In some places we prevented cache poisoning, in others we did not. We
also did not place any restriction on the minimum value for a revision.
This change introduces a new set of functions for configonly endpoints
which validates the revision numbers passed in. If the revision is
either too old, or too new, it is rejected and the file content is not
cached. The content is still served, but caching headers are not sent,
and any local storage caching is prevented.
The current time is used as the maximum version, with 60 seconds added
to allow for any clock skew between cluster nodes. Previously some
locations used one hour, but there should never be such a large clock
skew on a correctly configured system.
Co-authored-by: Andrew Nicols <andrew@nicols.co.uk>
With the new flow, users can go to the login page from the confirmed page,
and if the user successfully logs in, the user will be directed to the confirmed page.
To avoid that, the confirmed page can only be seen by users who are not logged in.
Within the forum_get_discussion_neighbours() function of forum
when the discussions timemodified (last post) is the same, there
is a bug that does strange things when calculating the prev and
next discussions.
Note that, in real life, this is really hard to achieve, but in tests,
when multiple discussions and post can be created by generators in the
same second (specially when the test machine is quick), chances of
facing that problem are higher.
By adding 1 second wait, we ensure that the discussions won't have
the same timemodified (last post) and workaround the problem.
No mater of that, the problem deserves an issue to be created
so we guarantee from code that it also works ok when the same
timemodified (last post) situation happens.
Note that normally this doesn't matter much, but there are situations
when we want the discussion list ordering fully consistent /
deterministic.
Specifically, when discussions (or forum posts )are created in the
same second, or when the discussion titles are repeated, or 2
discussions have the same number of votes... (any criteria in general),
in the context of testing, we don't want the order
to be non-consistent, so we need to provide an extra sorting
criterion to make it fully deterministic.
So, in this case, we are adding a sort by discussion.id <<DIRECTION>>
that is an unique value, primary key... so cheap to calculate and,
that way, when the 1st sorting column has repeated values, the id
will decide.
* For some reason, the `data-trigger="hover focus"` does not work when
rendering the popover via HTML. It just defaults to click, so the
feedback popover for qtype_multianswer questions does not show on hover.
Configuring the feedback popover's trigger via JS seems to work.
While, right now, sites using long (> 10 chars) $CFG->prefix
can continue working (because we still don't have any table
> 28 chars), as soon as some new table with long name is added,
it won't work with PostgreSQL anymore (if the 63 limit is raised).
Hence, this environmental check will verify on both install and
upgrade that the $CFG->prefix is always <= 10 chars.
Sites with longer prefixes will need to rename all their tables
(and maybe other objects, depending on the dbtype) to use a shorter
prefix.
Paul Holden