Re: On keybindings and the slow erosion of help's utility

>> Let's consider a simple case.
>> We want to bind C-k in foo-mode-map but only on specific lines in file.
>> When the point is elsewhere, we just want Emacs to call whatever is
>> bound to C-k as if C-k in foo-mode-map were not bound at all.
>> How would you do it with a command?
>
> The command would test if point is on one of those specific lines, and
> if not, call the binding of C-k in the "next" keymap in effect in the
> buffer.

This is indeed what some commands try to do nowadays, but it's hard to
find the "next" binding in a reliable way.  It comes with some
non-trivial problems:

- `C-h k` doesn't know about it, so it will only list "ourselves" and
  not the commands that will be used in the non-specific lines.
- `key-binding` will just return "ourselves", so you end up having to
  simulate the workings of `key-binding` but skipping "ourselves".
  This is tedious and bound to be not 100% correct and hard to maintain.
  If `foo-mode` is a minor mode you get lucky because you can turn it
  off while calling the real `key-binding`, but that doesn't help for
  other cases (and it still comes with the downside that it makes
  non-trivial assumptions about the link between "ourselves", `foo-mode`,
  and `foo-mode-map` which can break if when the users decide they want
  to add "ourselves" to some other map and/or use it in some other
  mode).
- If several commands do that in turn, their simulation of `key-binding`
  needs to be careful to skip not just themselves, but also the other
  previous commands, otherwise, they end up inf-looping delegating to
  each-other.
- Makes for weird behavior what the command is run via `M-x`.


=== Stefan