On 2021-05-20 08:51:00 +0200, Thomas Goirand wrote: > On 5/19/21 9:21 PM, Sebastian Ramacher wrote: > > Control: tags -1 moreinfo > > > > On 2021-05-07 10:56:51 +0200, Thomas Goirand wrote: > >> Package: release.debian.org > >> Severity: normal > >> > >> Hi, > >> > >> I need to discuss with the release team what to do in order to address > >> this bug: https://bugs.debian.org/987904 > >> > >> What happens is that each Horizon plugin is installing a bunch of python > >> files under /etc/openstack-dashboard/enable. > >> > >> When an Horizon plugin is removed, as the enable folder is in /etc, the > >> enable files of the plugins aren't removed. As a consequence, whenever > >> Horizon attemps to list its plugins (for example, when it tries to do a > >> "collect static" operation, which is kind of compiling all the JS files > >> into a single one, each time a plugin is added/removed or when Horizon > >> upgrades), it just fails, because the files in the enable folder are > >> referencing Python modules that do not exist anymore (since the plugin > >> package has been removed). > >> > >> The solution to fix this is strait forward: replace our symlink in > >> /usr/lib/python3/dist-packages/openstack_dashboard/enable by a folder, > >> and push the enable files in there instead of /etc. This way, the plugins > >> removal will also remove the enable files. > >> > >> The problem is that there are 20 Horizon plugins in Debian, and at this > >> point in the release cycle, it doesn't feel like it is a good time to > >> update 20 packages. > > > > Maybe I am missing some of the context, but it appears to me that in > > if the case the plugin package was removed but not purged, a > > ModuleNotFoundError is raised. So, wouldn't it be sufficient for horizon > > to ignore those plugins that raise a ModuleNotFoundError? > > > > Cheers > > Hi Sebastian, > > Thanks for your answer in this bug, I was desperate for an answer! :) > > I tried to do as Andreas suggested, and then it fails later on, the code > is harder to understand than just this. > > At this point, I've been able to stage the fix in Experimental, and I am > confident that I can backport the fixes to unstable. It's just a large > amount of packages to patch, but I'm confident I can do it. > > Your thoughts? Sorry for slacking on a reply. If you still feel that you can do that before the full freeze, please go ahead. Cheers -- Sebastian Ramacher
Attachment:
signature.asc
Description: PGP signature