[Core] Fix Kernel Destructor

[roigcarlo]

📝 Description
Kratos kernel now cleans the list of applications registered and components once destroyed, preventing pointers to deleted instances of Components and no longer available applications from existing after calling the kernel initialization and application registry without unloading the shared lib.

Related #11506

🆕 Changelog

• Fixing Kernel destruction leaving registered components and applications

[loumalouomega]

Thanks

[roigcarlo]

@loumalouomega I've been looking for the error of this and as I suspected is very hard to fix without revamping the test, so I am afraid this needs to be included in the GTests PR and cannot be split.

The error happens because some applications ( specifically the ones that have custom custom workflows in c++ like GeoMechanics) have custom kernels.

So:
1 - Python loads kratos (from the testing script)
2 - Python loads all apps (from the testing script)
3 - Python tests are run
4 - A call is made so all cpp tests are run (because are registered in the kernel regardless of the app so they run mixed)
5 - When a GeoMechanics tests or something using a custom instance of the kernel is called and ends, the kernel gets uninitialized
6 - As nobody re-initializes the kernel tests crash or segfault.

So in order to fix this I would have to disable all the cpp tests with cpp workflows or remade the script so instead of using the python instance that is created with python run_all_tests.py uses a subprocess that reloads the kernel and runs the tests based on a list of suits ( the ones uses in the CPP macro)

I don't want to disable those tests in the meantime and I will not make the changes in the resting script because I will ditch them when making the GTests. So, if you are ok with the changes here, I will just close the PR and consider this part reviewer in the other PR.

As a note, the current behavior is the wrong one as all tests are assuming that the kernel is in a correct state after a tests is executed, which is a very bold assumption and could cause interference between all the tests,

[loumalouomega]

@loumalouomega I've been looking for the error of this and as I suspected is very hard to fix without revamping the test, so I am afraid this needs to be included in the GTests PR and cannot be split.

The error happens because some applications ( specifically the ones that have custom custom workflows in c++ like GeoMechanics) have custom kernels.

So: 1 - Python loads kratos (from the testing script) 2 - Python loads all apps (from the testing script) 3 - Python tests are run 4 - A call is made so all cpp tests are run (because are registered in the kernel regardless of the app so they run mixed) 5 - When a GeoMechanics tests or something using a custom instance of the kernel is called and ends, the kernel gets uninitialized 6 - As nobody re-initializes the kernel tests crash or segfault.

So in order to fix this I would have to disable all the cpp tests with cpp workflows or remade the script so instead of using the python instance that is created with python run_all_tests.py uses a subprocess that reloads the kernel and runs the tests based on a list of suits ( the ones uses in the CPP macro)

I don't want to disable those tests in the meantime and I will not make the changes in the resting script because I will ditch them when making the GTests. So, if you are ok with the changes here, I will just close the PR and consider this part reviewer in the other PR.

As a note, the current behavior is the wrong one as all tests are assuming that the kernel is in a correct state after a tests is executed, which is a very bold assumption and could cause interference between all the tests,

Okay :')

[roigcarlo]

Closing. See #11506