This 'fix' is basically a work-around, but one that is already used in
some places. Now it is used in all similar places.
Exactly what causes the current file path in a file manager to become
unset is still unclear to me, but it seems to be an obscure race
condition that is very hard to reproduce.
However, when it happens, currently we are effectively defaulting to
path '/undefined/'. Using '/' is never worse, and is infinitely better
in the case where the filepicker does not allow folders to be used
(which includes important cases like essay questions in a quiz, which
some people use for exams!). In that case, falling back to non-top-level
path leads to dataloss.
Also, this is all historic JavaScript which will get replaced in the
grand de-YUI-fication, so I think a pragmatic fix is justified here.
The Open ID Connect plugin uses null for the password,
which makes the internal password update fail to proceed.
Allowing null resolved the problem.
As a note, there is a potential issue if the authentication method has
a false return for the prevent_local_password because it will trigger
the hash_internal_user_password() where the $password can not be null.
Since this only addresses the oauth2 issue, we should ignore it.
* Move the shared drag fixture out of the Atto plugin;
* Switch to using faster generators for populating private files;
* Clarify required fields for private files generator
It seems that the loading of the h5p content upsets other interactions
with the page in Firefox as it loads. Unfortunately I haven't found a
reliable way to handle this with pendingJS yet.
This is the poor man's fix and we should find a better solution.
If there's a driver error, for example from a step taking too long, then
this kills the entire Behat run and we have to start from scratch.
This change instead throws away the original connection and starts a new
one to try and continue the test.
W3C WebDriver Element::Click, Element::Clear, and Element::SendKeys all
state that the WebDriver implementation (chromedriver, geckodriver,
edgedriver) should scroll the element into view if it is not already
visible.
It is wrong for us to check if the element is visible or not before
calling these as it may not be but will during the click/clear/type
event.
