I'd like to put forth a proposal here, comments or other opinions very welcome. ;)
In EPEL4/5 we have a policy of: "EPEL won't ship anything that is in the Advanced Platform set of packages". This is easy to check, as all these have src.rpms on mirrors. This includes:
JBEAP JBEWS JBWFK os RHCERT RHDirServ RHDOCS RHEIPA RHEMRG RHHC RHNPROXY RHNSAT RHNTOOLS RHUI RHWAS SJIS
(see: http://mirrors.tummy.com/pub/ftp.redhat.com/pub/redhat/linux/enterprise/5Ser... )
For EPEL6, we can't do this as their is no Advanced Platform, and the secondary channels don't exist as src.rpms. We have:
6Server 6ComputeNode 6Client 6Workstation
and all the src.rpms in a single dir under those.
Additionally we have the following complications:
* Some packages only have binary rpms shipped for some arches. Ie, the entire virt stack is x86_64. There's no client/workstation stuff in ppc64. There's no java in ppc64.
* Some packages only have subpackages shipped in some arches (ie, pacemaker-cts and pacemaker-docs are shipped in server-optional, but the main pacemaker binary rpm is only in the HighAvailability channel.
I would like to propose the following:
EPEL6 will not ship any packages that have src.rpms on public mirrors under 6* directories with the following exception: If the binary rpm is only shipped in some arches in RHEL, EPEL may ship that exact same version (note that EPEL maintainer must keep up exactly with the RHEL src.rpm).
So, this would leave us with:
* someone could maintain java in EPEL and build the exact src.rpm version. If it took mods to work, I would say we should just not do so and excludearch our java stuff.
* folks could push packages that are x86_64 only into epel, but should keep them exactly the same as the rhel src.rpm.
* Items in other channels are fair game to ship in EPEL6.
Thoughts?
kevin
On Mon, Dec 20, 2010 at 01:41:25PM -0700, Kevin Fenzi wrote:
I'd like to put forth a proposal here, comments or other opinions very welcome. ;)
In EPEL4/5 we have a policy of: "EPEL won't ship anything that is in the Advanced Platform set of packages". This is easy to check, as all these have src.rpms on mirrors. This includes:
JBEAP JBEWS JBWFK os RHCERT RHDirServ RHDOCS RHEIPA RHEMRG RHHC RHNPROXY RHNSAT RHNTOOLS RHUI RHWAS SJIS
(see: http://mirrors.tummy.com/pub/ftp.redhat.com/pub/redhat/linux/enterprise/5Ser... )
For EPEL6, we can't do this as their is no Advanced Platform, and the secondary channels don't exist as src.rpms. We have:
6Server 6ComputeNode 6Client 6Workstation
and all the src.rpms in a single dir under those.
Additionally we have the following complications:
Some packages only have binary rpms shipped for some arches. Ie, the entire virt stack is x86_64. There's no client/workstation stuff in ppc64. There's no java in ppc64.
Some packages only have subpackages shipped in some arches (ie, pacemaker-cts and pacemaker-docs are shipped in server-optional, but the main pacemaker binary rpm is only in the HighAvailability channel.
I would like to propose the following:
EPEL6 will not ship any packages that have src.rpms on public mirrors under 6* directories with the following exception: If the binary rpm is only shipped in some arches in RHEL, EPEL may ship that exact same version (note that EPEL maintainer must keep up exactly with the RHEL src.rpm).
So, this would leave us with:
someone could maintain java in EPEL and build the exact src.rpm version. If it took mods to work, I would say we should just not do so and excludearch our java stuff.
folks could push packages that are x86_64 only into epel, but should keep them exactly the same as the rhel src.rpm.
Items in other channels are fair game to ship in EPEL6.
Thoughts?
I don't 100% like this but it seems like the best we can do with a messy situation. The only thought I have is we might want to modify some of this after we see what CentOS does. For instance, if they ship the virt stack on x86 but do have to make modifications to make it work there we should consider rebuilding with their packages or rebuilding with packages that are NEVR lower than the RHEL packages but include the CentOS changes.
-Toshio
On 12/20/2010 01:41 PM, Kevin Fenzi wrote:
I'd like to put forth a proposal here, comments or other opinions very welcome. ;)
Additionally we have the following complications:
Some packages only have binary rpms shipped for some arches. Ie, the entire virt stack is x86_64. There's no client/workstation stuff in ppc64. There's no java in ppc64.
Some packages only have subpackages shipped in some arches (ie, pacemaker-cts and pacemaker-docs are shipped in server-optional, but the main pacemaker binary rpm is only in the HighAvailability channel.
I would like to propose the following:
EPEL6 will not ship any packages that have src.rpms on public mirrors under 6* directories with the following exception: If the binary rpm is only shipped in some arches in RHEL, EPEL may ship that exact same version (note that EPEL maintainer must keep up exactly with the RHEL src.rpm).
So, this would leave us with:
someone could maintain java in EPEL and build the exact src.rpm version. If it took mods to work, I would say we should just not do so and excludearch our java stuff.
folks could push packages that are x86_64 only into epel, but should keep them exactly the same as the rhel src.rpm.
Using ExclusiveArch or ExcludeArch? Or just let them be in multiple repos assuming that they will really and truly be the same?
- Items in other channels are fair game to ship in EPEL6.
Thoughts?
kevin
Fine by me. My concern at the moment is a multilib issue - gcc-gfortran.i686 not being present in the x86_64 repository, which causes:
Broken deps for x86_64 ---------------------------------------------------------- getdata-devel-0.6.2-1.el6.i686 requires gcc-gfortran(x86-32)
Other than dropping the %{_isa} from the Requires, not sure there is anything else we can do. But I haven't seen much comment on it though
On Tue, 21 Dec 2010 12:31:30 -0800 Toshio Kuratomi a.badger@gmail.com wrote:
I don't 100% like this but it seems like the best we can do with a messy situation. The only thought I have is we might want to modify some of this after we see what CentOS does. For instance, if they ship the virt stack on x86 but do have to make modifications to make it work there we should consider rebuilding with their packages or rebuilding with packages that are NEVR lower than the RHEL packages but include the CentOS changes.
Yeah, sounds reasonable to me.
On Tue, 21 Dec 2010 13:32:58 -0700 Orion Poplawski orion@cora.nwra.com wrote:
Using ExclusiveArch or ExcludeArch? Or just let them be in multiple repos assuming that they will really and truly be the same?
Well, there's downsides to either. I guess just adding a ExclusiveArch isn't too much change, but it is change. Would it be bad just just import and rebuild them exactly from the src.rpm?
We should spell out what we want here exactly for sure.
Fine by me. My concern at the moment is a multilib issue - gcc-gfortran.i686 not being present in the x86_64 repository, which causes:
Broken deps for x86_64
getdata-devel-0.6.2-1.el6.i686 requires gcc-gfortran(x86-32)
Other than dropping the %{_isa} from the Requires, not sure there is anything else we can do. But I haven't seen much comment on it though
I don't see any other solution. We can't change the way RHEL composes their trees. I suppose you could file a bug and ask them to ship the 32bit one in the 64bit tree, but I am pretty sure they will just say no.
kevin
On Tue, 21 Dec 2010 15:32:04 -0700 Kevin Fenzi kevin@scrye.com wrote:
...snip...
Well, there's downsides to either. I guess just adding a ExclusiveArch isn't too much change, but it is change. Would it be bad just just import and rebuild them exactly from the src.rpm?
We should spell out what we want here exactly for sure.
Replying to myself here. Obviously we do have to modify them because the rhel ones have either Exclusivearch or Excludearch anyhow.
So, I think it would make sense to require adding:
ExclusiveArch: ppc64
to any rhel src.rpms that we rebuild for epel.
kevin
On Tuesday, December 21, 2010 04:57:53 pm Kevin Fenzi wrote:
On Tue, 21 Dec 2010 15:32:04 -0700 Kevin Fenzi kevin@scrye.com wrote:
...snip...
Well, there's downsides to either. I guess just adding a ExclusiveArch isn't too much change, but it is change. Would it be bad just just import and rebuild them exactly from the src.rpm?
We should spell out what we want here exactly for sure.
Replying to myself here. Obviously we do have to modify them because the rhel ones have either Exclusivearch or Excludearch anyhow.
So, I think it would make sense to require adding:
ExclusiveArch: ppc64
to any rhel src.rpms that we rebuild for epel.
kevin
We cant do that, because of the way koji works we would end up with only the ppc64 version available to build against, we build and ship it on all arches or none.
Dennis
On Tue, 21 Dec 2010 17:03:08 -0600 Dennis Gilmore dennis@ausil.us wrote:
We cant do that, because of the way koji works we would end up with only the ppc64 version available to build against, we build and ship it on all arches or none.
Ah right.
ok. That settles that.
kevin
On Tue, Dec 21, 2010 at 13:32, Orion Poplawski orion@cora.nwra.com wrote:
On 12/20/2010 01:41 PM, Kevin Fenzi wrote:
I'd like to put forth a proposal here, comments or other opinions very welcome. ;)
Question... who is using ppc64 on here to test/fix/help? From the sounds of it, openjdk is going to take a bunch of bootstrapping of building it from F12 rpms a couple of rounds until you have something 'working' on El6.
Does anyone have a need for it? [EG does IBM have some poor bas^W developer who can help us through this?]
Stephen John Smoogen píše v Út 21. 12. 2010 v 16:09 -0700:
On Tue, Dec 21, 2010 at 13:32, Orion Poplawski orion@cora.nwra.com wrote:
On 12/20/2010 01:41 PM, Kevin Fenzi wrote:
I'd like to put forth a proposal here, comments or other opinions very welcome. ;)
Question... who is using ppc64 on here to test/fix/help? From the sounds of it, openjdk is going to take a bunch of bootstrapping of building it from F12 rpms a couple of rounds until you have something 'working' on El6.
What about to import an older EL-6 build (java-1.6.0-openjdk-1.6.0.0-1.17.b16.el6 is built for all arches) into koji and then use it to build the latest one? We did that in Fedora/s390x.
Dan
epel-devel@lists.fedoraproject.org
-
Dan Horák -
Dennis Gilmore -
Kevin Fenzi -
Orion Poplawski -
Stephen John Smoogen -
Toshio Kuratomi