Bug 1563208

This fails to build against rawhide. I'm seeing errors like the following:
http://www.cpantesters.org/cpan/report/9f4359d6-38b1-11e8-9a96-3b1e7347484a

I cannot reproduce it locally. Does it happen in Koji only?

Interesting. This builds in koji but not in mock.

Koji: https://koji.fedoraproject.org/koji/taskinfo?taskID=26265832
Mock: http://people.parinux.org/~seyman/fedora/result/

The test tries to resolve TCP/localhost:www. Maybe the mock Koji is older and the newer local mock has broken /etc/hosts. Or it tries to resolve using /etc/resolv.conf that is properly blocked in Koji, but the new Mock discards the packets instead of rejecting them.

Running in local mock:

# perl -Ilib t/50resolver.t 
ok 1 - '$loop->resolver' isa 'IO::Async::Resolver'
ok 2 - '$future' isa 'Future'
ok 3 - getpwuid from future
ok 4 - getpwuid
ok 5 - getpwuid via $loop->resolve
ok 6 - getpwnam
ok 7 - getprotobyname
ok 8 - getprotobynumber
ok 9 - getaddrinfo_array - resolved
ok 10 - getaddrinfo_array - resolved addresses
ok 11 - getaddrinfo_hash - resolved
ok 12 - getaddrinfo_hash - resolved addresses
ok 13 - $resolver->getaddrinfo - resolved
ok 14 - $resolver->getaddrinfo - resolved addresses
ok 15 - '$future for $resolver->getaddrinfo' isa 'Future'
ok 16 - $resolver->getaddrinfo - resolved addresses
ok 17 - $resolver->getaddrinfo on numeric host/service is synchronous
ok 18 - $resolver->getaddrinfo resolved addresses synchronously
ok 19 - $resolver->getaddrinfo on numeric host/no service is synchronous
ok 20 - $resolver->getaddrinfo resolved addresses synchronously with no service
ok 21 - $resolver->getaddrinfo passive - resolved synchronously
ok 22 - $resolver->getaddrinfo passive - resolved addresses
ok 23 - '$future for $resolver->getaddrinfo numerical' isa 'Future'
ok 24 - $resolver->getaddrinfo resolved addresses synchronously

Here is hangs. And then it spits out:

Nothing was ready after 10 second wait; called at lib/IO/Async/Test.pm line 214
# Tests were run but no plan was declared and done_testing() was not seen.
# Looks like your test exited with 255 just after 24.

This hang is at t/50resolver.t:327 on:

    skip "Resolver has an answer for $missinghost", 1 if gethostbyname( $missinghost );

then the error about 10-second wait is from subsequent command:

    my $future = wait_for_future $resolver->getaddrinfo(
       host     => $missinghost,
       service  => "80",
       socktype => SOCK_STREAM,
    );

So the test tries to resolve nonexistent domain name and after a 40-second timeout in glibc resolver, the wait_for_future dies after 10 seconds with the error message.

If I comment out the "skip" test, it still dies after 10 seconds. It looks like the $resolver has 10-second timeout and the test expect a negative answer from a glibc resolved comes sooner. Sooner that glibc-in-mock's 40 seconds.

This makes the test to pass:

diff --git a/t/50resolver.t b/t/50resolver.t
index 7047b9d..587eb21 100644
--- a/t/50resolver.t
+++ b/t/50resolver.t
@@ -326,11 +326,13 @@ my $missinghost = "TbK4jM2M0OS.lm57DWIyu4i";
 SKIP: {
     skip "Resolver has an answer for $missinghost", 1 if gethostbyname( $missinghost );
 
-    my $future = wait_for_future $resolver->getaddrinfo(
+    my $future = $resolver->getaddrinfo(
        host     => $missinghost,
        service  => "80",
        socktype => SOCK_STREAM,
+       timeout => 60,
     );
+    wait_for { $future->is_ready } timeout => 60;
 
     ok( $future->failure, '$future failed for missing host' );
     is( ( $future->failure )[1], "resolve", '->failure [1] gives resolve' );

Maybe one could measure the "skip" command delay and use that period instead of the hard-coded 60 seconds. But if a resolver cached missed, and the cache expired, the measure would be unreliable.

Nevertheless I will need this patch to apply because the current code will fail in my weekly rebuilds.

Built for rawhide:
https://koji.fedoraproject.org/koji/buildinfo?buildID=1068750