SQP and QP update

[roccafrancesco]

QPDualActiveSolver
SQPOptimizerS2

[maisonobe]

I have pushed a few javadoc fixes in branch roccafrancesco-develop3.
There are still some two missing javadoc parameters in SQPLogger.logRow. As the parameters have very generic names (and are also out of order), I was not able to find their meaning.

There are also 33 failing tests.

[roccafrancesco]

Hi Luc,
on my side i have 18 tests failed and 7 are skipped;
the two ADMMQP optimizer tests fail due to tolerance issues; the tolerance needs to be relaxed;
the LineSearchTest fails because the class has been modified and the test is no longer applicable;
most of the failures in the Hock-Schittkowski suite are due to reaching a relative minimum; this accounts for 15 failed tests;

[maisonobe]

OK, I will then increase tolerance for ADMMQP and remove the linear search test.
Concerning the Hock-Schittkowski suite, I would prefer to keep at least the code of the tests, but flag them as @disabled with a comment, so if later improvements allow to identify the local minimum we could reactivate the tests?

[roccafrancesco]

Yes, Luc, that’s fine; the important thing is that we are aligned and both detect the same number of errors.

[maisonobe]

I have fixed numerous issues triggered by SonarQube after the merge.
I have also improved code coverage, but it is still below our very high thresholds (we want 95% code coverage on new code, we are at 90% after the improvements I made). In fact, despite the very large number of tests, some parts are never exercised. I guess it is because they are related to convergence issues or too stringent settings, and these do not occur in the current tests.
Could you have a look at these and see if we can improve this coverage? See SonarQube coverage for the remaining non-covered parts. Maybe reusing some existing tests but intentionally setting convergence to some too low values would enforce using these code paths?

[roccafrancesco]

Hi Luc,
I don't think the low coverage is due to the skipped test; rather, it stems from the fact that a specific execution path within a given class isn't being exercised. Take BFGSUpdate, for instance—the initial if condition is never met.