Message-ID: <35379A34.4E7E@CGSTE.MQ>
Date: Fri, 17 Apr 1998 13:06:44 -0500
From: HANRIGOU Philippe <HANRIGOU AT cgste DOT mq>
Reply-To: HANRIGOU AT cgste DOT mq
Organization: CONSEIL GENERAL DE LA MARTINIQUE / DGA2
MIME-Version: 1.0
To: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
CC: DJGPP <djgpp AT delorie DOT com>
Subject: Re: Problem with bash
References: <Pine DOT SUN DOT 3 DOT 91 DOT 980415192238 DOT 10767A-100000 AT is>
Content-Type: text/plain; charset=iso-8859-1
Content-Transfer-Encoding: 8bit
Precedence: bulk

Eli Zaretskii wrote:
> So I suggest to get the sources of the above two modules from
> djlsr201.zip, compile them with -g, link Bash with them, and step inside
> `read' and `__libc_termios_hook' with GDB.
Ok, that's what I've done!

> It figures.  IIRC Bash uses the termios functions to read console input.
> The DJGPP implementation of termios (written AFIK by Daisuke Aoyama and
> used almost exclusively by Bash) hooks calls to the `read' function and
> makes the library call `__libc_termios_read' instead.  See the library
> sources on src/libc/posix/unistd/read.c and
> src/libc/posix/termios/tminit.c, for the gory details.
In fact, there is already a "tminit.c" file in bash sources. It seems to
be
slightly different from the "unistd" one (I've done a quick diff).
Nevertheless
the "fatal" function __libc_termios_read_cooked_tty (see below) is the
same in both 
files. Moreover, I've tried to link bash with both "tminit" files: the
debug behaviour
is exactly the same.

>  It seems that the
> problem is somewhere inside `__libc_termios_read' or its subroutines.

You're right. More precisely at first call to __dpmi_yield in function
__libc_termios_read_cooked_tty.

Here is the source code of that function:

static ssize_t
__libc_termios_read_cooked_tty (int handle, void *buffer, size_t count)
{
  unsigned char *wp;
  ssize_t n;

  wp = buffer;
  n = count;

#if 0
  /* clear cooked queue */
  if (__libc_termios_exist_queue ())
    __libc_termios_clear_queue ();
#endif

  if (__libc_tty_p->t_lflag & ICANON)
    {
      /* get inputs (wait for NL or EOT) */
      if (! __libc_termios_exist_queue ())
	__libc_termios_fill_queue ();

      while (--n >= 0)
	{
	  if (! __libc_termios_exist_queue ())
	    break;

	  *wp++ = __libc_termios_get_queue ();
	}
    }
  else
    {
      /* block until getting inputs */
      while (! __libc_termios_exist_queue ())
	{
	  __dpmi_yield ();                  <<<<<< freezes here! (first call)
	  __libc_termios_fill_queue ();
	}

      while (--n >= 0)
	{
	  *wp++ = __libc_termios_get_queue ();

	  if (! __libc_termios_exist_queue ())
	    {
	      __libc_termios_fill_queue ();
	      if (! __libc_termios_exist_queue ())
		break;
	    }
	}
    }

  return (ssize_t) (wp - (unsigned char *) buffer);
}

> I hope you will be at least
> able to see what part there hangs on you, and we could then scratch our
> heads to try to figure out why.
Ouasp!!! I'm still trying! Unfortunately, I don't know anything about
DPMI. I'm going 
to study that, of course, but I'm sure you'll have an idea before me,
won't you.

Regards,
Philippe.
_______________________________________________________________________
  
     Philippe HANRIGOU           Conseil Général de la Martinique
  Ingénieur informatique              D.D.S.T. - service I.T.S.
                                Immeuble Concorde, route de la Folie
   Tél: (0596) 59-84-63                  97200 Fort de France
  E-mail: HANRIGOU AT cgste DOT mq                      FRANCE
_______________________________________________________________________