IRC channel logs
2014-03-27.log
back to list of logs
<spacebat>I haven't been able to install anything with guix, each time it follows dependencies until it fails to find the gc tarball on boehm's website <spacebat>perhaps I should be using guix from git instead of the 0.5 tarball *ggrant` wonders how far 1.0 is away. We are nearing that 3-4 month point. :^P <civodul>actually, you can also manually get the tarball with "guix download" <spacebat>guix pull gives: ERROR: In procedure put-bytevector: Wrong type argument in position 2 (expecting bytevector): #f <spacebat>I think the 0.5 tarball is broken, or the setup instructions are, or the diagnostics are <zacts>what can I do to help with the 0.6 release, in terms of importing packages? <mark_weaver>write recipes for packages you want that are missing :) <zacts>will guix use dmd, or systemd? <davexunit>"Gitian is a secure source-control oriented software distribution method." <davexunit>"Gitian uses a deterministic build process to allow multiple builders to create identical binaries." <viric>sometimes, deterministic = completely unflexible <davexunit>so I wanted to make sure that the guix community was aware of the tool, in case we could use it or learn from it. <viric>mark_weaver: I mean that it's easy to claim determinism, once you reduce flexibility or comfort. For example: install ubuntu, build X. Every time you do these two steps, the build will match. <viric>I agree, but did you understand me a bit more? :) <mark_weaver>but indeed, any deterministic build process that starts with downloading many megabytes of binaries is not very helpful. <phant0mas>for some reason while I target i686 I get this in the config.log <phant0mas>COLLECT_LTO_WRAPPER=/gnu/store/z8qifq9172ad17a7mxg9w8jq51axwwqr-gcc-4.8.2/libexec/gcc/x86_64-unknown-linux-gnu/4.8.2/lto-wrapper <phant0mas>shouldn't it normally refer to /gnu/store/nkrr16lw55p75483z2an0c67nnqdm4l5-binutils-cross-i686-pc-gnu-2.23.2 <bavier>I can't build python-3.3.3 on my machine <bavier>I'm getting, fairly deterministically, an "OSError: out of pty devices" when running the tests <bavier>looks like hydra is doing ok with python... <civodul>mark_weaver: has the daemon's --no-substitutes ever worked for you? <civodul>it seems it's always overridden by the client's settings <mark_weaver>I guess I don't know. it's possible that I simply never got any substitutes because there are none available for mips. <mark_weaver>but I guess I've been leaving myself open to attack by the NSA this whole time :-( <mark_weaver>(or anyone else who could intercept the connections) <mark_weaver>well, if things get fixed now, I'll be glad of that at least. <mark_weaver>civodul: one thing though: I've also built things on my x86_64 machine, and I don't recall it ever telling me that was downloading substitutes. <mark_weaver>but it's possible I simply wasn't paying enough attention. <civodul>actually i think it was a false alarm <civodul>i'm touching these things right now in my tree <phant0mas>it hasn't downloaded any substitutes here either, on my x86_64 machine <civodul>phant0mas: you've been running guix-daemon --no-substitutes <phant0mas>I have a question, sending it to the list with some logs <dfsdfds>civodul: A couple of days ago on mips64el, the substituter tried to download binaries while I was running the daemon with --no-substitutes. Adding --no-substitutes to guix package fixed the problem as far as I can tell. I was too busy when that happened, so I forgot to report it. <dfsdfds>civodul: Unrelated: I'm trying to understand when source tarballs get downloaded. Does it happen before the build process starts (in a batch), or does it happen on demand (download, configure, make, install)? <civodul>dfsdfds: about --no-substitutes: did you have Nix installed? i think that could happen if you had Nix installed, with Guix after commit 89faa5c <dfsdfds>I'm thinking of an option that would change the download process, but I'm not yet convinced that it's a good idea. <civodul>dfsdfds: re tarballs: downloads are just normal derivations <civodul>so they get downloaded when they're needed <dfsdfds>Yeah, I think it may be a good idea to download everything before building stuff. Some people travel frequently, so they aren't always connected to the network. <civodul>one could easily write a download-tarballs procedure <dfsdfds>civodul: wrt to --no-substitutes: Nix isn't installed on my system. However, I git clone'd it into nix-upstream before running ./bootstrap because ./bootstrap failed to do so for some reason (again, I was too busy to debug the issue). <civodul>i think the problem is that it would be a very specific use case <civodul>for instance, how do you choose what to download? <civodul>and one can basically run "guix build -S emacs foo bar baz" <civodul>dfsdfds: do report any such issues, even if the report is rough on the edges :-) <dfsdfds>civodul: Well, I would definitely use it. I'm on a slow machine and don't have network access at work. So I'd prefer to fetch everything at home, go to work, and continue building there. <civodul>it's better to have reports, otherwise we/i don't realize <dfsdfds>That's why I'm reporting it right now. <civodul>i actually do that before i take the train and have hack time, like when going to FOSDEM :-) <dfsdfds>civodul: I used it in the past, but it's not automatic (can't find a better word) enough. <dfsdfds>civodul: I simply want to run guix package --install={foo,bar,baz}, then pm-hibernate at some point, and be sure that if foo, bar, baz were downloaded, the build process would resume. <civodul>it's not possible to do it this way currently <civodul>what would be possible is something like "guix prefetch foo bar baz" <civodul>which would be like "guix build -S foo bar baz" plus "-S all the prerequisites" <dfsdfds>So, would we need to write a separate script from scratch? Is it that bad? <dfsdfds>Can't the existing code be modified? <civodul>well, it's the daemon that schedules derivations <civodul>and it doesn't care if it's a download or not <civodul>so it could be tweaked specifically for this use case <civodul>but i'm not convinced it's a good idea <civodul>anyway, these are just random thoughts <civodul>perhaps you could start a thread on the list? <civodul>now i'm trying to remain focused on 0.6ish things <civodul>(speaking of which, i'm happy with substitute signature checks!) <dfsdfds>Okay, will do if I have something concrete. Just to clarify, the only way to achieve this is to patch the daemon itself, right? <dfsdfds>civodul: WRT to the release, what are the remaining bits? Can I help somehow? <civodul>dfsdfds: if you want it to be magic, probably <civodul>but you should really post your idea, with example use cases <civodul>wrt. the release, mostly mark_weaver's union.scm and the substituter/signature stuff <dfsdfds>civodul: I've just realized that it may be better to have a separate script for that that you could pipe to guix package. After all you want to make sure that you actaully downloaded the needed sources before you leave. <dfsdfds>civodul: Perhaps it's already solved by the -S option. Could you demonstrate how you usually use it? *civodul doesn't scale so well :-) <civodul>dfsdfds: typically before taking the train, i'd do a bunch of "guix build" and "guix build -S" to have enough stuff in the store <phant0mas>civodul: take your time, I am looking into it so maybe I will find what's wrong