-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Is known yet what will happen to /etc/redhat-release?
In case the 2nd Severn Beta will be the first Fedora Core "test" release, will /etc/redhat-release be gone already and reappear as /etc/fedora-release?
Sort of
$ cat /etc/fedora-release Fedora Core release 0.90.94 (Severn)
or similar? Or will there be a different transition?
Any known details would be interesting in order to prepare some src.rpms in Fedora.us's Severn repository for Cambridge.
- --
Michael Schwendt (ms-nospam-0306@arcor.de) said:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Is known yet what will happen to /etc/redhat-release?
In case the 2nd Severn Beta will be the first Fedora Core "test" release, will /etc/redhat-release be gone already and reappear as /etc/fedora-release?
Not just yet.
The final solution will probably be:
package: fedora-release provides: redhat-release
Has a /etc/fedora-release and an /etc/redhat-release symlink.
The second test release will just have a redhat-release package that has "Fedora Core" in the name. Didn't want to break things just yet. :)
Bill
What about the versioning? Will xxx-release continue to have rpm-monotonic larger VRs than the previous releases?
Some repositories use the version of redhat-release package to form tags ensuring rpm-safe upgrade paths from previous RH releases (which wouldn't be a bad policy for RH itself).
On Tue, Sep 23, 2003 at 07:28:58PM -0400, Elliot Lee wrote:
On Wed, 24 Sep 2003, Axel Thimm wrote:
What about the versioning? Will xxx-release continue to have rpm-monotonic larger VRs than the previous releases?
EpochVR yes, VR no.
The beta will have redhat-release-0.94-1 (epoch 1).
That is ugly in multiple ways. Leaving all other reasons for not using epochs aside, this will break all upgrade paths from embedded disttags like 7.3, 8.0 or 9. The logical conduction would be to have these repos also bump up epochs to ensure rpm upgradability or invent their own versioning.
Any way it will be creating confusion and disorder. ;)
Could a better mechanism be created before the 25th that enables disttags to be carried on? Please use no epoch/release, and make the version be (rpm-)bigger than "9". Also keep redhat-release as a proper package carrying that version information.
Background: ===========
Up to now RH was not often in the need to release multiple versions of the same package for different RH releases. Even if the package did not change its release tag was bumbed (rebuilt for xxx).
Errata were either based on different upstream sources, so no upgrade clashes could come up, while in the very rare cases, where exactly the same package was rebuilt for different RH release the package maintainer had to invent a scheme that worked in that timeframe (e.g. the kernel errata releases for RH 7.3/8.0/9).
3rd party repos have been using the idiom of keeping the same upstream release current in multiple RH releases more often. The only portable way up to now to do so was to inspect the rpm-version value of redhat-release. Tags were created like
foo-fooversion-foorelease.rh8.0.i386.rpm
etc., ensuring sane upgrade paths for users of the 3rd party repo. Here are some google pointers:
http://www.google.com/search?q=rpm+%22--qf%22+version+redhat-release
My plea is therefore to
o keep redhat-release as a package of its own, so that `rpm -q --qf "%{VERSION}" redhat-release' still works. o Make the version of this package rpm-monotonic without using epoch (or even release tags), e.g. use a version rpm-larger than "9". o Put redhat-release into rawhide carrying the anaconda version ...
The alternative is to break the upgrade path mechanism of several repos or single package sites, please don't.
On Wed, 2003-09-24 at 03:45, Axel Thimm wrote:
That is ugly in multiple ways. Leaving all other reasons for not using epochs aside, this will break all upgrade paths from embedded disttags like 7.3, 8.0 or 9. The logical conduction would be to have these repos also bump up epochs to ensure rpm upgradability or invent their own versioning.
o keep redhat-release as a package of its own, so that `rpm -q --qf "%{VERSION}" redhat-release' still works. o Make the version of this package rpm-monotonic without using epoch (or even release tags), e.g. use a version rpm-larger than "9". o Put redhat-release into rawhide carrying the anaconda version ...
First, I would encourage you, and anyone else doing something like this to strongly consider doing 'rpm -qf --qf "%{VERSION}" /etc/redhat-release', as checking for the redhat-release package will fail on RHEL.
Second, it shouldn't be a big deal to come up with a versioning scheme that will satisfy your needs. Epochs are definitely not the answer. Your best bet, if you require the new package to compare favorably -18.rh80 would be to do -18.1fc, as that would be "newer" according to rpm. Recent rpm versions always will sort a numeric character higher than an alphabetic character.
Thanks. Peter
On Wed, 2003-09-24 at 03:57, Peter Bowen wrote:
First, I would encourage you, and anyone else doing something like this to strongly consider doing 'rpm -qf --qf "%{VERSION}" /etc/redhat-release', as checking for the redhat-release package will fail on RHEL.
I wish we could settle on a standard /etc/release that would encode these things in a shell-friendly manner, like /etc/lsb-release does.
Wil
On Wed, Sep 24, 2003 at 06:57:38AM -0400, Peter Bowen wrote:
On Wed, 2003-09-24 at 03:45, Axel Thimm wrote:
That is ugly in multiple ways. Leaving all other reasons for not using epochs aside, this will break all upgrade paths from embedded disttags like 7.3, 8.0 or 9. The logical conduction would be to have these repos also bump up epochs to ensure rpm upgradability or invent their own versioning.
o keep redhat-release as a package of its own, so that `rpm -q --qf "%{VERSION}" redhat-release' still works. o Make the version of this package rpm-monotonic without using epoch (or even release tags), e.g. use a version rpm-larger than "9". o Put redhat-release into rawhide carrying the anaconda version ...
First, I would encourage you, and anyone else doing something like this to strongly consider doing 'rpm -qf --qf "%{VERSION}" /etc/redhat-release', as checking for the redhat-release package will fail on RHEL.
Thanks for the tip. (I wouldn't like to put RHEL packages in the same upgrade path line as rh/fc packages, so I would poll the version explicitely from the corresponding rpm file and use a different disttag alltogether like rhel3).
Second, it shouldn't be a big deal to come up with a versioning scheme that will satisfy your needs. Epochs are definitely not the answer. Your best bet, if you require the new package to compare favorably -18.rh80 would be to do -18.1fc,
This would rather read
-18.1fc0.94
(or whatever the fc release might be). Adding the usual repo id tag, you land at
-18.1fc0.94.at
This my-release-tag-is-longer-than-your-release-tag madness must stop. It will push repo maintainers to either drop upgrade paths, or to come up with different versioning, e.g. rh10, to keep their repo structure intakt.
as that would be "newer" according to rpm. Recent rpm versions always will sort a numeric character higher than an alphabetic character.
Which is also another problem with upgrade paths as the starting release may have the toggling alphanumeric rpm bug (I think anything <= rpm-4.1, so at least the whole 7.x series is affected - not that is really makes sense to jump from 7.x to current rh/fc/rawhide).
How many packages actually *need* these different disttags? I can usually install packages for RH8 just fine on Rawhide, even tho most of the underlying software has newer packages.
I'd wager only a handful of packages really rely on the core OS version. Most probably rely on other packages, i.e., GNOME - a GNOME app shouldn't depend on RH8 for GNOME2 and RH9 for GNOME2.2 and FC1 for GNOME2.4 - it should just depend on the GNOME version. After all, an app made for GNOME2.0 will work just fine on GNOME2.4, and thus RH8 GNOME apps should run on FC1 no problem.
Instead of hacking up the distro's cleanliness to support poor packaging practices, maybe the packaging practices should just be corrected...
(I'll admit I could easily be missing some piece of the puzzle - if I'm babbling in such a way, have my apologies for being clueless ^^; )
On Wed, 2003-09-24 at 00:45, Axel Thimm wrote:
Could a better mechanism be created before the 25th that enables disttags to be carried on? Please use no epoch/release, and make the version be (rpm-)bigger than "9". Also keep redhat-release as a proper package carrying that version information.
I see no reason not to continue incrementing the version number from "Red Hat Linux" and just use 10. Yes, there have been big changes with changing the name to Fedora Linux and changing the development model from a commerical product to a community project, but it isn't a totally new distribution and the versioning shouldn't start from zero.
Wil
Axel Thimm (Axel.Thimm@physik.fu-berlin.de) said:
The beta will have redhat-release-0.94-1 (epoch 1).
That is ugly in multiple ways. Leaving all other reasons for not using epochs aside, this will break all upgrade paths from embedded disttags like 7.3, 8.0 or 9.
It's a new release model, hence a new release number. Fedora Core 10 doesn't make sense.
Also keep redhat-release as a proper package carrying that version information.
The package will still provide redhat-release, use that. Obviously you haven't tried to build packages for Red Hat Enterprise Linux. :)
Bill
On Wed, Sep 24, 2003 at 02:27:23PM -0400, Bill Nottingham wrote:
Axel Thimm (Axel.Thimm@physik.fu-berlin.de) said:
The beta will have redhat-release-0.94-1 (epoch 1).
That is ugly in multiple ways. Leaving all other reasons for not using epochs aside, this will break all upgrade paths from embedded disttags like 7.3, 8.0 or 9.
It's a new release model, hence a new release number. Fedora Core 10 doesn't make sense.
It does, if you want to retain upgrade paths ...
Also keep redhat-release as a proper package carrying that version information.
The package will still provide redhat-release, use that. Obviously you haven't tried to build packages for Red Hat Enterprise Linux. :)
Do I hear someone willing to fund a license? ;)
Axel Thimm (Axel.Thimm@physik.fu-berlin.de) said:
It's a new release model, hence a new release number. Fedora Core 10 doesn't make sense.
It does, if you want to retain upgrade paths ...
What do you mean by 'upgrade paths'? All packages provided in the release, sort newer-or-equal, in the method used by all sane upgrade tools (E:V-R). So, I don't see how we're not retaining an upgrade path.
Bill
On Wed, Sep 24, 2003 at 04:34:50PM -0400, Bill Nottingham wrote:
Axel Thimm (Axel.Thimm@physik.fu-berlin.de) said:
It's a new release model, hence a new release number. Fedora Core 10 doesn't make sense.
It does, if you want to retain upgrade paths ...
What do you mean by 'upgrade paths'? All packages provided in the release, sort newer-or-equal, in the method used by all sane upgrade tools (E:V-R). So, I don't see how we're not retaining an upgrade path.
This has not to do with fc isolated, but with the supporting 3rd party repos. It was discussed further up this thread, but I will rephrase with an example. There exist more than one repository building rpms for several RH releases (say RH7.3, 8.0 and 9). In order to ensure upgrade paths for the users of this repository the packages are being called
foo-1.2.3-1.rh7.3.i386.rpm foo-1.2.3-1.rh8.0.i386.rpm foo-1.2.3-1.rh9.i386.rpm
or similar variations like omitting the "rh" etc. Upgrades from one RH release with enabled repo to another were thus safe.
Just to give more food for thought: How would you version a kernel based on the same sources released for RH 7.x,8.0,9 and now additionally fc?
kernel-2.4.22-1.2082.7.i686.rpm kernel-2.4.22-1.2082.8.i686.rpm kernel-2.4.22-1.2082.9.i686.rpm kernel-2.4.22-1.2082.0.94.i686.rpm
The latter looses. You either have to rethink the first three or version the last with something rpm-higher than "9". Or start epochin all such packages occuring on multiple releases, which for somerpeos means all of the carried packages.
You cannot save the above scheme with epochs. See the previous messages in the thread with the "Background" information to see why this didn't hit RH so often in the past, but possibly will do yo rather often with shorter release cycles.
Or one decides to not support upgrade from older RH releases concerning non-RH repos, which would be unfortunate.
On Wednesday 24 September 2003 15:39, Axel Thimm wrote:
The latter looses. You either have to rethink the first three or version the last with something rpm-higher than "9". Or start epochin all such packages occuring on multiple releases, which for somerpeos means all of the carried packages.
Or you physically segment your updates directories.
Axel Thimm (Axel.Thimm@physik.fu-berlin.de) said:
Just to give more food for thought: How would you version a kernel based on the same sources released for RH 7.x,8.0,9 and now additionally fc?
kernel-2.4.22-1.2082.7.i686.rpm kernel-2.4.22-1.2082.8.i686.rpm kernel-2.4.22-1.2082.9.i686.rpm kernel-2.4.22-1.2082.0.94.i686.rpm
The latter looses. You either have to rethink the first three or version the last with something rpm-higher than "9". Or start epochin all such packages occuring on multiple releases, which for somerpeos means all of the carried packages.
'loses', not 'looses'. :)
You're arguing, as best I can tell, that it will break on upgrading some packages that haven't even been built yet, that are building by some automated release-querying script, correct? Because, if you're doing this versioning by hand, it's a non-starter. Anything you're modifying by hand can be modified differently.
Such a script can be *easily* modifed to work with Fedora Core in this case. It can key off of:
a) 'Fedora Core' in /etc/redhat-release b) the presence of /etc/fedora-release c) the fact that fedora-release provides /etc/redhat-release
Note that b) and c) will only be true in test release 3 and later.
Bill
On Wed, Sep 24, 2003 at 06:58:58PM -0400, Bill Nottingham wrote:
Axel Thimm (Axel.Thimm@physik.fu-berlin.de) said:
Just to give more food for thought: How would you version a kernel based on the same sources released for RH 7.x,8.0,9 and now additionally fc?
kernel-2.4.22-1.2082.7.i686.rpm kernel-2.4.22-1.2082.8.i686.rpm kernel-2.4.22-1.2082.9.i686.rpm kernel-2.4.22-1.2082.0.94.i686.rpm
The latter looses. You either have to rethink the first three or version the last with something rpm-higher than "9". Or start epochin all such packages occuring on multiple releases, which for somerpeos means all of the carried packages.
'loses', not 'looses'. :)
You're arguing, as best I can tell, that it will break on upgrading some packages that haven't even been built yet, that are building by some automated release-querying script, correct?
No, this is current practice on several 3rd party repos for RedHat including the one I maintain, as well as fedora itself ... (See also the google link in my first post pointing to ~200 similar sites)
Because, if you're doing this versioning by hand, it's a non-starter. Anything you're modifying by hand can be modified differently.
Well, I have heard of scripting before ...
Such a script can be *easily* modifed to work with Fedora Core in this case. It can key off of:
a) 'Fedora Core' in /etc/redhat-release b) the presence of /etc/fedora-release c) the fact that fedora-release provides /etc/redhat-release
Note that b) and c) will only be true in test release 3 and later.
You are missing the point, the "script" is rpm/apt/yum/<your favourite package resolver here>
Here is what is current practice for the upgraders (as opposed to "freshinstallers"): o Install RH8.0 o Install apt/yum o Use repo A, B and C (pointing to its RH8.0 section) o User decides to upgrade to RH9 o Upgrade the system with RH9 CDs o Adjust the apt/yum configuration to the new version o Update the system against the repos for RH9
There is no space for a custom script, unless you are thinking of anaconda specials, which is not desired.
On Thu, 2003-09-25 at 02:01, Axel Thimm wrote:
On Wed, Sep 24, 2003 at 06:58:58PM -0400, Bill Nottingham wrote:
Such a script can be *easily* modifed to work with Fedora Core in this case. It can key off of:
a) 'Fedora Core' in /etc/redhat-release b) the presence of /etc/fedora-release c) the fact that fedora-release provides /etc/redhat-release
Note that b) and c) will only be true in test release 3 and later.
You are missing the point, the "script" is rpm/apt/yum/<your favourite package resolver here>
Here is what is current practice for the upgraders (as opposed to "freshinstallers"): o Install RH8.0 o Install apt/yum o Use repo A, B and C (pointing to its RH8.0 section) o User decides to upgrade to RH9 o Upgrade the system with RH9 CDs o Adjust the apt/yum configuration to the new version o Update the system against the repos for RH9
There is no space for a custom script, unless you are thinking of anaconda specials, which is not desired.
There is a space for custom scripts, in the spec generation process. It is clear that you already have some custom script to help automate building packages for all the different distro versions. Just modify the script to look for one of the things that Bill mentioned, and either internally add 9 to the version or do something else to cause rpmvercmp to sort your EVR before the older EVR. No one is suggesting that rpm/yum/up2date/anaconda/apt/red-carpet should be modified in any way.
Thanks. Peter
Hi,
On Mi, 24 Sep 2003, Bill Nottingham sent incredible lines: [...]
Such a script can be *easily* modifed to work with Fedora Core in this case. It can key off of:
a) 'Fedora Core' in /etc/redhat-release b) the presence of /etc/fedora-release c) the fact that fedora-release provides /etc/redhat-release
[...]
why not /etc/release??
Fedora/Red Hat is part of the content of /etc/redhat-release. So I don't see the sense of this construct but I see an advantage of /etc/release if other distros will follow your this approach.
... may the Tux be with you! =Thomas=
On Oct 13, 2003, Thomas Bendler ml@bendler-net.de wrote:
why not /etc/release??
One reason I can think of is that some random package for another distro might fail to install because it happens to require a <distro>-release package, that would then conflict with the original <distro>-release should both of them try to install different /etc/release files. I admit this is a convoluted scenario that I don't think I've ever seen in practice. Besides, Requiring a specific <distro>-release is probably bad practice anyway, so the packager deserves the incompatibility (too bad it's the installer that gets punished :-(
That said, I do like /etc/release, at least as a soft-link, maybe not owned by RPM (to avoid the conflict described above)
Before this thread vanishes into nirvana:
I'd like to kindly request to set to release version to "10" or something higher than "9.0.94".
It would neither hurt marketing goals, nor any technical issues, but it helps a lot with 3rd party repos supporting more than just the latest release. Embedding (rpm-)increasing tags into the package release ensures sane upgrade paths (similar to some Red Hat errata releases like the kernel rpms). If you have questions on the technical background follow up the thread.
If there isn't any compelling reason for setting the release version to "1", please consider helping 3rd party repos with a sane release version. After all part of the new open source movement is to better support external developers and packagers.
Thank you.
On Thu, Sep 25, 2003 at 01:39:08AM +0300, Axel Thimm wrote:
This has not to do with fc isolated, but with the supporting 3rd party repos. It was discussed further up this thread, but I will rephrase with an example. There exist more than one repository building rpms for several RH releases (say RH7.3, 8.0 and 9). In order to ensure upgrade paths for the users of this repository the packages are being called
foo-1.2.3-1.rh7.3.i386.rpm foo-1.2.3-1.rh8.0.i386.rpm foo-1.2.3-1.rh9.i386.rpmor similar variations like omitting the "rh" etc. Upgrades from one RH release with enabled repo to another were thus safe.
Just to give more food for thought: How would you version a kernel based on the same sources released for RH 7.x,8.0,9 and now additionally fc?
kernel-2.4.22-1.2082.7.i686.rpm kernel-2.4.22-1.2082.8.i686.rpm kernel-2.4.22-1.2082.9.i686.rpm kernel-2.4.22-1.2082.0.94.i686.rpm
The latter looses. You either have to rethink the first three or version the last with something rpm-higher than "9". Or start epochin all such packages occuring on multiple releases, which for somerpeos means all of the carried packages.
On Tue, 30 Sep 2003, Axel Thimm wrote:
It would neither hurt marketing goals, nor any technical issues, but it helps a lot with 3rd party repos supporting more than just the latest release. Embedding (rpm-)increasing tags into the package release ensures sane upgrade paths (similar to some Red Hat errata releases like the kernel rpms).
With the Epoch: (which went from 0 to 1 on the redhat-release package), everything works out fine and is rpm-increasing.
-- Elliot Individualists, unite!
On Tue, Sep 30, 2003 at 07:31:59AM -0400, Elliot Lee wrote:
On Tue, 30 Sep 2003, Axel Thimm wrote:
It would neither hurt marketing goals, nor any technical issues, but it helps a lot with 3rd party repos supporting more than just the latest release. Embedding (rpm-)increasing tags into the package release ensures sane upgrade paths (similar to some Red Hat errata releases like the kernel rpms).
With the Epoch: (which went from 0 to 1 on the redhat-release package), everything works out fine and is rpm-increasing.
You have to do it for each package in a multiple concurrent release supporting repo that exists for multiple releases of RH. Read up this thread for details and examples.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Wed, 1 Oct 2003 12:12:47 +0300, Axel Thimm wrote:
With the Epoch: (which went from 0 to 1 on the redhat-release package), everything works out fine and is rpm-increasing.
You have to do it for each package in a multiple concurrent release supporting repo that exists for multiple releases of RH. Read up this thread for details and examples.
How many packages would that be? Consider the package naming scheme of fedora.us which uses the following target release tags:
rh80 < rh90 < rh90.93 < 0.94
No problems during upgrades. The other example (kernel) doesn't hold true for Fedora Core, because 0.94 contains a kernel version which is higher than Red Hat Linux 9, 8.0 and older releases:
2.4.20-20.9 < 2.4.22-1.2061.nptl 2.4.20-20.8 < 2.4.22-1.2061.nptl 2.4.20-20.7 < 2.4.22-1.2061.nptl
Even if there were a kernel upgrade for RHL<=9, I'm sure it would see a lower package release version than what is in Fedora Core, e.g. 2.4.22-2 in Fedora Core and 2.4.22-1.9 in Shrike Updates.
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Wed, Oct 01, 2003 at 01:56:24PM +0200, Michael Schwendt wrote:
On Wed, 1 Oct 2003 12:12:47 +0300, Axel Thimm wrote:
With the Epoch: (which went from 0 to 1 on the redhat-release package), everything works out fine and is rpm-increasing.
You have to do it for each package in a multiple concurrent release supporting repo that exists for multiple releases of RH. Read up this thread for details and examples.
How many packages would that be?
How many? All the ones from my repo for instance. Same for other repos. Some of the repos have intertwining dependencies that would also need to be updated.
No, I won't enter that epoch hell, and probably neither will any other repo maintainer.
Consider the package naming scheme of fedora.us which uses the following target release tags:
rh80 < rh90 < rh90.93 < 0.94
That's a kludge (BTW it should have been rh8.0 < rh9 < rh9.0.93). What about repos not using the rh tag and what about repos not willing to give up identification of the rpm as belonging to the rh family? What about Fedora Legacy, should part of it be distro-tagged and some not (like your example above)?
No problems during upgrades. The other example (kernel) doesn't hold true for Fedora Core, because 0.94 contains a kernel version which is higher than Red Hat Linux 9, 8.0 and older releases:
2.4.20-20.9 < 2.4.22-1.2061.nptl 2.4.20-20.8 < 2.4.22-1.2061.nptl 2.4.20-20.7 < 2.4.22-1.2061.nptl
Of course it works when the package version is newer, I am talking of keeping the same package built for different releases (e.g. built form the same specfile). Version bumping is the reason, why RH didn't encounter that problem that often (other than the few common errata).
Even if there were a kernel upgrade for RHL<=9, I'm sure it would see a lower package release version than what is in Fedora Core, e.g. 2.4.22-2 in Fedora Core and 2.4.22-1.9 in Shrike Updates.
Rawhide kernel specfile even has special precautions for RH7 (sevenexcompat macro), so the current intention obviously is to continue the common errata method, if the severn kernel turns out to be stable enough. I pitty the kernel rpm maintainer for having to come up with yet-another-custom-made-versioning-scheme.
Even if this should not be the case with the kernel rpm, it was only a RH-internal example to visualize to RH devels what exists in numbers of several hunderds in other repos. Any kludgy version numbering trick for the next common kernel rpm release may work for that one package, but we need a solution covering the general case.
On Wed, 2003-10-01 at 11:00, Axel Thimm wrote:
How many packages would that be?
How many? All the ones from my repo for instance. Same for other repos. Some of the repos have intertwining dependencies that would also need to be updated.
Well, this is a piece of legacy that you have created for yourself. You have opted to use an unreliable way of providing "upgradeability" for the packages you ship that was never guaranteed to be a foolproof solution. It is not fair to demand that this system should be continued simply because you made a choice that might have seemed a good one in the past, but ended up creating unnecessary legacy out of which you cannot get out (or are unwilling).
It was never guaranteed that the versioning of releases would remain consistent, and most packagers who have seen the jump from 8.0 to 9 have quickly realized that making the OS version part of their release is a very icky way of guaranteeing package upgradeability.
You've dug this hole for yourself, and I do not think it is fair to expect us all to jump down it.
Regards,
On Wed, Oct 01, 2003 at 12:36:34PM -0400, Konstantin Riabitsev wrote:
On Wed, 2003-10-01 at 11:00, Axel Thimm wrote: Well, this is a piece of legacy that you have created for yourself. You have opted to use an unreliable way of providing "upgradeability" for the packages you ship that was never guaranteed to be a foolproof solution. It is not fair to demand that this system should be continued simply because you made a choice that might have seemed a good one in the past, but ended up creating unnecessary legacy out of which you cannot get out (or are unwilling).
You are getting me wrong here. I simply present a solution for concurrent repositories, e.g. having the same package (sourcewise) in RH7.3, RH8.0, RH9 and the next N releases. This is a problem that the Fedora Legacy will face sooner or later, and it's better faced now, where things can get fixed.
If one is interested in keeping sane upgrade paths between the supported releases, one will find the only way to do so is to generate rpms with VRs following this scheme:
foo-1.2.3-4<suffix1> foo-1.2.3-4<suffix2> foo-1.2.3-4<suffix3>
Making sure the ordering of the suffixN corresponds to the rpm upgrade path algorithm.
The natural way to create this suffix is to use the distro versioning number, but this has now broken. Note: not only for my repo, but for others as well, including fedora.us!
fedora.us is saving their day with a kludge: While previously the distro release was prepended by alphabetic characters ("rh"), they skip it for the new releases, as anything starting with "f" is rpm-older.
But this is already broken for Fedora Legacy, as older rpms do not sort properly numbers and alphabetic entries (its a bug fixed only rather recently). And even if it were, it still is a kludge to remove the distro identification of the rpm.
It was never guaranteed that the versioning of releases would remain consistent, and most packagers who have seen the jump from 8.0 to 9 have quickly realized that making the OS version part of their release is a very icky way of guaranteeing package upgradeability.
You've dug this hole for yourself, and I do not think it is fair to expect us all to jump down it.
Please see it constructively. Maybe you are not interested in supporting multiple distros, but other who are ("Fedora Lagacy") need a versioning scheme they can live with.
So please provide a better working scheme.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Wed, 1 Oct 2003 18:00:51 +0300, Axel Thimm wrote:
How many packages would that be?
How many? All the ones from my repo for instance. Same for other repos. Some of the repos have intertwining dependencies that would also need to be updated.
A few observations: In your repository I don't see a consistent platform specific release tag scheme. The release tag scheme for RHL 7.3 and RHL 8.0 differs, e.g.
rpm-4.2-1_15.rh7.3.at.src.rpm mplayer-0.90-0_16rh8.0at.src.rpm
The mplayer package for Shrike doesn't have platform tag at all. Also note the higher package release:
mplayer-0.90-18.athlon.rpm
For Fedora Core, it seems to be %{version}-%{release}.rh9.0.94.at style again.
Other packages don't have any such release tag at all, regardless of whether they are in RHL already or not. Your repository contains packages that override core components of RHL and don't allow for a clean upgrade path, e.g.
bash-doc-2.05b-23.2.athlon.rpm
Should have been something like
bash-doc-2.05b-0.23.2.athlon.rpm
so it is guaranteed that Red Hat's bash packages always are newer than yours.
I'm not familiar with your repository. But it looks like one of your bigger problems is lack of a consistent and RH/Fedora compatible package release versioning scheme.
What about repos not using the rh tag and what about repos not willing to give up identification of the rpm as belonging to the rh family?
When either party is "not willing to give up" something, that sets off the alarm. ;)
What about Fedora Legacy, should part of it be distro-tagged and some not (like your example above)?
I haven't seen any work on guidelines for Fedora Legacy yet. But it would sound reasonable, if 3rd party projects adapted Fedora's .fdr release tag and vepoch.
2.4.20-20.9 < 2.4.22-1.2061.nptl 2.4.20-20.8 < 2.4.22-1.2061.nptl 2.4.20-20.7 < 2.4.22-1.2061.nptl
Of course it works when the package version is newer, I am talking of keeping the same package built for different releases (e.g. built form the same specfile).
2.4.20-20.9 = kernel for Shrike 2.4.20-20.8 = kernel for Psyche 2.4.20-20.7 = kernel for Valhalla
doesn't differ much from
2.4.20-20.rh9 = kernel for Shrike 2.4.20-20.rh8 = kernel for Psyche 2.4.20-20.rh7 = kernel for Valhalla
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Wed, Oct 01, 2003 at 07:19:08PM +0200, Michael Schwendt wrote:
A few observations: In your repository I don't see a consistent platform specific release tag scheme.
check the dates and the discussion on fedora.us in March/April (yes, I was once a fedora member).
The fact is that there still is no versioning scheme one can rely upon. The scheme we discussed with fedora.us in March/April is now broken.
The main issue is, that RH need to come up with a consistent scheme (not only for fedora but also including fedora legacy), or bless an existing one.
Currently there is no schem that will work: o either the upgrade paths are broken, e.g. rh9 -> fc1/fdr1 o or rpm bugs are triggered, e.g. rhXX -> numbers o Red Hat policy is violated, e.g. rh9 -> rh10
I understand RH's marketing and company policy, I won't join the "free beer" group, but I would like the policy to not spoil technical issues.
What about Fedora Legacy, should part of it be distro-tagged and some not (like your example above)?
I haven't seen any work on guidelines for Fedora Legacy yet.
This thread _is_ about Fedora Lagacy ... ;)
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Mon, 6 Oct 2003 10:50:52 +0200, Axel Thimm wrote:
On Wed, Oct 01, 2003 at 07:19:08PM +0200, Michael Schwendt wrote:
A few observations: In your repository I don't see a consistent platform specific release tag scheme.
check the dates and the discussion on fedora.us in March/April (yes, I was once a fedora member).
Been there before I joined the fedora-devel@fedora.us list. Going through list archives is a time-consuming process. I prefer summaries of ready-to-use concepts which can be commented on as a whole. Continueing Fedora Core with Red Hat Linux specific release numbers does not sound reasonable. It is like an implicit epoch.
You don't answer why some of your packages have no distribution specific release tag at all and why other packages override versions found in Red Hat Linux.
The fact is that there still is no versioning scheme one can rely upon. The scheme we discussed with fedora.us in March/April is now broken.
What had been the plans for RHL 9 => RHL 10? I don't remember, maybe rh90 => rh100 or anything like that? And why is the "rh" in front of the version? And why is it "rh90.93" instead of "9.0.93.rh"? It's too late to change a broken system and additionally make it play nicely with future changes. But Fedora Core's start at "1" is the chance for a change. I'm not that interested in learning about old broken versions of RPM and trying to make them fit into a general concept nevertheless.
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Mon, Oct 06, 2003 at 05:51:07PM +0200, Michael Schwendt wrote:
On Mon, 6 Oct 2003 10:50:52 +0200, Axel Thimm wrote:
On Wed, Oct 01, 2003 at 07:19:08PM +0200, Michael Schwendt wrote:
A few observations: In your repository I don't see a consistent platform specific release tag scheme.
check the dates and the discussion on fedora.us in March/April (yes, I was once a fedora member).
Been there before I joined the fedora-devel@fedora.us list. Going through list archives is a time-consuming process. I prefer summaries of ready-to-use concepts which can be commented on as a whole. You don't answer why some of your packages have no distribution specific release tag at all and why other packages override versions found in Red Hat Linux.
What I wanted to say is that the versioning scheme evolved, partly on its own, partly in discussion within (the old) fedora. So when you find inconsistency in the repo you are seeing at different layers of history. (I am hesitating to rebuild packages simply for re-versioning them to avoid bandwidth consumption at end users, especially while the versioning scheme was/is still at flux).
Continueing Fedora Core with Red Hat Linux specific release numbers does not sound reasonable. It is like an implicit epoch.
OTOH you are right, but if the packages from the Fedora Project are to be related to Red Hat Linux ones (e.g. you want them to be rpm-newer), you need to come up with a compatible scheme.
Did you check Warren's proposal on doing the reverse? E.g. instead of continuing RHL versioning into Fedora, one can back-continue Fedora guidelines to the past, e.g. treat RHL8.0 as Fedora Core 0.8.0.
The fact is that there still is no versioning scheme one can rely upon. The scheme we discussed with fedora.us in March/April is now broken.
What had been the plans for RHL 9 => RHL 10?
On fedora.us? I think the final consent was to "wait 'til it comes, it's always different than you planned". It turns out to be more than right. But that has to be fixed now, otherwise we will be having the same problems with "Fedora New Core XP2004" ... ;)
I don't remember, maybe rh90 => rh100 or anything like that? And why is the "rh" in front of the version? And why is it "rh90.93" instead of "9.0.93.rh"? It's too late to change a broken system and additionally make it play nicely with future changes. But Fedora Core's start at "1" is the chance for a change. I'm not that interested in learning about old broken versions of RPM and trying to make them fit into a general concept nevertheless.
That's why I changed the Subject on the main thread to contain "Fedora Legacy". If one doesn't care about past releases, you don't see the problem.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Mon, 6 Oct 2003 20:47:48 +0200, Axel Thimm wrote:
A few observations: In your repository I don't see a consistent platform specific release tag scheme.
check the dates and the discussion on fedora.us in March/April (yes, I was once a fedora member).
Been there before I joined the fedora-devel@fedora.us list. Going through list archives is a time-consuming process. I prefer summaries of ready-to-use concepts which can be commented on as a whole. You don't answer why some of your packages have no distribution specific release tag at all and why other packages override versions found in Red Hat Linux.
What I wanted to say is that the versioning scheme evolved, partly on its own, partly in discussion within (the old) fedora. So when you find inconsistency in the repo you are seeing at different layers of history.
I fail to see that. Here's one of the packages for Shrike, which was touched long after the disttag discussions:
mplayer-0.90-18.athlon.rpm http://atrpms.physik.fu-berlin.de/dist/rh9/mplayer/mplayer-0.90-18.athlon.rp...
Or this one from July (0.9 = Shrike?):
perl-DateManip-5.42-0.9.noarch.rpm http://atrpms.physik.fu-berlin.de/dist/rh9/perl-DateManip/perl-DateManip-5.4...
Another one for Shrike, from the same period of time, but has a disttag:
libapt-pkg-devel-0.5.5cnc6-34.rh9.at.i386.rpm http://atrpms.physik.fu-berlin.de/dist/rh9/apt/libapt-pkg-devel-0.5.5cnc6-34...
So much about consistency. There are more examples like that. Back to the topic.
Continueing Fedora Core with Red Hat Linux specific release numbers does not sound reasonable. It is like an implicit epoch.
OTOH you are right, but if the packages from the Fedora Project are to be related to Red Hat Linux ones (e.g. you want them to be rpm-newer), you need to come up with a compatible scheme.
Packages in Fedora Core 1 will be newer than any packages in Red Hat Linux. Only a few cases (e.g. comps, maybe comps-extras, redhat-release => fedora-release) need special treatment (probably an increased epoch).
I understand that Red Hat have never supported an upgrade path from installations other than Red Hat Linux. The Fedora Project is a new thing. It doubt it changes history by treating old 3rd party repositories for Red Hat Linux as if they had been official Fedora Extras/Alternatives/Legacy repositories before. In other words, they are unsupported. Starting with Fedora Core 1 you may need to implement a compatible versioning scheme, i.e. bump the release versions or vepochs (if any) of all your old packages before you put them into your Fedora add-ons repository. As I've pointed out before, some of your package versions conflict with Red Hat Linux anyway (e.g. bash, perl-DateManip, mozilla).
Did you check Warren's proposal on doing the reverse? E.g. instead of continuing RHL versioning into Fedora, one can back-continue Fedora guidelines to the past, e.g. treat RHL8.0 as Fedora Core 0.8.0.
Yes, sounds good. It's not the first time a "0." prefix is useful.
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Mon, Oct 06, 2003 at 09:54:23PM +0200, Michael Schwendt wrote:
On Mon, 6 Oct 2003 20:47:48 +0200, Axel Thimm wrote:
A few observations: In your repository I don't see a consistent platform specific release tag scheme.
check the dates and the discussion on fedora.us in March/April (yes, I was once a fedora member).
Been there before I joined the fedora-devel@fedora.us list. Going through list archives is a time-consuming process. I prefer summaries of ready-to-use concepts which can be commented on as a whole. You don't answer why some of your packages have no distribution specific release tag at all and why other packages override versions found in Red Hat Linux.
What I wanted to say is that the versioning scheme evolved, partly on its own, partly in discussion within (the old) fedora. So when you find inconsistency in the repo you are seeing at different layers of history.
I fail to see that. Here's one of the packages for Shrike, which was touched long after the disttag discussions:
[senseless examples skipped]
Exactly what are you trying to prove? "My repo is more consistent than your repo"? Didn't I mention that there was evolution in the versioning scheme, _parts of which_ were discussed with fedora.us?
(and the examples are crap also, i.e. you are ranting about mozilla's versioning scheme, which is a verbatim copy of rawhide ...)
While the discussion with fedora.us back in March/April did create some first specifications, others and I broke with fedora.us due to the increased non-tolerance against other 3rd party repos.
So the maintainers of the old repos stepped back and kept and evolved their own versioning schemes (This is a bit oversimplified, in reality there were and are coordination efforts to keep the repos compatible).
Please don't repeat the same intolerance attitude.
So much about consistency. There are more examples like that. Back to the topic.
Continueing Fedora Core with Red Hat Linux specific release numbers does not sound reasonable. It is like an implicit epoch.
OTOH you are right, but if the packages from the Fedora Project are to be related to Red Hat Linux ones (e.g. you want them to be rpm-newer), you need to come up with a compatible scheme.
Packages in Fedora Core 1 will be newer than any packages in Red Hat Linux. Only a few cases (e.g. comps, maybe comps-extras, redhat-release => fedora-release) need special treatment (probably an increased epoch).
You trimmed (and maybe didn't read) the following from my previous reply: "That's why I changed the Subject on the main thread to contain "Fedora Legacy". If one doesn't care about past releases, you don't see the problem."
This applies here also, don't let the thread loose its topic.
Did you check Warren's proposal on doing the reverse? E.g. instead of continuing RHL versioning into Fedora, one can back-continue Fedora guidelines to the past, e.g. treat RHL8.0 as Fedora Core 0.8.0.
Yes, sounds good. It's not the first time a "0." prefix is useful.
I know, I introduced it into fedora.us for consistency, shortly before I gave up ...
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Mon, 6 Oct 2003 23:11:47 +0200, Axel Thimm wrote:
[senseless examples skipped]
Ha!
Exactly what are you trying to prove? "My repo is more consistent than your repo"?
I don't have a repository.
Didn't I mention that there was evolution in the versioning scheme, _parts of which_ were discussed with fedora.us?
(and the examples are crap also,
Wrong choice of words here for my taste. Aggression causes me to block off.
i.e. you are ranting about mozilla's versioning scheme, which is a verbatim copy of rawhide ...)
mplayer isn't.
While the discussion with fedora.us back in March/April did create some first specifications, others and I broke with fedora.us due to the increased non-tolerance against other 3rd party repos.
So the maintainers of the old repos stepped back and kept and evolved their own versioning schemes (This is a bit oversimplified, in reality there were and are coordination efforts to keep the repos compatible).
How is your current versioning scheme defined?
Packages in Fedora Core 1 will be newer than any packages in Red Hat Linux. Only a few cases (e.g. comps, maybe comps-extras, redhat-release => fedora-release) need special treatment (probably an increased epoch).
You trimmed (and maybe didn't read) the following from my previous reply: "That's why I changed the Subject on the main thread to contain "Fedora Legacy". If one doesn't care about past releases, you don't see the problem."
I refused to quote it. Does Fedora Legacy cover "old releases of Fedora Core" (quote from fedora.redhat.com) or also old release from Red Hat Linux or also pre-Fedora 3rd party repositories?
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Monday 06 October 2003 14:32, Michael Schwendt wrote:
I refused to quote it. Does Fedora Legacy cover "old releases of Fedora Core" (quote from fedora.redhat.com) or also old release from Red Hat Linux or also pre-Fedora 3rd party repositories?
Until pried otherwise from my fingers, Fedora Legacy will provide backports for 7.3 and 9 when they respectively go EOL.
On Mon, Oct 06, 2003 at 11:32:40PM +0200, Michael Schwendt wrote:
How is your current versioning scheme defined?
Extracting kernel-modules, beta/cvs and other non-trivial examples the versioning scheme I suggest to use (note not only for myself, and not only for RH/FC) is:
<name>-<upstream_version>-<buildnumber>_<disttag>_<optional repo id>
Where
o _ is a seperator, that could be "_", "." or under certain circumstances even ""
o disttag is a combination of a distid and a distversion, e.g. rh9, rhel3, fdr1, lsb1.3, mdk8, suse9 should be rpm-sortable within a family, e.g. rhl/fdr should be sortable, rhel also, but possible rhl/fdr and rhel should not compare, etc.
o repoid is an identity marker like "at", "fr", "dag", "fdr" They are at the least significant position to not have rpm sort on them. repoid is optional and should not be used by vendors or first tier packaging repos like FC. disttag
You trimmed (and maybe didn't read) the following from my previous reply: "That's why I changed the Subject on the main thread to contain "Fedora Legacy". If one doesn't care about past releases, you don't see the problem."
I refused to quote it.
No need to comment on that, I guess ...
Does Fedora Legacy cover "old releases of Fedora Core" (quote from fedora.redhat.com) or also old release from Red Hat Linux or also pre-Fedora 3rd party repositories?
On Mon, 6 Oct 2003, Axel Thimm wrote:
On Wed, Oct 01, 2003 at 07:19:08PM +0200, Michael Schwendt wrote:
A few observations: In your repository I don't see a consistent platform specific release tag scheme.
check the dates and the discussion on fedora.us in March/April (yes, I was once a fedora member).
The fact is that there still is no versioning scheme one can rely upon. The scheme we discussed with fedora.us in March/April is now broken.
The main issue is, that RH need to come up with a consistent scheme (not only for fedora but also including fedora legacy), or bless an existing one.
Currently there is no schem that will work: o either the upgrade paths are broken, e.g. rh9 -> fc1/fdr1 o or rpm bugs are triggered, e.g. rhXX -> numbers o Red Hat policy is violated, e.g. rh9 -> rh10
I understand RH's marketing and company policy, I won't join the "free beer" group, but I would like the policy to not spoil technical issues.
This has to do with marketing.
However, to stay in tune with the actual problem you're having....
Reading the thread, it is aparent to me that many people who read it and responded, did not completely listen to the problem which you are describing. Numerous people completely misunderstood what you are trying to say, or purposefully chose to not read everything and attempt to understand it. I read your emails fairly carefully however and believe I understand you correctly. Without renaming or reversioning the distribution (which is just straight plain and simple not going to happen period for any reason short of an Act of God(TM)), there is another extremely simple solution. Simply add an additional digit in front of the normal modifier you use to indicate your special packages.
For example:
RHL 8.0: foo-1.2.3-3.rh80 RHL 9 : foo-1.2.3-3.rh90 FC 1 : foo-1.2.3-3.1fc1 or foo-1.2.3-3.0fc1 or foo-1.2.3-3.1.fc1
Using this scheme, all of the packages with fc1 in their names are considered newer by RPM than the RHL 9 or 8.0 packages above them. No Epoch specifier is needed, no reason that the distribution naming/versioning scheme we've chosen to go with will break your packages. It may require some getting used to, but the problem is trivially solveable by you for your package repositories, and by others with their package repositories.
The reason this works, is because when rpm goes to compare 2 packages to see which is newer, it compares based on name + epoch + version + release. The version and release fields are compared like this (this is fairly oversimplified):
It breaks the release field into multiple parts on field boundaries. A field boundary occurs when the next character transitions from alpha to numeric or numeric to alpha, or by an optional "." or "_" delimiter character[1]. Numerals are considered higher than alphabetic characters in a given field number.
So taking the 3 examples above and splitting them up into fields and comparing them, we have:
Release Field1 Field2 Field3 Field4 3.rh80 3 rh 80 N/A 3.rh90 3 rh 90 N/A 3.0fc1 3 0 fc 1 3.1fc1 3 1 fc 1 3.1.fc1 3 1 fc 1
Field 1 matches for all of them so comparison proceeds to field 2, in which case 1 is newer than 0, which is newer than "rh", so the fc1 packages are all newer than both the rh80 and rh90 packages.
Problem solved.
Hope this helps. TTYL
[1] There may be other delimiters I'm forgetting also, but I'm not trying to be 100% complete here to what the rpm source code handles, just trying to give a fairly terse example. In general, it is best to stick to using only numeric and alphabetic characters and ".". When in doubt, consult RPM documentation, and source code for the complete gory details if one requires more advanced knowledge than my terse description provides.
On Mon, 6 Oct 2003, Mike A. Harris wrote:
I understand RH's marketing and company policy, I won't join the "free beer" group, but I would like the policy to not spoil technical issues.
This has to do with marketing.
^ "nothing"
Oops, I left that word out.
On Mon, Oct 06, 2003 at 03:50:49PM -0400, Mike A. Harris wrote:
On Mon, 6 Oct 2003, Axel Thimm wrote:
The fact is that there still is no versioning scheme one can rely upon. The scheme we discussed with fedora.us in March/April is now broken.
The main issue is, that RH need to come up with a consistent scheme (not only for fedora but also including fedora legacy), or bless an existing one.
Currently there is no schem that will work: o either the upgrade paths are broken, e.g. rh9 -> fc1/fdr1 o or rpm bugs are triggered, e.g. rhXX -> numbers o Red Hat policy is violated, e.g. rh9 -> rh10
Reading the thread, it is aparent to me that many people who read it and responded, did not completely listen to the problem which you are describing. Numerous people completely misunderstood what you are trying to say, or purposefully chose to not read everything and attempt to understand it. I read your emails fairly carefully however and believe I understand you correctly. Without renaming or reversioning the distribution (which is just straight plain and simple not going to happen period for any reason short of an Act of God(TM)), there is another extremely simple solution. Simply add an additional digit in front of the normal modifier you use to indicate your special packages.
For example:
RHL 8.0: foo-1.2.3-3.rh80 RHL 9 : foo-1.2.3-3.rh90 FC 1 : foo-1.2.3-3.1fc1 or foo-1.2.3-3.0fc1 or foo-1.2.3-3.1.fc1
This breaks with all rpm versions shipped until the beginning of this year (including RH8.0). On rpm-list we rediscussed in January the rpm non-symmetric bug (or letter-number toggling bug) that invalidated ordering of segments of different types.
While there do exist backported rpm versions with this bug fixed on rpm.org, they have not been approved to enter RH errata, yet, and a Legacy project needs to take into account non-updated setups.
Therefore this solution, as well as dropping the letters completely (which would have worked with the same semantics) is flawed for a Legacy project (See also the renamed "Fedora and Fedora Legacy package versioning schemes", where this is discussed in more detail.)
Thanks for the suggestion, though, and for taking your time to go through it. As you noticed, only few do.
Currently I think the best way would be for RH to allow to tag the old releases with tags like 0.7.3, 0.8.0, 0.9 to make the versioning ordered again.
Using this scheme, all of the packages with fc1 in their names are considered newer by RPM than the RHL 9 or 8.0 packages above them. No Epoch specifier is needed, no reason that the distribution naming/versioning scheme we've chosen to go with will break your packages. It may require some getting used to, but the problem is trivially solveable by you for your package repositories, and by others with their package repositories.
The reason this works, is because when rpm goes to compare 2 packages to see which is newer, it compares based on name + epoch
- version + release. The version and release fields are compared
like this (this is fairly oversimplified):
It breaks the release field into multiple parts on field boundaries. A field boundary occurs when the next character transitions from alpha to numeric or numeric to alpha, or by an optional "." or "_" delimiter character[1]. Numerals are considered higher than alphabetic characters in a given field number.
So taking the 3 examples above and splitting them up into fields and comparing them, we have:
Release Field1 Field2 Field3 Field4 3.rh80 3 rh 80 N/A 3.rh90 3 rh 90 N/A 3.0fc1 3 0 fc 1 3.1fc1 3 1 fc 1 3.1.fc1 3 1 fc 1
Field 1 matches for all of them so comparison proceeds to field 2, in which case 1 is newer than 0, which is newer than "rh", so the fc1 packages are all newer than both the rh80 and rh90 packages.
Problem solved.
Hope this helps. TTYL
[1] There may be other delimiters I'm forgetting also, but I'm not trying to be 100% complete here to what the rpm source code handles, just trying to give a fairly terse example. In general, it is best to stick to using only numeric and alphabetic characters and ".". When in doubt, consult RPM documentation, and source code for the complete gory details if one requires more advanced knowledge than my terse description provides.
On Mon, Oct 06, 2003 at 10:50:40PM +0200, Axel Thimm wrote:
While there do exist backported rpm versions with this bug fixed on rpm.org, they have not been approved to enter RH errata, yet, and a Legacy project needs to take into account non-updated setups.
Of course, whatever the naming scheme is, one of the first Legacy updates should be a version of rpm that doesn't have random failure modes requiring manual intervention (i.e. the one on ftp.rpm.org or fedora),
On Tue, Oct 07, 2003 at 12:15:21AM +0300, Pekka Pietikainen wrote:
On Mon, Oct 06, 2003 at 10:50:40PM +0200, Axel Thimm wrote:
While there do exist backported rpm versions with this bug fixed on rpm.org, they have not been approved to enter RH errata, yet, and a Legacy project needs to take into account non-updated setups.
Of course, whatever the naming scheme is, one of the first Legacy updates should be a version of rpm that doesn't have random failure modes requiring manual intervention (i.e. the one on ftp.rpm.org or fedora),
Yes, most certainly, but you do see the chicken-and-egg problem? Better to not depend on the order of the packages installed.
(OT: Unfortunately all current rpm versions still have the "random failure modes" you refer to, even in severn2. This is probably the reason for rpm not making it to errata. Unfortunately the bug is very difficult to reproduce, usually happens during large updates).
On Oct 6, 2003, Axel Thimm Axel.Thimm@physik.fu-berlin.de wrote:
Yes, most certainly, but you do see the chicken-and-egg problem? Better to not depend on the order of the packages installed.
Err... Isn't this what rpm dependencies are for?
Sure enough, you may have to run rpm -F more than once, to get the rpm updated in the first round, and everything else later, but I can't see other problems. Do you?
On Mon, Oct 06, 2003 at 08:31:56PM -0300, Alexandre Oliva wrote:
On Oct 6, 2003, Axel Thimm Axel.Thimm@physik.fu-berlin.de wrote:
Yes, most certainly, but you do see the chicken-and-egg problem? Better to not depend on the order of the packages installed.
Err... Isn't this what rpm dependencies are for?
Sure enough, you may have to run rpm -F more than once, to get the rpm updated in the first round, and everything else later, but I can't see other problems. Do you?
If rpm itself is broken and needs to be upgraded before performing any further steps, you loose. Unless of course you make all packages depend on rpm >= 4.1.1, but even then a buggy rpm would sort the dependencies according to its buggy ordering algorithm and install the wrong bits.
To summarize the problem: Some suggestions for versioning concurrent packages for different RH/FC releases were to use the recent letters-are-older-than-numbers idiom from rpm, e.g.
foo-1.2.3-4_rh9 < foo-1.2.3-4_1 or foo-1.2.3-4_rh9 < foo-1.2.3-4_0fdr1
which is broken for rpm up to the one shipped with RH8.0.
I think the best suggestion was to set
"Fedora Core 0.<value>" := "Red Hat Linux <value>",
so that the resulting disttags compare letters to letters and numbers to numbers (so the rpm non-symmetry bug is not triggered), and the disttags sort still in chronological order, e.g.
foo-1.2.3-4_fdr0.9 < foo-1.2.3-4_fdr1
Unfortunately nobody from RH commented on this. Bill?
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Mon, 6 Oct 2003 15:50:49 -0400 (EDT), Mike A. Harris wrote:
Reading the thread, it is aparent to me that many people who read it and responded, did not completely listen to the problem which you are describing. Numerous people completely misunderstood what you are trying to say, or purposefully chose to not read everything and attempt to understand it. I read your emails fairly carefully however and believe I understand you correctly.
Interesting observation. But the solution you've come up is one that has been covered before. It's used by fedora.us, and Axel calls it a "kludge".
rh73 < rh80 < rh90 < 0.94 < 1 rh73 < rh80 < rh90 < 1
Whether you add something after the 0.94 (or 1), doesn't matter.
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Mon, Oct 06, 2003 at 11:02:54PM +0200, Michael Schwendt wrote:
On Mon, 6 Oct 2003 15:50:49 -0400 (EDT), Mike A. Harris wrote:
Reading the thread, it is aparent to me that many people who read it and responded, did not completely listen to the problem which you are describing. Numerous people completely misunderstood what you are trying to say, or purposefully chose to not read everything and attempt to understand it. I read your emails fairly carefully however and believe I understand you correctly.
Interesting observation. But the solution you've come up is one that has been covered before. It's used by fedora.us, and Axel calls it a "kludge".
Please Michael, stop misquoting me, it has become a bad habit of yours. While it definitely is a kludge, you (deliberately?) skip my main concerns about this scheme being broken by all rpm versions up to this January, making this idiom unsuitable for "Fedora Legacy".
You know the war 1870/71 between the German states and France was triggered because the Prussian cancler deliberately misquoted statements of the French king towards the Prussian king?
rh73 < rh80 < rh90 < 0.94 < 1 rh73 < rh80 < rh90 < 1
Whether you add something after the 0.94 (or 1), doesn't matter.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Tue, 7 Oct 2003 00:02:20 +0200, Axel Thimm wrote:
Please Michael, stop misquoting me, it has become a bad habit of yours. While it definitely is a kludge, you (deliberately?) skip my main concerns about this scheme being broken by all rpm versions up to this January, making this idiom unsuitable for "Fedora Legacy".
Rest assured, I'm out of this thread. I'm back when there are serious attempts at developing a versioning scheme for Fedora Legacy, such as using %{release}.0.X.Y (where 0.X.Y is e.g. 0.7.3 for Valhalla and 0.9 for Shrike), instead of trying to find a compromise for current 3rd party repositories.
My final comments here. You're complicating matters. I'm aware that for older versions of RPM (<= 4.1 or 4.1.1, I don't know),
rh73 < 1fdr *and* 1fdr < rh73
so that would be a problem for upgrade chanells which don't update RPM to a better version. But where has been defined that Fedora Legacy will carry packages with rh* disttags? Fedora Legacy has other problems that must be considered. Obviously, packages for a Fedora Legacy supported distribution must have a higher version than all previous packages (E:V-R based version comparison) for that distribution *and* all packages of older distributions. At the same time, Fedora Legacy update packages for one distribution may not have a higher version than stock packages of newer distributions, so the upgrade path is not disturbed. You can only avoid that with backported fixes, which keep the V in E:V-R below what is shipped with later distributions, or if you break upgradability. Whether or not that may be necessary also depends on the feasibility of backporting security fixes in time.
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Tue, Oct 07, 2003 at 01:13:49AM +0200, Michael Schwendt wrote:
Rest assured, I'm out of this thread. I'm back when there are serious attempts at developing a versioning scheme for Fedora Legacy, such as using %{release}.0.X.Y (where 0.X.Y is e.g. 0.7.3 for Valhalla and 0.9 for Shrike), instead of trying to find a compromise for current 3rd party repositories.
You possibly haven't read the thread ...
My final comments here. You're complicating matters. I'm aware that for older versions of RPM (<= 4.1 or 4.1.1, I don't know),
rh73 < 1fdr *and* 1fdr < rh73
so that would be a problem for upgrade chanells which don't update RPM to a better version. But where has been defined that Fedora Legacy will carry packages with rh* disttags?
Where has anything concerning Fedora Legacy been defined? Isn't that what this list members are supposed to develop?
Fedora Legacy has other problems that must be considered. Obviously, packages for a Fedora Legacy supported distribution must have a higher version than all previous packages (E:V-R based version comparison) for that distribution *and* all packages of older distributions. At the same time, Fedora Legacy update packages for one distribution may not have a higher version than stock packages of newer distributions, so the upgrade path is not disturbed.
You are describing the same problem.
You can only avoid that with backported fixes, which keep the V in E:V-R below what is shipped with later distributions, or if you break upgradability. Whether or not that may be necessary also depends on the feasibility of backporting security fixes in time.
Not true, that's exactly when an rpm-sortable disttag in the release tag would help. Check the rest of the thread.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Tue, 7 Oct 2003 08:36:06 +0200, Axel Thimm wrote:
Fedora Legacy has other problems that must be considered. Obviously, packages for a Fedora Legacy supported distribution must have a higher version than all previous packages (E:V-R based version comparison) for that distribution *and* all packages of older distributions. At the same time, Fedora Legacy update packages for one distribution may not have a higher version than stock packages of newer distributions, so the upgrade path is not disturbed.
You can only avoid that with backported fixes, which keep the V in E:V-R below what is shipped with later distributions, or if you break upgradability. Whether or not that may be necessary also depends on the feasibility of backporting security fixes in time.
Not true, that's exactly when an rpm-sortable disttag in the release tag would help. Check the rest of the thread.
Wrong. Please consider thinking about it a bit longer before you pipe out your usual quick replies.
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
On Monday 06 October 2003 23:36, Axel Thimm uttered:
Where has anything concerning Fedora Legacy been defined? Isn't that what this list members are supposed to develop?
Dude, I've replied to you a couple times about this...
http://www.fedora.us/wiki/FedoraLegacy
Legacy will have 7.3 and 9 support initially, possibly more.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Tue, 7 Oct 2003 07:44:17 -0700, Jesse Keating wrote:
On Monday 06 October 2003 23:36, Axel Thimm uttered:
Where has anything concerning Fedora Legacy been defined? Isn't that what this list members are supposed to develop?
Dude, I've replied to you a couple times about this...
http://www.fedora.us/wiki/FedoraLegacy
Legacy will have 7.3 and 9 support initially, possibly more.
No secret. The list announcements and things like that are everywhere.
It may sound like splitting hairs, but do you have official confirmation from Red Hat or the Fedora Project committee that "Fedora (!) Legacy" will carry updates for old release of _Red Hat_ (!) Linux?
You try to fit RHL Legacy under the Fedora Legacy umbrella.
Fedora Legacy -- as defined at fedora.redhat.com -- and a project that extends the Legacy concept to Red Hat Linux, are two different things, albeit with a similar goal.
- --
Michael Schwendt wrote:
It may sound like splitting hairs, but do you have official confirmation from Red Hat or the Fedora Project committee that "Fedora (!) Legacy" will carry updates for old release of _Red Hat_ (!) Linux?
There is a new Fedora-legacy-list:
Fedora Legacy list list is for the discussion of the Fedora Legacy project. This project exists to maintain security/bug backports for EOLd Red Hat Linux and Fedora Core releases.
https://lists.dulug.duke.edu/mailman/listinfo/fedora-legacy-list
On Tuesday 07 October 2003 11:51, Michael Schwendt wrote:
On Tue, 7 Oct 2003 07:44:17 -0700, Jesse Keating wrote:
On Monday 06 October 2003 23:36, Axel Thimm uttered:
Where has anything concerning Fedora Legacy been defined? Isn't that what this list members are supposed to develop?
Dude, I've replied to you a couple times about this...
http://www.fedora.us/wiki/FedoraLegacy
Legacy will have 7.3 and 9 support initially, possibly more.
No secret. The list announcements and things like that are everywhere.
It may sound like splitting hairs, but do you have official confirmation from Red Hat or the Fedora Project committee that "Fedora (!) Legacy" will carry updates for old release of _Red Hat_ (!) Linux?
Fedora Legacy was coined by Warren of Fedora.US. Warren was doing nothing with Legacy, mostly because it was there for the community to pick up. I've picked it up, and with Warren's blessings, I'm running with it. There isn't much to "confirm" with Red Hat, legacy is what the community wants Legacy to be.
You try to fit RHL Legacy under the Fedora Legacy umbrella.
Fedora Legacy -- as defined at fedora.redhat.com -- and a project that extends the Legacy concept to Red Hat Linux, are two different things, albeit with a similar goal.
The definition of Fedora Legacy at fedora.redhat.com needs to be updated. It was made long ago, prior to my involvement with the project.
On Tue, Oct 07, 2003 at 07:44:17AM -0700, Jesse Keating wrote:
On Monday 06 October 2003 23:36, Axel Thimm uttered:
Where has anything concerning Fedora Legacy been defined? Isn't that what this list members are supposed to develop?
Dude, I've replied to you a couple times about this...
Thanks for the pointer (are you sure you are not mistaken me with someone else? We didn't have that much communication until now).
Legacy will have 7.3 and 9 support initially, possibly more.
That's what I also thought. Since RHL is the precursor of Fedora Core/Project, one could define RHL <= 9 as "Fedora prenatal release" and say that Fedora Legacy will be supporting (part of) the EOL'd Fedora releases (including the prenatal ones ;).
On Sep 24, 2003, Bill Nottingham notting@redhat.com wrote:
It's a new release model, hence a new release number. Fedora Core 10 doesn't make sense.
FWIW, I don't see why not. dBase II and RHEL 2.1 come to mind :-)
On Thu, 2003-09-25 at 12:06, Roozbeh Pournader wrote:
On Thu, 2003-09-25 at 03:44, Alexandre Oliva wrote:
FWIW, I don't see why not. dBase II and RHEL 2.1 come to mind :-)
Or worse than that. The first version of LaTeX was called 2.09!
I think it's a good idea, especially a nice random sounding number like "2.09", just for the Discordian value.
Metacity did the same thing; from its README: "The first release of Metacity was version 2.3. Metacity has no need for your petty hangups about version numbers."
Bwahahaha...
"less" has a great scheme, too - a wholly integer based release numbering system. The current version of less is 378. :)
- jck
On Sep 25, 2003 at 12:21, Jens Knutson in a maddening rage wrote:
On Thu, 2003-09-25 at 12:06, Roozbeh Pournader wrote:
On Thu, 2003-09-25 at 03:44, Alexandre Oliva wrote:
FWIW, I don't see why not. dBase II and RHEL 2.1 come to mind :-)
Or worse than that. The first version of LaTeX was called 2.09!
I think it's a good idea, especially a nice random sounding number like "2.09", just for the Discordian value.
How about 8.1. This would surely satisfy those ppl who were waiting for RH8.1 and got RH9.
I am changing the subject to avoid loosing the context again. I am also top posting, as the quoted part below is for reference only and not a reply. I hope this longish post clarifies the problem and stimulates more constructive discussions (no, the thread until now was not constructive ;)
If you are a "Fedora Legacy" champion take your time please to get through it. I hope the problem becomes more transparent that way, and that a good solution is found before everything gets hit in stone (e.g. November).
Thank you for your time!
The problem: ============
The aim is to have packages for Red Hat Linux and Fedora (Core) which should be naturally (speak rpm-wise) upgradeable. For Red Hat's packages from release to release this is ensured by
o Updated upstream versions o Updated release tags due to different packaging or simply due to rebuilding for the current release o Epoch bumping where the above is not possible, e.g. downgrading a buggy package.
When common errata were issued, either the packages were having different upstream versions, or the errata was using an ad hoc method for ensuring upgrade paths, usually with suffixes derived from the distro version, like .7x, .80, .90, or .7, .8 and .9.
There was no policy on how to deal with it, as it was not really an issue within RH.
The true trouble comes up, when you want to maintain the same package for multiple different releases, e.g. RH7.3, RH8.0, RH9, FC/FDR 1 etc.
This is a common task in the "Fedora Legacy" subcomponent.
These packages will have the same upstream version and will (whenever possible) be based on the same sources (same SourceN/PatchN/specfile).
This had been discussed on fedora.us half a year before. The viable solutions that emerged were to have suffixes at the rpm release tag that order the rpms according to the distro version, e.g.
foo-1.2.3-4_rh7.3 foo-1.2.3-4_rh8.0 foo-1.2.3-4_rh9
There are variations on this scheme, like dropping "rh" or using 73, 80, 90 instead of 7.3, 8.0 and 9.
The current decision to rewind the distro release is breaking such schemes, and "Fedora Legacy" like repos need a solution.
Defining the problem domain for Fedora Legacy: ==============================================
a) We need a versioning scheme that ensures upgradability without obfuscating too much the VR scheme. b) It should be possible to use the same sources/patches/specfiles with macro controlled differences c) It should work with older rpm versions, e.g. don't let rpm compare numbers with letters.
The easiest way to achieve a) and b) is to have
Release: 4_%{disttag}
in the specfile and let %{disttag} expand to values in the right rpm-order for the given distributions.
A proposal on the fedora.us side is to allow rpm to compare "rh9" versus "1", which under newer rpms will resolve as older, but which is not the case for RH7.x (possibly even stock RH8.0, but I may be mistaken). There are updated rpm rpms fixing this issue, but they are neither available via RHN, nor can Fedora Legacy assume that those rpms are available on the system to be upgraded.
Therefore it is far better to compare numbers only with rpm.
So we need a versioning scheme f that numerically sorts
... < f("7.3") < f("8.0") < f("9") < f("1") < ...
Possible solutions: ===================
a) Call _internally_ the next release ("Fedora [Core] 1") "RawHide 10" or "RawHide 9.1". It looks compatible with RH's policy and would allow for tags like "rh10" to complete the above scheme.
b) Drop upgradability withing Fedora Legacy from RHL to FC/FDR.
c) Drop upgradability from rpm < 4.1 and use kludgy "rhnumber" < "number" idiom.
d) Use tags with something larger than "rh", e.g. "tfp" ("The Fedora Project")
X) Your suggestion here.
It is important to have Red Hat's consent on any sollution, as well as a common versioning scheme across repositories of different maintainers to allow for properly formulating cross-repository dependencies and having easy identification by end users.
Optional good-to-have semantics: ================================ o Use the same versioning for Fedora.
It would be good to have a scheme applicable to both Fedora Legacy as well as Fedora (current release). That way Fedora releases could automatically become Fedora Legacy without rebuilding and re-versioning.
o Use the same versioning for RawHide/Betas.
This would help building more often for RawHide. E.g. the following scheme could be used until now:
foo-1.2.3-4_rh8.0 foo-1.2.3-4_rh8.0.93 (beta) foo-1.2.3-4_rh8.0.94.20030325 (rawhide) foo-1.2.3-4_rh9 (release)
On Wed, Sep 24, 2003 at 10:45:45AM +0300, Axel Thimm wrote:
On Tue, Sep 23, 2003 at 07:28:58PM -0400, Elliot Lee wrote:
On Wed, 24 Sep 2003, Axel Thimm wrote:
What about the versioning? Will xxx-release continue to have rpm-monotonic larger VRs than the previous releases?
EpochVR yes, VR no.
The beta will have redhat-release-0.94-1 (epoch 1).
That is ugly in multiple ways. Leaving all other reasons for not using epochs aside, this will break all upgrade paths from embedded disttags like 7.3, 8.0 or 9. The logical conduction would be to have these repos also bump up epochs to ensure rpm upgradability or invent their own versioning.
Any way it will be creating confusion and disorder. ;)
Could a better mechanism be created before the 25th that enables disttags to be carried on? Please use no epoch/release, and make the version be (rpm-)bigger than "9". Also keep redhat-release as a proper package carrying that version information.
Background:
Up to now RH was not often in the need to release multiple versions of the same package for different RH releases. Even if the package did not change its release tag was bumbed (rebuilt for xxx).
Errata were either based on different upstream sources, so no upgrade clashes could come up, while in the very rare cases, where exactly the same package was rebuilt for different RH release the package maintainer had to invent a scheme that worked in that timeframe (e.g. the kernel errata releases for RH 7.3/8.0/9).
3rd party repos have been using the idiom of keeping the same upstream release current in multiple RH releases more often. The only portable way up to now to do so was to inspect the rpm-version value of redhat-release. Tags were created like
foo-fooversion-foorelease.rh8.0.i386.rpm
etc., ensuring sane upgrade paths for users of the 3rd party repo. Here are some google pointers:
http://www.google.com/search?q=rpm+%22--qf%22+version+redhat-release
My plea is therefore to
o keep redhat-release as a package of its own, so that `rpm -q --qf "%{VERSION}" redhat-release' still works. o Make the version of this package rpm-monotonic without using epoch (or even release tags), e.g. use a version rpm-larger than "9". o Put redhat-release into rawhide carrying the anaconda version ...
The alternative is to break the upgrade path mechanism of several repos or single package sites, please don't.
Axel Thimm writes:
c) Drop upgradability from rpm < 4.1 and use kludgy "rhnumber" < "number" idiom.
A little observation I haven't seen so far. (But I can have missed it in the long thread, sorry if so.):
It will matter that "rhnumber" < "number" only for systems from FC1 and on. And those systems come rpm > 4.2. Systems with rpm < 4.1 should never have to see the packages without rh.
So it ought to work out right in practice. It's still kludgy, though.
On Mon, Oct 06, 2003 at 12:04:44PM +0200, Göran Uddeborg wrote:
Axel Thimm writes:
c) Drop upgradability from rpm < 4.1 and use kludgy "rhnumber" < "number" idiom.
A little observation I haven't seen so far. (But I can have missed it in the long thread, sorry if so.):
It will matter that "rhnumber" < "number" only for systems from FC1 and on. And those systems come rpm > 4.2. Systems with rpm < 4.1 should never have to see the packages without rh.
Well, not for "Fedora Legacy", which is about supporting systems before FC1 (and later FC1 also etc.).
So it ought to work out right in practice. It's still kludgy, though.
On Mon, Oct 06, 2003 at 11:39:16AM +0200, Axel Thimm wrote:
A proposal on the fedora.us side is to allow rpm to compare "rh9" versus "1", which under newer rpms will resolve as older, but which is not the case for RH7.x (possibly even stock RH8.0, but I may be mistaken).
Small addendum: I just looked it up for another communication on fedora.us list. This bug was fixed 9 months ago in rpm 4.2-0.55:
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=50977#c20
So that means that even RH8.0 is affected. I assume the 4.1.1 backport available from rpm.org does contain the bug-fix, but I know nothing about 4.0.5.
There are updated rpm rpms fixing this issue, but they are neither available via RHN, nor can Fedora Legacy assume that those rpms are available on the system to be upgraded. [...] c) Drop upgradability from rpm < 4.1 and use kludgy "rhnumber" < "number" idiom.
This should then probably be "rpm < 4.1.1".
Anyway, the presence of this bug prohibits mixing numerical and alphabetical segments in rpm comparision for any system running an rpm from before 2003, unless rpm is upgraded to a version without this bug before any other upgrading operation.
Therefore IMHO mixing numerical and alphabetical segments in comparisons should be avoided for legacy considerations.
Le lun 06/10/2003 à 12:13, Axel Thimm a écrit :
On Mon, Oct 06, 2003 at 11:39:16AM +0200, Axel Thimm wrote:
Anyway, the presence of this bug prohibits mixing numerical and alphabetical segments in rpm comparision for any system running an rpm from before 2003, unless rpm is upgraded to a version without this bug before any other upgrading operation.
Therefore IMHO mixing numerical and alphabetical segments in comparisons should be avoided for legacy considerations.
At one point you should consider if providing a fixed rpm version for all affected systems and have all updates depend on this rpm version is not easier than trying to come with the perfect backwards-compatible macro.
Cheers,
Nicolas Mailhot wrote:
Le lun 06/10/2003 à 12:13, Axel Thimm a écrit :
On Mon, Oct 06, 2003 at 11:39:16AM +0200, Axel Thimm wrote:
Anyway, the presence of this bug prohibits mixing numerical and alphabetical segments in rpm comparision for any system running an rpm from before 2003, unless rpm is upgraded to a version without this bug before any other upgrading operation.
Therefore IMHO mixing numerical and alphabetical segments in comparisons should be avoided for legacy considerations.
At one point you should consider if providing a fixed rpm version for all affected systems and have all updates depend on this rpm version is not easier than trying to come with the perfect backwards-compatible macro.
+1 Exactly. And given rpm-4.1 from RH8 as an example, why would anyone WANT to not upgrade it... =)
Warren
On Mon, Oct 06, 2003 at 01:08:26AM -1000, Warren Togami wrote:
Nicolas Mailhot wrote:
Le lun 06/10/2003 à 12:13, Axel Thimm a écrit :
On Mon, Oct 06, 2003 at 11:39:16AM +0200, Axel Thimm wrote:
Anyway, the presence of this bug prohibits mixing numerical and alphabetical segments in rpm comparision for any system running an rpm from before 2003, unless rpm is upgraded to a version without this bug before any other upgrading operation.
Therefore IMHO mixing numerical and alphabetical segments in comparisons should be avoided for legacy considerations.
At one point you should consider if providing a fixed rpm version for all affected systems and have all updates depend on this rpm version is not easier than trying to come with the perfect backwards-compatible macro.
+1 Exactly. And given rpm-4.1 from RH8 as an example, why would anyone WANT to not upgrade it... =)
Then why isn't is offered on RHN?
Thank you for your well written proposal.
Axel Thimm wrote:
Defining the problem domain for Fedora Legacy:
a) We need a versioning scheme that ensures upgradability without obfuscating too much the VR scheme. b) It should be possible to use the same sources/patches/specfiles with macro controlled differences c) It should work with older rpm versions, e.g. don't let rpm compare numbers with letters.
a + b are perfectly valid, but I see no reason to continue to support the ugly behavior of c, and here's why:
1) jbj has long indicated that he wishes all distributions to use the same upgraded version of rpm in order to simplify maintenance. 2) Fedora Legacy is the time to FORCE everyone to upgrade RPM if they wish to continue to use the repository. This means that we will be able to avoid a lot of unpredictable behavior (#1) in older versions of rpm, making Legacy distributions easier to support in the long term. 3) RH 8.0 Psyche rpm *needs* to be upgraded anyway because the default rpm is too broken to use.
(Read the bottom of this message for a summary of my %{release} naming proposal.)
The easiest way to achieve a) and b) is to have
Release: 4_%{disttag}in the specfile and let %{disttag} expand to values in the right rpm-order for the given distributions.
A proposal on the fedora.us side is to allow rpm to compare "rh9" versus "1", which under newer rpms will resolve as older, but which is not the case for RH7.x (possibly even stock RH8.0, but I may be mistaken). There are updated rpm rpms fixing this issue, but they are neither available via RHN, nor can Fedora Legacy assume that those rpms are available on the system to be upgraded.
Therefore it is far better to compare numbers only with rpm.
So we need a versioning scheme f that numerically sorts
... < f("7.3") < f("8.0") < f("9") < f("1") < ...
Possible solutions:
a) Call _internally_ the next release ("Fedora [Core] 1") "RawHide 10" or "RawHide 9.1". It looks compatible with RH's policy and would allow for tags like "rh10" to complete the above scheme.
-1 Too confusing to have different numbers and names in different places.
b) Drop upgradability withing Fedora Legacy from RHL to FC/FDR.
-999 We should always maintain upgradability between distributions, and Fedora will strengthen this and prevent problems by stricter packaging standards than those enforced by RH internally in the past. There is currently absolutely no reason to break upgradability.
c) Drop upgradability from rpm < 4.1 and use kludgy "rhnumber" < "number" idiom.
-1
d) Use tags with something larger than "rh", e.g. "tfp" ("The Fedora Project")
-500 The whole reason why we had those dist tags was because fedora.us (or freshrpms, or dag, or axel, etc.) were not official repositories. Now Fedora is official, so I believe we can drop this ugliness that makes the package filename unnecessarily longer.
X) Your suggestion here.
It is important to have Red Hat's consent on any sollution, as well as a common versioning scheme across repositories of different maintainers to allow for properly formulating cross-repository dependencies and having easy identification by end users.
Optional good-to-have semantics:
o Use the same versioning for Fedora.
It would be good to have a scheme applicable to both Fedora Legacy as well as Fedora (current release). That way Fedora releases could automatically become Fedora Legacy without rebuilding and re-versioning.
+1
o Use the same versioning for RawHide/Betas.
This would help building more often for RawHide. E.g. the following scheme could be used until now:
foo-1.2.3-4_rh8.0 foo-1.2.3-4_rh8.0.93 (beta) foo-1.2.3-4_rh8.0.94.20030325 (rawhide) foo-1.2.3-4_rh9 (release)
+1 I dislike your specific examples that contain "rh", but the numbers I fully agree.
Warren's Release Naming Proposal (draft 1) ========================================== All distributions in Fedora Legacy must upgrade to the latest version of rpm for that distribution.
%{release} name structure ------------------------- X.%{non-numeric version}.%{distrel}
X RPM package release number, always increments with every update, even a rebuild. %{non-numeric version} Non-numeric characters at the end of upstream's %{version}. Since %{version} must (almost) never contain non-numeric characters in order to avoid the Epoch treadmill (like mozilla package), the non-numeric portion can optionally be contained within the %{release} tag after the release number. Examples: mozilla-1.5-5.rc2 foobar-2.0-7.cvs20031005
%{distrel} Upgrade Path from fedora.us -------------------------------------- 0.fdr.X.rh80 -> 0.fdr.X.rh90 -> 0.fdr.X.0.94 -> 0.94 -> 1
fedora.us currently has an even longer "0.fdr.X" prepended before the suggested fedora.redhat.com scheme. Thus absolutely anything containing numbers in the second glob will "win" against "fdr".
Examples: foobar-1.5-6.rc3_1.9.2 (Fedora Core 1.92 beta) foobar-1.5-6.rc3_1 (Fedora Core 1) foobar-1.5-6.rc3_0.9.4 (Fedora Core 0.94) foobar-1.5-6.rc3_0.8 (RHL 8.0 Psyche) foobar-1.5-6.rc3_0.7.3 (RHL 7.3)
Note that I used "_" instead of "." in this particular example. Either of them would work. The former would be a clearer seperator between the %{distrel} and the rest of the tag, while the latter is slightly easier to type. I personally don't care which one is used. Opinions?
Warren
On Mon, Oct 06, 2003 at 01:06:30AM -1000, Warren Togami wrote:
Axel Thimm wrote:
Defining the problem domain for Fedora Legacy:
a) We need a versioning scheme that ensures upgradability without obfuscating too much the VR scheme. b) It should be possible to use the same sources/patches/specfiles with macro controlled differences c) It should work with older rpm versions, e.g. don't let rpm compare numbers with letters.
a + b are perfectly valid, but I see no reason to continue to support the ugly behavior of c, and here's why:
- jbj has long indicated that he wishes all distributions to use the
same upgraded version of rpm in order to simplify maintenance. 2) Fedora Legacy is the time to FORCE everyone to upgrade RPM if they wish to continue to use the repository. This means that we will be able to avoid a lot of unpredictable behavior (#1) in older versions of rpm, making Legacy distributions easier to support in the long term. 3) RH 8.0 Psyche rpm *needs* to be upgraded anyway because the default rpm is too broken to use.
I partly agree, but imagine the first time initiation of a existing RH7.3 cluster of machines. The admin would have to manually update the rpm rpms and then point his up2date/yum/apt config to a "Fedora Legacy" repo.
Red Hat has been very conservative to replace the rpms in RH8.0 (and RH7.3) and even when a QA rpm release is pushed into RHN, "Fedora Legacy" should work with non-updated systems, too (just like you have to build apt/yum/up2date against a non-updated system).
a) Call _internally_ the next release ("Fedora [Core] 1") "RawHide 10" or "RawHide 9.1". It looks compatible with RH's policy and would allow for tags like "rh10" to complete the above scheme.
-1 Too confusing to have different numbers and names in different places.
Yes, but even Windows users could cope with different maket and technical versions, and technically it is the cleanest solution.
b) Drop upgradability withing Fedora Legacy from RHL to FC/FDR.
-999
I agree, but this is still a possible solution ;)
c) Drop upgradability from rpm < 4.1 and use kludgy "rhnumber" < "number" idiom.
-1
d) Use tags with something larger than "rh", e.g. "tfp" ("The Fedora Project")
-500 The whole reason why we had those dist tags was because fedora.us (or freshrpms, or dag, or axel, etc.) were not official repositories. Now Fedora is official, so I believe we can drop this ugliness that makes the package filename unnecessarily longer.
You are talking about the .fdr. tag, e.g. the repo tag. Drop that by all means. I am referring to the disttag like rh9 etc.
X) Your suggestion here.
It is important to have Red Hat's consent on any sollution, as well as a common versioning scheme across repositories of different maintainers to allow for properly formulating cross-repository dependencies and having easy identification by end users.
Optional good-to-have semantics:
o Use the same versioning for Fedora. [...]
+1
o Use the same versioning for RawHide/Betas. [...]
+1 I dislike your specific examples that contain "rh", but the numbers I fully agree.
Note that redhat/fedora packages are also distributed on larger mirrors containing Mandrake, suse etc. rpm. A lot of people get the wrong rpm. An identification string there is nice-to-have (but nothing more, the true problem lies not here and this should not distract).
Examples: foobar-1.5-6.rc3_1.9.2 (Fedora Core 1.92 beta) foobar-1.5-6.rc3_1 (Fedora Core 1) foobar-1.5-6.rc3_0.9.4 (Fedora Core 0.94) foobar-1.5-6.rc3_0.8 (RHL 8.0 Psyche) foobar-1.5-6.rc3_0.7.3 (RHL 7.3)
Note that I used "_" instead of "." in this particular example. Either of them would work. The former would be a clearer seperator between the %{distrel} and the rest of the tag, while the latter is slightly easier to type. I personally don't care which one is used. Opinions?
I am already using _ for that purpose, and will even ask Jeff in the future to give _ a higher seperating pririty than . (but that's another issue, don't want to get off-topic).
I see another suggestion in your examples (which I didn't clearly read through the description):
e) Prepend "0." to previous RH releases, e.g.
0.7.3 < 0.8.0 < 0.9 < 0.9.0.93 < ... < 1.0
I personally like that. so how about conaming "RHL X" to "Fedora Core 0.X" resulting to tags like foo-1.2.3-4_fdr0.7.3 (non-numeric upstream versions, kernel modules and other special cases aside)?
On Monday 06 October 2003 05:17, Axel Thimm uttered:
I personally like that. so how about conaming "RHL X" to "Fedora Core 0.X" resulting to tags like foo-1.2.3-4_fdr0.7.3 (non-numeric upstream versions, kernel modules and other special cases aside)?
I think the "fdr" part should be removed all together. Packages are kept in different directory trees for the relative distro. If some poor user pulls a rpm from a FedoraCore/ directory and expects it to work on Mandrake, thats his problem, not ours.
I've always disliked the text tags (rhl9, fdr, fr2, etc..) and I feel that they are indicitive of 3'rd party rpms. Fedora.us and Fedora Legacy are less 3'rd party and thus shouldn't suffer the headache of text tags.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Mon, 6 Oct 2003 14:17:14 +0200, Axel Thimm wrote:
d) Use tags with something larger than "rh", e.g. "tfp" ("The Fedora Project")
-500 The whole reason why we had those dist tags was because fedora.us (or freshrpms, or dag, or axel, etc.) were not official repositories. Now Fedora is official, so I believe we can drop this ugliness that makes the package filename unnecessarily longer.
You are talking about the .fdr. tag, e.g. the repo tag. Drop that by all means. I am referring to the disttag like rh9 etc.
Fedora.us was separate from Red Hat [Linux]. And fedora.us could have offered packages for Mandrake Linux (.mdk), for instance. Hence the distribution tag to emphasize the target platform.
The .fdr tag and the 0. release prefix should be adapted by 3rd party repositories. (http://fedora.redhat.com/participate/terminology.html)
- -- Michael, who doesn't reply to top posts and complete quotes anymore.
Le lun 06/10/2003 à 20:51, Michael Schwendt a écrit :
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On Mon, 6 Oct 2003 14:17:14 +0200, Axel Thimm wrote:
d) Use tags with something larger than "rh", e.g. "tfp" ("The Fedora Project")
-500 The whole reason why we had those dist tags was because fedora.us (or freshrpms, or dag, or axel, etc.) were not official repositories. Now Fedora is official, so I believe we can drop this ugliness that makes the package filename unnecessarily longer.
You are talking about the .fdr. tag, e.g. the repo tag. Drop that by all means. I am referring to the disttag like rh9 etc.
Fedora.us was separate from Red Hat [Linux]. And fedora.us could have offered packages for Mandrake Linux (.mdk), for instance. Hence the distribution tag to emphasize the target platform.
Well, we've been happily providing mandrake and redhat java packages without any distributions tags for months at jpackage.
People just put their distro-specific repository entries into apt and never ever see other packages. Same for ximian (of course for jpackage it helps 90% of the packages are noarch and have no overlap with whatever is already in the distributions, but anyway we've never needed any special flags to differenciate distro-specific packages).
And BTW I'm deeply sceptical about Fedora getting mandrake acceptance short term. urpmi means the mdk communities had an head start, and we're only now catching up with apt/yum. RedHat is pretty late in the community game.
Cheers,
Michael Schwendt wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Is known yet what will happen to /etc/redhat-release?
In case the 2nd Severn Beta will be the first Fedora Core "test" release, will /etc/redhat-release be gone already and reappear as /etc/fedora-release?
Sort of
$ cat /etc/fedora-release Fedora Core release 0.90.94 (Severn)
or similar? Or will there be a different transition?
Any known details would be interesting in order to prepare some src.rpms in Fedora.us's Severn repository for Cambridge.
-----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.3 (GNU/Linux)
iD8DBQE/cMik0iMVcrivHFQRAj6GAJoC2iiPUGkpBomgLnYtBJ5UCy3G7ACffma2 TrM3jChftfTtzwwTUVClpRM= =ljtj -----END PGP SIGNATURE-----
-- fedora-devel-list mailing list fedora-devel-list@redhat.com http://www.redhat.com/mailman/listinfo/fedora-devel-list
When the new beta comes out, is the rhn_applet gonna let us know?