Re: About zcomplete

[Ergus]

On Sun, Feb 20, 2022 at 11:11:00AM +0000, Philip Kaludercic wrote:
> Ergus <spacibba@aol.com> writes:
>
> > Hi recently I have been trying to move back to default completion. As
> > I had many issues with fido and icomplete.
> >
> > The default completion system received a nice improvement recently with
> > the completion-autoselect... and I am wondering a simple mode like
> > this may be added to vanilla (or at least it gives an idea to a better
> > lisper to implement something better, this is just a proof of concept.)
> >
> > ```
> > (require 'simple)
> > (require 'minibuffer)
> >
> > (defvar-keymap zcomplete-map
> >    :doc "Keymap used in *Completions* on zcomplete"
> >    :parent completion-list-mode-map
> >    "z" nil
> >    "n" nil
> >    "p" nil
> >    "q" nil
> >    "g" nil
> >    "h" nil
> >    "DEL" nil
> >    )
>
> When completion-auto-select was first added, there was a discussion
> about adding automatic narrowing in the *Completions* buffer.  Given a
> mode like zcomplete, that decides to unbind all single-character
> bindings, this might be possible (as an alternative to what you do in
> zcomplete--try-on-minibuffer).
>
> Whether or not this should be coupled to removing the mode line of the
> completion buffer is a different question.
Actually a more `urgent` thing in my opinion may be to allow
*Completions* to update without needing to narrow first. Then I could
add a post-command-hook in the minibuffer to update *Completions* when
it is visible.

With that we only need a highlight and the zsh experience may be almost
done with minimal changes.

> > (defun zcomplete--try-on-minibuffer ()
> >    "Try to execute the binding on minibuffer."
> >    (switch-to-minibuffer)
> >    (if-let ((command (lookup-key (current-active-maps)
> >                                  (this-single-command-keys))))
> >        (progn
> >          (minibuffer-hide-completions)
> >          (call-interactively command)
> >          t)
> >      ;; back to completions
> >      (switch-to-completions)
> >      nil))
> >
> > (defun zcomplete--completions-pre-hook ()
> >    "Try on minibuffer when the command is not in *Completions* map."
> >    (when (eq this-command 'undefined)
> >      (zcomplete--try-on-minibuffer)))
> >
> > (defun zcomplete--hack (data context signal)
> >    "Alternative to command-error-default-function.
> > This will try to execute on minibuffer, else emits the error"
> >    (unless (and (string= (buffer-name) "*Completions*")
> >                 (zcomplete--try-on-minibuffer))
> >      (command-error-default-function data context signal)))
> >
> > (defun zcomplete--completions-setup-hook ()
> >    "To call on completions setup."
> >    (add-hook 'pre-command-hook #'zcomplete--completions-pre-hook nil t)
> >
> >    (setq-local command-error-function #'zcomplete--hack)
> >    (setq-local mode-line-format nil)
> >    (use-local-map zcomplete-map))
> >
> > (define-minor-mode zcomplete-mode
> >    "Completion highlight mode to enable candidates highlight in the minibuffer."
> >    :global t
> >    :group 'minibuffer
> >
> >    (if zcomplete-mode
> >        (progn
> >          (setq completion-auto-select t)
>
> This should probably be reverted when zcomplete-mode is disabled.
Agree. But we need to cache the original value tho... I have been
thinking to implement a general solution for that issue similar to
connection-local-variables, because it is a general issue many packages
need to deal with.

> >          ;; (overlay-put zcomplete-overlay 'face 'zcomplete)
> >         (add-hook 'completion-setup-hook #'zcomplete--completions-setup-hook t))
> >
> >      (remove-hook 'completion-setup-hook #'zcomplete--completions-setup-hook)))
> >
> > (provide 'zcomplete)
> > ```
> >
> > It lacks some features (like highlight the current candidate or
> > automatically update completion buffers when visible) to be really a
> > zsh-like feature, but at least we don't need to press C-g every time we
> > want to edit the minibuffer.
> >
> > With this a tab shows the completions and goes there, and any attempt to
> > edit or not defined command tries to execute on the minibuffer...
> >
> > Parts of this could be implemented with advises, but I know we try to
> > avoid those on vanilla code.
>
> If something like this would be added to the core, it would be better to
> patch the relevant sections directly.
>
> > WDYT?
> > Best,
> > Ergus
>
> --
>         Philip Kaludercic