Compile quic_server on MacOS

net/tools/quic/quic_socket_utils.cc

Index: net/tools/quic/quic_socket_utils.cc
diff --git a/net/tools/quic/quic_socket_utils.cc b/net/tools/quic/quic_socket_utils.cc
index 3cd12ed9549e8a338568363207e87b6112eca125..31078bb413c54efcd862305dd2af91594eff051a 100644
--- a/net/tools/quic/quic_socket_utils.cc
+++ b/net/tools/quic/quic_socket_utils.cc
@@ -5,6 +5,13 @@
 #include "net/tools/quic/quic_socket_utils.h"
 
 #include <errno.h>
+// Apple's "in6.h" header file says:
+//  *  RFC 3542 define[s] the following socket options in a manner incompatible
+//  *  with RFC 2292:
+//  *    IPV6_PKTINFO
+// and in order to get IPV6_PKTINFO defined at all,
+// you must pick a mode, as follows:
+#define __APPLE_USE_RFC_3542

[Ryan Hamilton, 2015/03/12 03:34:38]

Should this be wrapped in an #if apple_or_darwin_or whatever?
Should we also undef it after we use it?

[dougk, 2015/03/12 12:39:10]

Personally I find it less ugly not to add preamble and postamble noise on top of
the required noise.
- It seems to be a deliberately obscure (system-internal) name chosen not to
conflict with anything.
- It's done being used after <netinet/in> is included, but putting an #undef
there is another wart. 
Makes no difference to me though.
Are these convincing enough arguments to leave it alone?

[Ryan Hamilton, 2015/03/12 17:55:41]

I'm happy with adding a #undef but leaving off an #if is_whatever

 #include <netinet/in.h>
 #include <sys/socket.h>
 #include <sys/uio.h>