[Bug target/126782] [Darwin] sys/_types/_rsize_t.h does not define rsize_t with -fmodules

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126782

--- Comment #5 from Bjorn Pagen <hello at bjornpagen dot com> ---
USE_CLANG_STDDEF tells these SDK headers to obtain selected fundamental
typedefs from the compiler-provided <stddef.h> rather than defining them
directly in the SDK header.

In this case, _rsize_t.h defines __need_rsize_t and includes <stddef.h>.
Clang's <stddef.h> recognizes that request and defines:

typedef __SIZE_TYPE__ rsize_t;

GCC's <stddef.h> supports the same selective-inclusion pattern for types such
as size_t, ptrdiff_t, and wchar_t, but it does not currently recognize
__need_rsize_t. Consequently, the include returns without defining the type.

This does not require GCC or the C library to implement all of Annex K. It only
provides the rsize_t typedef when the system header explicitly requests it. For
an ordinary full inclusion of <stddef.h>, it could continue to expose rsize_t
only when __STDC_WANT_LIB_EXT1__ >= 1, matching Clang's behavior.

I agree that supporting __need_rsize_t in GCC's <stddef.h> is preferable to a
Darwin fixincludes rule, particularly because the SDK can be selected at build
time.