vty/DESIGN

On the design of vty

    It appears to me that there are two kinds of graphical
applications, regardless of the output form; the synchronous and the
asynchronous.  Synchronous displays update as changes occur; a good
example of this type is nethack, with its many newsym() calls embedded
in the logic.  Synchronous applications use very little abstractable
code, and in practice all use low level interfaces such as terminfo.

    Asynchronous screen programs, OTOH, do not have update code within
the main logic.  Instead, they perform output "lazily", only computing
it at periodic refresh points.  Because "backtracking" is not
rendered, asynchronous screen programs use less bandwidth, and can
(but usually don't) use less CPU.  Asynchronous programs have their
update logic centralized in such a way that it can be abstracted as a
library; this is what both vty and curses are.

    In the past, vty has had considerable confusion and race
conditions due to the fact that screen resizes can occur
asynchronously with respect to output.  Vty 3.0 handles this in an
very elegant (IMO) way, by treating resizes as just another input
event; the size of the picture being output at any time need have no
relation to the screen, though of course corruption will result if
they are different.

    On a "real" terminal (termcap, not xcurses), output and input can
be completely separated; they can occur concurrently, and do not
effect each other.  Because of this we simplify the internal structure
by using entirely different mechanisms for input and output.  This is
also a great benefit because of the differing characteristics of input
code (complicated, best table driven, etc) versus output code
(performance critical).