On Sun, 24 Aug 2008, Ed Schouten wrote:
> * Robert Watson <rwatson@FreeBSD.org> wrote:
>> So users using slightly old versions of screen, etc, shouldn't appear in
>> finger(1), w(1), or receive messages from biff(1), talk(1), write(1),
>> wall(1), shutdown(8), and dump(8), all of which (I believe) rely on utmp(5)
>> to determine who is logged in and where? I'm sure that quite a few of
>> these are of diminishing significance in the current world order (certainly
>> biff is), but I'm not convinced that we should exclude users on historic
>> tty devices from receiving advance notice of system shutdowns or dump
> Right now we're actually digging up the entire dynamic vs static linkage
> discussion again. If people run a dynamically linked version of screen,
> xterm, etc, they are not affected (except libc.so.6 of course).
I'm not sure I see such a tight congruence: historical applications don't use
the POSIX PTY calls, since they didn't exist or were unreliably implemented
for many years. Instead, applications embedded the pty allocation policy in
the same way they embed the BPF allocation policy, which is to search a series
of hard-coded names until they find a match.
> The current /etc/ttys already seemed like an improvement when compared to
> the old one, where we spent 2 out of 3 entries on commonly unused PTY names.
> What kind of ratio do you propose?
For 256 lines in /etc/ttys, you can keep people's systems working with older
applications. Doesn't seem like a big sacrifice -- it's not like we're
forcing Giant to be kept on part of the kernel, etc.
Robert N M Watson
University of Cambridge