SyN registration tests pass without running a registration (argc guard rejects CMake argv; return value discarded)

[hjmjohnson]

itkSyNImageRegistrationTest and itkBSplineSyNImageRegistrationTest pass without ever running a registration: an argument-count guard rejects the CMake-supplied argv, and the helper's return value is discarded so the failure is swallowed. Both complete in ~0.06 s in Release and Debug alike.

itkSyNImageRegistrationTest

Modules/Registration/RegistrationMethodsv4/test/itkSyNImageRegistrationTest.cxx:112-118

PerformDisplacementFieldImageRegistration(int argc, char * argv[])
{
  if (argc != 5)
  {
    std::cout << "ERROR: incorrect number of arguments" << std::endl;
    return EXIT_FAILURE;
  }

test/CMakeLists.txt:415-428 invokes the test with 7 arguments (test name, dimension, fixed, moving, output prefix, iterations, learning rate), so argc == 7 and the guard always fires.

The failure is then discarded — itkSyNImageRegistrationTest.cxx:373-380:

  switch (std::stoi(argv[1]))
  {
    case 2:
      PerformDisplacementFieldImageRegistration<2>(argc, argv);   // return value ignored
      break;

Observed output:

Name of Class = SyNImageRegistrationMethod
ERROR: incorrect number of arguments
Test finished.
1/1 Test #3193: itkSyNImageRegistrationTest ......   Passed    0.07 sec

itkBSplineSyNImageRegistrationTest

Same shape. itkBSplineSyNImageRegistrationTest.cxx:80 asserts ITK_TEST_EXPECT_TRUE(argc == 4) while test/CMakeLists.txt:431-443 supplies 7 arguments; the helper's return value is likewise discarded at line 376.

Passes in 0.06 s.

Impact and suggested fix

Only ITK_EXERCISE_BASIC_OBJECT_METHODS on the registration object actually executes. The SyN and B-spline SyN registration paths — multi-level optimization, displacement-field update, convergence — have no effective test coverage, and no baseline comparison is performed.

Fix has two parts, both needed:

1. Correct the argc guards to match the actual invocations (or correct the invocations).
2. Propagate the helper's return value out of the switch so a failure fails the test.

Part 2 should land regardless — without it, any future argument-handling drift is silent again. Once the registrations actually run, baselines will need to be established and the runtimes will no longer be ~0.06 s.

Found while investigating the Debug-mode failure of itkSimpleImageRegistrationTest{Float,Double} (separate issue: the #ifdef NDEBUG 1-iteration reduction combined with the removal of the Debug-specific .3 baseline in PR #6569).