Tatsuo Ishii <ishii(at)postgresql(dot)org> writes:
> ! #define iswordchr(c) (t_isalpha(c) || t_isdigit(c) || (lc_ctype_is_c() && !t_isspace(c)))
This seems entirely arbitrary. It might "fix" things in your view
but it will break the longstanding behavior for other people.
I think a more appropriate type of fix would be to expose the
KEEPONLYALNUM option as a GUC, or some other way of letting the
user decide what he wants.
regards, tom lane