PEP 3144 added ipaddress in Python 3.3, which is great!
My question: Is a multi-network container type like netaddr.IPSet in scope for ipaddress?
Background:
I happen to be exploring a case of checking whether requests are coming from any major cloud provider’s datacenter, which means checking against a few thousand ip ranges. The simple version of this is pretty expensive (a few milliseconds for 3k non-overlapping networks) with ipaddress when you have thousands of networks to check:
any(ip in network for network in networks)
whereas netaddr.IPSet has an O(1) implementation of this same check.
An O(1) implementation with comparable performance (about 2x slower, but identical O(1) scaling) can be achieved with the standard library, assuming you are starting with a set or dict of networks:
def ip_in_network_set(ip, network_set):
"""Return whether `ip` is in a set of networks
This is O(1) regardless of the size of network_set
"""
check_net = ipaddress.ip_network(ip)
while check_net.prefixlen:
if check_net in network_set:
return True
check_net = check_net.supernet(1)
return False
Does it make sense for the standard library to include a similar construct of higher level “collection of ips” like IPvXNetwork, but not necessarily represented by a single CIDR string? It would logically have the various iterator, __contains__, overlap methods, but not those that refer to a single CIDR string (prefixlen, etc.).