Hi Martin,
On Mon, Jul 25, 2011 at 9:02 PM, Martin Langhoff martin.langhoff@gmail.comwrote:
Peter, Fedorans,
I'm playing with test builds using F14 for arm. And I get strange errors of f14 python packages that depend on python 2.6. And yet, when I look at koji, it clearly has newer RPMs...
So for example, see python-decorator: http://arm.koji.fedoraproject.org/koji/packageinfo?packageID=7698
Right now, it lists python-decorator-3.2.0-2.fc14 pbrobinson 2011-07-23 10:55:56 python-decorator-3.2.0-3.fc14 pbrobinson 2011-05-25 19:47:30
The -3 package is newer, but was build earlier than the -2 package. Looking at the changelog, the -3 rev is where the Python 2.7 rebuild happened.
Now if I grab the pkg list, it offers python-decorator-3.2.0-2.fc14 http://arm.koji.fedoraproject.org/repos/dist-f14-build/latest/arm/pkglist
I've had similar problems with other Python packages. Anybody can help me understand why? Any fixes? Workarounds?
Its a known problem. Mash (or what ever it is that generates the koji build repo) uses some very random criteria for tagging packages into the build repo. This is causing problems for noarch packages that are just imported by build-previous as opposed to rebuilt esp for python. The problem is that some of the packages, like the one you mention, were rebuilt/imported explicitly to get the build root for koji up to a F-14 level. It should hopefully just disappear once we have build-previous go to completion. The real fix it to add a check for the script to check if a later release is already in the tree.
Peter
On Tue, Jul 26, 2011 at 4:04 AM, Peter Robinson pbrobinson@gmail.com wrote:
Its a known problem. Mash (or what ever it is that generates the koji build repo) uses some very random criteria for tagging packages into the build repo.
It looks like the criteria is "most recently built" - which isn't very useful.
My current workaround is to *also* use the dist-f14-py27-rebuild repo -- so it's not a crippling issue. cheers,
m
Martin Langhoff píše v Út 26. 07. 2011 v 10:02 -0400:
On Tue, Jul 26, 2011 at 4:04 AM, Peter Robinson pbrobinson@gmail.com wrote:
Its a known problem. Mash (or what ever it is that generates the koji build repo) uses some very random criteria for tagging packages into the build repo.
It looks like the criteria is "most recently built" - which isn't very useful.
the criteria is "most recently tagged" to be precise, the latest tagged build goes into the build repo
Dan
On Tue, Jul 26, 2011 at 3:10 PM, Dan Horák dan@danny.cz wrote:
Martin Langhoff píše v Út 26. 07. 2011 v 10:02 -0400:
On Tue, Jul 26, 2011 at 4:04 AM, Peter Robinson pbrobinson@gmail.com
wrote:
Its a known problem. Mash (or what ever it is that generates the koji
build
repo) uses some very random criteria for tagging packages into the
build
repo.
It looks like the criteria is "most recently built" - which isn't very
useful.
the criteria is "most recently tagged" to be precise, the latest tagged build goes into the build repo
There's also added confusion with inheritance, so that a newer tagged package on dist-f13 doesn't necessarily toast an older one on dist-f14.
Peter
-
Dan Horák -
Martin Langhoff -
Peter Robinson