MDL-73727 aimed to improve the performance of the
get_conversions_for_file SQL. Unfortunately it also
changed the behvaiour of the SQL.
On further investigation, there is nothing wrong with
the SQL. This patch returns it back to its initial state.
The "Services and support" and "Contact site support" links were
missing from the footer in the Classic theme (which now has its own
footer template), so there was no way to navigate to the relevant links.
The mnet_environment->keypair array contains the following
elements (and more, just focussing on these):
- keypair_PEM : textual representation of the private key.
- certificate : textual representation of the public key.
- privatekey : OpenSSLAsymmetricKey representation of the private key,
generated from keypair_PEM. See get_private_key().
- publickey : OpenSSLAsymmetricKey representation if the public key,
generated from certificate. See get_public_key().
The last 2 elements in the array are only used as "caching", to avoid
having to call to openssl_pkey_get_private() and
openssl_pkey_get_public() to convert from the textual representation
to the OpenSSLAsymmetricKey representation that is the one required
by a number of openssl functions.
Problems arrive when, as part of the MNet protocol, the mnet_environment
is serialised, because, since PHP 8.0 those OpenSSLAsymmetricKey objects
aren't serialisable any more.
So, as far as they are only used for internal caching it's perfectly ok
to remove the caching bits and use the openssl_pkey_get_xxx() methods
to calculate them under demand.
The alternative to this would be to implement into the mnet_environment
some custom serialisation, skipping those OpenSSLAsymmetricKey
instances, using __sleep(), the Serializabla interface or __serialize(),
but that seems unnecessary because, as explained above, the uses are
really limited and easily replaceable.
That's what this patch does.
This is basically the manual background that was removed in the
previous commit, with 4 attempts here. Note it includes the 2
waits that are REQUIRED to make it pass in too-quick environments.
If they are removed, they begin to fail because it seems that
the attempts need a minimum of separation between them.
This moves away from manual attempts and uses the activity
attempts generator instead. Quicker and not exposed to the
problems with clicks, refreshes and missing attempts.
Note that there is a small change in the attempt generator
so, now, when a new attempt is created, if the activity has
tracking enabled, the final result / grades are recalculated.
That's the real behaviour when the activity settings are
changes in the edition form and it's natural to get them
recalculated without having to visit the form. Specifically,
this allows the 1st scenario to pass, because in that scenario
no changes to the form are being performed.
Also, little detail, amend a dupe scenario name.
To avoid getting some random errors with Oracle, the order for the
students have been removed from the tests (to check they are there,
regardless the order they are returned, because in that case, the
order is not important).
This commit adds a new Grunt task which generates a jsconfig.json file.
This file is described at
https://code.visualstudio.com/docs/languages/jsconfig and can be used
to configure vscode and other similar editors to understand our module
loading.
This task is an optional task for Grunt and can be run with:
npx grunt jsconfig
The file is git-ignored and will not affect normal usage.
