Use manually constructed IPv6 socket addresses for tests, rather than system created ones.

The advantage is that GURL's parsing of IPv6 addresses works on all systems, whereas getaddrinfo(ipv6_literal) only succeeds on IPv6 enabled systems.

This allows the tests to run consistently on all systems, including our own WinXP buildbots (which do not support IPv6).

BUG=http://crbug.com/16452
TEST=[net_unittests] SOCKS5ClientSocketTest.IPv6Domain, SOCKSClientSocketTest.SOCKS4AIfDomainInIPv6

Review URL: http://codereview.chromium.org/155618

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@21053 0039d316-1c4b-4281-b951-d872f2087c98

1 files changed, 6 insertions, 1 deletions

diff --git a/net/base/host_resolver_impl_unittest.cc b/net/base/host_resolver_impl_unittest.cc
index 0ff345c..350a25d 100644
--- a/net/base/host_resolver_impl_unittest.cc
+++ b/net/base/host_resolver_impl_unittest.cc
@@ -318,7 +318,12 @@ TEST_F(HostResolverImplTest, NumericIPv6Address) {
   }
 }
 
-TEST_F(HostResolverImplTest, EmptyHost) {
+// TODO(eroman): This test is disabled because it is bogus. It used to pass
+// solely because of a bug in the RuleBasedHostMapper -- (empty replacements
+// would map to a failure). However when using the actual host resolver
+// (getaddrinfo), this is not necessarily the case. On windows getaddrinfo("")
+// gives you the address of your machine.
+TEST_F(HostResolverImplTest, DISABLED_EmptyHost) {
   scoped_refptr<RuleBasedHostResolverProc> resolver_proc =
       new RuleBasedHostResolverProc(NULL);
   resolver_proc->AllowDirectLookup("*");