ENH: Make default-constructors of RGBPixel and RGBAPixel constexpr

[N-Dekker]

Tested by means of CheckConstexprBeginAndEndOfContainer(). Note that these default-constructors were already constexpr implicitly, as implied by = default, when ITK_FUTURE_LEGACY_REMOVE would be enabled.

[N-Dekker]

• @thewtex @dzenanz @hjmjohnson This is my very first ITK commit after the commit/pre-commit/clang-format/pixi infrastructure upgrade (including PR Update to clang format 19 support #5015) So it works on my side as well 🎉

I did this pull request by a regular git push. Is that OK? Or should we still use our custom git review-push? What is the difference between the two commands?

[dzenanz]

I use TortoiseGit, which implicitly uses plain push. I forgot what review-push does, but I am sure it could be found in the documentation.

[thewtex]

🚀

@N-Dekker awesome!!

Yes, both ways work for creating the PR.

[N-Dekker]

With this force-push, replacing {} with (), I'm hoping to address the warnings at https://open.cdash.org/viewBuildError.php?type=1&buildid=10096694 saying:

In file included from Modules/Core/Common/test/itkNumericTraitsTest.cxx:42:
/.../s/Modules/Core/Common/include/itkNumericTraitsRGBPixel.h: In function 'void numeric_traits_test::CheckVariableLengthArrayTraits(const T&) [with T = itk::RGBPixel<char>]':

Modules/Core/Common/include/itkNumericTraitsRGBPixel.h:118:17: warning: '<anonymous>' may be used uninitialized in this function [-Wmaybe-uninitialized]

[N-Dekker]

It appears that with the old Linux GCC 9.4.0, the constexpr default-constructors that I wrote just don't properly zero-initialize the components, as they should. I'll think of another workaround...

Update: almost reproduced the old compiler bug at https://godbolt.org/z/rzdv3GGG4:

struct FixedArray
{
    FixedArray() = default;
    FixedArray(int);      
    int m_InternalArray[3];
};

struct RGBPixel: FixedArray
{
    RGBPixel(): FixedArray() {}
};

int main()
{
    const auto x = RGBPixel();
    return x.m_InternalArray[0];
}

Saying:

 warning: 'x.RGBPixel::<anonymous>.FixedArray::m_InternalArray[0]' is used uninitialized in this function [-Wuninitialized]

[N-Dekker]

Another attempt, initializing by the seemingly redundant expression Superclass(Superclass()): force-push

[dzenanz]

Now might be a reasonable time to reconsider which compiler versions we support in ITK 6?

[N-Dekker]

@dzenanz It looks like this particular bug (as I reproduced with the example code at #5087 (comment)) is fixed with GCC 12.1 (which was released in May 2022): https://godbolt.org/z/of6Tdbzcj GCC 11 still appears to have the compiler bug. I'm not sure if we can already drop GCC 11 now. Ubuntu-22.04 still comes with GCC 11.4 🤷

Having said so, I think we have a workaround for this one now 😃

[dzenanz]

My Linux workstation is on Ubuntu 22.04. So yes, we can't drop it yet 😄

[hjmjohnson]

@N-Dekker I only use git push. The git review-push looks like it does extra work that is probably no longer needed.

[hjmjohnson]

I think that "".github/workflows/itk_dict.txt"" needs to have "Wmaybe" added.

[hjmjohnson]

I created an issue to track how this may be done automatically: #5092