Kamil Dudka wrote: > On Saturday, November 4, 2017 11:16:17 PM CET Felix Janda wrote: > > Below is a hacky patch that makes elinks using gnutls load > > https://tracker.debian.org/mlterm. > > > > It would be cleaner to use gnutls_handshake_set_timeout(). > > > > diff --git a/src/network/ssl/socket.c b/src/network/ssl/socket.c > > index 23273046..4627629b 100644 > > --- a/src/network/ssl/socket.c > > +++ b/src/network/ssl/socket.c > > @@ -553,8 +553,10 @@ ssl_read(struct socket *socket, unsigned char *data, > > int len) #endif > > > > #ifdef CONFIG_GNUTLS > > - if (err == GNUTLS_E_REHANDSHAKE) > > - return -1; > > + if (err == GNUTLS_E_REHANDSHAKE) { > > + while (ssl_do_connect(socket) == SSL_ERROR_WANT_READ); > > + return ssl_read(socket, data, len); > > + } > > #endif > > > > if (err == SSL_ERROR_WANT_READ || > > Not that I know gnutls API but the fix does not look correct... > > Would not it busy loop until the needed data arrives over the network?
Essentially yes, which is why I called it a "hacky patch". Note that ssl_do_connect is the same thing as gnutls_handshake and SSL_ERROR_WANT_READ is GNUTLS_E_AGAIN. I am too inexperienced with gnutls and elinks to fix this cleanly. Felix -- http://lists.linuxfromscratch.org/listinfo/elinks-dev Unsubscribe: See the above information page
