Udeb handling by britney -- a proposal
Hi all,
Let me first briefly outline how britney currently handles udebs:
- Well, it doesn't really, there's simply a plain text file in heidi
format, called 'HeidiDI', that lists the testing versions for udebs,
and commits that daily to the database. That's it, it requires manual
updating.
This is of course not an ideal situation. Not only requires it manual
intervention every time something changes, it also makes the 'sources go
in along with all the corresponding binaries' property not true for d-i
binaries. When unstable slugs on, udeb's that one might want to have in
testing, could be already being removed due to this.
My proposal is to maintain the 'sources go in along with all
corresponding binaries' property for testing at all times. There are two
major cases for d-i binaries:
- Source packages with only udebs
- Source packages with both debs and udebs
Both could be handled just like the other source packages in principle,
but a number of assumptions why britney keeps testing in an mostly
releaseable state at all times don't go for udebs: dependencies are
often inadequate, and so is testing/filing of RC bugs on the correct
packages.
With the current hints available to the release team though, this can be
handled adequately and consistently by them, in stead of via HeidiDI. I
think that for now[1] udeb-only sources definitely should get a
semi-permanent block hint, and only updated manually via an unblock
hint. Mixed packages can go either way, if their maintainers are well
aware of the implications of modifying the udeb's, they can be handled
as any non-udeb related package. But when gearing up for a release, at
least when entering the stage where one thinks about release candidates,
those mixed packages are better off with block hints too (a partial
freeze).
Does anyone see any disadvanteage of this approach to udeb handling?
Otherwise, with a realatively simple patch to britney udeb disparities
will become something of the past, and the d-i team has by means of Joey
Hess the possibility to push the testing set of udeb's into shape
easily.
--Jeroen
[1] While britney cannot accurately enough propgate udeb's automatically
--
Jeroen van Wolffelaar
Jeroen@wolffelaar.nl (also for Jabber & MSN; ICQ: 33944357)
http://Jeroen.A-Eskwadraat.nl
Reply to: