[Pkg-sysvinit-devel] Bug#327865: bootlogd status?

[Thomas Hood]

Zlatko Calusic wrote:
> Yes, I think you're right on that one, theoretically! But for some
> strange reasons openpty() doesn't work well such early in the boot
> sequence. I think that fact is even documented somewhere in the
> bootlogd documentation.


bootlogd.c itself says:
/*
 *      For some reason, openpty() in glibc sometimes doesn't
 *      work at boot-time. It must be a bug with old-style pty
 *      names, as new-style (/dev/pts) is not available at that
 *      point. So, we find a pty/tty pair ourself if openpty()
 *      fails for whatever reason.
 */


The statement that /dev/pts is not early when bootlogd is started is
no longer true if you have udev 0.080-1 and initscripts 2.86.ds1-10.


> When I was debugging all that I doublechecked that /dev/pts is mounted
> at the time bootlogd starts, but it still doesn't work, openpty()
> fails, then bootlogd falls back and tries opening old terminals in
> succession.


/dev/ptmx is also needed.  udev ought to create this in its S:S03udev initscript.

Md: Is /dev/ptmx guaranteed to exist before S:S03udev finishes?
-- 
Thomas Hood