Thread: BUG #18712: inet value ::2 handling goes not as expected

The following bug has been logged on the website:

Bug reference:      18712
Logged by:          Denis Feklushkin
Email address:      denis.feklushkin@gmail.com
PostgreSQL version: 17.1
Operating system:   Debian
Description:

I think that statement

select '::0.0.0.2'::inet;

should return '::0.0.0.2' value, not '::2'

I think this is what was intended in the Postgres sources. But it seems that
this line is never executed:
https://github.com/postgres/postgres/blob/7b88529f4363994450bd4cd3c172006a8a77e222/src/port/inet_net_ntop.c#L260

PG Bug reporting form <noreply@postgresql.org> writes:
> I think that statement
> select '::0.0.0.2'::inet;
> should return '::0.0.0.2' value, not '::2'

> I think this is what was intended in the Postgres sources. But it seems that
> this line is never executed:
> https://github.com/postgres/postgres/blob/7b88529f4363994450bd4cd3c172006a8a77e222/src/port/inet_net_ntop.c#L260

Hmm ... you are right that that code doesn't do what it seems to
intend to, because by the time that we get to this logic we have
i == 7 not i == 6.  But I'm not sure if we should change it after
all this time.  We're about as likely to get complaints as kudos,
I fear.  And the output isn't incorrect, just not-per-style.

I wonder whether ISC ever changed their version?

            regards, tom lane

Denis Feklushkin <feklushkin.denis@gmail.com> writes:
> вс, 17 нояб. 2024 г. в 20:05, Tom Lane <tgl@sss.pgh.pa.us>:
>> Hmm ... you are right that that code doesn't do what it seems to
>> intend to, because by the time that we get to this logic we have
>> i == 7 not i == 6.
>> But I'm not sure if we should change it after
>> all this time.  We're about as likely to get complaints as kudos,
>> I fear.  And the output isn't incorrect, just not-per-style.
>>
>> I wonder whether ISC ever changed their version?

> https://github.com/openbsd/src/blob/f7304f605db8e3a4de0a0d1c1488830678f77517/sys/netinet/inet_ntop.c#L196
> Looks, they don't used 0x0001 check, so, they version was always as intended

Well, that's OpenBSD's copy.  I thought that the original code was
in BIND, but it doesn't seem to be there anymore, though there's
this tantalizing hint in the changelog from around 9.7.0:

bind-9.18.28/ChangeLog:2635.    [bug]           isc_inet_ntop() incorrectly handled 0.0/16 addresses.

However, I tried probing the actual behavior of various systems
with the attached test program.  I got

::ff -> ::ff
::ff:ff -> ::0.255.0.255

on Linux, macOS, FreeBSD, NetBSD, and OpenBSD.  So regardless of
original intent, this behavior seems to be the de facto standard.
I'm disinclined to make us deviate from it.

            regards, tom lane

#include <stdio.h>
#include <string.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>

int main(int argc, char **argv)
{
    char buf[16];
    char formatted[128];

    memset(buf, 0, sizeof(buf));

    buf[15] = 0xff;
    inet_ntop(AF_INET6, buf, formatted, sizeof(formatted));
    printf("::ff -> %s\n", formatted);

    buf[13] = 0xff;
    inet_ntop(AF_INET6, buf, formatted, sizeof(formatted));
    printf("::ff:ff -> %s\n", formatted);

    return 0;
}