After our first attempt at doing a 6.0 -> 6.1 upgrade
failed with a signal 11 and subsequent system halt at
the apparent end of the install-updated-RPMs stage (see
bug #6913), we felt we had little choice but to retry
the upgrade. Doing so resulted in a significantly
scrambled RPM database, to wit:
The crash of the first install process had apparently
left the old versions of a number of upgraded RPMs not
removed (at least in the database). When the installer
re-ran, it apparently decided that it needed to upgrade
them again (despite the presence of the upgraded versions
on the system), but then failed to remove the old versions
The result, after the upgrade 'completed', was that we
had an RPM database and a system with one old copy and
two new copies of the same package for quite a few
I think that the installer should make more effort to
handle previous installations that, for whatever reason
(ranging from an installer crash to an inopportune power
failure!) have aborted in the middle and are incomplete.
As a minimum I would expect it not to re-upgrade packages
when the upgraded package is on the system already (it
seems likely that this can happen normally, even without
the installer failing, with such packages as the kernel).
If the installer must bypass ordinary RPM checks, it should
make sure that it's not doing evil in the process.
This issue has been forwarded to a developer for further action.
This issue has been added to a list of features for future releases.
This is a very difficult problem but we will consider it for future development.