Signed-off-by: Paul Gofman pgofman@codeweavers.com --- As testing involves waiting for timeouts it doesn't look good for inclusion into the test suite. I've tested this change separately: https://gist.github.com/gofman/bd2f3b6fd6cbe993e97353560ce81e04 This test succeeds on Windows for me but fails without this patch in Wine: the server close status received during closing handshake becomes WINHTTP_WEB_SOCKET_PROTOCOL_ERROR_CLOSE_STATUS (1002).
Sending ping instead of pong causes the server to send a pong reply as expected although I still see status 1002 on close after. Which probably suggests that Windows doesn't send the ping as well (I also tried that with WINHTTP_OPTION_WEB_SOCKET_KEEPALIVE_INTERVAL socket option set to 15000 and Sleep to 30000).
Also, while looking at this I spotted that we currently ignore PING application data (which may be present according to rfc6455 and should be echoed back with PONG. But I didn't see any server sending pings yet and didn't do anything with this part so far.
dlls/winhttp/request.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/dlls/winhttp/request.c b/dlls/winhttp/request.c index 62de9f5cff9..385a4fc4e55 100644 --- a/dlls/winhttp/request.c +++ b/dlls/winhttp/request.c @@ -3600,7 +3600,11 @@ static DWORD socket_receive( struct socket *socket, void *buf, DWORD len, DWORD { if (!(socket->opcode & CONTROL_BIT) || (ret = handle_control_frame( socket ))) break; } - else if (ret == WSAETIMEDOUT) ret = socket_send_pong( socket ); + else if (ret == WSAETIMEDOUT) + { + WARN( "timeout, retrying.\n" ); + continue; + } if (ret) break; } }