In the netperf4 code I'm porting-over to glib, it used to call sendmsg
with a two entry iovec - the first entry was the message length, and the
second was the message body. The reason to use sendmsg was to present
logically associated data to the transport at the same time, as one
really should (must) do for efficient networking. Setting TCP_NODELAY
is just a kludge.
At first I figured "Ok, no vectored operations, but there are buffered
operations and I suppose I can live with the extra copies and having to
flush" but I have a question - on Windows a channel is made
non-blocking when a watch is set (IIRC) - are there any other "gotchas"
related to that that might make writing unbuffered as well? I _really_
don't want the message length going-out separately from the message
body... it may not matter all that much for file I/O, but in networking
that is doubleplusungood.
sincerely,
rick jones
_______________________________________________
gtk-list@xxxxxxxxx
http://mail.gnome.org/mailman/listinfo/gtk-list
