Hi there,
after cursing my head off a few times when trying to help with debugging
the unstable dist due to really nasty problems with rather basic packages like
X or libstdc++ I'd like to make two proposals:
+ Firstly, shouldn't we assign, lets say five volunteers to every maintainer
of a base package (including X for that matter) to make sure those vital
packages have been run sucessfully on at least five different (preferrably
librarywise non-tweaked) systems before upload to master ?
I gained the impression that the systems used by core package developers
are for obvious reasons not quite what one would expect from vanilla
dpkg/apt/dselect upgrades, which might explain how the recent catastrophies
could happen. I know X is undergoing restructuring and there's a lot of
code turnover in glibc, but I still don't understand how packages working
on the developer's system can fail so badly on others unless we have
a conceptual mistake in the project here...
+ Secondly we need a crash-proof upgrading process. For instance we would need
a dpkg option 'Install with crash recovery' which repacks a package
to be upgraded and stores it for automatic downgrading to a working state
if the upgrade causes severe trouble... This would be a rather lengthy
process for large upgrades when the number of packages is large and also
takes up substantial space but I can't think of a better approach at the
moment. Much too often my systems were really screwed up after upgrading, as
the working .deb's of lower version number disappeared from all mirrors out
there.
I seem to remember repacking is already implemented so perhaps we should
just write a HOWTO on safely using the unstable distribution for Dev Corner.
Regards,
--
/(__ __|\ Lars Steinke, Research Student @
( \/ __)_ www.fmf.uni-freiburg.de, Germany
) (_____ / for PGP PKey and WWW-Page finger
/___________/ steinke@mibm.ruf.uni-freiburg.de
Attachment:
pgpcp5rv1nSnP.pgp
Description: PGP signature