<Jookia>iyzsong: Ludovic has spoken and it looks like we'll need to figure out a way to find all packages that specify both gdk-pixbuf and gdk-pixbuf+svg indirectly/directly and fix them to use just one :|
<mark_weaver>grafting needs to first download or build everything that's needed without grafts, use that to find the set of runtime references, and then create new versions with references to buggy things replaced with the fixed versions.
<Jookia>http://sprunge.us/VhPS this is the script, the problem is that numix-icon-theme-circle depends on numix-icon-theme and does this using relative symlinks, assuming a certain set of directories will be in the output as if I did do a union, but I haven't
<civodul>mark_weaver: i've started a security-updates evaluation
<Jookia>The package installs fine and sets my profile, but I get issues when I install another package
<civodul>Jookia: the relation between these two package looks tricky
<Jookia>Well, there's numix-icon-theme, and numix-icon-theme-circle which assumes numix-icon-theme is propagated but also symlinks to ../../Numix/, which would be where numix-icon-theme would be if you installed it in a traditional setup in ~/.themes since they go side by side
<wingo>iyzsong: "guix package -p /tmp/foo -i gnome" has lots of conflicts in gschemas.compiled; is that a problem?
<mark_weaver>jmarciano: putting the name before the hash would slow down all filesystem lookups in /gnu/store, and to no useful end because you need to know the hash of the one you're looking for anyway, or else you might get something old with security holes or whatever
<mark_weaver>janneke: Guix records are enhanced with extra features, they are not plain SRFI-9 records. I don't know that 'set-field' works on them. I suspect not.
<rain1>this is not blocked though, I guess it's because it's a subdomain
<rain1>not sure if that counts as a bug in that service or not
<raphmaster>I got a black screen after following the instructions to install GuixSD in Guix manual section 7. I am using the desktop operating system configuration example. After running "guix system init /mnt/etc/desktop.scm /mnt", it starts to download files. The black screen appear during the linux-libre-3.14 download. I am using Oracle VirtualBox.
<mark_weaver>civodul: fyi, I just tried "guix system build --no-grafts" before "guix system build" (with grafts), and the results were much nicer. I wonder if this could be done by default?
<mark_weaver>(also, I see no way to do the equivalent for "guix package -u")
<mark_weaver>in particular, I got notification ahead of time of what would be built and downloaded, and it avoided the repeated "updating list of substitutes" messages without useful progress information
<mark_weaver>ACTION switched back to xfce for now, so I can suspend-to-ram during LP
<mark_weaver>and I think --substitute-urls='' should be one of those ways :)
<mark_weaver>more generally, if --substitute-urls= is passed, and there's some kind of syntax error in that specification, it would be better to error out or assume an empty list than to revert to the default list.
<mark_weaver>I apologize if I reacted a bit strongly here, but I'm sensitive to the security issues around having a substitute server being used when I asked it not to.
<mark_weaver>NiAsterisk: [[:space:]] should work, but two things: as I said before (did you miss it?) you need to remove the parens from around the replacement expression, and putting [[:space:]] in the replacement expression will just result in that exact string "[[:space:]]" being put there.
<mark_weaver>the replacement is not a regular expression, it is just a string
<mark_weaver>or rather, a scheme expression that must evaluate to a string
<bavier>good-point, I missed that part. paren-blindness I guess
<mark_weaver>("#define CMDSEP[[:space:]]*'§'") as a scheme expression fails, because the parens make that into a procedure call, with the first item being the procedure. but a string literal is not a procedure, so it will fail
<NiAsterisk>hm. I thought it was file (string) (replacement)
<NiAsterisk>does guix have an option like gentoo where you can integrate every .patch and ignore it when there are none
<mark_weaver>I'm not entirely sure I understand what that means, but I'm fairly sure the answer is "no"
<NiAsterisk>hm. makes sense, the packaging architecture is different
<mark_weaver>I suppose in theory, the 'patches' field could be an expression that scans a directory for filenames matching a certain pattern and return that list of files, but that's not how we do things in Guix, and I don't think we'd want to.
<mark_weaver>for one thing, we often have multiple variants of the same package with different sets of patches applied
<mark_weaver>e.g. for 'guile' we have a bootstrap guile with at least one extra patch, and we often have multiple versions of the same package with different patches needed for them, etc.
<mark_weaver>and from a stylistic standpoint, I prefer to see from looking at the package definition whether there are any patches, rather than also having to look in a directory to see if there are any files starting with the package name.
<mark_weaver>IMO, anyway, feel free to propose something on guix-devel if you disagree :)
<NiAsterisk>it's open in gentoo and it's what you write and can include regexp patches in a folder, and also have specific patches. I'm okay with how it works here.
<gnudevotee>Happy GNUvidad! May He be with us for a long time!
<mik_>I want to modify a package definition in the guix source tree, then actually be able to install it. Someone earlier told me to use pre-inst-env script when developing, but I actually want to be able to install the package