[EMAIL PROTECTED] writes:
> jerenkrantz 01/07/23 20:02:58
>
> Modified: . CHANGES configure.in
> network_io/unix sendrecv.c
> Log:
> Add Solaris 8 sendfilev() support.
>
> Index: sendrecv.c
> + /* If we are in non-blocking mode, we need to make sure we wait until
> + * the other side says it is okay. */
> + if (apr_is_option_set(sock->netmask, APR_SO_NONBLOCK) == 1 ||
> + apr_is_option_set(sock->netmask, APR_SO_TIMEOUT) == 1)
> + {
> + rv = apr_wait_for_io_or_timeout(sock, 0);
> + }
1) bug
We shouldn't wait for writability when the app has set the APR
nonblock option. We only do that when the app has set the APR
timeout option to non-zero.
2) performance...
You should check for writability *after* sendfilev() returns
EAGAIN. Otherwise, we get extra syscalls from time to time.
Yeah, rbb made the FreeBSD apr_sendfile() check for writability
first, but it is preferable to call sendfilev() first *then* wait
if needed.
Since a web server will use sendfile() often on small files (e.g.,
.png) you can bet that this will have a bunch of unnecessary
select() calls.
> +
> + /* Actually do the sendfilev */
> + do {
> + /* socket, vecs, number of vecs, bytes written */
> + rv = sendfilev(sock->socketdes, sfv, vecs, &nbytes);
> + } while (rv == -1 && errno == EINTR);
> +
> + /* Solaris returns EAGAIN even though it sent bytes on a non-block
> sock */
> + if (rv == -1 && errno != EAGAIN) {
> + rv = errno;
> + return rv;
> + }
--
Jeff Trawick | [EMAIL PROTECTED] | PGP public key at web site:
http://www.geocities.com/SiliconValley/Park/9289/
Born in Roswell... married an alien...
