Due to the bug, the no_teaching model might have been marked as
not-trained in the database. Static predictions models (i.e. those using
a target based on assumptions, not facts) are always considered as
trained. If they were marked as not-trained, the prediction scheduled
task would skip them and they would produce no predictions.
Ideally, such a fix should be done for all static models. But there is
no easy way to do it during the upgrade where accessing the analytics
API is not possible. I don't think there are many models out there that
would be affected by this so this seems to be good enough solution for
now (and the future ability to reset models will cover the rest).
The useridfrom would normally be a foreign key to the users table. But
it can also contain some extra negative values with special semantics
(-10 or -20 for no-reply or support user). Therefore we can't make it a
foreign key and the index must be created explicitly.
The capabilities changed ('contextlevel' => CONTEXT_COURSE changed to
'contextlevel' => CONTEXT_MODULE) are:
* moodle/site:accessallgroups
* moodle/site:viewfullnames
* moodle/site:trustcontent
* moodle/site:viewuseridentity
This list came from reviewing the _get_extra_capabilities functions in
all core activities. They were all somewhat inconsistent, but I think it
makes sense that these capabilities are consistently overridable in all
activities. E.g. moodle/site:accessallgroups affects conditional
availability even if there is no other user of groups, and
moodle/site:viewuseridentity and moodle/site:viewfullnames affect the
logs report, if nothing else.
As a result of this, several _get_extra_capabilities functions are no
longer needed, and all the rest have been simplified.
