On Tuesday November 2 2010 16:40:49 Rob McMahon wrote:
The fix to NetAddr::IP seems to be as simple as --- NetAddr/IP/Lite.pm.bak Fri Oct 29 00:33:06 2010 +++ NetAddr/IP/Lite.pm Tue Nov 2 15:18:05 2010 @@ -740,7 +740,7 @@ if ($mval == 128) { # cidr 128 $mask = Ones; }- elsif ($ip =~ /^\d+\.\d+\.\d+\.\d+$/) { # corner case of ipV4 with new6 + elsif ($ip =~ /^\d+(\.\d+(\.\d+(\.\d+)?)?)?$/) { # corner case of ipV4 with new6$mask = shiftleft(Ones,32 -$mval); } elsif ($mask < 128) { # small cidr Having made that change SpamAssassin runs `make test' cleanly. Before it failed on four tests in cidrs.t, including ok !tryone "128.0.0.1", "127/8"; ok !tryone "128.0.0.1", "127.0/16"; ok !tryone "128.0.0.1", "127.0.0/24"; and one in recreate.t (complaining about the "netset: cannot include" warnings).
Thanks. The bug has been reopened now: https://rt.cpan.org/Public/Bug/Display.html?id=62521 you may want to attach your solution there. Philip Prindeville wrote:
You might be able to get better results with: Net-Patricia-1.18 which I released earlier this week.
Thank you for bringing it to our attention. Now that you have added the AF_INET6 support is became useful. I toyed with it for a while, looks good and fast (and a bit memory fat compared to NetAddr::IP). Lookups are more than 16 times faster for a set of 30 networks (our real setup) using 750.000 real SMTP client's addresses collected from our mailer log (using AF_INET6 database variant to store both inet and inet6 addresses) - compared to the current SpamAssassin solution. It does allow storing data with each network, which we need to implement includes/excludes. One thing that we need to figure out is how to test if one network (not a single IP address) is in the trie - this is needed to check validity of config files, but is not used later for mail checking, so it need not be very fast. One suggestion: currently it is not possible to store 0 and 1 as a data item associated with each net, because a 0 is treated the same as undef and replaced by the key. And the AF_NET6 argument to new() needs to be documented in a POD. Thanks for your efforts! Mark
