On Mon, 30 Sep 2019 at 14:34, Jonathan Wakely <jwakely.gcc@xxxxxxxxx> wrote:
>
> On Mon, 30 Sep 2019 at 14:20, Edward Diener
> <eldlistmailingz@xxxxxxxxxxxxxx> wrote:
> >
> > On 9/30/2019 6:08 AM, Jonathan Wakely wrote:
> > > On Sat, 28 Sep 2019 at 04:29, Edward Diener wrote:
> > >>
> > >> Given this program:
> > >>
> > >> #define PP_THIRD_ARG(a,b,c,...) c
> > >> #define VA_OPT_SUPPORTED_I(...) PP_THIRD_ARG(__VA_OPT__(,),1,0,)
> > >> #define VA_OPT_SUPPORTED() VA_OPT_SUPPORTED_I(?)
> > >>
> > >> #include <iostream>
> > >>
> > >> int main()
> > >> {
> > >> int result = VA_OPT_SUPPORTED();
> > >> std::cout << result;
> > >> return 0;
> > >> }
> > >>
> > >> as a test for __VA_OPT__ support in a C++ compiler ( taken from
> > >> https://stackoverflow.com/questions/48045470/portably-detect-va-opt-support
> > >> )
> > >> I have discovered that __VA_OPT__ support started with gcc-8.1.
> > >
> > > You could also consult
> > > https://gcc.gnu.org/projects/cxx-status.html#cxx2a which notes that
> > > the support is incomplete.
> > >
> > >> However
> > >> I have also discovered that the support occurs no matter what the C++
> > >> standard level is used for the compilation and not just when the option
> > >> is 'std=c++2a'. In other words I can compile the program with
> > >> 'std=c++03', link and run the program and the program will output 1,
> > >> showing __VA_OPT__ support, rather than 0, which shows that __VA_OPT__
> > >> is not supported.
> > >>
> > >> If I compile the above with any version of gcc lower than gcc-8.1 the
> > >> program will outpyt 0 no matter what -std mode I use.
> > >>
> > >> Is this intended, that the C++20 __VA_OPT__ support works in all modes
> > >> for gcc-8.1 and higher ?
> > >
> > > Yes, it's available as a GNU extension. You can get a diagnostic with
> > > -std=c++14 -Wpedantic, but not with any -std=gnu++NN modes.
> >
> > I would strongly argue that if it is available as a GNU extension
> > outside of normal C++20 support in gcc 8.1 on up then it should be
> > available when specifying '-std=gnu++nn' but not available when
> > specifying '-std=c++nn' for any C++ level except for '-std=c++2a'. Isn't
> > the idea of gnu extensions that they should only be available when the
> > programmer is using a gnu compiler mode, but not available if a
> > programmer is using a c++ compiler mode ?
>
> No, absolutely not. The manual seems clear on that point:
>
> "The compiler can accept several base standards, such as ‘c90’ or
> ‘c++98’, and GNU dialects of those standards, such as ‘gnu90’ or
> ‘gnu++98’. When a base standard is specified, the compiler accepts all
> programs following that standard plus those using GNU extensions that
> do not contradict it. For example, -std=c90 turns off certain features
> of GCC that are incompatible with ISO C90, such as the asm and typeof
> keywords, but not other GNU extensions that do not have a meaning in
> ISO C90, such as omitting the middle term of a ?: expression."
>
> Since __VA_OPT__ has no meaning in C++ prior to C++20 (and is a
> reserved name so can't conflict with anything else declared by users)
> it clearly falls into the second category, "GNU extensions that do not
> contradict it".
As I already said, if you want a diagnostic use -pedantic (or
-pedantic-errors to make it ill-formed).
This is consistent with how other extensions have always been handled.
