At today's FESCo meeting Fedora's policy of enabling updates-testing
on prereleases came into question. The crux of the matter is that
right now updates-testing is enabled by default on all branched but
unreleased releases until shortly before release, which causes two
problems that they would like to hear input or decisions from QA and
Rel-eng on:
* Systems running Fedora betas that update to stable end up with
effectively-orphaned packages installed on their systems because they
never made it out of testing, which can cause problems when installing
packages later on. Once correctly diagnosed, the problem is easily
solved with yum distro-sync, but the cause is frequently non-obvious
even to people who know how Fedora's repos interact.
* The non-testing prerelease gets minimal testing as a whole before
the RC phase. Since most systems run testing instead of non-testing
prereleases, the only time non-testing gets much QA as a whole is at
alpha/beta/RC compose times. Even then, immediately after one
installs an alpha or beta release it gets updated to testing anyway.
This largely defeats the purpose of running bodhi for branched but
unreleased releases.
Some possible solutions for this that came up at the FESCo meeting were:
* Keep things as-is, but document the need to manually downgrade packages at GA
* Switch updates-testing off at beta and create similar documentation
for alpha users
* Never enable updates-testing by default
Decisions or discussions about this would be greatly appreciated.
Thanks,
--
Garrett Holmstrom
Show replies by date