Re: term.el and undo

[David Kastrup]

Richard Stallman <address@hidden> writes:

>     Personally I'd be happy with *no* undo in term buffers.  Undo
>     with the patch works as expected when in line mode and editing a
>     command-line.  That's the only time that undo seems useful.
>     With the patch using undo at any other time makes the cursor
>     jump around from line to line which is confusing.
> 
> I do not understand what you mean.  Could you describe a precise
> test case in which this jumping occurs?

I'd suppose that the terminal has one idea of the cursor location, and
undo happens at other points.

However, since the whole point of term is to provide an
ncurses-capable terminal, and since undo or any other output or cursor
movement operation not initiated from the device writing into the
pseudotty would lead to a discrepancy between the screen contents and
the application's expectations, _any_ editing by the user including
undo would seem completely pointless.  term probably keeps track of
its own cursor position, so at least you can use copy and paste to do
screenshots as long as the application is not writing to the screen,
but apart from read-only operations like that, user access does not
seem sensible.  "Undo" might be a curiosity to go back in time in the
term, but just like it would seem necessary to move the cursor to the
expected place before inserting new output, it would seem necessary to
completely redo all changes before the application writes new
material.

Since this would require specialized undo semantics, I think we are
better off with disabling it altogether by default.

-- 
David Kastrup, Kriemhildstr. 15, 44793 Bochum