Let IOComponent::CHAR represents signed char

[N-Dekker]

• Added a note about IOComponent::CHAR representing signed char
• Let file readers interpret the CHAR enum as signed char (instead of just plain char)
• Adjusted tests accordingly

Triggered by a discussion with Bradley (@blowekamp) at https://discourse.itk.org/t/long-double-support-for-imageio-meshio-pixels/7570/11 and his comment at #5449 (comment)

[blowekamp]

Thank you for this change. This change now matches how I have seen this enum type used for ImageIO. Getting this cleanup up and documented is a good step forward.

[N-Dekker]

For the record, signed char was already mapped to IOComponentEnum::CHAR before, by calling IMAGEIOBASE_TYPEMAP(signed char, IOComponentEnum::CHAR), at:

ITK/Modules/IO/ImageBase/include/itkImageIOBase.h

Line 900 in 0dd31a6

IMAGEIOBASE_TYPEMAP(signed char, IOComponentEnum::CHAR);

This logic was syntactically restyled (but not altered, sematically) into std::is_same_v<TComponent, signed char> ? IOComponentEnum::CHAR by pull request #5438, at:

ITK/Modules/IO/ImageBase/include/itkImageIOBase.h

Line 611 in 09305ce

: std::is_same_v<TComponent, signed char> ? IOComponentEnum::CHAR

[N-Dekker]

As far as I can see, signed char and std::int8_t denote the very same type, on all supported platforms:

static_assert(std::is_same_v<std::int8_t, signed char>);

On the other hand, signed char and char are always distinct types, by definition, even when char is a signed type:

static_assert(! std::is_same_v<signed char, char>);

https://godbolt.org/z/3dGcx4vdP

[blowekamp]

I just saw that GCC has the -funsigned-char and -fsigned-char it might be an interesting experiment to see if ITK works with both states. I am not sure if any of the support platform as char as signed.

[N-Dekker]

I just saw that GCC has the -funsigned-char and -fsigned-char it might be an interesting experiment to see if ITK works with both states.

@blowekamp The test failures of my experiment show that ITK currently does not work with -funsigned-char (letting the default char be an unsigned type)

• pull request WIP: [Do not merge] Change the default char type to unsigned #5455
  Note that the test failures are partially caused by the tests, rather than by the library!

The failures can typically be fixed by explicitly using signed char, instead of just plain char. So this pull request (#5450) is a step in the right direction, I think 😇