Yeah, a year ago it would have failed on Linux and Mac too, but I put in a hack that fixes it for "unix" but not Windows (http://git.gnome.org/browse/libsoup/commit/?id=ef7fc0585990b9c2e056596d1850bdb3f6f6dae9)

This will hopefully be fixed in 2.32/3.0.

If you need a fix now, try this: add a method soup_socket_get_iochannel() to return the socket's iochannel, and in soup_connection_get_state(), on windows create the GPollFD using g_io_channel_win32_make_pollfd() on the iochannel. (And then pass it into g_poll just like in the unix case.) If that works then I can add something like that to the 2.30 branch.

I've tried to add the proposed fix.
Unfortunately it seems to crash the application the second time g_io_channel_win32_make_pollfd() is called.

hm... can you attach your patch?

Here it comes

In soup-socket.h:
GIOChannel *   soup_socket_get_iochannel      (SoupSocket         *sock);


In soup-socket.c:

GIOChannel *
soup_socket_get_iochannel (SoupSocket *sock)
{
        GIOChannel * iochannel = NULL;
	g_return_val_if_fail (SOUP_IS_SOCKET (sock), NULL);

	if(SOUP_SOCKET_GET_PRIVATE (sock)->sockfd != -1)
		{
		iochannel = SOUP_SOCKET_GET_PRIVATE (sock)->iochannel;
		}
	return iochannel;
}

In soup-connection.c:
In function soup_connection_gt_state(), just after the G_OS_UNIX stuff

#ifdef G_OS_WIN32
	if (priv->state == SOUP_CONNECTION_IDLE) {
	        GIOChannel * iochannel;
	        GPollFD pfd;
		iochannel = soup_socket_get_iochannel (priv->socket);
		if(iochannel != NULL)
			{
			g_io_channel_win32_make_pollfd(iochannel, G_IO_IN, &pfd);
			if (g_poll (&pfd, 1, 0) == 1)
				soup_connection_set_state (conn, SOUP_CONNECTION_REMOTE_DISCONNECTED);
			}
		}
#endif

hm... g_io_channel_win32_make_pollfd is kinda weird. it's possible you can only call it once per iochannel. So I guess instead of soup_socket_get_iochannel(), you'd want soup_socket_get_pollfd(), where the first time you call it it would call g_io_channel_win32_make_pollfd() on its iochannel, and it would store that result in socket->priv, and return the same pollfd later when it was called again...

(this is all part of why I didn't bother trying to fix it on windows before :)

I have now tried this too.
Unfortunately with same bad result the second time g_io_channel_win32_make_pollfd() is called.

Looking into g_io_channel_win32_make_pollfd() shows that WSACreateEvent() is called.
WSACreateEvent() do not take any parameters. So it doesn't seems to have any clue about iochannel.
From the WSACreateEvent() description on MSDN, I don't really get any clue about how it is supposed to be used and why it crashes.


Added in soup-socket.c:
(Both priv->iochannel and priv->pfd are initialized to NULL in soup_socket_init() )

GPollFD *
soup_socket_get_pollfd (SoupSocket *sock)
{
	SoupSocketPrivate *priv;
	g_return_val_if_fail (SOUP_IS_SOCKET (sock), NULL);
	priv = SOUP_SOCKET_GET_PRIVATE (sock);
	g_return_val_if_fail (priv->sockfd != -1, NULL);
	g_return_val_if_fail (priv->iochannel != NULL, NULL);
	if(priv->pfd == NULL)
		{
		priv->pfd = g_new0 (GPollFD, 1);
		g_io_channel_win32_make_pollfd(priv->iochannel, G_IO_IN, priv->pfd);
		}
	return priv->pfd;
}

this will be fixed by the port to using GSocketConnection. I'm marking this as a dup of the "use gio tls" bug, since that's the thing that's blocking the move to GSocketConnection

*** This bug has been marked as a duplicate of bug 634425 ***