Per Backlog Refinement.

These are notes I took when looking last into this on Feb 25th, before I had to look into other things:

• the container seems to actually be destroyed, as placing a breakpoint on the exception being thrown reveals
• I do not know why that container is destroyed
• there are two variants for each test executed, plain and with ApiBase::GET_VALUES_FOR_HELP and only the latter encounters these issues
• If I move the test out of the data-provider, then it no longer errors out! => could there somehow be a contamination from a different test?
• if I reverse the order of the tests in the data-provider, then that does not make a difference

• next thing to try for me: only return the data for this particular api-endpoint from the data-provider

Looking more into it now.

A few more observations:

• the error still occurs if I narrow the provider down to only return the two test-cases for the 'visualeditoredit' path.
  • => so it is not a contamination from any other API endpoint
• the error goes away if I switch around the two values in

$argsets = [
	'plain' => [],
	'for help' => [ ApiBase::GET_VALUES_FOR_HELP ],
];

• => so it is about *something* happening in the first case that affects the second
• the error also goes away If I ensure that there is a new ApiMain instance every time by doing something like

self::$main = null;
$main = self::getMain();

• => so somehow something somewhere is holding on to an old instance of the services?
• the error goes away if I comment out 'plain' => [], or reverse the order, and add additional paths; the same is true if I comment out 'plain' => [], but duplicate 'for help' => [ ApiBase::GET_VALUES_FOR_HELP ],
  • => so it is not "something runs before our for help" test-case, but specifically 'plain' => [], for the same 'visualeditoredit' endpoint runs before it.

Further debugging revealed that the GrowthExperimentsTaskTypeHandlerRegistry is not being instantiated a second time when running only the tests for 'visualeditoredit', thus using a service container that was destroyed in the teardown of the first testcase with 'plain'. But why?

Change #1125153 had a related patch set uploaded (by Michael Große; author: Michael Große):

[mediawiki/core@master] test(ApiStructureTest): use a fresh ApiMain instance for each test

https://gerrit.wikimedia.org/r/1125153

Change #1125155 had a related patch set uploaded (by Michael Große; author: Michael Große):

[mediawiki/extensions/GrowthExperiments@master] test(ApiStructureTest): reenable test for visualeditoredit integration

https://gerrit.wikimedia.org/r/1125155

Change #1125153 merged by jenkins-bot:

[mediawiki/core@master] test(ApiStructureTest): use a fresh ApiMain instance for each test

https://gerrit.wikimedia.org/r/1125153

Change #1125155 merged by jenkins-bot:

[mediawiki/extensions/GrowthExperiments@master] test(ApiStructureTest): reenable test for visualeditoredit integration

https://gerrit.wikimedia.org/r/1125155

There should be no production impact from this, it mainly touched test-files.