Re: History completion

[André A . Gomes]

Juri Linkov <juri@linkov.net> writes:

> Wouldn't it be nice to allow completion on previous input
> in the minibuffer.
>
> There are two unbound commands ‘previous-complete-history-element’ and
> ‘next-complete-history-element’ but their completion is limited
> only to the beginnings of history items.
>
> So tried to set the minibuffer completion table to the history list:
>
> #+begin_src emacs-lisp
> (defun minibuffer-setup-history-completions ()
>   (unless (or minibuffer-completion-table minibuffer-completion-predicate)
>     (setq-local minibuffer-completion-table (symbol-value 
> minibuffer-history-variable))))
> (add-hook 'minibuffer-setup-hook 'minibuffer-setup-history-completions)
> #+end_src

This is extremely valuable!!!  I have thought about it myself in the
past.  My inspiration, perhaps yours too, came from
`comint-previous-matching-input-from-input' and friends.

> It works nicely with ‘icomplete-mode’ and ‘fido-vertical-mode’
> by automatically displaying completions from the history
> depending on input in the minibuffer.  But this also has drawbacks:
> while typing a completely new command that doesn't exist in the history,
> it permanently says at the end of the minibuffer: [No matches]
>
> So instead of this, tried to show history completions only after
> typing a special key.  TAB can't be reused because in some minibuffers
> such as ‘M-!’ (shell-command), TAB completes the command/file names.
>
> There is ‘C-tab’ bound to file-cache-minibuffer-complete.
> When file-cache is not used, then ‘C-tab’ could be rebound
> to a command that completes on the minibuffer history:
>
> #+begin_src emacs-lisp
> ;; Adapted from ‘minibuffer-complete’:
> (defun minibuffer-complete-history ()
>   (interactive)
>   (completion-in-region (minibuffer--completion-prompt-end) (point-max)
>                         (symbol-value minibuffer-history-variable)
>                         nil))
> (define-key minibuffer-local-map [C-tab] 'minibuffer-complete-history)
> #+end_src

Hopefully I was able to follow you thoroughly.

I think that such a completion should always be triggered by the user,
and not automatically.

To cycle the history ring with a given prefix string, I suggest C-n/p.
The minibuffer is always a single line after all, or am I missing
something?  M-n/p can't be taken, and the same applies to C-M-n/p.
Another suggestion would be to follow comint mode's bindings: C-c M-r/s.
But, unlike comint's case, C-c M-p/n aren't being used so it's also an
option.


--
André A. Gomes
"Free Thought, Free World"