Should M-SPC clear selected candidate preview?

[jdtsmith]

Often I scroll down a bit in the list, and then realize I'm going to need to search instead. But the candidate preview is in the way. If I haven't gone far I can go back up to the first candidate. Or I can reset the input with ESC-ESC-ESC, but that's a lot to type.

I wonder if we could overload M-SPC to call corfu-reset? Right now it does nothing in this scenario (well, it flashes the popup). And this is the key combo I mentally reach for when I think "OK, I need to sharpen my search". If there is no user-input at all yet, probably omitting the actual separator is also the right thing to do.

[minad]

Yes, let's do that. Feel free to commit this right away if the change is simple enough.

If there is no user-input at all yet, probably omitting the actual separator is also the right thing to do.

You mean, if M-SPC is pressed at the prompt with no user input? I prefer to keep the current behavior - this way the user can initiate an Orderless search right away.

Or I can reset the input with ESC-ESC-ESC, but that's a lot to type.

Regarding ESC-ESC-ESC, I think this is one of the bad Emacs defaults. My suggestion is to bind <escape> to keyboard-escape-quit, but this makes sense on the GUI only of course.

[jdtsmith]

You mean, if M-SPC is pressed at the prompt with no user input?

No user input, and previewing a selected candidate. I actually don't recall how we test the latter. corfu--preview-current-p?

So M-SPC would:

• just reset if previewing without any user input. I'm clearing out the selection to begin a search.
• insert separator if not previewing (like now). I'm starting or refining a search.
• reset + insert separator if previewing with (hidden) user input. I.e. I'm continuing to refine an ongoing search.

[minad]

No user input, and previewing a selected candidate. I actually don't recall how we test the latter. corfu--preview-current-p?

I am not sure if such a mixed behavior is ideal. I would have to test. I think the simplest solution would be to call corfu-reset always and never insert the separator. This might be unexpected for M-SPC users though? Alternatively always jump back to the prompt in corfu-insert-separator and always insert.

[minad]

After some testing I find DWIM behavior better. What about this?

(defun corfu-insert-separator ()
  "Insert a separator character, inhibiting quit on completion boundary.
See `corfu-separator' for more details."
  (interactive)
  (if (not (corfu--preview-current-p))
      (insert corfu-separator)
    (corfu--goto -1)
    (unless (or (= (car completion-in-region--data) (point))
                (= (char-before) corfu-separator))
      (insert corfu-separator))))

[jdtsmith]

This latest version seems great to me; it DoesWIM ;).

[minad]

Okay, added ac505ab.

[jdtsmith]

BTW, do you know if eglot's CAPF's caching behavior changed? Now it seems to respond to a space separator by completing ignoring the user-entered component before the space, which I don't think used to happen.

I am not wrapping its CAPF in in capf-buster anymore for a long time (which must have been for a reason). No worries if not; I know you're not an eglot user.

[minad]

No, I am not aware of any such change. I'd prefer if we don't touch this area for the time being if you don't mind.