its UI.
<f1refly>hello, i'm trying to set up guix on luks and lvm (in the luks volume). grub asks me for my password, for the right partition according to the uuid. but when i enter my password and press enter, it tells me "access denied. no such cryptdisk found." and then it tells me that it couldn't find "disk 'lvmid/XXXX..XXX" (which really does not exist, according to lvshow)
<cehteh>i wrote down some notes/plan about deletion: https://github.com/cehteh/rmrfd ... question is only who/when/how this gets implemented. I picked that up because its some use case i eventually need to be fixed too, but its not high on my prio list
<cehteh>is that still the deletion or some gc scanning now?
<efraim>civodul: I'll report when it finishes the build phase
<efraim>perhaps compiling it with -O2 isn't the best option
<rekado>civodul: I started the du process yesterday night and it finished hours later. So... it's not clear how close this value is to the actual value now.
<civodul>rekado: /gnu/store/trash was almost empty at the time i got your message, but it's filling up now
<cehteh>rekado: any opinions about my rmrfd plan? i am just trying a small POC implementation in rust creating an inventory DB on my backups, wanna see how much ram that takes when just using std collections instead some disk based database
<nckx>roptat: -ENOMUA, but from the Web interface: ‘This library implement’, I think changing from url-fetch to git-fetch should be logged, adding properties is also unlogged and should be separate, no point in trailing #t at this late point IMO.
*nckx makes a note to call it /luks.<device>.key, duh.
<jpoiret>thinking about mapped-device, isn't it a bit complicated to have luks and lvm-specific fields in the definition? seems like quite some things in guix are screaming *inheritance*, although there's no interface for that (except GOOPS)
<jpoiret>also thinking about package/inferior-package
<jpoiret>but seeing how goops handles inheritance using generics, i'm not too stoked about it either. has a very "impure" side effect feel to it
<singpolyma>I love GOOPS but you can always model inheritance without it. With enough macros you might even reinvent GOOPS ;)
<civodul>jpoiret: we could have record type inheritance without pullin in GOOPS