On Fri, 2006-09-08 at 11:27 +0200, Tilman Sauerbeck wrote:
Mark Rosenstand [2006-09-08 10:44]:
Hi,
One of my personal likings of CRUX is that it's easy to customize and extend. Because it (used to?) keep as close to upstream as possible, there tend to be fewer distribution-specific bugs.
We _still_ do. But it's probably related to the fact that we don't read all of the code of the software we're packaging and that we don't fix any and all bug (-> Debian ;D).
Totally fair, and that's also why I argue that keeping close to (if not verbatim) upstream is the best bet for a project the size of CRUX. Personally I've had great success getting things such as DESTDIR fixes included upstream, making it a joy to maintain packages in the long run. Unfortunately I've had to fork some opt packages lately. For example, I depend largely on GTK+ (GNOME desktop) so I just don't dare to patch it for something I have no clue what does (http://crux.nu/svnweb/CRUX/revision?rev=1701)
I've been using/packaging the dchroot utility from Debian for a while. It's a small setuid utility to allow regular users to chroot into a specified list of directories. Because the path of 'su' is hardcoded to /bin/su and CRUX keeps su in /usr/bin/su, I've only used it for Debian chroots (running on CRUX) so far. Now, the other day I accidently looked at the shadow Pkgfile, and to my surprise su is actually installed in /bin and then moved manually to /usr/bin.
Per probably had a good reason to do this... Would be okay by me to undo that change though.
Looking forward, it'd be nice if packagers would comment why they do such things in the Pkgfiles (unless it's obvious, say "mv $PKG/usr/share/man $PKG/usr/man") I think that recently there have been a growing number of patches in core and opt - are upstreams just getting insane? I get the feeling that the heavy patching is also the primary reason we're stucked with years old versions of things such as grub. I recently made an attempt to make udev less distro-specific (and hence more maintainable) but it seems that by now (6 revisions later) things are pretty much back to the pain it was before.
To avoid these kinds of distribution-specific problems, I'm kindly asking packagers to think twice before deciding to divert from upstream - perhaps even limiting it to "required for essential functioning of the package" and basic CRUX integration (no nls, man-pages in /usr/man, etc.) :)
Uh uh, I actually thought twice before doing that Openbox commit. I even contact upstream, ziomg! ;)
Ah, true, you said that. I shouldn't have used that as an example, it was rather a case of "shit happens" :) The autoconf 2.60 mandir patch was probably a better example.