Re: Problems with whole buffer Custom functions.

[Richard M. Stallman]

    I'd reserve M-n and M-p in editable fields for history navigation.

I don't think this is necessary, because there is already an undo
mechanism (the "backup value").

Instead I think it should preserve the ordinary undo list when you set
the variable.  Currently, setting the variable discards the undo list,
which is more or less a bug.

However, I now see that TAB does work for moving from an editable
field to the next button.  So we don't need M-n to do that.
We can stick with TAB as the one recommended way to do that.

    S-TAB is the preferable key binding, but currently it's impossible to
    specify a preferable key binding in the docstring.

The way to do that is to use two different command names.  Look at
`advertised-undo', for instance.  We could use that same technique to
make sure that S-TAB is what appears in a doc string.

However, arranging to present only S-TAB would have a drawback,
that it would not give any way to do this on a tty.

I find that M-TAB is bound to complete-symbol in custom buffers.  That
is not useful, so we could change it to widget-backwards.  However,
M-TAB in an editable field is bound to widget-complete, and we
certainly do not want to change that.

This means that no variant of TAB is suitable to be the universal way
to move to the previous widget, from all buffer positions, on all
terminals.  That is too bad.

Maybe we should define M-n and M-p for this, just to have a consistent
set of commands that move in both directions.

Or maybe we should settle for having S-TAB available only on graphics
terminals.