Add missing add() method to NullEngine to prevent undefined prefix key error - #19568

CI failure here is unrelated to your change. It's the rector step in the "Coding Standard & Static Analysis" job blowing up on a dependency mismatch:

PHP Fatal error:  Uncaught Rector\Exception\Reflection\MissingPrivatePropertyException:
Property "$container" was not found in "PHPStan\Parser\RichParser" class
in vendor/rector/rector/src/Util/Reflection/PrivatesAccessor.php:82
Script vendor/bin/rector process --dry-run handling the rector-check event returned with error code 255

composer rector-setup runs an unpinned update, which pulls rector 2.4.6 against phpstan 2.2.6, and rector's PHPStanContainerMemento::removeRichVisitors() reaches for a private property that no longer exists there. phpstan, phpcs and the split-package analysis all pass. The same failure hits #19565, and this has been red since the last green 5.x run on 2026-07-23, so there is nothing to fix on your branch.

On the change itself: the diagnosis is correct. NullEngine::init() overrides the parent without calling setConfig(), so _configInitialized stays false and _config stays empty. CacheEngine::_key() then reads $this->_config['prefix'] directly instead of going through getConfig(), so the lazy default merge never happens, hence the undefined key. Returning true from add() is consistent with the rest of the class as well: get() returns the default and set() returns true, so the fall-through result would have been true anyway.

Two things worth considering:

• A test in tests/TestCase/Cache/Engine/NullEngineTest.php covering add().
• The override fixes this one method, but the underlying fragility stays: any future CacheEngine method touching _config directly breaks on NullEngine the same way, which is exactly how add() slipped through when it was introduced. Either NullEngine::init() calling parent::init($config), or _key() using getConfig('prefix'), would close the whole class.

You are right, better to ensure parent init is called..Added the test as well to ensure it behaves as expected.

This one failed: CI / testsuite (8.2, mysql, highest) (pull_request) because of "Mysql Server has gone away"..not sure if you can re-run the job

Lets see if 5.x becomes green, it should.