[libcxx-commits] [libcxx] [libc++] Fix checks for terminal and flushes in std::print() (PR #70321)

[Nikolas Klauser via libcxx-commits]

================
@@ -236,26 +236,11 @@ __vprint_nonunicode(FILE* __stream, string_view __fmt, format_args __args, bool
 // terminal when the output is redirected. Typically during testing the
 // output is redirected to be able to capture it. This makes it hard to
 // test this code path.
-template <class = void> // TODO PRINT template or availability markup fires too eagerly (http://llvm.org/PR61563).
-_LIBCPP_HIDE_FROM_ABI inline void
-__vprint_unicode_posix(FILE* __stream, string_view __fmt, format_args __args, bool __write_nl, bool __is_terminal) {
-  // TODO PRINT Should flush errors throw too?
-  if (__is_terminal)
-    std::fflush(__stream);
-
-  __print::__vprint_nonunicode(__stream, __fmt, __args, __write_nl);
-}
 
 #      if _LIBCPP_HAS_WIDE_CHARACTERS
 template <class = void> // TODO PRINT template or availability markup fires too eagerly (http://llvm.org/PR61563).
 _LIBCPP_HIDE_FROM_ABI inline void
-__vprint_unicode_windows(FILE* __stream, string_view __fmt, format_args __args, bool __write_nl, bool __is_terminal) {
-  if (!__is_terminal)
-    return __print::__vprint_nonunicode(__stream, __fmt, __args, __write_nl);
-
-  // TODO PRINT Should flush errors throw too?
-  std::fflush(__stream);
----------------
philnik777 wrote:

I'm still not sure I understand the point. The paper seems to say that synchronization between `stdout` and the native API should be required, but then requires outright flushing of the stream. AFAICT that means if you want to call a special unicode API, but that synchronizes with `stdout` on its own you still need to flush `stdout`. In other words, you're not allowed to keep the output of any non-unicode-api data in a buffer if you use a unicode API even though that API could write into the very same buffer.

https://github.com/llvm/llvm-project/pull/70321