Completion for use-package macro (e.g., within :custom)

[leinfink]

I would like to have completion-at-point for use-package. In particular, it would be great if there was a way to let emacs know that inside :custom, it should expect variables, not functions, even though point might look as if it was in a funcall position. If not easily fixable, could someone guide me in the correct direction to work on this?

#277 apparently tried to do something like this, but it didn't seem to lead anywhere.

[montchr]

I also would like to have improved completion candidates for use-package. I actually switched away from using use-package primarily for this reason. IMHO, calling the underlying functions and macros (e.g. setopt, add-hook) provide a much better sense of discoverability and even (counterintuitively?) save me time and effort because those completions just work. Clearly I rely a lot on completions to help me figure out what I am trying to do 😅

I guess a lot of the issue is use-package's extensive use of unquoted lists. But still, even if the lists in :custom, for example, were quoted, you would still get completion candidates for any symbol, not just variables. Callables like setopt and keymap-set offer out-of-the-box completion appropriate to the context, and they don't require everything to be contained in extraneous lists like :custom and :bind.

[conao3]

Since when, I think Emacs assumes that the :custom symbols are functions, not variables, and makes code jumps.
The same issue is happening in leaf and I don't know how to solve this problem.
As a work-around, for example, when there is a use-package like this,

(use-package comint
  :custom
  (comint-buffer-maximum-size 20000 “Increase comint buffer size.”)
  (comint-prompt-read-only t “Make the prompt read only.”))

(use-package comint
  :custom
  (a comint-buffer-maximum-size 20000 “Increase comint buffer size.”))
  (comint-prompt-read-only t “Make the prompt read only.”)))

Change it this way, and for comint-buffer-maximum-size, M-. would allow me to jump properly, but it's too bad work-around.

[yonta]

The default capf used in elisp-mode, elisp-completion-at-point, filters completion candidates based on context (variables, functions, ...).
The problem is that this filter cannot recognize variables in :custom of use-package.

This is not a complete solution, but I am working around the issue by setting (cape-capf-inside-code #'cape-elisp-symbol) to local variable completion-at-point-function in elisp mode.
I distinguish whether a completion candidate is a function or a variable using icon information provided by nerd-icons-corfu.
If you are using company, you can achieve the same with company-box.

Here is a part of my init.el and screenshot:

(defun my/elisp-mode-init ()
  "Set completion function to cape"
  (setq-local completion-at-point-functions
              (list (cape-capf-inside-code #'cape-elisp-symbol))))
(add-hook 'emacs-lisp-mode-hook #'my/elisp-mode-init)