[libcxx] Don't truncate intermediates to wchar_t when widening

[mstorsjo]

Summary

On windows, wchar_t is 16 bit, while we might be widening chars to char32_t.

[ldionne]

libcxx/include/__locale

1410

Do you know why this was added in the first place? Did you try this out on non-windows systems without the cast?

[mstorsjo]

libcxx/include/__locale

1410

Not sure why it was there; it's been present since the initial libc++ commit. On linux at least, this change makes no difference in check-cxx.

[ldionne]

Can you please rebase on top of master and update the diff? It will re-trigger the CI, which we should see passing. LGTM once that's done.

[mstorsjo]

Rebased

[ldionne]

CI doesn't trigger a second time -- maybe it's because the diff hasn't changed, or because the revision was accepted. @goncharov Do you know?

In all cases, please feel free to ship it.

[goncharov]

In D90228#2357218, @ldionne wrote:

CI doesn't trigger a second time -- maybe it's because the diff hasn't changed, or because the revision was accepted. @goncharov Do you know?

Yes, in some cases revisions do not trigger herald rules at all (like this one) or have not repository set. I believe that individual configuration or how one uploads diff can cause that.
@mstorsjo have you used "arc" to create the second diff (rebased one)? Have you set any config parameters of it that might affect this?

In all cases, please feel free to ship it.

libcxx/include/__locale

operator()(_OutputIterator __s, const char* __nb, const char* __ne) const
         result __r = ok;
         mbstate_t __mb;
         while (__nb < __ne && __r != error)
         {
             const int __sz = 32;
             char16_t __buf[__sz];
             char16_t* __bn;
             const char* __nn = __nb;
             __r = do_in(__mb, __nb, __ne - __nb > __sz ? __nb+__sz : __ne, __nn,
                         __buf, __buf+__sz, __bn);
             if (__r == codecvt_base::error || __nn == __nb)
                 __throw_runtime_error("locale not supported");
             for (const char16_t* __p = __buf; __p < __bn; ++__p, ++__s)
-                *__s = (wchar_t)*__p;
+                *__s = *__p;
             __nb = __nn;
         }
         return __s;
     }
 };
 
 template <>
 struct _LIBCPP_TEMPLATE_VIS __widen_from_utf8<32>
operator()(_OutputIterator __s, const char* __nb, const char* __ne) const
             char32_t __buf[__sz];
             char32_t* __bn;
             const char* __nn = __nb;
             __r = do_in(__mb, __nb, __ne - __nb > __sz ? __nb+__sz : __ne, __nn,
                         __buf, __buf+__sz, __bn);
             if (__r == codecvt_base::error || __nn == __nb)
                 __throw_runtime_error("locale not supported");
             for (const char32_t* __p = __buf; __p < __bn; ++__p, ++__s)
-                *__s = (wchar_t)*__p;
+                *__s = *__p;
             __nb = __nn;
         }
         return __s;
     }
 };
 
 // template <class charT> class numpunct