Cider completion style is invalid

[minad]

Hi Bozhidar!

Cider adds a custom cider completion style to the completion-styles-alist. However the entry that is being added does not follow the required specification of completion-styles. This issue has been found by @manuel-uberti (see minad/corfu#8 for the original issue). I wonder how cider gets away with the current implementation. Does the default completion-at-point function work with Cider?

Daniel

---

(defun cider-company-unfiltered-candidates (string &rest _)
  "Return CIDER completion candidates for STRING as is, unfiltered."
  (cider-complete string))

(add-to-list 'completion-styles-alist
             '(cider
               cider-company-unfiltered-candidates
               cider-company-unfiltered-candidates
               "CIDER backend-driven completion style."))

• The first function should follow the calling convention of completion-try-completion, it should return t, nil or a pair (newstr.newpoint)
• The second function should follow completion-all-completions (I think it follows the spec, but you are not returning a base size in the last cdr which is okay I guess?)

[bbatsov]

Does the default completion-at-point function work with Cider?

Yeah, it does. This code is one of the oldest parts of CIDER and has seen minimal changes in the past 9 years, so I'm assuming that some APIs might have changed in Emacs. Is there some documentation on the topic that I can peruse?

[minad]

The docstrings of completion-try-completion and completion-all-completions should have all the necessary information.

[bbatsov]

I see. So basically we need some cider-completion-try-completion function alongside the existing one. I guess I'll also have to check what are the benefits/use-cases of all the params we've suppressed for cider-company-unfiltered-candidates, as I see the actual definition has a lot of params:

(completion-all-completions STRING TABLE PRED POINT &optional METADATA)

[bbatsov]

Btw, I've also forgotten what's the difference between completion-at-point-functions and completion-styles-alist. Perhaps @dgutov can help by suggesting what'd be the appropriate changes.

[dgutov]

@bbatsov Completions functions return some data from their own set, completion styles filter the data provided by the former.

I haven't written any completion styles myself, though, so I'm not going to help with the exact diff.

[bbatsov]

I was mostly wondering if those completion styles have anything to do with company-mode, but I guess I'll figure this out myself when I find a bit of time.

[bbatsov]

Looking at the surrounding code - this was some hack to enable fuzzy completion in company-mode:

(defun cider-company-unfiltered-candidates (string &rest _)
  "Return CIDER completion candidates for STRING as is, unfiltered."
  (cider-complete string))

(add-to-list 'completion-styles-alist
             '(cider
               cider-company-unfiltered-candidates
               cider-company-unfiltered-candidates
               "CIDER backend-driven completion style."))

(defun cider-company-enable-fuzzy-completion ()
  "Enable backend-driven fuzzy completion in the current buffer."
  (setq-local completion-styles '(cider)))

[minad]

It is okay to effectively disable the completion style given that your backend already performs the filtering. However the functions should still follow the specifications, as I noted only the return value of the try-completion function is incorrect. cider-company-unfiltered-candidates is only valid as all-completions function.

[manuel-uberti]

Looking at the surrounding code - this was some hack to enable fuzzy completion in company-mode:

(defun cider-company-unfiltered-candidates (string &rest _)
  "Return CIDER completion candidates for STRING as is, unfiltered."
  (cider-complete string))

(add-to-list 'completion-styles-alist
             '(cider
               cider-company-unfiltered-candidates
               cider-company-unfiltered-candidates
               "CIDER backend-driven completion style."))

(defun cider-company-enable-fuzzy-completion ()
  "Enable backend-driven fuzzy completion in the current buffer."
  (setq-local completion-styles '(cider)))

It was probably depending on my settings, but without hooking cider-company-enable-fuzzy-completion into cider-mode-hook I was never able to get Company completion in my CLJ/CLJC/CLJS buffers.