Retrying truncated UDP queries over TCP leads to discrepancies in the $socks{$fileno} hash...

[nhuillard]

I stumbled upon a crash that I hunted down to echo '196.29.35.79' | jdresolve --debug=3 --timeout=10 -. I can see in my improved debug the trace that freesocket uses a different $fileno than the corresponding query. The tcpdump trace shows an UDP query and a subsequent TCP query. This specific IP address resolves to 34 names... My solution was to add $res->igntc(1); just after the resolver instanciation, which prevents TCP retries and the added socket:

347 my $res = new Net::DNS::Resolver; # used for sending dns requests 348 $res->igntc(1); # If true, truncated packets will be ignored. If false, the query will be retried using TCP. 349 my $sel = new IO::Select; # controls the open sockets