"Stefan van der Eijk" <[EMAIL PROTECTED]> writes: > Hello,
Hello, > I'm having some issues with urpmi lately. Take a look here: > > http://eijk.homelinux.org/build/cooker/urpmi/i586/k3b-0.9-2mdk There is a strange things with your rpm database, at the very beginning, libexpat0 is installed and in another transaction later, fontconfig and libfontconfig1 installation fails due to libexpat.so.0 not available which are provided by the previous installation of libexpat0 normally... > As you can see 2 packages are basically being installaed by urpmi: > > kdelibs-devel (fails) and libcdda0-devel (is installed) > > I can't judge the interals of urpmi (I'm a no-no in perl) but the > difference I see with urpmi shipped with 9.1 is that urpmi now splits up > the job into smaller bits and tries to install those, seperately. > > The nice things of the version shipped with 9.1 are: > - it does the calculation first; > - if something isn't right then it gives the reason why something has failed; > - when something is wrong in the calculation it doesn't install any packages. You can get the previous behaviour by using --split-level 0 > With the current version it goes ahead and tries to install packages > anyway, leaving the user with without the package he wanted to install. > There is also no clear error message that the package he was trying to > install didn't make it, or why it didn't make it. > > There are some messages (for the individual packages) why they didn't make > it, but these often overlap --> because package A failed to install, > package B also failed to install. Instead of giving the combined error, > which in most cases will be LESS, now the user can go ahead and puzzle > what is going on. I will get back the whole error list at the end, this should make things more clear ? > May I ask what the long term strategy is for urpmi? Are these issues known > and being worked on? Is there a possibility to get the 9.1 version back? Snif, this split has been added to help upgrading a whole box when /var is not the big, or to help upgrading when cooker is moving during the upgrade (because it could be a long time to download packages and some packages may have disapeared during the download, causing the whole upgrade to fails. With the newer system, some packages have been upgraded (but the whole database is kept consistent, even if some transaction have failed), now an urpmi error code is returned describing correctly what has occurred. It may not be synced with all urpmi.8 man pages, but here is a list of error code due to installation failed (from the C man pages) : 10 Some files are missing for installation. 11 Some transactions failed but not all. 12 All transactions failed. 13 Some files are missing and some transactions failed but not all. 14 Some files are missing and all transactions failed. When error code is 0, everything worked correctly, ie all transactions where successfull. > Why I'm saying this: urpmi isn't helping me (with slbd) at this momment. > I'm capable to install kdelibs-devel by hand on a system and the system > ends up without any dependency issues, and only with cooker packages. Why > can't urpmi do the same? It _used_ to be possible... It could be a bug, there were many bugs in perl-URPM recently which caused weird behaviour of urpmi. Regards, Fran�ois. > PS: this issue isn't limited to kdelibs-devel. The number of packages that > can't be installed due to urpmi not being able to install the > BuildRequires is increasing, on a daily basis. See: > > Packages with "missing BuildRequires": > http://eijk.homelinux.org/build/cooker/i586/ > > Urpmi install logs: > http://eijk.homelinux.org/build/cooker/urpmi/i586/ I have read some error, some are caused by libexpat.so.0 or similar problems, I don't know what is going wrong, but packages seems not to be installed correctly, there could be a much stronger bug somewhere else. It is the same with libfontconfig.so.1, the error is propagating with some others as well, it looks like transaction error, assuming the install of the packages were effective (which seems to be). Fran�ois.
