ipaddress IPv4Network and IPv6Network need specialized reverse_pointer property

[pemensik]

Bug report

Bug description:

# Add a code block here, if required
from ipaddress import ip_network

# Generates now:
# '8.4./.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa'
assert(ip_network("2001:db8::/48").reverse_pointer == ['0.0.0.0.8.b.d.0.1.0.0.2.ip6.arpa'])

# Generates now:
# '0/29.2.0.192.in-addr.arpa'
assert(ip_network("192.0.2.0/29").reverse_pointer == [
'0.2.0.192.in-addr.arpa', '1.2.0.192.in-addr.arpa', '2.2.0.192.in-addr.arpa', '3.2.0.192.in-addr.arpa', '4.2.0.192.in-addr.arpa',
'5.2.0.192.in-addr.arpa', '6.2.0.192.in-addr.arpa', '7.2.0.192.in-addr.arpa'])

IPv4Network and IPv6Network get their implementation from _BaseAddress. But for network, it does not work as it should. It returns broken nonsense instead of helping data. Problem is unlike normal IP address, for a network range, it may return just single string for prefix lengths divisible by 8 for IPv4 and 4 for IPv6. But for other prefixes, it should return a list of domain names used.

I have implemented working generator for a list result at https://github.com/InfrastructureServices/dnsconfd/pull/70 I think something similar should be used in base ipaddress directly. If reverse_pointer should not be fixed, it should be removed from networks instead.

Related to #123409, but that is not exactly about networks. Can be verified a bit using ipcalc --reverse-dns, but even that crashes on undivisible ipv6 prefixes.

CPython versions tested on:

3.13

Operating systems tested on:

Linux

[Zheaoli]

Seems, we need to get more detail about the reverse pointer from the RFC.

@picnixz plz assign this issue to me, I will take handle of it .

[picnixz]

You can take care of the issue, I'm not here for the next days. I didn't know that the reverse pointer for network was different (clearly not my field of research) so I'm happy if you can fix the bug!

[pemensik]

There is no such thing as a reverse pointer for networks in dns itself. But it is related to how those domains can be delegated. Meaning where RFC 1035 can cut between zones operated by someone else. rfc2317 specifies recommended classless delegation using DNAME or CNAMEs in DNS.

Classless in-addr.arpa delegation describes how to delegate 1.0/25.2.0.192, therefore 192.0.2.0/25 network. Example from that page.

   $ORIGIN 2.0.192.in-addr.arpa.
   @       IN      SOA     my-ns.my.domain. hostmaster.my.domain. (...)
   ;...
   ;  <<0-127>> /25
   0/25            NS      ns.A.domain.
   0/25            NS      some.other.name.server.
   ;
   1               CNAME   1.0/25.2.0.192.in-addr.arpa.
   2               CNAME   2.0/25.2.0.192.in-addr.arpa.
   3               CNAME   3.0/25.2.0.192.in-addr.arpa.

This means that in 2.0.192.in-addr.arpa. zone, one needs to send ['0.2.0.192.in-addr.arpa.', '1.2.0.192.in-addr.arpa.', ... '127.2.0.192.in-addr.arpa.'] domains to some other domain. Those are the same, what I want generated by my proposal of . Special name 0/25.2.0.192.in-addr.arpa. is done to be able to delegate part of 2.0.192.in-addr.arpa. domain to DNS server operated by someone else. In that names 1.0/25.2.0.192.in-addr.arpa. to 127.0/25.2.0.192.in-addr.arpa. can be administered by different entity than original zone 2.0.192.in-addr.arpa..

There is no similar proposal for IPv6, because common IPv6 prefixes assigned to people are divided by 4 with zero remainder. That is /48, /52, /56, /60, /64. But prefix like /54 is still valid, but needs multiple domains to hold reverse addresses contained in that prefix.

[ZeroIntensity]

This is affecting all bugfix+ versions (3.12+), right?

[Zheaoli]

I think we just need to backport to 3.13 if this is necessary.