> The second idea is to have only ptys be tty devices in the current
> sense (stty settings and all), with the hardware device drivers being
> optimized for high throughput to user-land and no tty goop.
> Interactive logins are then handled with a user-land daemon that
> interfaces between a hardware line device and a pty master half.
> 
> The real hardware drivers would then be just byte shovelers with a
> little extra stuff to handle modem control lines and speed changes and
> similar hardware issues.  I currently have no proposed design for the

This one interests me more, mostly because it deals appropriately with
the fact that the most common current use of the tty drivers (as well as
the most demanding in terms of performance) is no longer interactive
logins, but rather the data-movement-intensive support of networking
protocols.  It seems silly to retain a hardcoded intimacy between the
device drivers and the fancy interactive terminal support when the latter
most often goes unused.

The ironic part of this is that this is precisely the thing that the
original streams implementation (the research v8 one, not the thing system V
turned it into) did more than 10 years ago, with the sole additional
enhancement of allowing you to eliminate the userland stub between
kernel procedures, if you wanted, by providing a generic interface for
wiring things directly together inside the kernel instead.  Some ideas
are just good.

Dennis Ferguson