Differences in printing UTF-8 strings to stdout vs. stderr under Windows

It seems that under Windows, some UTF-8 strings that print OK to
stdout do not print correctly to stderr.  To reproduce:

x <- "\ub124"
cat(x, file = stdout())
## ?
cat(x, file = stderr())
## <U+B124>

Original motivating problem here:
https://stackoverflow.com/questions/32696241/how-to-display-a-message-warning-error-with-unicode-characters-under-windows

How does printing to stderr differ from printing to stdout?

And more importantly, is there any way I can ensure correct printing
of Unicode characters when I need to write to stderr (when throwing
errors or warnings)?

The answer to the last question is certainly "no".  If you are running
Rterm, you are limited to printing in the native encoding, and that
won't cover all Unicode characters.

In Rgui we make more of an effort to convert characters to UTF-16, which
covers most Unicode characters.  (I think we still don't handle
surrogate pairs, but they're rarely used.)  Apparently this is done for
stdout() but not for stderr().  I don't know the rationale for that design.

Duncan Murdoch