Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!samsung!usc!henry.jpl.nasa.gov!elroy.jpl.nasa.gov!mahendo!jplgodo!seila!edison!don
From: don@edison.UUCP (Don Kossman)
Newsgroups: comp.unix.ultrix
Subject: blocking read()
Summary: how to make read() block?
Keywords: read(), SYSTEM_FIVE, O_NDELAY
Message-ID: <295@edison.UUCP>
Date: 10 Nov 89 02:31:56 GMT
Reply-To: don@seila.UUCP (Don Kossman)
Followup-To: comp.unix.ultrix
Organization: SEI Information Technology, Whittier, CA
Lines: 66

environment: ultrix 3.0, microvaxII, SYSTEM_FIVE environment (cc -Y ...)

problem: read() returns 0 bytes, even though the O_NDELAY flag is NOT set.

according to the SVID spec, read() should BLOCK under these circumstances,
and return -1 or the number of bytes read.  however, it seems to
return 0 if there is nothing in the read buffer.  which forces my
application to poll.  which is silly, and expensive on the cpu.

according to the ultrix man page:

     When you use the System V environment, note the following:

     +    If your program is compiled in this environment, a read
          and readv system call returns 0 if the file has been
          set up for non-blocking I/O and the read would block.

which i read to mean that if i'm set up for blocking I/O, it should
NOT return 0!.

is there something i don't understand here or is ultrix broken in
this respect?

i've checked the actual state of the flags with an fcntl() call,
and the O_NDELAY flag is indeed clear.

other relevant info: modem control is turned on, but carrier IS
present.  the driver is in RAW mode (i'm reading/writing binary data).
i want loss of carrier to generate a hangup signal.  and i want to
hangup the line on close.  i have a SIGALRM handler which sets
sig_state and returns.  code (much abridged) is something like:

    if((tty_in = open(device,O_RDWR|O_FSYNC)) < 0)
    ...
    if(ioctl(tty_in,TIOCGETP,&sg) == ERROR)
    ...
    sg.sg_flags = 00000040;     /* RAW mode */
    if(ioctl(tty_in,TIOCSETP,&sg) == ERROR)
    ...
	if(ioctl(tty_in,TIOCLBIC,000400) == ERROR)
    ...
    if(ioctl(tty_in,TIOCHPCL,0) == ERROR)
    ...
    alarm(5);
    cc=read(tty_in,buf,count);
    alarm(0);
    switch(cc)
    {
        case ERROR:
            if (errno == EINTR && sig_state == SIGALRM)
            [handle timeout]
        case 0:
            [? i handle this by re-trying, but why?...]
        case count:
            [success- handle input]
        default:
            [?]
    }

ok folks, what am i doing wrong??

-- 
------
don kossman, sei information technology, los angeles
...sun.com!suntzu!seila!don
...uunet!mahendo.jpl.nasa.gov!jplgodo!seila!don