As spotted by Brent, libwebrtc socket factory is not assigning the right socket type.
Created attachment 301358 [details] Patch
Comment on attachment 301358 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=301358&action=review It seems like we should have some kind of test for these code paths; I only stumbled upon this by chance. > Source/WebKit2/WebProcess/Network/webrtc/LibWebRTCSocketFactory.cpp:64 > + auto socket = std::make_unique<LibWebRTCSocket>(*this, ++s_uniqueSocketIdentifier, LibWebRTCSocket::Type::UDP, socketAddress, rtc::SocketAddress()); I approve this change! :-) > Source/WebKit2/WebProcess/Network/webrtc/LibWebRTCSocketFactory.cpp:81 > + auto socket = std::make_unique<LibWebRTCSocket>(*this, ++s_uniqueSocketIdentifier, LibWebRTCSocket::Type::ClientTCP, localSocketAddress, remoteSocketAddress); Hey! You caught one I missed! :-)
(In reply to comment #2) > Comment on attachment 301358 [details] > Patch > > View in context: > https://bugs.webkit.org/attachment.cgi?id=301358&action=review > > It seems like we should have some kind of test for these code paths; I only > stumbled upon this by chance. I totally agree! I would like to test the whole code path: WebKit2 WebProcess creating webrtc sockets and calling NetworkProcess to do that. But then, the bots might not like that we do networking, although maybe 127.0.0.1 networking might work and be good enough for that purpose.
Comment on attachment 301358 [details] Patch Clearing flags on attachment: 301358 Committed r212232: <http://trac.webkit.org/changeset/212232>
All reviewed patches have been landed. Closing bug.