Axel Thimm wrote:
On Mon, Dec 31, 2007 at 06:44:08AM -0600, Johnny Hughes wrote:
Everyone is entitled to their own opinion ... mine is that the yum from CentOS is a critical package and should not be replaced with out a very, very good reason. Yours is different. Neither is right or wrong ... they are just different approaches.
Sure, and ATrpms' policy wrt to RHEL is to move replacing packages into the testing repo until a better solution is found. Only that yum is not part of RHEL4 end even worse, different clones of RHEL use different yum versions and different sets of (sometimes home-made) plugins. E.g. when talking about "replacing base packages" in RHEL & clone worlds it becomes quite obfuscated.
The yum in ATrpms is there for two reasons:
o making sure RHEL users also have a yum, but more important o several yum bugs that are triggered by some ATrpms packages have been fixed in later yum version w/o a backport. ATM some plugins and repo policies had unconvered yet another pile of yum bugs that were fixed with 3.2.8 and affected many ATrpms packages (the infamous: "Your installed kernel is not installed" installonly bug).
We are working on a yum-3.2.8 version for CentOS-5 as well, as there is a major bug in the 3.0.x branch that causes problems with file paths used with file dependency calculations. However, just like we don't roll newer KDE changes back into CentOS-4 and CentOS-3, we will probably not upgrade the yum in centos-4 or centos-3 to yum-3.2.8.
We have added a new version of yum (the same version in centos-4) into the CentOS-3 CentOSPlus repo. The problem is that the new versions of yum require new versions of python ... and python is not something that we recommend updating .. EVER :)
That means that either we have to change the newer yum versions to work with the older python or create newer a python to use in conjunction with the older python on older versions of CentOS.
We are backporting things to older centos versions ... the c3 plus version of yum is one example, another is the new centos-metadata-parser rolled back into centos-4 from centos-5. However we do this with much testing and it takes much time to modify these to work with the older pythons.
But we have thousands of CentOS-3 users who are perfectly happy with the old yum 2.0.8 and it works for them, so we did not put the yum upgrade for CentOS-3 into base, but into centosplus.
We do this since one of the main purposes of Enterprise software, once it is installed, is to not break (or even change) anything it already does ... so there is a GOAL to not upgrade anything unless it is absolutely required once released.
So wrt to yum and a couple similar infratsructure packages it would be nice to have a canonical clone (e.g. from my POV a merged CentOS/SL universe, something I've been advocating) to define as base and then either invest in backporting bugfixes (which is difficult given that yum is at a fast pace and the authors almost never do backports), or update yum more often to remove these bugs (which involves testing yum on CentOS3-5).
We have to mix that with not changing any released package ... and if it changes, not changing the ABI/APIs but only the bugs. We can't break update scripts written to use our yum by people who have 10,000 clients because we want a shiny new feature ... or maybe even if we want to fix a bug.
But this gets off the centos-user list charter a bit, I do hope that working together on the merged 3rd party repo will have side-effects like bringing CentOS/SL even closer and at the end not even have to worry about 3rd party repos. Maybe we should move part of this discussions to other lists. As a short term fix we could discuss on centos/sl-devel whether a new common yum infrastructure could be shared by centos/sl and atrpms (and maybe other 3rd party repos, I think maybe Dag or Dries may have yum shipping, too) removing its own?
Yes, but their's (dag/dries) is older than the one in CentOS. I recommend using the priorities plugin against rpmforge and kbs-CentOS-Extras too.
I am an equal opportunity filterer :D
Thanks, Johnny Hughes