The patch reimplements the core_string_manager::get_language_dependencies()
so that potentially mis-configured language packs with circular
dependencies or self dependency do not make the site unavailable.
Unit tests for the expected behaviour are added.
This step is widely used by JS scenarios,
with Javascript disabled this step fails tests
as tree nodes al already expanded; we can
reuse JS scenarios for no-JS tests skiping
this step in no-JS mode.
Chaning also the adapt_ prefix for non-direct
data generator uses to process_ because not all
of behat's data generators would have a lib/testing
data generator to share with phpunit tests.
This function used to be in the user selector, but it is useful in
other areas where we want to search for users as it handles a number
of search options (including extra fields, etc.) and returns SQL
which can be included in a joined query.
A unit test has been added to check the results of these searches.
- Course categories caches are purged by event now
- session cache has additional 10 minutes ttl to clear itself and accomodate for permission changes that do not trigger event purging
- additional request-level cache for coursecat::get()
- We store only children of one category in one cache key
- Function coursecat::get_children() can return results sorted and/or paginated. Added tests
- Retrieve and cache only often-used fields of course category
- Removed function coursecat::get_all_visible() as potentially causing performance issues
- removed function coursecat::get_all_parents() as ineffective and unnecessary, replaced with get_parents()
- retrieve all fields from course_categories when unretrieved field is accessed
Also some code improvements:
- rename functions starting with _ , rename arguments, etc.
For statslib tests was enough to require cronlib.php as
far as all the tests there are using ob-capture / output
expectations, so the new function does not bork anything.
For pluginlib tests finally the use was deleted because
it's a part of cron not interesting (should be always "cheap")
and to keep it there we should be adding a bunch of output
expectations under some cases, for practically nothing.
