[Fedora-packaging] Package rename breakage - wat do?

Toshio Kuratomi a.badger at gmail.com
Thu May 2 18:03:19 UTC 2013


On Thu, May 02, 2013 at 08:36:37AM -0400, Stephen Gallagher wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> I accidentally replied only to the node mailing list, but this is my
> interpretation of the situation:
> 
> On 05/02/2013 07:41 AM, Stephen Gallagher wrote:
> > On 05/02/2013 06:55 AM, Tom Hughes wrote:
> >> On 02/05/13 08:38, T.C. Hollingsworth wrote:
> > 
> >>> So, should we: 1. provide a compatibility symlink so older 
> >>> versions of nodejs-request aren't broken at all
> >>> 
> >>> 2. add Conflicts to nodejs-cookie-jar specifiying the older 
> >>> versions of nodejs-request that it will break
> >>> 
> >>> 3. do nothing; if you install only part of a single bodhi
> >>> update and it breaks you get to keep both pieces
> > 
> > 
> > FYI, this is not technically permissible. The packaging guidelines 
> > require clean upgrade paths. If you need to install multiple pieces
> > of a Bodhi update for things to work, they need to be arranged
> > with Requires/Obsoletes/Conflicts appropriately so that they are
> > pulled in automatically. Anything else is a bug in the packaging.
> > 
I'm not sure about this... I seem to recall other cases where the FPC ruled
that end users *could* install pieces of an update and get broken systems.
The right thing to do was to update all of the pieces instead.

If you could point to some guidelines that imply the opposite it might be
that there's some finer grained distinction when it's appropriate and when
it's not.  Or alternately, someone needs to reopen some tickets saying that
the precedent actually is for us to take care of these.

Here's the latest example I could find:
https://fedorahosted.org/fpc/ticket/241

and Timestamp 17:47:49
http://meetbot.fedoraproject.org/fedora-meeting-1/2013-02-06/fpc.2013-02-06-17.02.log.html

This is slightly different as it is crossing a release boundary (f17+
updates to f18) but it's similar in nature.

> >> Another option is to say that a rename of a node module is not
> >> what the rename guidelines call a "compatible enough replacement"
> >> which would mean the new package would not provide the old name,
> >> but would still obsolete it.
> > 
> > 
> > That's not necessarily true. If it's a one-to-one replacement
> > (except for the name), then I'd suggest that option 1 is the best
> > plan for the short term, but that we should open Bugzilla tickets
> > against all known packages depending on the old name to update in
> > their next releases.
> > 
> > If just adding a symlink to make it a full replacement is enough to
> > do the job, that's definitely the least impact. No need to force
> > an update until the other package is ready for it.
> > 
One note here:  From the sounds of the original post, we're talking about a
very few (just one?) package here and that package has already been updated.
If that's true, adding a versioned Requires/Conflict pair or telling people
that all the pieces have to be updated seems like the right thing.


-Toshio
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.fedoraproject.org/pipermail/packaging/attachments/20130502/339fa9f7/attachment.sig>


More information about the packaging mailing list