On Tuesday 28 April 2015 07:33:39 Light, John J wrote: > Thank you, Thiago, for pointing that out. > > I will add that many robust, well-written C/C++ programs don't set > O_NONBLOCK yet never block, through careful use of select.
That is true, though I still advise using O_NONBLOCK. The one case that you may want O_NONBLOCK despite careful use of select is an API "I've been told to send this packet without blocking, can I do it now?", like the use of send/sendto/sendmsg. You have to query whether the socket can take the packet now. One way is to do select/poll for writing, with a zero timeout, and see if you get the permission to write, then write. But that's two system calls. With O_NONBLOCK, you can simply write and handle the EWOULDBLOCK error if it couldn't send -- one system call. -- Thiago Macieira - thiago.macieira (AT) intel.com Software Architect - Intel Open Source Technology Center
