On a RH5.1 box.
I'm getting dual printing of every error message.
Almost panicked when I ran the regression test, but the only fails were
for different error messages in int, some float handling function not
overflowing (exp pow), and some differences in float8 as far as unit in
the last place. -DEJ