Functionality tests utils dev

[stonecoldhughes]

Linked issue: #193

note: this PR has been opened and closed a few times due to insufficient testing before pushing. The others did not pass the checks. This PR applies a testing helper class to remove boilerplate from a single testing file

[superwhiskers]

some minor things that stuck out to me that may be worth considering

[superwhiskers]

the name checkRefactorizationResult doesn't seem to be too appropriate here. while the only current exception is lusol, it may be more appropriate to name this checkResult, checkFactors, or something along those lines if this is to be more general in the future

[pelesh]

checkResult seems simple and self-explanatory.

[stonecoldhughes]

I changed this to checkResultNorms() to be more specific

[pelesh]

Checking error norms is how one would check the result, so checkResult will be as informative and less redundant, imho.

[stonecoldhughes]

changed to checkResult

[pelesh]

Can we make this call in testSysRefactor.cpp rather than here? Perhaps there will be an instance where we want to just check the result and not norm computation.

[stonecoldhughes]

this function is now invoked in the main testing file

[pelesh]

I would invoke it before checkResult. We want to establish first that SystemSolver computes error norms correctly before we test its solution.

[stonecoldhughes]

done

[pelesh]

The call to checkNormOfScaledResiduals should probably be made from testSysRefactor.cpp instead of here. I would make this function check the result only.

[stonecoldhughes]

this function is now invoked in the main testing file

[pelesh]

See my comment above.

[stonecoldhughes]

done

[stonecoldhughes]

note: the following improvements extend the work so far, but should probably be in a separate PR:

• use same workspace in test_helper that is used in main test
• create class to contain the matrix and vectors in the problem
• re-order function calls in the test file to flow more logically
• improve RAII in this file for code robustness; one ax=b problem and one testhelper exist for one test result. Use move constructors to preserve memory efficiency
• configure test helper to calculate norms upon construction; as an explicit step, norm calculation is easy to forget and would make subsequent checks involving the norms ill-formed
• use smart pointers in the test helper instead of raw pointers; this removes need for manual new/delete style memory management
• break up the testing file into more functions with file-scope in the test to make main() more succinct. The top level naturally lends itself to a test_factorize() function followed by a test_refactorize()