Enrico Forestieri wrote: > On Sun, Nov 12, 2006 at 09:34:16PM +0100, Jean-Marc Lasgouttes wrote: > >> This is not a regression for windows anyway, right?
No. On windows the code stays as broken as it is. It simply moves the pressure to implement a working solution to those who care for windows. > Yes, neither mingw nor cygwin have support for anything else than > the C locale, and I doubt that MSVC has any, too. > >> I'd say to put the >> patch in, except that it might be better to define a >> USE_LIBC_WCHAR_SUPPORT (or something more explicit) define instead of >> +#if defined(HAVE_WCHAR_T) && SIZEOF_WCHAR_T == 4 Yes, I did that and put it in config.h. I did not use types.h, since what we really need to know here is whether the wctype functions use UCS4 as encoding. What we need to know in types.h is a bit different: We can use wchar_t as char_type if it is wide enough. We do not care how libc interprets wchar_t normally, because we do not use the libc functions anywhere else than what I introduced. > I tried to see whether the #ifdef's could be avoided altogether, > and simply use the wctype functions anyway. This would be possible > on cygwin, as the attached test program produces: > > sizeof(wint_t) = 4 > towupper(97) = 65 > towupper(4294967295) = 4294967295 > > but it turned out that it is not possible with mingw, as the result was: > > sizeof(wint_t) = 4 > towupper(97) = 65 > towupper(4294967295) = 65535 That test is not enough anyway. We can only use the wchar_t classification functions from libc if wchar_t is a 32bit type and they use UCS4 as encoding. Otherwise they would not do what we want. Georg
