Re: bash vs. screen vs. SIGINT

[Thomas Schwinge]

Hello!

On Sat, Oct 03, 2009 at 06:06:35PM +0200, I wrote:
> There are issues with the bash 4.0 packages, and new Debian screen
> packages w.r.t. handling of SIGINT.  I don't think that glibc is at
> fault, but instead bash and screen should be looked at.  Details are at
> <http://www.bddebian.com:8888/~hurd-web/open_issues/bash_vs_screen_vs_sigint/>.

Some data about this issue: C-c doesn't make it to the foreground process
in a screen session.

      * in a *screen* session (Debian package 4.0.3-14)
          * shell prompt: **no reaction (nothing printed)**; `$?` thus is 
**unchanged**.
          * `sleep 10` running: **`^C` printed**, SIGINT is **not** sent.

This is a issue in the screen package, that arised when the Debian screen
and GNU/kFreeBSD guys made screen work for them (c.f.
<http://bugs.debian.org/522689>).

This is the old code, from tty.c:

    int
    fgtty(fd)
    int fd;
    {
    #ifdef BSDJOBS
      int mypid;
    
      mypid = getpid();
    
      /* The next lines should be obsolete. Can anybody check if they
       * are really needed on the BSD platforms?
       *
       * this is to avoid the message:
       *    fgtty: Not a typewriter (25)
       */
    # if defined(__osf__) || (BSD >= 199103) || defined(ISC)
      if (separate_sids)
        setsid();   /* should be already done */
    #  ifdef TIOCSCTTY
      ioctl(fd, TIOCSCTTY, (char *)0);
    #  endif
    # endif
    
    # ifdef POSIX
      if (separate_sids)
        if (tcsetpgrp(fd, mypid))
          {
            debug1("fgtty: tcsetpgrp: %d\n", errno);
            return -1;
          }
    # else /* POSIX */
      [...]
    #endif /* BSDJOBS */
      return 0;
    }

... and this is their new version:

    int
    fgtty(fd)
    int fd;
    {
    #ifdef BSDJOBS
      int mypid;
    
      mypid = getpid();
    
      /*
       * Under BSD we have to set the controlling terminal again explicitly.
       */
    # if defined(__FreeBSD_kernel__) && defined(TIOCSCTTY)
      ioctl(fd, TIOCSCTTY, (char *)0);
    # endif
    
    # ifdef POSIX
      if (separate_sids)
        if (tcsetpgrp(fd, mypid))
          {
            debug1("fgtty: tcsetpgrp: %d\n", errno);
            return -1;
          }
    # else /* POSIX */
      [...]
    #endif /* BSDJOBS */
      return 0;
    }

The setsid call indeed doesn't seem to be needed / is done before, but it
seems that on the Hurd we also still need the TIOCSCTTY one -- if I cover
__GNU__ in the #if, then screen works again as expected.  Is this the
correct thing to do then?

The whole screen code that I read (the job control / controlling terminal
/ etc. stuff) is a bit of a mess, and my undestanding of these topics
also is mostly non-existent...


On <http://www.bddebian.com/~hurd-web/open_issues/screen/> I recorded all
the screen issues that I saw.


Regards,
 Thomas

signature.asc
Description: Digital signature