Per this gripe:
http://archives.postgresql.org/pgsql-bugs/2003-11/msg00233.php
I think that the recent IPv6 changes made an undocumented and
un-agreed-to change in the semantics of netmask().
The previous behavior of netmask() was that it always delivered
a result with a masklen of 32. It seems
Tom Lane wrote:
The previous behavior had been agreed to at one point:
http://archives.postgresql.org/pgsql-hackers/2000-10/msg01132.php
I can't find anyplace in the archives where the change was discussed
at all.
hostmask() was not in 7.3 so its behavior is a bit up for grabs,
but I think
