The machinery to fix orphaned calendar events that were broken by MDL-67494.
The solution consists of:
1) Upgrade step that checks if this site has executed the problematic upgrade steps and
if positive, it will schedule a new run for calendar_fix_orphaned_events adhoc task.
2) Adhoc task that will self-spawn calling the recovery machinery, running until
all the orphaned calendar events are fixed. It also sets the maximum runtime of
60 seconds. It is also possible to override that number by specifing the desired
number setting the ->calendareventsmaxseconds in your config.php
3) CLI script that will look for all the calendar events which userids
where broken by a wrong upgrade step, affecting to Moodle 3.9.5
and up.
It performs checks to both:
a) Detect if the site was affected (ran the wrong upgrade step).
b) Look for orphaned calendar events, categorising them as:
- standard: site / category / course / group / user events
- subscription: events created via subscriptions.
- action: normal action events, created to show common important dates.
- override: user and group override events, particular, that some activities support.
- custom: other events, not being any of the above, common or particular.
By specifying it (--fix) try to recover as many broken events (missing userid) as
possible. Standard, subscription, action, override events in core are fully supported but
override or custom events should be fixed by each plugin as far as there isn't any standard
API (plugin-wise) to launch a rebuild of the calendar events.
4) Unit tests and helper functions to generate calendar events. We have decided to
keep the tests simple, testing only true and false and not using data generators because
the code is purely to recover the calendar events and won't turn into an API or something
and also due to the urgency of this issue.
The helpers have been created in calendar/tests/helpers.php since there are no data generators
for calendar.
In PHP8.0 using `ksort` was producing incorrect results by sorting
keys differing only in case in the wrong order. This change makes
sorting consistent between PHP versions.
Co-Authored-By: Tim Hunt <T.J.Hunt@open.ac.uk>
Some recent tests do set a date time element
to ##now## or tomorrow and, immediately after that
the look if, effectively, ##now## and #tomorrow#
have been set (with minutes resolution).
Problem is that, between the field is set and the field
is verified, it can happen that the time advances to
next minute (from H:M:59 to H:M+1:00) and then the
assertion fails.
To avoid this, we could have lowered the resolution to be
hours... but that doesn't solve the problem just makes it
to happen less often.
So, instead of that... we are setting the 2 now and tomorrow
cases to be "today noon" and "tomorrow noon" (12:00:00) so
we ensure they won't be ever in the risk of jumping of minute.
Even if a prediction is hidden from the report once is flagged,
it can be flagged several times if the user visits detailed view via URL.
We remove the checkbox to select a prediction and flag it
once it has already been flagged.
Due to Solr bug SOLR-15039, uploading files for indexing can fail if
it uses multipart upload. This changes it to use direct binary upload.
Unfortunately, the direct binary version in PHP curl only supports a
string, so we have to load the file into memory. I added extra code to
restrict the size of files indexed to (memory limit - 100MB), which is
usually 284MB unless configured differently because cron runs under
MEMORY_EXTRA.
Some errors raised when calling external services were ignored.
Displaying this information helps to debug and find the real reasons
why some action, like connecting to an external backpack, can't be
done.
