Try to remove ParticipantImpl::mesh(const std::string &meshName)

[BenjaminRodenberg]

I stumbled across the public getter ParticipantImpl::mesh(...) when working on the documentation of ParticipantImpl in #1249 and think we should remove this getter, because it is only used in tests and has no proper use-case to justify a public function. This is mostly motivated from a class design perspective trying to simplify ParticipantImpl.

As a compromise, we could also add this function to a fixture or the WhiteboxAccessor, but I think improving the tests would be the better solution. However, for most of the tests I cannot judge whether removing ParticipantImpl::mesh(...) would be ok in the context of the original goal of the test.

I found the following tests that use this function:

tests/serial/whitebox/TestExplicitWithDataScaling.cpp

Easy to remove, because testing::WhiteboxAccessor::impl(cplInterface).mesh("Test-Square-One").vertices().size() always returns nVertices = 4. No need for this function.

tests/parallel/TestFinalize.cpp

What is it actually testing? Can we remove this test?

tests/parallel/NearestProjectionRePartitioning.cpp

I don't understand the rationale here. But is related to #371

/home/benjamin/Programming/precice/tests/serial/mapping-scaled-consistent/helpers.cpp

Do we need to go that low here? For some functions I have the impression the purpose of these helpers lies somewhere else and not on testing the correct creation of the mesh. (Can we just remove the BOOST_REQUIRE calls?)

tests/serial/mapping-nearest-projection/helpers.cpp

At different places. Sometimes a similar situations like above (BOOST_REQUIRE). testQuadMappingNearestProjectionTallKite and testQuadMappingNearestProjectionWideKite are testing mesh::edgeLength(edge). Should this test maybe go somewhere else as a unit test, where accessing the mesh is easier?

[fsimonis]

Looking at the use-cases, they normally ensure that some test conditions are met. This is not strictly necessary and should probably be checked in unit tests instead.
Feel free to remove the checks if you dislike mesh() that much 😁

[BenjaminRodenberg]

I updated SolverInterfaceImpl to ParticipantImpl following #1643