COMP: Avoid Superclass type alias GCC compile errors

[thewtex]

GCC is not resolving object->Superclass in some cases even though the
definition is available and other compilers are resolving the
definition.

xref #2676

[jhlegarreta]

Although I do not have a better proposal (sorry), I am not sure avoiding to print the superclass name is what we would like as a mid-/long-term fix.

I find particularly worrying that this is only happening for a few classes in the review module; the macro is used extensively throughout the code and the faulty statements are working well. So I think there is something more behind it that we are not being able to identify.

As for the proposed fix, if this is what we aim at for now, am I wrong when saying that the only statements that should need to be within the if/else are those related to the superclass? i.e. That the rest of the code is exactly the same, and we could avoid duplicating it?

[Leengit]

In #2676, the object is a smart pointer rather than a C++ pointer, and perhaps that is the reason that object->Superclass::GetNameOfClass() is confusing to the compiler. Would a simple switch to (*object).Superclass::GetNameOfClass() help? Unfortunately I can't test out this or other solutions on my own system because the original code compiles just fine on my system.

[thewtex]

Would a simple switch to (*object).Superclass::GetNameOfClass() help?

This results in the same error with GCC 9.3.0.

[Leengit]

Looks good.

[jhlegarreta]

Windows complaints might have to do with the comments in #2720 (comment):
https://open.cdash.org/viewBuildError.php?buildid=7444074

[Leengit]

Grasping at straws, but ... in, e.g., Modules/Nonunit/Review/include/itkLabelGeometryImageFilter.h, should itkTypeMacro be invoked before itkNewMacro? That might be helpful if, deep inside, itkNewMacro can print debug output that makes use of the GetNameOfClass method defined by itkTypeMacro. (Yeah, the compiler should be figuring this out, even if out of order, ....)

[dzenanz]

Would a simple switch to (*object).Superclass::GetNameOfClass() help?

I am trying this on my Linux box.

[dzenanz]

Lee's suggestion does not work:

In file included from /home/dzenan/Dashboards/Tests/ITK/Modules/Nonunit/Review/test/itkLabelGeometryImageFilterTest.cxx:26:
/home/dzenan/Dashboards/Tests/ITK/Modules/Nonunit/Review/test/itkLabelGeometryImageFilterTest.cxx: In function ‘int LabelGeometryImageFilterTest(std::string, std::string, std::string, std::string, std::string)’:
/home/dzenan/Dashboards/Tests/ITK/Modules/Core/TestKernel/include/itkTestingMacros.h:68:53: error: ‘Superclass’ has not been declared
   68 |   std::cout << "Name of Superclass = " << (*object).Superclass::GetNameOfClass() << std::endl;                         \
      |                                                     ^~~~~~~~~~
/home/dzenan/Dashboards/Tests/ITK/Modules/Core/TestKernel/include/itkTestingMacros.h:68:53: note: in definition of macro ‘ITK_EXERCISE_BASIC_OBJECT_METHODS’
   68 |   std::cout << "Name of Superclass = " << (*object).Superclass::GetNameOfClass() << std::endl;                         \
      |                                                     ^~~~~~~~~~
/home/dzenan/Dashboards/Tests/ITK/Modules/Core/TestKernel/include/itkTestingMacros.h:78:30: error: ‘Superclass’ has not been declared
   78 |   if (!std::strcmp((*object).Superclass::GetNameOfClass(), #SuperclassName))                                           \
      |                              ^~~~~~~~~~
/home/dzenan/Dashboards/Tests/ITK/Modules/Core/TestKernel/include/itkTestingMacros.h:78:30: note: in definition of macro ‘ITK_EXERCISE_BASIC_OBJECT_METHODS’
   78 |   if (!std::strcmp((*object).Superclass::GetNameOfClass(), #SuperclassName))                                           \
      |                              ^~~~~~~~~~

[thewtex]

/azp run ITK.Windows

[jhlegarreta]

Grasping at straws, but ... in, e.g., Modules/Nonunit/Review/include/itkLabelGeometryImageFilter.h, should itkTypeMacro be invoked before itkNewMacro? That might be helpful if, deep inside, itkNewMacro can print debug output that makes use of the GetNameOfClass method defined by itkTypeMacro. (Yeah, the compiler should be figuring this out, even if out of order, ....)

I was tempted to push a PR with that, but I found many counter-examples (classes whose RTTI macro declaration comes after the New macro and whose tests exercise the basic object methods, and the compiler seems not to complain), e.g.:
https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Core/Common/include/itkImportImageFilter.h#L65
https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Core/Common/test/itkImportImageTest.cxx#L44

Even in the Review module itself:
https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Nonunit/Review/include/itkAreaClosingImageFilter.h#L99
https://github.com/InsightSoftwareConsortium/ITK/blob/master/Modules/Nonunit/Review/test/itkAreaClosingImageFilterTest.cxx#L52

[jhlegarreta]

Not sure how to make the linter happy.

Thanks for doing this @thewtex.

[dzenanz]

If we can't make clang-format happy, there is /* clan-format: off */ (or something similar) to disable checking.

[dyollb]

Maybe something like this would work:

#if defined(__GNUC__)
#define ITK_EXERCISE_BASIC_OBJECT_METHODS(object, ClassName, SuperclassName)                                           \
  object->Print(std::cout);                                                                                            \
  std::cout << "Name of Class = " << object->GetNameOfClass() << std::endl;                                            \
  using self_t = std::remove_reference<decltype(*object)>::type; \
  using parent_t = typename self_t::Superclass; \
  std::cout << "Name of Superclass = " << dynamic_cast<parent_t&>(*object).GetNameOfClass() << std::endl; \
  ITK_MACROEND_NOOP_STATEMENT