While reading N5032 [print.fun], I noticed what appears to be a
wording defect in
  void vprint_nonunicode_buffered(FILE* stream, string_view fmt,
format_args args);

Its Effects are currently specified as:
  string out = vformat(fmt, args);
  vprint_nonunicode("{}", make_format_args(out));

That call appears to select the two-argument overload
vprint_nonunicode(string_view, format_args), whose Effects are
specified as
  vprint_nonunicode(stdout, fmt, args);

If so, the caller-supplied stream is discarded and
vprint_nonunicode_buffered is specified to write to stdout.

By contrast, the parallel vprint_unicode_buffered says:
  string out = vformat(fmt, args);
  vprint_unicode(stream, "{}", make_format_args(out));

This seems user-visible because print(FILE* stream, ...) can dispatch
to vprint_nonunicode_buffered(stream, ...) on the non-UTF-8 buffered
path.

For what it's worth, libstdc++ and MSVC STL also appear to implement
the intended behavior by passing stream through, so this may be a
wording defect rather than an implementation bug
https://godbolt.org/z/Kjvvxejhx.

Am I reading this correctly, and is this already a known LWG issue? I
did not find a matching issue in the active or defect lists. If this
is not already tracked, I can send a formal library issue writeup to
lwgchair_at_gmail.com.

Regards
Abhinav Agarwal