Assignable users should respect database collation
mysql collation: utf8_general_ci
['a','u','č'] should be sorted as ['a','č','u'] but because of ruby sort it's reordered back as ['a','u','č']
def assignable_users users = project.assignable_users(tracker).to_a users << author if author && author.active? if assigned_to_id_was.present? && assignee = Principal.find_by_id(assigned_to_id_was) users << assignee end users.uniq.sort end
I can provide a patch, are you interested or is it desired behaviour?
Environment (not important, all redmine versions and databases are affected): Redmine version 3.3.3.devel.16557 Ruby version 2.1.5-p273 (2014-11-13) [x64-mingw32] Rails version 4.2.8 Environment production Database adapter Mysql2 SCM: Subversion 1.9.5 Git 2.11.0 Filesystem Redmine plugins: no plugin installed
#6 Updated by Holger Just about 1 month ago
In the updated patch, you removed the line
return @assignable_users unless @assignable_users.nil? from the original patch in #25867#note-1. Without this line, the caching of the result in
@assignable_users becomes useless.
I'm actually unsure if it's worthwhile to introduce caching here at all.
I tend to say: we do not need it as the method does not appear to be regularly called multiple times per request. As such, I think, we can get rid of the caching and its associated possibility for inconsistencies. Turns out, it is called multiple times in the
issues/_attributes.html.erb partial. Thus, we still might want caching... In any case though, we should either remove the instance variable caching completely, or use it if present.
As a slight improvement, it might also be useful to also remove the
to_a at the end and to return a query object. That way, callers might chain other query refinements to it without affecting the current use-case.
Finally, it might also be useful to extract the fetching of the (unsorted) user ids into a separate method, e.g.
assignable_user_ids, which might make checks such as those in the Issue model to check if the assignee is allowed less expensive by avoiding the final fetch of the Principal objects. Only these ids might then possible be cached?