On May 1,  2:23pm, khendricks@ivey.uwo.ca (Kevin B.Hendricks) wrote:
-- Subject: Re: pipes from FreeBSD and the NetBSD async I/O bug

| Hi,
| 
| I tried everything I could with another test program similar to the one you 
| just sent to me to get the read from the pipe to generate a sigio even when 
| the write had produced an EAGAIN and I can never see a sigio at all.
| 
| This is funny.  Perhaps pipes are special cases?
| 

I don't recall if I mentioned this before, but the only file
descriptor types I've seen SIGIO implemented is sockets and ttys.
The rest just don't implement SIGIO. The reason jdk works is probably
because you never get SIGIO on pipes. One would argue that O_ASYNC
and perhaps O_SETOWN should fail on file descriptors that do not
implement SIGIO.

Once we move to the new code in NetBSD, we'll have to decide if we
are going to make SIGIO work with them [and we probably should].

Since the only implementation where SIGIO actually works is in BSD
sockets, and there is really no spec for it, I guess that makes it
the reference implementation :-(

christos