bug#44016: 28.0.50; Add new "gnus-search" search interface to Gnus

[Eric Abrahamsen]

On 11/01/20 18:50 PM, Stefan Monnier wrote:
>> The expandable search keys are kept in `gnus-search-expandable-keys'.
>> The programmatic completion part looks like:
>>
>> --8<---------------cut here---------------start------------->8---
>> +(defun gnus-search-query-expand-key (key)
>> +  (cond ((test-completion key gnus-search-expandable-keys)
>> +     ;; We're done!
>> +     key)
>> +    ;; There is more than one possible completion.
>> +    ((consp (cdr (completion-all-completions
>> +                  key gnus-search-expandable-keys #'stringp 0)))
>> +     (signal 'gnus-search-parse-error
>> +             (list (format "Ambiguous keyword: %s" key))))
>> +    ;; Return KEY, either completed or untouched.
>> +    ((car-safe (completion-try-completion
>> +                key gnus-search-expandable-keys
>> +                    #'stringp 0)))))
>> --8<---------------cut here---------------end--------------->8---
>
> IIUC this function is used to expand unambiguous abbreviations, right
> (rather than the more usual "completion" which is done as the user is
> typing)?

Right, at this stage the string is already out of the user's hands.

> It just happens to reuse the completion machinery to do the work.

I wrote a function to do the expansion, then realized it was just a poor
implementation of half of what the completion functions already do. I
understand that these functions are aimed at interactive use (the
"(consp (cdr (completion-all-completions" bit above definitely felt like
I was holding the tool upside down), but they're absolutely useful in a
programmatic setting. I wonder if there could be some version of these
functions that could be "blessed" for use in Elisp programs.

> If so, it looks OK (I guess you could try and reuse the output from
> completion-all-completions in the last branch instead of calling
> completion-try-completion, but it might be more trouble than it's worth).
>
>> --8<---------------cut here---------------start------------->8---
>> +(defvar gnus-search-minibuffer-map
>> +  (let ((km (make-sparse-keymap)))
>> +    (set-keymap-parent km minibuffer-local-map)
>> +    (define-key km (kbd "SPC") #'self-insert-command)
>> +    (define-key km (kbd "TAB") #'gnus-search-complete-key)
>> +    km))
>> +
>> +(defun gnus-search-complete-key ()
>> +  "Complete a search key at point.
>> +Used when reading a search query from the minibuffer."
>> +  (interactive)
>> +  (when (completion-in-region
>> +     (save-excursion
>> +       (if (re-search-backward " " (minibuffer-prompt-end) t)
>> +           (1+ (point))
>> +         (minibuffer-prompt-end)))
>> +     (point) gnus-search-expandable-keys)
>> +    (insert ":")))
>> +
>> +(defun gnus-search-make-spec (arg)
>> +  (list (cons 'query
>> +          (read-from-minibuffer
>> +           "Query: " nil gnus-search-minibuffer-map
>> +           nil 'gnus-search-history))
>> +    (cons 'raw arg)))
>> --8<---------------cut here---------------end--------------->8---
>
> Hmm... here I think instead of calling `completion-in-region` yourself,
> you'd want to do something like:
>
>     (defvar gnus-search-minibuffer-map
>       (let ((km (make-sparse-keymap)))
>         (set-keymap-parent km minibuffer-local-map)
>         (define-key km (kbd "SPC") #'self-insert-command) ;; Isn't this 
> redundant?

Somewhere I'd gotten the idea that SPC was bound to
`minibuffer-complete-word', I don't know how.

>         (define-key km (kbd "TAB") #'completion-at-point)
>         km))
>
>     (defun gnus-search--complete-key-data ()
>       "Return completion data for gnus-search keys."
>       (list (save-excursion
>               (if (re-search-backward " " (minibuffer-prompt-end) t)
>                   (1+ (point))
>                 (minibuffer-prompt-end)))
>             (point)
>             gnus-search-expandable-keys))
>
>     (minibuffer-with-setup-hook
>         (lambda ()
>           (add-hook 'completion-at-point-functions
>                     #'gnus-search--complete-key-data nil t))
>       (read-from-minibuffer ...))
>
> See `read--expression` (used to provide completion on Elisp function
> and var names when reading an ELisp expression in the minibuffer, such
> as in `M-:`) for an example.
>
> If you want to auto-insert a `:` you'd then do it via an `:exit-function`.

Ah, of course! And if `gnus-search--complete-key-data' is conservative
about when it fires, it leaves the door open for other completion
functions. This function could complete "cont" to "contact:", at which
point (for example) an EBDB-specific capf function could take over and
complete names or email addresses to search for.

Thanks!
Eric