corfu and ESS: accessing list elements without having first to select list

[maxecharel]

Hi @minad,

I've tried corfu with eglot, works pretty well (amazing work, as always).
I decided to give it a try with ESS, which relies on CAPF + some Company-specific backends (so I also took that as an opportunity to try cape; also a really cool package).

In R, dataframes are named lists, and their elements (i.e. features/columns/variables) can be accessed using the $ operator.
In the example below, I consider iris, a builtin dataset, but the result is the same if I assign it to foo or bar.

With Company, I can type (in one shot) iris$, and the popup will suggests the different elements it contains:

However, I failed to have Corfu behaving the same way. If I type iris$ in one shot, Corfu will only suggest the corresponding object candidate, that is, iris. I then have to select it, and then use the $ operator to have Corfu suggesting the different elements of the list:

I tried to play with corfu-preselect, but didn't succeed to get the desired behavior. Same when I tried to change the order the different backends were added to completion-at-point-functions via cape-company-to-capf.

Is there a way to make corfu behave the way company does? Could my problem be due to the fact that corfu treats as $ as the regex metacharacter?

Thanks a lot for the great work!

Best,
M

Emacs 29.3
Corfu 20240409.438 (Melpa)
Cape 20240411.2207 (Melpa)

[maxecharel]

The more I think about it, the more I think the problem is due to $ being treated as the 'end of string' metacharacter; also note that I use orderless

[minad]

Yes, likely.

[maxecharel]

I will have a look at how to prevent corfu/orderless to treat it as such; I'll keep you posted; thanks

[maxecharel]

(NB: as expected disabling orderless 'solves' the problem; next step is to keep orderless enabled and make it treat $ as a standard character)

[maxecharel]

Hi,
I ended up implementing the following workaround:

(defun mpger/cape-capf-ess ()
  (dolist (elmnt
	   (mapcar #'cape-company-to-capf
		   (list #'company-R-library #'company-R-args #'company-R-objects)))
    (add-to-list 'completion-at-point-functions elmnt))
  (setq-local orderless-matching-styles '(orderless-literal))) ;; default is '(orderless-literal orderless-regexp)
(use-package cape
  :hook
  (ess-r-mode . mpger/cape-capf-ess))

So, at the end of the day, to get the intended behavior I had to locally disable orderless-regexp matching style.
However, this is unfortunate because this is a pretty cool functionality of orderless. @minad would it be worth adding a feature request to orderless, something like a custom variable allowing to exclude $ (or other characters) from the metacharacters in order to keep orderless-regexp as a matching style?

I guess that would be a substantial amount of work for a corner case, this is the reason I prefer to ask directly here instead of 'polluting' the issue tracker of orderless.

[maxecharel]

(Note to self) problems seems to come from the combination orderless + company backends via cape: latter not enabled in inferior-ess-r-mode, while matching styles include orderless-regexp, and I have no problem having the list elements suggested by Corfu after having typed e.g. df$ (and without preselecting it) --> to be investigated

[minad]

Max Pger ***@***.***> writes:

So, at the end of the day, to get the intended behavior I had to locally disable `orderless-regexp` matching style. However, this is unfortunate because this is a pretty cool functionality of `orderless`. @minad would it be worth to add a feature request to `orderless`, like a custom variable allowing to exclude `$` from the metacharacters in order to keep `orderless-regexp` as a matching style?

I don't think it would be worth it. But you can probably already achieve your desired result with a custom Orderless matching style or style dispatcher.

[maxecharel]

Ok thanks. I will try to investigate the Orderless customization you suggest. If you're ok I'll let this discussion open. If I succeed in achieving the desired result I'll post the solution here.