IRC channel logs
2015-05-05.log
back to list of logs
<phant0mas>civodul: I posted the backtrace with the patch so you can have a look <Phlogistique>do you know if people are using Guix without GuixSD to install X server + graphical environments? (e.g. XFCE, if you have that in Guix) <rekado_>Is there a nicer interface to hydra (e.g an Emacs mode) and/or a way to restrict the list of failing builds to only one architecture and only to direct failures (not propagated failures from dependencies)? <rekado_>Phlogistique: on the cluster I installed graphical tools (e.g. Thunar) for a user, but not all of XFCE. <rekado_>Phlogistique: on my main laptop I use Guix on top of Fedora but still use the GNOME shell; I have installed XFCE there, but haven't had the time to actually try it yet. <iyzsong>Phlogistique: the X server won't work <iyzsong>it build a service for dmd, and need root permission <rekado_>iyzsong: dmd can be run in addition to the existing init, even as root, to manage the required service, no? <rekado_>(although that's getting eerily close to running the system distribution) <iyzsong>rekado_: yes, but I wonder who will do that ;-) <rekado_>I wonder if it would be possible to create packages on the fly for Emacs packages from ELPA. <rekado_>It seems like a lot of work to duplicate the package database ina different format. <rekado_>I would really like to guixify all of my Emacs stuff, because none of the package management methods available for Emacs are really satisfying. <rekado_>guix package --from=elpa --install god-mode geiser ido-vertical-mode ... <paroneayea>rekado_: it's an interesting idea to have on-the-fly-generated packages <paroneayea>since you can build package objects, I don't see why not <rekado_>I guess the biggest hurdle is that there are no published hashes for elpa packages afaics. <rekado_>I would really like to use something like on-the-fly packages for R packages as well. <rekado_>a guess only: from the "mirror://gnu..." line? <jmd>rekado_: Could be ... It would explain why it thinks that gnome and R are not part of Gnu. <civodul>paroneayea: the initial idea was to have 'guix import' produce <package> objects instead of sexps, so that one could run, say, "guix import pypi foo --build" <civodul>we could have mirror-type return 'gnu for mirror://cran, but that wouldn't be accurate for non-GNU packages available on CRAN <jmd>What about mirror://gnome ? <jmd>Or better still, just import /gd/gnuorg/maintainers.bypkg from fencepost <effa>Hi rekado_, re on-the-fly generated packages: I guess the first step would be to write an ELPA importer. <civodul>jmd: i've just fixed these two cases :-) <civodul>fencepost is not publicly accessible so it's not an option <civodul>the problem is that importers currently don't do much with dependency information <civodul>they just import a single package and try to guess the variable name for dependencies <civodul>something more sophisticated would be needed for general on-the-fly generation <effa>Some repositories provide a full list of dependencies (e.g. hackage). So, if the name mapping is consistent, then it should work. <effa>Personally I'm not entusiastic about on-the-fly package generation because it makes things non reproducible. <effa>I prefer importers toghether with a way to refresh things. *davexunit also prefers an importer to on-the-fly generation. <davexunit>it's just a matter of getting people to maintain these packages. <jmd>civodul: gnumeric should also be in that list? <rekado_>I was thinking about the case when a user would like to have software managed by guix (--> rollbacks), but is not comfortable enough or willing to hack on package definitions. <rekado_>to have guix build and install stuff is still better than using some external means. <rekado_>(obviously, for the project it makes more sense to have more packages and have them maintained, but maintining packages isn't for everyone.) <davexunit>I'm not particular enthused about that direction, but if someone makes it work, that's cool. <civodul>iyzsong: BTW, there's some Bash syntax in our /etc/profile and with environment-variable-definition in (guix search-paths) i'm about to introduce some more <civodul>i guess zsh actually understands that syntax though <civodul>davexunit: i've just pushed changes to 'guix environment' to consolidate search path handling <iyzsong>civodul: yeah, /etc/profile should work for both bash and zsh, and we could make another file for fish :-) <civodul>iyzsong: so apparently "export foo=bar" works for zsh, right? <davexunit>civodul: oh good, I remember making a not-so-great patch for that awhile ago. <civodul>iyzsong: doesn't it need "FOO=bar; export FOO"? <civodul>davexunit: it's just that search paths are in cache now for me, so everything looked clearer today :-) <civodul>iyzsong: also, does zsh support "${FOO:+bar}"? <rekado_>python-pillow does not seem to work properly. I get this error on "from PIL import Image" (python3): ImportError: libopenjp2.so.7: cannot open shared object file: No such file or directory <effa>However, the package builds fine locally. <rekado_>effa: would it be possible to uild to build numpy against openblas? <civodul>effa`: oh, i thought it did build on Hydra at some point <phant0mas>civodul: when bootstraping and a package origin has a non-empty 'patches' field it uses %bootstrap-patch-inputs from bootstrap.scm to patch them, right? <bavier>civodul: I looked a bit at the hop build, I think the currect failure is due to a change in the way that bigloo 4.1a handles rpath embedding in executables <civodul>bavier: i looked at it a bit, and i think bigloo just doesn't try to deal with rpath <civodul>i tried to have it pass -Wl,-rpath to gcc, but failed <bavier>civodul: yeah, so the current patch-rpath phase I think needs to move from making substitutions in the current rpath to adding libdir to the rpath <civodul>maybe what you suggest will work better <rekado_>building texlive-texmf without substitutes and with very big timeout and max-silent-time values takes an incredibly long time :-( <rekado_>downloading the substitute doesn't work for me because of corruption. <rekado_>I wished we could split this huge package. <rekado_>I'm now stuck at "phase `texmf-config' succeeded after 64 seconds" since several hours. <civodul>and why do you get that corruption when i don't? :-) <civodul>would anyone be willing to help with the html->sxml conversion of the new web site? <civodul>it would be great to put it on-line quickly, before the release <effa`>rekado_: I never used OpenBLAS. I picked ATLAS because that's what is mentioned in the numpy build instructions that I did follow. <effa`>I'm not against chaning that, but hopefully not at the expenses of performance :-) <civodul>davexunit: just clone the guix/guix-artwork.git repo, manually do html->xhtml, and then run xml->sxml on that, pretty-print, and voilĂ :-) <civodul>from there we can gradually factorize/improve things <davexunit>but yeah, I can try to squeeze some time in for this conversion. <civodul>but i think we can use utf-8 all the way and not worry about escapes no? <davexunit>civodul: I was told that by someone else, too. <civodul>as long as the page's content-type is utf-8 and the server's HTTP content-type is utf-8 too (which it is), we're fine <effa`>civodul: hydra does build Numpy for i686, but fails for x86_64 <davexunit>civodul: hmm, okay, but writing a general procedure that assumes utf-8 seems risky. <civodul>davexunit: nothing :-) the html pages already have charset=utf-8 <civodul>in general yes, but here we can safely assume utf-8 *civodul advocates best practices ;-) <davexunit>I'm still worried about the general case, because I think Guile core's SXML modules should handle HTML, too. :) <civodul>so i'm not surprised xml->sxml doesn't handle html <davexunit>I think a separate procedure should, though. <civodul>but then again, i'm not the html expert here ;-) <davexunit>because you can represent HTML with sexps, too :) <civodul>there's an (htmlprag) module in guile-library <civodul>meaning it tries to deal with hand-written, broken html from the 90's <davexunit>I had a friend very recently wishing he could do this with guile <davexunit>it's so easy to make mistakes that web browsers have to put lots of effort in working around it. <rekado_>civodul: I don't know if it is normal. But the corruption is something I get every time I need to install texlive :( <effa`>I also got corrupted TeXLive. Now I usually do not substitute <rekado_>I'm dreading core update merges, because this usually means I have to mess with texlive building again (because that's needed for numpy) <civodul>i don't think we have an open bug report for that, do we? <rekado_>I'm not sure if it's a network/firewall issue. <rekado_>I only run into this problem with texlive-texmf-2014. <civodul>i would have thought that, but if effa` is having the same issue... <rekado_>also, the download is usually very slow, but it seems much faster when downloaded with wget. <civodul>could you just file the backtrace/error that you get with whatever details are relevant? <civodul>and regarding speed, it would be nice to check that <bavier>I've noticed the speed difference with wget as well <civodul>long ago there was this kind of problem because Guile was setting an wrongfully small input buffer on the socket <civodul>bavier: could you report it, stating the Guile version, Guix commit, etc.? <taylanub>when some sources are patched, guix downloads the patched version from hydra (as a substitute) instead of downloading the raw sources and patching them locally, right? so maybe it downloads slower than with-wget/from-upstream because hydra is slower <bavier>taylanub: yes, but we don't patch texlive <civodul>guix always downloads sources from hydra when substitutes are enabled <civodul>the test would be to compare 'guix download foo' with 'wget foo' <rekado_>I was referring to downloading the binary substitute, the nar file. <rekado_>fetching it with wget is faster. (I'll try to reproduce this.) <rekado_>another thing I find odd is that even when the daemon runs with --debug there is no output for a long time (> 1min) after printing the list of files that will be downloaded. <rekado_>I have no idea what Guix is doing at this point and why it takes so long. <rekado_>it's actually considerably longer than 1 min. <rekado_>strace shows me that "guix build" is waiting for input: "read(10, " <civodul>when the daemon runs with --debug, it prints tons of stuff <civodul>so i'm not surprised that makes it slower, but still <bavier>I just got some atlas/openblas performance data to share <civodul>RUNPATH challenges of the day: webkitgtk and avidemux <rekado->a patch for webkitgtk has already been sent to the ML. <davexunit>civodul: the 'guix environment' changes look good, btw.