Fixed by recently submitted patch.
> Vadim Mikheev <vadim@krs.ru> writes:
> > Yes, I reproduced this (Solaris 2.5/sparc).
> > Seems like CIDR problem(??!):
>
> Yes. Looks like the low-order bits of a CIDR address are garbage,
> but network_cmp() compares them as though all bits are significant.
> So, indeed, it may think two different instances of '1.2.3/24'
> are not equal.
>
> The regular inet comparison functions at least *try* to mask out
> garbage bits, but I think they get it wrong too --- they should be
> taking the smaller of ip_bits(a1) and ip_bits(a2) as the number of
> bits to compare. They don't. Thus, for example,
>
> regression=> select '1.2.5/16'::cidr < '1.2.3/24'::cidr;
> ?column?
> --------
> f
> (1 row)
>
> which looks wrong to me.
>
> In short, it's a bug in the inet data types, not a generic problem
> with unique indexes.
>
> regards, tom lane
>
> ************
>
--
Bruce Momjian | http://www.op.net/~candle
maillist@candle.pha.pa.us | (610) 853-3000
+ If your life is a hard drive, | 830 Blythe Avenue
+ Christ can be your backup. | Drexel Hill, Pennsylvania 19026