During the bootstrap of PHPUnit we ensure that the database has been
reset to its initial state.
We do this by checking the internally-stored DB write count between
runs. If the count is not yet set (null), or it has been increased, we
force a reset.
When running an isolated test the test runner resets the database, it
then sets up a new isolated test environment by writing a new PHPUnit
test case and passing it to a new PHP Process using standard in. As part
of this, the bootstrap is run for that process.
Because we are in a new process, the db write count is fresh and not yet
set. This has been leading to an additional db reset before the isolated
test.
To handle this we want to _not_ perform a reset during the
initialisation for isolated runs. We know that the DB is in a fresh
state before we start the run.
To support this we need to know whether the test is an isolated test
during the bootstrap, which means we cannot use the previous approach to
calculating this.
Instead we look at the PHP_SELF value. PHP sets this to "Standard input
code" when run from stdin, instead of running a file.
There should not be any other legitimate reason to run a PHPUnit
bootstrap via this stdin approach.
Unfortunately this approach is a little bit risky as it depends on the
presence of a specific string, however this string has been in place
since 2016, and there is no legitimate way of calculating this.
I did consider looking at whether the called script included `/vendor/`
and `/phpunit`, but this is also likely a risky approach if someone
calls PHPUnit in an unexpected way.
This approach is itself unit tested so any change to PHP's stdin string
before we deprecate this approach entirely in 12 months time will be
caught.
Statically cache list of loaded reports during request lifecycle,
this ensures that computationally heavy initialisation routines
in system reports are only executed once (e.g. the access tab).
This performs all the changes needed in the util generator to
produce XML files compliant with the new schema (see previous
commit for description of changes).
- All the occurrences in code of filter => coverage.
- All the occurrences in code of whitelist => include.
- Apply all the changes to comply with the new schema.
- Remove processUncoveredFilesFromWhitelist attribute, useless (false
is its default value, and now have another name).
- Move from 4-spaces indented XML to 2-spaces indented.
- Small linefeed tweaks to generate better-looking XML.
We are already cleaning all sort of caches, statics, singletons
there and it's better to ensure they are always reset to avoid
myterious failures @ distance.
Thanks to MDL-49398, we can separate the combined user session and role
definition cache clearing function into two separate functions. At the
same time, we want to identify and remove mark_dirty() calls that were
added for role definition changes but were incorrectly left behind.
Change highlights:
- Remove unnecessary mark_dirty() calls performed after
assign_capability(), unassign_capability(), delete_role(),
deleted contexts, brand new contexts
- Move role definition cache clear from the user-centric
accesslib_clear_all_caches() to the newly created,
role-dedicated accesslib_reset_role_cache()
Also deprecated the following functions -
1. message_move_userfrom_unread2read - It is not necessary
for us to mark a message as read on user deletion.
2. message_get_blocked_users - Horrible logic used to
determine if a user is blocked via reference on some
randomly chosen 'isblocked' variable.
3. message_get_contacts - The same as above. This can be
done in a much nicer way.
4. message_mark_message_read - We want two functions to do
this to avoid confusing messages and notifications.
5. message_can_delete_message - This assumed the variable
$message contained the 'useridto' property, which
was present in the old table structure. We do not want
future usages where a query is done on the new table
and is simply passed as this won't contain this property.
6. message_delete_message - Same as above.
Remove unittest-specific callbacks for checking access and displaying
the calendar events on the dashboard.
This will allow plugin developers unittest the full behavior
of how their plugins add events to the dashboard.
Reset all static caches between unittests.
Before this patch behat_dataroot for parallel runs
were created at same level as ->behat_dataroot
After this patch, it will be created 1 level under
->behat_dataroot with BEHAT_PARALLEL_SITE_NAME
prefix and run number as suffix. This will allow
common files as test enabled and parallel run info
to be saved in parent directory and access it easily.
Created a new phpunit util function run_all_adhoc_tasks which runs any
pending tasks, for use in unit tests. Added new recyclebin and course
unit tests covering the new functionality.
By specifying the filter on all those partial phpunit.xml files
we are restricting in a 99% the effort (time and memory) needed
by phpunit to process xdebug execution coverage, retricting it
to the target files within a component.
Also the get_property_definition() was created to get the property without retrieve the whole definition cache and
unit tests were created to tests those new methods.
Introducing both APIs in moodle along with:
- search_box widget to add a tiny search box
- admin settings with setup steps helper
- cache for search results
- template for a search result
- php unit stuff
Many thanks to Tomasz Muras, Prateek Sachan and Daniel Neis for their contributions, for starting this development
and for pushing for it to be completed. Also thanks to other contributors: Jonathan Harker and eugeneventer.
The mdeploy.php standalone script (used to download and unzip plugins
into the dirroot) and the \core\update\deployer class (as a
communication bridge between the core and the mdeploy.php) was
originally designed and implemented with the assumption that it would be
eventually used for updating the Moodle core itself, too. Therefore it
was written as standalone utility without dependency on the Moodle core
libraries.
However, it never happened and there is no real demand for that. So now
there is no need to have and maintain a completely parallel solution for
common things like fetching and unzipping plugin ZIPs.
Additional reasoning for mdeploy.php was that the core is not very
reliable during the core upgrade and we could run into various troubles.
This does not seem to be that bad. We rely on a lot of core
functionality (such as output rendering, DB access etc) and plugins
deployment seems to work well (and better) with common core libraries.
So long mdeploy, and thanks for all the hard work you did for us.
The testsuite name is fed straight into a class_exists() in
PHPUnit_Framework_TestSuite. In some cases, e.g. cachestore_static, the
class does exist, it's then fed into a ReflectionClass, and the test fails.
The testsuite needs to not conflict with any classes which could be
autoloaded in Moodle core.
Andrew Nicols