On Sat, Jun 01, 2019 at 05:39:30AM +0800, Alec Liu wrote: > I have the merged code pull, and rerun some basic test, ends up failed. > here is the testing configuration: > * server Apache2_Via_SocksProxy_HTTP_SP 192.168.101.227:49980 > send-proxy verify none socks4 127.0.0.1:1080 check-via-socks4 check > inter 30000 fastinter 1000 > > Looks like it is something related to the send and recv you have mentioned.
Ah, bad :-( One of the problems I faced after integrating it was that conn_fd_handler() was called in loops until recv() succeeded. I found that fd_cant_recv() was initially not expected to be called from there (eventhough all other handshakes do it as well) and that it was later cancelled by with the conn_sock_* calls done in conn_fd_handler(). I managed to work around this by using both conn_sock* and fd_cant*, but I'm not very proud of this. > Should we just leave it in this way for now, and wait for the issue to > be fixed? I'll have a look at it. If it ends up being a limitation of the current polling mechanism that's going to be dropped by Olivier's update, I won't insist too much but I'd rather try to fix it anyway. Thanks for the report! Willy
