Handle minibuffer completion table, predicate and metadata

[clemera]

This will allow us to access the metadata for purposes as discussed in #15 and #82 and allows other packages and commands to access those variables (see #94). With this patch we can retrieve the metadata like:

(completion-metadata input
                     minibuffer-completion-table
                     minibuffer-completion-predicate)

We could also decide to only pass the metadata instead (which would probably reduce garbage collection when exiting the minibuffer I guess). But the metadata can depend on the current input and the predicate passed to the completion table. Having metadata dependent on the input or predicate sounds like a crazy idea and I don't know if it actually happens in practice.

[clemera]

This PR fails with CI for emacs-master but I can't check why because I don't want to give them read and write access for all my repos which they apparently require to show me the results for some reason.

[raxod502]

You should also be able to test locally, make docker VERSION=master.

[clemera]

Thanks, it's fixed now.

[raxod502]

See #94 (comment).

[raxod502]

Okay, seems fine. Could you add a changelog entry?

[raxod502]

This should probably explicitly mention that passing the keyword arguments will dynamically bind the corresponding variables, as otherwise you have to be extremely familiar with binding semantics and the implementation of cl-defun to guess that it has this effect.

[clemera]

Good idea, have done it.

[clemera]

The embark package has made a lot of progress and can be used with selectrum for occur and actions. Currently there is one last issue with selectrum: One of the tricks embark uses is to call a command, insert a candidate into the minibuffer and exit with it. With default completion there is no extensive computation when a completing-read session starts because the candidates are only computed on TAB. With selectrum the candidates get currently computed beforehand which introduces a noticeable delay especially for slow/large completion tables.

With this PR standard completion tables are now computed once in selectrum--minibuffer-post-command-hook. This way the general behaviour does not change but one can easily skip the computation if needed, for example with embark the delay issue is gone using this PR and the following config for embark:

(add-hook 'embark-setup-hook
          (defun embark-selectrum-handle-inject ()
            (when selectrum--active-p
              ;; use the injected candidate and make sure selectrum doesn't
              ;; update afterwards
              (setq selectrum--current-candidate-index -1)
              (setq selectrum--previous-input-string
                    (buffer-substring
                     selectrum--start-of-input-marker
                     selectrum--end-of-input-marker)))))

Another benefit of computing the table in post command hook is that we could offer an selectrum-exhibit function (analogue to icomplete-exhibit):

(defun selectrum-exhibit ()
  "Refresh completion."
  (when-let ((mini (active-minibuffer-window)))
    (with-selected-window mini
      (when minibuffer-completion-table
        ;; trigger recomputation and display update
        (setq selectrum--preprocessed-candidates nil)
        (setq selectrum--previous-input-string nil)
        (selectrum--minibuffer-post-command-hook)))))

This can for example be used for creating simple async completion tables.

If you want to checkout embark yourself here is my current setup to configure it for selectrum (at current PR), you should omit the which-key section if you don't use it.

[clemera]

I also added support for display-sort-function metadata now. You can test with:

(completing-read "Check: "
                 (lambda (string pred action)
                   (if (eq action 'metadata)
                       `(metadata
                         (display-sort-function
                          . ,(lambda (seq) (sort seq #'string<))))
                     (complete-with-action action '("c" "b" "a") string pred))))

[clemera]

Thanks, I finished updating the CHANGELOG and also added the mentioned selectrum-exhibit function from above.

[clemera]

Turned out that completion-extra-properties annotations also need to be handled for completing-read (see proced-send-signal for a built-in example). I previously thought that they would only be relevant for completion-in-region.

[raxod502]

Thanks very much!

[clemera]

Thanks for the feedback and review!