When using a try/catch for a find if the first match is not met then we
have an implicit 6 second delay waiting for the search to time out.
We can avoid this by combining the searches in a comma-separated
selection.
I have also taken the opportunity to convert the use of ->click() to the
i_click_on function which has the benefit of supporting wait for
pendingJS.
The standard NodeElement functions for getAttribute, getTagName,
getParent, and friends go back to WebDriver and parse the DOM for each
request. This is insanely slow per request, and in the case of forms we
do a lot of checking to determine the field type.
This change modifies the form field detection to copy the entire node
content into a DOMDocument and parse the document locally.
This is significantly faster - in some cases where there are large
documents minutes faster.
I believe that this should be a safe change as the document fetched from
the browser is normalised to match the doctype specified.
With the default 5 sections course, after editing the activities, the
"log out" link is exactly on the edge of the view port. It's one of
those cases where Chrome fails to click/press it (have tried both).
So just make the course to have 3 sections is enough to avoid the
problem. Some day Chrome will fix that problem but, until then...
we cannot do much apart from changing the window size or try to
configure things to show differently.
All the scenarios failing with Chrome zero-size errors are
getting the @skip_chrome_zerosize to be able to keep them apart from
regular runs. See MDL-71108 for more information about
that error.
They will be run by another job, specifically using that tag to
keep them failing and under control.
Current anchor link offset fix is not working with 'display:flex' elements
Note this is a backport of MDL-71152 exclusively to 39_STABLE because
the original issue already has been applied to 310, 311 and master.
And seems to workaround a bug in Chrome 93 that is hitting Moodle.
We should ensure that users being exported are enrolled on the
course being exported from. In courses where the current user can't
access all groups we should ensure that users being exported belong
to the same groups as them.
LaTeX documents have a preamble section and admins can use a \newcommand
statement to define new commands there (or to give an alias to another
command). This commit makes sure no blocked command can escape
sanitization by being used in a new seemingly harmless command that is
defined in the LaTeX preamble.
Before PHP 8.0.10, the time is lost when datetime->modify() is used
with weekdays. However, with PHP 8.0.10, the time is kept.
In consequence, the test_yearly_every_20th_monday_forever method
started failing.
For fixing it, the offset has been removed and the time has been set
to 00:00, so the behaviour will be the same regardless the PHP
version.
When npm-shrinkwrap.json was regenerated few days ago by MDL-72014
running audit-fix on it, it seems that some checksums in the "lodash"
package were missing.
Now they are properly set and that's leading to npm-shrinkwrap.json
changes when people runs npm install (note npm ci is the recommended
way to install all the dev dependencies).
So, just reconcile the file to include the new checksums.
This optional environmental check will look if:
- the site is running php72.
- the site has igbinary extension enabled.
- the igbinary extension version is a buggy one >=3.2.2 <= 3.2.4
- the bug is reproducible.
And will warn if all the conditions are met.
