Hi! I tested dpkg with my friend and we noticed that ldconfig runs when needed. This should be implemented to pacman too. Bye, Nagy Gabor PS.: We found that pacman is not definitely slower than dpkg.
On 3/31/07, Nagy Gabor <ngaba@petra.hos.u-szeged.hu> wrote:
Hi!
I tested dpkg with my friend and we noticed that ldconfig runs when needed. This should be implemented to pacman too.
If you noticed it, perhaps you can tell us how it is done? I don't have an easy way of testing dpkg and I sure don't want to spend 8 hours looking at the source trying to figure it out. -Dan
If you noticed it, perhaps you can tell us how it is done? I don't have an easy way of testing dpkg and I sure don't want to spend 8 hours looking at the source trying to figure it out. Sorry, I can't tell you... We just concluded it from the ldconfig's access time. Probably it checks, if */lib directories are touched or not (with the help of ld.so.conf and similar). I may will look into its source later. Bye
On Sun, 01 Apr 2007 20:09:52 +0200, Nagy Gabor <ngaba@petra.hos.u-szeged.hu> wrote:
If you noticed it, perhaps you can tell us how it is done? I don't have an easy way of testing dpkg and I sure don't want to spend 8 hours looking at the source trying to figure it out. Sorry, I can't tell you... We just concluded it from the ldconfig's access time. Probably it checks, if */lib directories are touched or not (with the help of ld.so.conf and similar). I may will look into its source later. Bye
_______________________________________________ pacman-dev mailing list pacman-dev@archlinux.org http://archlinux.org/mailman/listinfo/pacman-dev
dpkg is also in aur: http://aur.archlinux.org/packages.php?do_Details=1&ID=6201&O=0&L=0&C=0&K=dpkg&SB=n&SO=a&PP=25&do_MyPackages=0&do_Orphans=0&SeB=nd
On 4/1/07, Nagy Gabor <ngaba@petra.hos.u-szeged.hu> wrote:
If you noticed it, perhaps you can tell us how it is done? I don't have an easy way of testing dpkg and I sure don't want to spend 8 hours looking at the source trying to figure it out. Sorry, I can't tell you... We just concluded it from the ldconfig's access time. Probably it checks, if */lib directories are touched or not (with the help of ld.so.conf and similar). I may will look into its source later.
Stuff like this is not very helpful. If you did a scientific test, please provide the steps you took.
Hi! I prefer a noldconfig (or similar) flag in .PKGINFO (or similar) which would indicate that with this package we don't need to run ldconfig. And makepkg should take care of filling in this entry properly (which can do a dpkg-ported autodetection (user could override this in PKGBUILD) or whatever). ldconfig runs after all package removal or adding. This is not so time-wasting because of disk cache, however in case of many files this is noticeable. So I think _alpm_ldconfig should be called from alpm_trans_release for example (combined with a needldconfig flag if needed, see above). Yes, I know, I may provide a patch ;-), I just ask first, what do you think about these ideas. Bye, ngaba
Some notes for the last idea. Because of install scripts we may need to run ldconfig after a package install/removal (glibc for example?) otherwise we couldn't run other packages' install scripts. So at least 3 options needed for "needldconfig" (0: not needed, 1: needed at the and of transaction, 2: needed after install/removal (default) for example). Here I mention, that many packagers forget about adding packages to dependency list which is needed for their install script (see jre for example: it uses grep in its install script, however it doesn't depend on grep.) Bye, Nagy Gabor
Nagy Gabor wrote:
Some notes for the last idea. Because of install scripts we may need to run ldconfig after a package install/removal (glibc for example?) otherwise we couldn't run other packages' install scripts. So at least 3 options needed for "needldconfig" (0: not needed, 1: needed at the and of transaction, 2: needed after install/removal (default) for example). Here I mention, that many packagers forget about adding packages to dependency list which is needed for their install script (see jre for example: it uses grep in its install script, however it doesn't depend on grep.)
What's the point? In most cases ldconfig is needed. All this would cause is headaches for packagers and headaches for the pacman developers. Say, for example, the detection didn't detect that you need an ldconfig. That introduces bugs because Joe User installs that package which was uploaded to the repo, and finds that it won't work because it can't find a library. That means that both his time and developer's time will be wasted in fixing and testing the issue. Thanks, Alex -- Alex Smith Frugalware Linux developer - http://www.frugalware.org
Hi! I think that the packager should be able to decide whether his package needs ldconfig-run or not, and fill in PKGBUILD to override the autodetection. If he cannot decide, he can set the safest option. You say, this option makes our life more difficult. I say: yes, and probably we can make more bugs after this, but mankind usually selects difficult but more efficient things over "simple" ones. (For example we are using cars, which are not the safest things, but they are efficient. Sorry for this stupid example ;-) And this works somehow in debian. Bye, Nagy Gabor
participants (5)
-
Aaron Griffin
-
Alex Smith
-
Dan McGee
-
kfs1＠online.no
-
Nagy Gabor