From: Daniel Berger <djberg96@...>
Date: 2007-08-06T23:19:07+09:00
Subject: Re: Bug in Ruby's PTY extension



On Aug 5, 11:28 am, "ronald braswell" <rpbrasw...@hotmail.com> wrote:
> I'm new to the ML so please forgive me if this has been discussed
> previously.
>
> Ruby version:  1.8.5
> OS:  Solaris 10 x64 (11/06)
> CPU:  Dual core Opteron
>
> When using expect or RExpect I was getting an intermittent premature
> EOFError on the first call to expect.  Catching this exception and then
> going on was an ugly work around.
>
> pty.c:
>
> I noticed that Solaris 10 does not support TIOCSCTTY so in function
> establishShell() the child process closes the slave device and reopens it to
> force it to be the controlling tty.  The parent process also closes the
> slave device because it does not need it.  I added a call to nanosleep for
> 20 milliseconds just before the parent process closes the slave fd at the
> end of establishShell() and I now both expect and RExpect work reliably.  I
> suspect that just keeping a reference to the open file object in the parent
> process while the child process closes and reopens the slave device prevents
> the EOF from occurring -- but I am far from an expert on Solaris internals.
> This may not be the best way to solve the problem and just adding the small
> delay does not ensure that the problem will never happen again but so far,
> so good.

This quote from comp.programming may be useful:

"In System V, the controlling terminal is the first tty that's opened
that is not already associated with a session.  If you don't want that
tty to be the controlling terminal, use the O_NOCTTY flag on open().

You don't need the TIOCSCTTY ioctl in SVR4 systems."

http://tinyurl.com/2af8ob

Regards,

Dan